TABLE OF CONTENTS

Concept Note 1
Plenary Speakers 2
Technical Program (Energy Session) 4
Technical Program (Environment and Infrastructure Session) 5
Technical Program (Information and Communications Technology Session) 7
Technical Program (Manufacturing and Machinery Session) 8
Technical Program (Semiconductor, Materials and Electronics Session) 9
Book of Abstracts

Code Title of Paper Author(s) Page No.

Energy
Bjorn Ivan G. Ofrasio, Alexander F. Padilla
Jr., Tristan Jeorge Perez, Daryl Matthew
Energy Audit on Lighting and Air-conditioning
Tongson, Sophirak Nou, Karlo Jed Pestaño,
EGY_01 of Laboratories for Energy Cost Savings by 11
Jabar Mangandiri, Joseph Gerard T. Reyes,
Simulations in DIALux and eQuest
Carl Michael F. Odulio, Justin Jesse L.
Seranilla
Production and Characterization of Sol-Gel
Processed Microencapsulated Lauric Acid- Paolo Yves L. De Silos, Menandro S.
EGY_02 20
Based Phase Change Material from Treated Berana, Rizalinda L. De Leon
Waste Coconut Oil
Investigation of Heat Transfer Efficiency of
Tungsten Carbide and Cobalt Oxide Jhone Mychale I. Mariquit, Edgar Alan A.
EGY_03 24
Nanoparticles Dispersed on Distilled Water as Donasco
Base Fluid
Development of Hybrid Convection Solar Reniel Z. Rocaberte, Marvin M. Cinense,
EGY_04 31
Dryer for Seaweeds (Kappaphycus alvarezii S.) Ruel G. Peneyra, Jeffrey A. Lavarias
GIS-based Site Suitability Analysis for
Wenyville Nabor Galang, Ian Dominic
EGY_05 Biomass Energy Plant in Bohol's Rice and 36
Tabañag, Michael Loretero
Coconut Producer Municipalities
Environment and Infrastructure
Edgar Clyde R. Lopez, Jeanette Vivien R.
Colorimetric Detection of Copper Ions in Water Jalalon, Mary Faith C. Delas Alas, Cheska
EI_01 42
using Polymer-decorated Silver Nanoparticles Mae P. Aldana, Marlon L. Mopon, Jem
Valerie D. Perez
Geotechnical Characterization of Quarry Waste
EI_02 Limestone as an Alternative Road Base Course Emerzon S. Torres, Mary Ann Q. Adajar 44
Material
A Nonlinear Optimization Model for Subsidy
Allocation to Disrupted Infrastructures and Its
EI_03 Egberto F. Selerio, Jr. 46
Application in the Philippines during the
COVID-19 Pandemic
ERDT National Conference 2020

Theme:

BOUCING BACK INTO THE NEW NORMAL

CONCEPT NOTE

The pandemic has undoubtedly shaken our foundations. Our usual routines where we get to be
surrounded by peers and colleagues in the academe and in the workplace were replaced by new means of
conduct that are heavily anchored online.

As with precedent, science and technology continues to be more relevant in every passing day, but more
so during adversities. From the basic requirement of identifying the distance that should be observed by
people in order to prevent the transmission of the virus, to the larger demand for vaccine creation--- we
always rely in the field of science and technology for answers and solutions.

In this time of crisis, we are called upon to come back to the maxim, “Science and Technology for the
People,” to ensure that our outputs are relevant and easily transferable for the benefit of other members
of society.

We enjoin our researchers, scientists, and engineers to continue with their noble pursuit of helping the
country bounce back into the new normal, and recognize that their efforts will undoubtedly ripple into
waves that will enable us to create a better and more secure future.

1
 TECHNICAL PROGRAM

Meeting ID : 843-5345-2720
Passcode : ENC2020EGY
Energy Session
07 December 2020
Session Chair : Dr. Luzvisminda M. Bellotindos
Moderator : Engr. Isabelo A. Rabuya
Panelists : Dr. Reynaldo C. Guerrero, Mr. Robert Alfie S. Peña,
Mr. Raymark C. Parocha, Dr. Marvin M. Cinense

09:10AM – 09:20AM Log-in and Roll Call

09:20AM – 09:25AM Welcome Remarks and Introduction of Plenary Speaker
09:25AM – 09:55AM Plenary Talk: DR. ERWIN P. ENRIQUEZ
09:55AM – 10:10AM Open Forum
10:10AM – 10:15AM Preparatory Break

Energy Audit on Lighting and Air-conditioning of BJORN IVAN G. OFRASIO

10:15AM – 10:30AM Laboratories for Energy Cost Savings by University of the Philippines
Simulations in DIALux and eQUest Diliman

10:30AM – 10:35AM Q&A

Production and Characterization of Sol-Gel

PAOLO YVES L. DE SILOS
Processed Micro-encapsulated Lauric Acid-Based
10:35AM – 10:50AM University of the Philippines
Phase Change Material from Treated Waste
Diliman
Coconut Oil

10:50AM – 10:55AM Q&A

JHONE MYCHALE I.
Investigation of Heat Transfer Efficiency of
MARIQUIT
10:55AM – 11:10AM Tungsten Carbide and Cobalt Oxide Nanoparticles
Mindanao State University-Iligan
Dispersed on Distilled Water as Base Fluid
Institute of Technology

11:10AM – 11:15AM Q&A

Development of Hybrid Convection Solar Dryer RENIEL Z. ROCABERTE

11:15AM – 11:30AM
for Seaweeds (Kappaphycus alvarezii S.) Central Luzon State University

11:30AM – 11:35AM Q&A

GIS-based Site Suitability Analysis for Biomass

WENYVILLE N. GALANG
11:35AM – 11:50AM Energy Plant in Bohol’s Rice and Coconut
University of San Carlos
Producer Municipalities

11:50AM – 11:55AM Q&A

11:55AM – 12:00NN Deliberation and Announcement of Best Paper

4
 TECHNICAL PROGRAM

Meeting ID : 817-9034-7704
Passcode : ENC2020EI
Environment and Infrastructure Session
08 December 2020
Session Chair : Dr. Sheila K. Ramos
Moderator : Dr. Bryan G. Alamani
Panelists : Dr. Mary Donnabelle L. Balela, Dr. Candy C. Mercado, Dr. Annaliza Rollon,
Dr. Maria Antonia Tanchuling
09:10AM – 09:20AM Log-in and Roll Call
09:20AM – 09:25AM Welcome Remarks and Introduction of Plenary Speaker
09:25AM – 09:55AM Plenary Talk: DR. ENRICO C. PARINGIT
09:55AM – 10:10AM Open Forum
10:10AM – 10:15AM Preparatory Break

Colorimetric Detection of Copper Ions in Water EDGAR CLYDE R. LOPEZ

10:15AM – 10:25AM
using Polymer-decorated Silver Nanoparticles University of the Philippines Diliman
10:25AM – 10:30AM Q&A

Geotechnical Characterization of Quarry Waste EMERZON S. TORRES

10:30AM – 10:40AM Limestone as an Alternative Road Base Course De La Salle University
Material

10:40AM – 10:45AM Q&A

A Nonlinear Optimization Model for Subsidy

Allocation to Disrupted Infrastructures and Its EGBERTO F. SELERIO, JR.
10:45AM – 10:55AM
Application in the Philippines during the University of San Carlos
COVID-19 Pandemic

10:55AM – 11:00AM Q&A

Earthquake Resistant Building Through Genetic LISA T. GUIEB

11:00AM – 11:10AM Algorithm Optimization: A Case Study Using Mindanao State University-Iligan
Ground Motion Data Institute of Technology

11:10AM – 11:15AM Q&A

Forensic Wind Hazard Analysis of Typhoon JOSHUA C. AGAR
11:15AM – 11:25AM
Rolly University of the Philippines Diliman
11:25AM – 11:30AM Q&A

Environmental Analysis and Estimation of

Evaporation Rate using Pennman Numerical MICHELLE D. ENRIQUEZ
11:30AM – 11:40AM
Simulation in the Salt Production Industry in De La Salle University
Occidental Mindoro

11:40AM – 11:45AM Q&A

5
 Simulating the Impacts of Deficit Irrigation on
Corn Yield and Water Productivity using MARILYN S. PAINAGAN
11:45AM – 11:55AM
Coupled DSSAT CERES-Maize and Aquacrop University of the Philippines Los Baños
Models
11:55AM – 12:00NN Q&A
12:00NN – 12:55PM Lunch Break
Chi and Hypsometric Analysis for Watershed IMELIDA G. TORREFRANCA
12:55PM – 01:05PM
Prioritization and Conservation University of San Carlos
01:05PM – 01:10PM Q&A
Automated Furrow Irrigation System for
SHEILA G. GRIENGO
01:10PM – 01:20PM Maize (Zea mays L. var. saccharata)
Central Luzon State University
Production
01:20PM – 01:25PM Q&A
Physical and Mechanical Properties
MARY ANN N. AHALAJAL
Assessment of Lignocellulosic Rice Straw-
01:25PM – 01:35PM Mindanao State University-Iligan Institute
based Rigid Polyurethane Foam for Structural
of Technology
Sandwich Panel Cores
01:35PM – 01:40PM Q&A
DAVE JOSEPH E. ESTRADA
Silanated Coconut Oil-based 2K Waterborne
01:40PM – 01:50PM Mindanao State University-Iligan Institute
Polyurethane Coating for Corrosion Protection
of Technology
01:50PM – 01:55 PM Q&A
CIELO D. FRIANEZA
01:55PM – 02:05PM Qualitative Risk Assessment of Tailings Dam
De La Salle University
02:05PM – 02:10PM Q&A
02:10PM – 02:20PM Deliberation and Announcement of Best Paper

6
 TECHNICAL PROGRAM

Meeting ID : 819-3345-4662
Passcode : ENC2020ICT
Information and Communications Technology Session
09 December 2020
Session Chair : Dr. Edwin Sybingco
Moderator : Engr. Jose Martin Maningo
Panelists : Dr. Kathleen Aviso, Dr. Argel Bandala, Engr. John Anthony Jose
08:45AM – 08:55AM Log-in and Roll Call
08:55AM – 09:00AM Welcome Remarks and Introduction of Plenary Speaker
09:00AM – 09:30AM Plenary Talk: DR. JOHN-JOHN CABIBIHAN
09:30AM – 09:45AM Open Forum
09:45AM – 09:50AM Preparatory Break

Predicting Stag and Hare Hunting Behaviors REX BRINGULA

09:50AM – 10:05AM
Using Hidden Markov Model Ateneo de Manila University

10:05AM – 10:10AM Q&A

Marketing and Engagement Tool for JOZELLE C. ADDAWE

10:10AM – 10:25AM
Educational Institutions University of the Philippines Diliman

10:25AM – 10:30AM Q&A

Semi-immersive Virtual Reality Game for

JOSIAH CYRUS BOQUE
10:30AM – 10:45AM Upper Body Rehabilitation of Pediatric
University of the Philippines Diliman
Cerebral Palsy Patient

10:45AM – 10:50AM Q&A

The Effects of Quality of Service Mechanisms

JOSIAH ELEAZAR T. REGENCIA
10:50AM – 11:05AM in a Software-Defined Networking
Ateneo de Manila University
Environment on HTTP Requests

11:05AM – 11:10AM Q&A

Philippine License Plate Localization Using PATRICK MATTHEW J. CHAN

11:10AM – 11:25AM
Genetic Algorithm and Feature Extraction De La Salle University

11:25AM – 11:30AM Q&A

Evolution of Social Networks in a Social ORVEN E. LLANTOS

11:30AM – 11:45AM
Learning Management System Ateneo de Manila University

11:45AM – 11:50AM Q&A

11:50AM – 12:00NN Deliberation and Announcement of Best Paper

7
 TECHNICAL PROGRAM

Meeting ID : 892-1144-3819
Passcode : ENC2020MM

Manufacturing and Machinery Session

10 December 2020
Session Chair : Dr. Arthur L. Fajardo
Moderator : Dr. Napoleon Enteria
Panelists : Dr. Omar F. Zubia, Dr. Sherwin Guirnaldo, Dr. Carl Salaan

09:10AM – 09:20AM Log-in and Roll Call

09:20AM – 09:25AM Welcome Remarks and Introduction of Plenary Speaker
09:25AM – 09:55AM Plenary Talk: MR. RAFAEL NESTOR MANTARING
09:55AM – 10:10AM Open Forum
10:10AM – 10:15AM Preparatory Break

Development of an Oyster Mushroom JORDAN L. ABAD

10:15AM – 10:30AM
(Pleurotus Spp.) Substrate Compactor-Bagger Central Luzon State University

10:30AM – 10:35AM Q&A

Development of a Compost Tea Brewer ELMAN C. TORRES

10:35AM – 10:50AM
Machine Central Luzon State University

10:50AM – 10:55AM Q&A

Development of an Impact-Type Cacao HERALD KAM G. HERNANDO

10:55AM – 11:10AM
(Theobroma cacao L.) Bean Huller Central Luzon State University

11:10AM – 11:15AM Q&A

Development of Dry-Scale Cleaner cum JAMEL M. SALO

11:15AM – 11:30AM
Grader for Stored Onion (Allium cepa L.) Central Luzon State University

11:30AM – 11:35AM Q&A

11:35AM – 11:45AM Deliberation and Announcement of Best Paper

8
 TECHNICAL PROGRAM

Meeting ID : 875-5922-7833
Passcode : ENC2020SME
Semiconductor, Materials, and Electronics Session
11 December 2020
Session Chair : Dr. Jonathan W.L. Salvacion
Moderator : Engr. Ericson D. Dimaunahan
Panelists : Dr. Michael C. Pacis, Dr. Jennifer C. Dela Cruz,
Engr. Febus Reidj G. Cruz
09:10AM – 09:20AM Log-in and Roll Call
09:20AM – 09:25AM Welcome Remarks and Introduction of Plenary Speaker
09:25AM – 09:55AM Plenary Talk: DR. JOCELYN F. VILLAVERDE
09:55AM – 10:10AM Open Forum
10:10AM – 10:15AM Preparatory Break

i-LAB-u: Internet-connected Laboratory Units

ALEXANDER C. ABAD
10:15AM – 10:30AM with Over-The-Air (OTA) Programming
De La Salle University
Capability

10:30AM – 10:35AM Q&A

Design and Development of a Neural Network-

NEMILYN A. FADCHAR
10:35AM – 10:50AM based Coconut Maturity Detector using Sound
Mapua University
Signatures

10:50AM – 10:55AM Q&A

Transfer Learning with Convolutional Neural

JOSH DANIEL LAO ONG
10:55AM – 11:10AM Network Towards Fungi Classification in
Ateneo de Manila University
Microscopy Images

11:10AM – 11:15AM Q&A

Vision-based Approach for Checkpoint Status CZERITONNIE GAIL V. YA-ON

11:15AM – 11:30AM
Analysis De La Salle University

11:30AM – 11:35AM Q&A

11:35AM – 11:45AM Deliberation and Announcement of Best Paper

PROF. FERDINAND G.
11:45AM – 12:00NN Closing Remarks MANEGDEG
Program Leader, ERDT

9
 Energy
Session Chair : Dr. Luzvisminda M. Bellotindos
Moderator : Engr. Isabelo M. Rabuya
Panelists : Dr. Reynaldo C. Guerrero
Dr. Robert Alfie S. Peña
Mr. Raymark C. Parocha
Dr. Marvin M. Cinense
 Energy audit on lighting and air-conditioning of
laboratories for energy cost savings by
simulations in DIALux and eQUEST
Bjorn Ivan G. Ofrasio1,a *, Alexander F. Padilla1,b, Jr., Tristan Jeorge Perez1,c, Daryl Matthew Tongson1,d, Sophirak
Nou1,e, Karlo Jed Pestaño1,f, Jabar Mangandiri1,g, Joseph Gerard T. Reyes1,2,h, Carl Michael F. Odulio1,3,i & Justin Jesse
L. Seranilla1,4,j
1
Energy Engineering Program, National Graduate School of Engineering, University of the Philippines Diliman
2
Department of Mechanical Engineering, University of the Philippines Diliman
3
Electronics and Electrical Engineering University of the Philippines Diliman
4
Institute of Civil Engineering, University of the Philippines Diliman
a
bjornivanofrasio09f@gmail.com bafpadilla@up.edu.ph ctj.perez23@yahoo.com.ph d darryl_tongson@hotmail.ph
e
nousophirak001@gmail.com f karlojedpestano@gmail.com g jabarmangandiri@gmail.com hjtreyes2@up.edu.ph
i
carl.odulio@eee.upd.edu.ph jjlseranilla@up.edu.ph

Abstract— Energy audit was conducted in two research
facilities in the University of the Philippines Diliman, the Digital
Signals Processing Laboratory (DSPLAB) and Microelectronics
Laboratory (MICROLAB). DSPLAB was then divided into two
energy activity centers (EAC). Inventory of products were
conducted as a guide to the energy consumption of each laboratory.
Illuminance and temperatures were sampled on grid-divided
laboratories for three different times (0900H, 1200H and 1500H) to
determine the lighting and temperature distributions of each
laboratories at different times. It was found out that most lighting
did not meet the 300-lux standard due to inefficiencies to lighting
caused by low rated lamp output, walls with low light reflectance,
layout of desks and layout of luminaires among others.
Temperatures also did not pass the 23-25°C standard due to the
overrating or underrating of air-conditioning (AC) units compared
to the laboratory heat load. The layout of AC units also caused
certain areas to overcool and other areas uncooled. Likewise, energy
patterns suggest that energy consumption happens by noon to
afternoon due to increase in number of occupants and sunlight
during morning. As visualized in DIALux™, recommended
measures for lighting include changing the wall paint, changing the
desk layout, rewiring the lights and replacing lamps into efficient
ones that saves 312-1398 Wh/year and 1.3 to 7.2-year payback of
investment. Recommended measures for air-conditioning was
simulated in eQuest™ which could save up to Php 197,363.00 and
with maximum payback period of 5.1 years.
Several problems in the UP Electrical and Electronics
Institute (EEEI) laboratories, Microelectronics and Microprocessors
laboratory (MICROLAB) and the Digital Signal Processing
Laboratory (DSPLAB), include dark spots, hot spots and thermal
discomfort to occupants overall. As such, energy audit was
conducted on these rooms (1) to determine the causes of lighting
distributions on each room with the respective energy efficiency
index (EEI); (2) to determine the causes of temperature distributions
on each room with the respective EEI; (3) to determine the cause of
energy consumption profiles and inefficiencies on different times;
and (4) to recommend upgrading measures by computation and
simulation based from the following criteria: lower EEI, monthly
energy savings and adhering to lighting and temperature standards.
II. METHODOLOGY
A. Materials
A Bosch™ GLM 40 Laser Distance Meter was used to get
the room dimensions. A Smart Sensor digital lux meter AS403 with
a sampling rate of 1.5x per second and resolution of 1 lux was used
to record the illuminance in lux. Benetech™ GM816A anemometer
was used to obtain both wind speed and temperature with 1-m from
the air-conditioning unit. Creston™ FN-8830 clamp meter (double
integral A/D switch implement measuring method) was used to
measure electric current and voltage.

Keywords—energy conservation, energy audit, DIALux, B. Planning and pre-inspection

eQuest
I. INTRODUCTION
Energy consumption management in school institutions
and research facilities is crucial in sustaining public education
within financial limit while ensuring comfort of students, faculty or
personnel within the workspace. Research facilities consumes large
energy because of electrical and electronic equipment testing and
operations, time-laborious computer simulations and researchers
working in 24 hours at most. In addition, selection of proper lighting
and air-conditioning equipment should match the appropriate heat
load of routinary activities for each laboratory. The goal is to
optimize energy by saving in financial aspect while giving the
proper lighting and thermal comfort within the service life of the
luminaire or air-conditioning unit, respectively.
Audit schedule was as follows: (1) obtain floor plan; (2)
interview and survey of the occupants; (3) pre-audit inspection and
measurement; (4) inventory of lighting, heating, ventilation, and air
conditioning (HVAC) units; (5) obtain temperature, wind speed and
lighting data; (6) finalization of measurements; (7) analysis of data
and simulation; and (8) forming recommendations based on the
energy activity of the laboratories.
The locations, floor levels and access areas of
MICROLAB and DSPLAB were identified. Their dimensions,
distances from adjacent facilities and accessibility were measured
and assessed to plan the energy audit. In particular, the floor area of
MICROLAB and DSPLAB were divided into 1.5-by-1.5-m grids to
record one datum per parameter (temperature, light intensity) per
grid.
Pre-inspection of the facility occurs prior to the actual
lighting and temperature measurements. The dark spots were
observed when there are lighting deficiencies or shadowing on the

11
area. Air leakage sources were identified wherein holes or unsealed
parts of the room were observed within the conditioned space
boundary. Inventory of equipments (such as computers, laptops,
battery charger, lamps and other electrical equipment) that
contributed to the heat load were recorded. Checking and listing the
power rating of each equipment provided an estimate on how high
or low is the typical power consumption of the facility. The
condition of the equipment was checked for performance, leakage
for air-conditioning units, dimming for lamps and other issues.
The activity of the occupants for HVAC operation
schedule, lighting use and receptacle outlet use were also identified.
C. Energy audit and EEI measurement
The energy audit was conducted from April 7 to May 16,
2019. The audit schedule covered within operation days of Monday
to Friday only by which the measurements were taken three fixed
hours a day: 09:00 AM, 12:00 PM and 03:00 PM. Power
consumption of the facility was obtained by taking current
measurements and voltage measurements at the breaker room. Then,
the lighting and receptacle loads of the facilities were measured on
the breaker room. Furthermore, the air conditioning loads are
measured beside their respective condenser units. Also, the current
of the air conditioning units went from maximum rated current of
17A to minimum operating current of 0.3A (operating blower). The
cycle time for the air conditioning units were identified to estimate
its power consumption. Aside from the energy measurements, the
number of users of the laboratory during the audit period were
counted to estimate the average occupancy and the heat load of the
room.
Room temperatures of DSPLAB and MICRLAB were
taken in the center of each grid at 0.76-m height (or table height)
using the anemometer. Other temperature measurements taken
include outside temperature and hallway temperature.
The lux levels of the facility were taken simultaneously
with the temperature measurement. In light intensity measurement,
lux meter was placed at the center of each grid at the same level and
height as of the anemometer.
EEI was obtained by plotting the raw data measurements
of temperature and luminance, tracing the best fit line and getting
the value of power consumption at the reference illuminance level
(300 lux) or room temperature (24 ᵒC). The EEI from the raw data
was used as the baseline for the recommendations.
D. Lighting calculation and simulation
To verify the lighting measurements, Zonal Cavity
calculation was used to determine the variance between the
luminance of the installed lamp and the measured value [1].
Parameters include room height, ceiling cavity height, room area,
room length and width and lamp rated output. (1)-(3) were used for
calculating light loss factor (LLF) and total initial lamp lumens.
𝜙𝑡𝑜𝑡𝑎𝑙 × 𝐶𝑈 × 𝐿𝐿𝐹
𝐸𝐼𝑁𝑆𝑇 = (1)
𝐹. 𝐴.
𝐸 × 𝐴𝑟𝑒𝑎
𝑇𝐼𝐿𝐿 = (2)
𝐶𝑈 × 𝐿. 𝐿. 𝐹.
𝑇𝐼𝐿𝐿
𝑁𝑅𝐿 =
𝑛𝑜. 𝑜𝑓 𝑙𝑎𝑚𝑝𝑠 𝐼𝑛𝑖𝑡𝑖𝑎𝑙 (3)
[ ]×[ ]
𝑝𝑒𝑟 𝑙𝑢𝑚𝑖𝑛𝑎𝑖𝑟𝑒 𝑙𝑢𝑚𝑒𝑛𝑠 𝑝𝑒𝑟 𝑙𝑎𝑚𝑝
𝐸𝐼𝑁𝑆𝑇 is the average illuminance on the surface-of-
measurement. 𝜙𝑡𝑜𝑡𝑎𝑙 is the total lamp lumen output. 𝐶𝑈 is the
coefficient of utilization. 𝐿𝐿𝐹 is the light loss factor. 𝐹. 𝐴. is the area
of the surface. 𝑇𝐼𝐿𝐿 is the total initial lamp lumens required. 𝐸 is the
recommended minimum illuminance. 𝑁𝑅𝐿 is the number of
remaining luminaires assuming the same lumens of the lamps.
DIALux™ lighting simulator software was used as a
simulation method for calculating LLF. The actual lighting
conditions were simulated to verify dark spots, to check the factors
affecting the lighting and to propose measures to improve it.
Daylight effects is set to none to reduce the sunlight effects on the
windows. This was also done for recommendations on the
assumption that blinds and window tints will be applied for
reduction of heat load.
E. Heat load calculation and simulation
Heat loads are computed using (4)-(10) for solar heat
gains (4), envelope heat gain (5) with resistance calculations if
conductive (6) or convective (7), internal heat for luminaires (8),
internal heat for people (9) and internal heat for others (10) [2].
Cooling load factor (CLF) for appliances is 1.0 since cooling system
is not operated 24 hours. The equipment inventories (working
luminaires, idle and working computers, energized stand fans and
water dispenser) and occupancies of each laboratory were
considered as internal heat sources. External heat sources were
disaggregated into incident light and envelope conduction.
Approximate values were used for the thermal resistance associated
to convection, these being for free air at a value of 2 𝑊𝑚−2 𝐾 −1 and
static air at a value of 1 𝑊𝑚−2 𝐾 −1 . The average value of the EAC
was used as the temperature of the room. Recorded temperatures of
the hallway and the bridge were used as a proxy for hallway and
outside temperatures, respectively. A temperature of 24 ᵒC was used
for the adjacent rooms. Heat load was compared against the installed
cooling capacity of the air conditioning unit to classify whether the
installed AC unit is overrated or underrated. A thermal circuit
diagram was used to visualize the influences of heating load for its
adjacent areas.
𝑄=𝐴 𝑥 𝑆𝐻𝐺𝐹 𝑥 𝑆𝐶 𝑥 𝐶𝐿𝐹 (4)
𝑄̇ = ∆𝑇/𝑅 (5)
R=t/kA (6)
R=1/hA (7)
𝑄=3.41* ( 𝑞𝑖 )*( 𝐹𝑢 )*( 𝐹𝑠 )*(𝐶𝐿𝐹) (8)
𝑄= n * [(SHG *CLF) + (LHG)] (9)
𝑄 = Heat Gain x CLF (10)
𝑄 is the heat load (W). A is the window area (m2). SHGF
is the solar heat gain factor. CLF is the cooling load factor. SC is the
shading coefficient. ∆𝑇 is the difference in temperatures (°C) in and
out of the room. 𝑅 is the resistance. t is the wall thickness. k is the
thermal conductivity of the wall. h is the convective heat transfer
coefficient. qi is the sensible heat gain from lights. Fu is the usage
factor. Fs is the service allowance factor. n is number of peopler.
SHG is the sensible heat gain of a person. LHG is the latent heat gain
of a person.
eQUEST™ was used to generate a more apt model. The
room was considered in isolation. The floor plan was copied into the
software. Wall construction was designated as 6-inch concrete,
likewise with floor construction, ceiling construction and roof
construction. Floor insulation was set to be vinyl tiles as all EAC’s
have vinyl tiles, whereas ceiling insulation was set to either be lay-
in acoustic tiles, or none. Doors were set to be of wooden
construction with a hollow core, flushed at 1¾” and window
construction was set to single paned, clear at ⅛” thickness. A
weather data for Manila was provided by the software. Parameters
12
for occupancy and lighting was put into the software with lighting matches patterns of the color maps – indicating that the illumination
power consumption being expressed as watts per floor area. model for the MICROLAB is valid. The simulated lighting has a
However, it can neither model air flow nor specify layout of LLF of 0.51.
equipment and furniture within a room. As a result, it assumes that
the room is of even temperature all throughout and optimized air
flow.

F. Recommendations
The recommendations for energy efficiency were based
on the obtained EEI. The objective of the recommendation is to
maintain or lessen the energy consumption while increasing the
output of the energy utility appliance to its recommended level. The
simulations of lighting and air-conditioning were conducted using
DIALux™ and eQUEST™, respectively to determine the projected
consumption given a specific parameter. The floor plan, objects,
ceiling and walls were designed in the software. Recommendation
was classified as no cost, low cost or high cost depending on the
simple payback period as computed from (11) and (12).

𝑆𝑎𝑣𝑖𝑛𝑔𝑠
= (𝑘𝑊ℎ 𝑎𝑐𝑡𝑢𝑎𝑙 𝑎𝑛𝑛𝑢𝑎𝑙 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛
− 𝑘𝑊ℎ 𝑟𝑒𝑐𝑜𝑚𝑚𝑒𝑛𝑑𝑒𝑑 𝑎𝑛𝑛𝑢𝑎𝑙 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛) (11)
𝑃ℎ𝑝
×( )
𝑘𝑊ℎ

𝐼𝑛𝑠𝑡𝑎𝑙𝑙𝑎𝑡𝑖𝑜𝑛 𝑐𝑜𝑠𝑡
𝑆𝑖𝑚𝑝𝑙𝑒 𝑝𝑎𝑦𝑏𝑎𝑐𝑘 𝑝𝑒𝑟𝑖𝑜𝑑 = (12) Fig. 1. Sample luminance data of MICROLAB (April 30, 2019) at (a) 9-10
𝑆𝑎𝑣𝑖𝑛𝑔𝑠
A.M., 115.8 lux average; (b) 12-1PM, 113.3 lux average; (c) 3-4PM, 117.6
lux average; and (d) average lighting simulation, 115 lux average.
III. RESULTS AND DISCUSSION

A. Lighting distribution issues

The illuminance on MICROLAB per grid supplied by a
15-W lamp (Omni LT8G DL) was represented by color mapping
with corresponding values as shown on sample data presented on
Fig. 1. The luminance on the workstation ranges from 50 to 200 lux
(yellow) while the illuminance on the meeting room reached at 250-
350 lux (red). When the lights are on at the meeting room, the
luminance exceeds the requirement of 300 lux due to the fenestration
of the room. A window-to-wall ratio (WWR) of 0.40 suggests that
the window is too large that it accommodates too much sunlight
causing to overheat the meeting room (ASHRAE 90.1-2007) [2][3].
Even when the lights are off at the meeting room, the luminance
reaches from 100 to 250 lux which is already favorable for the users
of the meeting room. The EEI of MICROLAB is at 360 Wh/80-120
lux only since the lighting cannot reach 300 lux.
The workstation however did not reach the recommended
luminance of 300 lux. The lighting levels were not properly
distributed as well due to the following observed factors: (1) low
reflectance (55.7 %) of the brown wall paint (#B66D3E) on the left
side of the room causes low luminance ranging from 50 to 100; (2)
shadowing due to the cubicles and positioning of its panels which
obstructs the light; and (3) dirt accumulation on the lamp which
would obstruct the light resulting to low lux output. It can also be
observed that the lighting is concentrated on the center and a little
amount of luminance is distributed along the pathway leading to the
meeting room. Even so, the shadowing of panels and positioning of
the tables have decreased the effective luminance that the user is
receiving.
MICROLAB lighting was simulated with 18 working
lamps with an individual lamp rating of 1500 lumens (T8LED Fig. 2. Sample luminance data of DSPLAB EAC1 (April 30, 2019) at (a) 9-
10 A.M., 115.6 lux average; (b) 12-1 P.M., 114.5 lux average; (c) 3-4 P.M.,
OMNI 15W), 2’ x 4’ x 3” PARAMAX parabolic troffer and a 120.9 lux average; and (d) average lighting simulation, 120 lux average.
mounting height of 2.7 m. As shown in Fig. 1d, the dark spots are
mostly located on the right side of the room with lux levels reaching
up to 50 Lux which is shown in red line. The simulation map

13
 aging of the lamp which would affect the light loss factor resulting
to low lux output. This is apart from the occasional flickering of
some lamps.
DSPLAB EAC2 lighting was simulated with 12 lamps
with individual lamp rating of 2500 lumens (OSRAM 36W), 4
lamps with individual lamp rating of 2750 lumens (GE F36W T8),
2’ x 4’ x 3” PARAMAX parabolic troffer and a mounting height of
2.7 m with ceiling cavity of 0.65m. As shown in Fig. 3d, the dark
spots are mostly located on the portions of the room with busted
lamps showing lux levels reaching up to 25 lux as indicated on red
lines. Comparing with the values on the grid, the measured lux levels
are reaching 13 Lux on the portion of L10 and 60 Lux on the L8.
The simulation map matches patterns of the color maps – indicating
that the illumination model for the DSPLAB EAC2 is valid. The
LLF is at 0.49.

B. Temperature distribution issues

Using color mapping, the room temperature per each
measurement were represented on each grid with their
corresponding values as shown on Fig. 4. The EEI of MICROLAB
is 3582.72 Wh / 24 °C
MICROLAB did not reach the target temperature within
23-25 °C. Several observed factors have caused the thermal
discomfort. AC2 could not reach 16-17 °C during daytime as cycle
Fig. 3. Sample luminance data of DSPLAB EAC2 (April 30, 2019) at (a) 9- time is not observed during the audit while AC3 unit was not
10 A.M., 120.7 lux average; (b) 12-1 P.M., 136.5 lux average; (c) 3-4 P.M.,
working and undergone maintenance during the audit. This had
110.3 lux average; and (d) average lighting simulation, 138 lux average.
caused AC2 unable to deliver the proper cooling. Also, location of
DSPLAB was divided into two energy activity centers AC2 cannot deliver proper cooling for the whole room. A
(EAC1 and EAC2) since one EAC was observed to have (1) condensate leak was observed in AC2 unit which may be caused by
different occupant activity than the other; and (2) a clear boundary a clogged drain line or evaporator blockage. Aside from AC
between them. Using color mapping, the luminance per grid are problems, air leakage from unsealed ceiling conduits and lacking
shown on a sample measurement (Fig. 2). The EEI of DSPLAB door louver contribute to cooling inefficiency.
EAC1 and EAC2 are 216 Wh/80-130 lux and 432 Wh/100-140 lux, Power consumption issues in the MICROLAB includes
respectively.The DSPLAB EAC1 did not reach the recommended the use of electric stand fans which are turned on even when the air
illuminance of 300 lux. A WWR of 0.1911 shown on Table 7 conditioning units are set at low temperatures. Personal computers
suggests that Room 408 is poorly lighted (ASHRAE 90.1-2007) (PC) are also left in idle state or plugged.
[2][1][3]. The illuminance on DSPLAB EAC1 ranges from 60 to 160 The room temperatures of DSPLAB EAC1 and EAC 2 per
lux (yellow) while there are some dark spots below 90 lux (gray) can grid are shown as color map on Fig. 4. The EEI for EAC1 and EAC2
be observed. The lighting levels were not properly distributed as are 1727.76 and 2128.76 Wh / 24 °C, respectively.
well due to the following observed factors: (1) low reflectance
(55.7%) of wall paint (blue) on the west side of the room; (2)
shadowing due to the cabinets which obstructs the light; (3) dirt
accumulation and aging of the lamp which would affect the light loss
resulting to low lux output; and (4) lamp positioning. Lamp L1 is
positioned on the area where there are no users. The lighting on the
workstation of DSPLAB EAC1 is solely provided by lamps L2 and
L3.
DSPLAB EAC1 lighting was simulated with 6 lamps with
individual lamp rating of 2500 lumens (OSRAM 36W), 2’ x 4’ x 3”
PARAMAX parabolic troffer and a mounting height of 2.7 m with
ceiling cavity of 0.65m. As shown in Fig. 2d, the dark spots are
mostly located on left side of the room with lux levels reaching up
to 25 lux shown in red line. Comparing with the values on the grid,
the measured lux levels reached 76 lux below the audio recording
room (marked as n/a on the grid). The simulation map matches
patterns of the color maps – indicating that the illumination model
for the DSPLAB EAC1 is valid. The LLF is at 0.50.
Some areas of DSPLAB EAC2 reached 600 lux (red) due
to high illuminance attributed to the glass windows with partially
opened vertical blinds. Its 0.30 WWR suggests that Room 410
conforms to standards (ASHRAE 90.1-2007) [2][3]. However, it did
not also reach the recommended illuminance of 300 lux even if the Fig. 4. Sample room temperature data (April 30, 2019) of MICROLAB at
illuminance caused by the fenestration have reached above 600 lux. (a) 9-10 A.M.; (b) 12-1 P.M.; (c) 3-4 P.M; and DSPLAB at (d) 9-10 A.M.;
The lighting levels were not properly distributed as well due to the (e) 12-1 P.M.; and (f) 3-4 P.M.
following observed factors: (1) low reflectance wall paints (blue and
beige); (2) four lamps are not working (L8 and L10) where the dark
spots on the mapping can be observed; and (3) dirt accumulation and

14
 Table 1. Evaluation of existing air-conditioning equipment based on the cooling loads of each area.

Area Heat Load Heat Load Installed Recommended Remarks

(W) (hp) AC (hp)*
(hp)
Microlab Workstation 4934 6.62 2 x 14 8.0 Overrated
Microlab Meeting room 1120 1.50 3 2.0 Overrated
DSP EAC 1 2765 3.71 3 5.0 Underrated
DSP EAC 2 4327 5.80 14 & 3 8.0 Overrated

None of the facilities reached the 24 °C target temperature. Table 2. Estimated energy consumption of each area.
This has caused thermal discomfort due to several factors. First,
doors between rooms are kept open as a space cooling practice of Description MICROLAB DSPLAB DSPLAB
the occupants. The sliding door between EAC1 and EAC2 was kept EAC 1 EAC 2
open with oscillating fans in between to keep the cold air circulating Average energy 5.004 1.993 0.861
between these areas. The doors leading the faculty are also kept consumption per
opened while the AC units are turned ON. hour (kWh)
Second, some of possible air leak sources include wall Average power in 39.3 15.9 6.9
conduits without proper sealing, edges of doors have no sealing and 8-h operating time
louvers leading to maintenance areas are not completely closed as (W)
well. Overtime instances 50 100 100
Third, placement of AC units are not efficient for even during audit (%)
cooling distribution of the space. In EAC1, the unit AC1 is placed at Hours of overtime 9 15 15
near the sliding door and at the end of the workstation which does (h)
not distribute air flow evenly within the workstation. In EAC2, the
Estimated daily 98.3 47.8 20.6
coldest spots are in the intersection of air flows of AC2 and AC3.
consumption
Most of the work desks are the coldest spots in the facility and
with overtime
reaching below 23 °C by afternoon. However, remaining areas did
(kWh)
not reach near 24 °C target (either lower than 23 °C or higher than
Cost per month* 15,654 10,839 4,680
25 °C) since placement of AC2 and AC3 and tables arrangement
(PhP)
made cooling uneven for the whole workspace. * 22 working days per month, 10 PhP/kWh.
Lastly, the unit AC3 does not have a definite temperature
as set point making temperature control difficult.

C. Cooling load and energy consumption

The cooling load requirement (heat load of the room)
compared to the installed AC equipment are compared in Table 1. It
shows that the AC equipment per area either have underrated or
overrated units.
Majority of heat load of MICROLAB workstation goes to
computers with 57 % contribution. This was followed by the
occupants with 24%, other equipment with 10%, luminaires with 6%
and external sources with 3% contribution. Two AC units (LG,
floor-mounted) are installed in this area but are overrated based on
the total heat load. The meeting room (with 51% heat from
occupants, 25% from laptops used, 5% from luminaire and
remaining 19% from external sources and envelope) also has
overrated AC unit (Daikin, wall-mounted). However, heat load in
the meeting room still varies due to the flexibility of the number of
occupants using the meeting room and the mobility of the laptop
computers.
Computers have one of the largest contributions on
DSPLAB. Heat load of DSPLAB EAC1 is mostly on occupants
(35%) and computers (34%) which is followed by external sources
(15%), luminaires (8%), windows (8%) and other equipment (4%).
For DSPLAB EAC2, heating load is contributed mostly by
computers (42%), followed by the occupants (28%), luminaires
(12%) other equipment (8%), windows (5%) and adjacent sources
(5%). It was also observed that DSPLAB cooling practice is to
increase cool air distribution between EAC1 and EAC2. This is
mainly due to an underrated AC unit (Daikin, wall-mounted) for
EAC1 whereas AC units (Daikin, wall-mounted; and Carrier, floor-
mounted) are overrated for EAC2.
Fig. 5. Simulation of recommended settings for (a) MICROLAB (342 lux),
(b) DSPLAB EAC1 (306 lux) and (c) DSPLAB EAC2 (320 lux).

15
 Table 3. Energy efficiency measures for lighting.

No. Facility Energy Conservation Measures Total Average EEI Initial Annual Payback
Installed Illuminance (W- Cost Energy Period
Rated (lux) hr/300 (Php) Savings (years)
Power lux) (W-hr/year)
(W)
EEM- MICROLAB Rearrange tables; lowered luminaire 300 348 300 23,092.80 312 7.2
L1 mounting to 2.3 m; wall paint
application; rewiring connection of
switch of L3; replace lamps with 15W
led 2000 lm
EEM- DSPLAB Add 2 luminaires; replace all lamp of 180 335 180 18,754.00 449.28 4
L2 EAC1 luminaires to 18 W, 2100 lm; rearrange
mounting of lamps; apply white wall
paint; installed ceiling panels
EEM- DSPLAB Repaired 4 busted lamps; applied white 352 320 352 16,587.00 1,397.76 1.3
L3 EAC2 wall paint; replace lamps with LED
2640-lm 22-W; installed ceiling panels

It was observed that the three areas have similar energy
consumption characteristics. MICROLAB, DSPLAB EAC1 and
EAC2 have 4.48, 1.55 and 0.63 kWh morning consumptions,
respectively. Morning (0900H-1000H) has the lowest energy
consumption. This was due to the cool temperature felt in the
morning which delays AC starting operation, closed lamps if the
EAC is hit by sunlight, low morning activity of the occupants and
low occupancy overall.
Noontime (1200H-1300H) and afternoon (1500H-1600H)
energy consumptions for MICROLAB (5.35 to 5.20 kWh),
DSPLAB EAC1 (2.21 to 2.25 kWh) and EAC2 (1.29 to 1.42) are
almost similar to each other but are larger than the morning energy
consumption. This is due to the maximum occupancy of these areas
resulting to maximum power output from the computers, full AC
and lighting operations.
Accounting for all energized equipment (lighting,
receptacle and AC), the estimated energy consumption within the
audit period was determined. Basted from Table 1, MICROLAB has
the most energy consumption with 5 kWh. This is due to the desktop
computers operating for 24 hours on active and idle modes.
However, daily consumption increases due to overtime activities of
the occupants doing research and study activities. This has led to
daily monthly electricity costs to around PhP 15654, PhP 10839 and
PhP 4680 for MICROLAB, DSPLAB EAC1 and DSPLAB EAC2,
respectively.
D. Lighting recommendations
Table 3 lists the recommended measures of the three areas.
A high cost alternative is proposed for the MICROLAB (EEM-L1)
as illustrated in Fig. 5a. This recommendation involves low-cost
methods: rearranging desks and lowering of light mount; and high-
cost methods: replacement of lamps to 15-W 2000 lm (LIFE
BULB™ 4-ft. T8 LED Tube), painting all walls with white latex to
increase LRV, and rewire lamp L3 to the workstation lamp circuit.
This will lower EEI by 60 Wh into 300Wh/300 lux with a uniform
distribution of 300 lux on the workstation. The lighting simulation
results are shown on Fig. 5.
A recommendation for the DSPLAB EAC1 is a high-cost
alternative (EEM-L2) illustrated in Fig. 5b. Low cost measures
include rearrangement of luminaires by 90°. Adding luminaires and
rearranging them is inevitable since the tables are placed on the
edges of the room. Rearrangement of computers is not applicable as
well as the room width is cramped. High cost measures include
replacing lamps to New Lights™ 18-W, 2100-lumen LED, adding 2
luminaires and applying white wall paint. This has led to a lower
power consumption and uniform distribution of 300 lux on the
workstation. This will lower EEI to 180 Wh/300 lux from 216
Wh/300 lux. It requires an investment of PhP 18,564 but the energy
savings per month will be at 449.28 Wh.
Lighting recommendation for DSPLAB EAC2 (EEM-L3)
is illustrated in Fig. 5c. This alternative includes replacing all lamps
to WY™ 4FT T8 LED Tube (22-W, 2640-lumen LED), painting of
walls into white and installing ceiling panels. This will cause an
average luminance of 320 lux and lower EEI to 352 Wh/300 lux.
This is a better alternative than a low-cost measure to replace the
busted lamps L8 and L10 but will increase EEI to 648 Wh/300 lux
aside from the 192-lux average illuminance. The cost of upgrade is
Php 16,588 with 1.3-years payback and 1398-Wh monthly energy
savings.
E. Air-conditioning recommendations
Average data for MICROLAB (115.6 lux, 25.9 °C) was
used in eQUEST™ to get the annual energy consumption (16,309.09
kWh). Savings were based on differences in power consumption
from the simulations applying the energy conservation measures.
Energy consumption is assumed to have a rate of Php 10/kWh.
EEM-T1. All energy conservation measures are
summarized in Table 4. Tint and roll-up blinds were applied to
windows to reduce heat fenestration. Door, window and conduit
seals were applied to solve the problems on air leakage. Simulation
of air-flow optimization will have a projected consumption of
13,590.91 kWh per year. The air-flow optimization alone will cost
Php 43,053.60, will lower EEI to 2613.63 W-h/24 °C and will have
a 0.6-year payback.
EEM-T2. A high cost solution involves replacing of AC2
and AC3 by a 2-TR Carrier™ ceiling cassette AC unit (FP-
53CCFASB024308-2T). This will give appropriate cooling load to
the MICROLAB and give proper air flow to working areas as shown
in Fig. 6a. Simulation of AC installation will have a projected
consumption of 8,981.82 kWh per year. Air-flow optimization with
installation of cassette-type ACU will cost Php 27,181.82, will lower
EEI to 1727.27 W-h/24 °C and will have a 5.1-years payback.
3.5.2 DSPLAB EAC1

16
 Table 4. Energy efficiency measures for thermal control and air flow.

No. Facility Energy Conservation Measures Annual EEI Initial Cost Annual Payback
Consumption (W- (PHP) Savings Period
(kW-hr/year) hr/24ᵒC) (PHP) (years)
EEM- MICROLAB Airflow Optimization (Seals) 13,590.91 2613.63 43,053.60 73,272.73 0.6
T1 Addition of Tints and Blinds

EEM- AC3 Replacement (2TR cassette AC) 8,981.82 1727.27 139,892.40 27,181.82 5.1
T2 Airflow Optimization (Seals)
Addition of Tints and Blinds
EEM- DSPLAB Airflow Optimization (Seal) 4,466.09 715.72 27,420.00 29,675.45 0.9
T3 EAC1 AC1 Relocation
EEM- Ceiling Insulation 561.73 710.417 39,084.00 30,006.36 1.3
T4 Airflow Optimization (Seals)
Addition of Blinds
AC1 Relocation
EEM- Lowered Ceiling Insulation 4,801.73 721.78 40,812.00 29,687.00 1.4
T5 Airflow Optimization (Seals)
AC1 Relocation
EEM- DSPLAB AC Repair 12,290.91 1969.697 37,314.00 185,545.50 0.2
T6 EAC2 Airflow Optimization (Seals)
Addition of Tints and Blinds
EEM- Lowered Ceiling 12,172.73 1950.758 68,850.00 186,727.30 0.4
T7 AC Repair
Airflow Optimization (Seals)
Addition of Tints and Blinds
EEM- AC3 Replacement (floor-mounted) 11,109.09 1780.303 146,848.80 192,363.60 0.8
T8 Lowered Ceiling
Airflow Optimization (Seals)
Addition of Tints and Blinds
EEM- AC3 Replacement (3TR cassette AC) 11,109.09 1780.303 208,768.80 197,363.60 1.1
T9 Lowered Ceiling
Airflow Optimization (Seals)
Addition of Tints and Blinds
EEM- AC3 Replacement (two 1.5TR cassette 11,109.09 1780.303 188,550.00 197,363.60 1.0
T10 AC)
Lowered Ceiling Insulation
Airflow Optimization (Seals)
Addition of Tints and Blinds

Using actual average data of 175.5 lux lighting and 22.1
°C temperature, the power consumption simulated in eQUEST™ is
7,433.64 kWh per year. Energy consumption rate is assumed to be
Php10/kWh. Savings were based from differences in power
consumption from the simulations applying the energy conservation
measures. Air flow optimization is assumed by following the
proposed layout in Fig. 6c and d.
EEM-T3. One-meter roll-up blinds were applied to
windows without tint to reduce heat fenestration. Door, window and
conduit seals were applied to solve the problems on air leakage.
Simulation of air-flow optimization alone will have a projected
consumption of 4,466.09 kWh per year. The air-flow optimization
alone will cost PHP 27,420.00, will lower EEI to 715.72 W-h/24 °C
and will have a 0.9-of-a-year payback.
EEM-T4. Installing ceiling insulation will give better
airflow around DSPLAB EAC1 and insulating the room further
from external heat source. Simulation of AC installation will have a
projected consumption of 39,084.00 kWh per year. Air-flow
optimization with ceiling insulation will cost PHP 30,006.36, will
lower EEI to 710.417 W-h/24 °C and will be paid back in 1.3 years.
EEM-T5. Applying ceiling insulation 2.7-m above with
conduit seals, window blinds and tints will give DSPLAB EAC1 a
projected consumption of 40,812.00 kWh per year. EEI will be
721.78 W-h/24 °C. This will cost PHP 40,812.00 and will be paid
back in 1.4 years.
For DSPLAB (122.5 lux, 24.4 °C), the annual energy
consumption simulated is 30,845.45 kWh in a year. The layouts in
Fig. 6b, c and d are assumed to have an optimized air flow.
EEM-T6. Tint and roll-up blinds were applied to windows
to reduce heat fenestration. Door, window and conduit seals were
applied to solve the problems on air leakage. AC3 repair will also
optimize air flow. Simulation of air-flow optimization alone will
have a projected consumption of 12,290.91 kWh per year. The air-
flow optimization alone will cost Php 37,314.00, will lower EEI to
1969.697 W-h/24 °C and will have a 0.2-of-a-year payback. Layout
shown in Fig. 6b.
EEM-T7. Aside from AC3 repair, addition of ceiling
insulation 2.3-m high will give better airflow around DSPLAB
EAC1 and insulating the room further from external heat source.
Simulation of AC installation will have a projected consumption of
12,172.73 kWh per year. Air-flow optimization with ceiling
insulation will cost Php 68,850.00, will lower EEI to 1950.758 W-
h/24 °C and will be paid back in 0.4-of-a-year. Layout shown in Fig.
6b.
EEM-T8. Replacing of AC3 with a 3TR QUBE™ floor-
mount split-type (SKU11415) AC unit with ceiling installation will

17
 Fig. 6. Proposed layout on (a) MICROLAB; (b) DSPLAB with AC3 repair or replacement, (c) DSPLAB with cassette-type AC installation, and (d)
DSPLAB with two ceiling-mounted installation.
have a projected consumption of 11,109.09 kWh per year. It will
reduce the EEI to 1780.303 W-h/24 °C by spending Php 146,848.80
in total and will be paid back in 0.8-of-a-year. Layout shown in Fig.
6b.
EEM-T9-10. Replacing of AC3 with a 3-TR Carrier™
cassette-type AC unit (FP-53CAV036308) (EEM-T9) or two 1.5-
TR LG™ cassette-type AC unit (JTNQ18GPLE5) (EEM-T10) will
have a projected consumption of 11,109.09 kWh per year same as
replacing with floor-mount AC. This will reduce the EEI to
1780.303 W-h/24 °C by spending Php 208,768.80 in total and will
be paid back in 1 to 1.1 years. Layouts show in Fig. 6c and d.
IV. CONCLUSION
In conclusion, energy audit was done to recommend
energy saving upgrades visualized by their costs. Lighting
distributions is generally inefficient because of low lamp lighting
output, unreplaced busted lamps, low reflectance of wall colors and
overall layout of the office desks. Temperature distribution
inefficiencies are found due to the AC layout, overrating or
underrating of the installed AC units and service length of AC units.
This has caused inefficiencies in power use and, therefore, increase
in monthly electricity costs. Energy consumptions are the lowest in
daytime and highest during noon to afternoon. This activity extends
when occupants file overtime occurring after 1700H.
Energy efficiency measures (EEM) were recommended
for lighting and temperature. These measures adhere to the
following criteria: lower EEI, monthly energy savings and outputs
with at least 300 lux and within 23-25 °C. The keys for an energy-
efficient light distribution include the following: desk alignment to
luminaires, high reflective wall color, installation of ceiling panels,
and replacement of lamps to a lower power rating but can supply
300 lux at least. An energy-efficient temperature distribution
includes the following upgrades: sealing air leaks, window tints and
blinds, ceiling insulation, layout of AC unit, correct AC unit
18
installation based on the heat load of the room, and (optional) use of
a cassette-type AC unit for a more even cooling distribution.

ACKNOWLEDGEMENT

The authors would like to thank the UP Electronics and

Electrical Engineering Institute, Dr. Franz de Leon of the Digital
Signal Processing Laboratory (DSP), and Dr. Maria Theresa de
Leon of the Microelectronics Laboratory (MICROLAB) for
permitting the energy audit on their respective laboratories.

REFERENCES

[1] Department of Energy. (2007). Manual practice on efficient

lighting. Philippines: Department of Energy. Retrieved at May
2, 2019 from https://documents.pub/document/manual-on-
efficient-lighting-5584560dce502.html.
[2] American Society of Heating, Refrigerating and Air-
Conditioning Engineers. (2009). 2009 ASHRAE Handbook
Fundamentals. New York: The Society.
[3] Tiew, S.Y. (2012). Chapter 4.0 Window to wall ratio.
Retrieved at May 2, 2019 from
http://studentsrepo.um.edu.my/3598/6/04_chap_4.pdf

19
Production and Characterization of Sol-Gel Processed
Microencapsulated Lauric Acid-Based Phase Change
Material from Treated Waste Coconut Oil
Paolo Yves L. De Silos1, a *, Menandro S. Berana,b, and Rizalinda L. De Leon3,c
1
Energy Engineering Program, College of Engineering,
University of the Philippines Diliman, Quezon City, Philippines
2
Department of Mechanical Engineering, College of Engineering,
University of the Philippines Diliman, Quezon City, Philippines
3
Department of Chemical Engineering, College of Engineering,
University of the Philippines Diliman, Quezon City,
a
pldesilos1@up.edu.ph, bmenandro.berana@coe.upd.edu.ph, crizalinda.deleon@coe.upd.edu.ph
* please mark the corresponding author with an asterisk
Abstract— The study dealt with the production and fluctuations of temperature and reduce the load of the cooling
characterization of microencapsulated lauric acid from treated waste equipment for energy efficiency [1].
coconut oil. The treatment of the waste coconut oil included Coconut oil is widely used in the Philippines as a primary
filtration, water removal, acid-catalyzed esterification, and alkali- cooking material. Because of the abundance in food processing
catalyzed transesterification. For the extraction of lauric acid, the plants in the country, a large amount of waste cooking oil is
method of Ezigbo & Mbaegbu which used sodium hydroxide was produced in a day-to-day period. Likewise, there is a need to
applied. The extracted lauric acid was encapsulated using Sol-Gel determine possible uses of this waste product to promote sustainable
process. In the preparation of the mPCM, the effect of drying time and interconnected waste management and renewable source of
and drying temperature on the product weight, diameter, energy. The formulation and production of cost-effective,
encapsulation ratio, and latent heat of melting was determined. For environmental-friendly, and sustainable ways in creating PCM is
the characterization, thermal properties were determined using vital to ensure waste minimization and energy security. There is a
Differential Scanning Calorimetry (DSC), presence of possible compelling reason to use waste cooking oil from food processing
functional groups were identified using Fourier Transform Infrared plants in order to extract lauric acid which can be microencapsulated
(FTIR) Spectroscopy, and the morphology were investigated using to form thermal energy systems [2][3].
Scanning Electron Microscopy (SEM). In FTIR, a greater detection In general, this study aims to produced sol-gel processed
of lauric acid may be inferred from the higher transmittance. microencapsulated lauric acid-based phase change material from
Likewise, a lower detection of lauric acid in the microencapsulated treated waste coconut oil and determine its properties as thermal
PCM may be attributed to the ineffective encapsulation which may energy system (TES). Specifically, this study aims to: determine the
result to leakage. The peaks for the functional groups of lauric acid effect of drying temperature and drying time in the product weight,
and silicate were found. On the other hand, a faster drying with high diameter, encapsulation ratio, and latent heat of melting of the
temperature result to a high yield. Among the lauric acid sources, microencapsulated phase change material (mPCM) from the
the one from technical grade lauric acid obtained the highest due to extracted lauric acid obtained from the treated waste coconut oil;
purity. In the DSC, a fast drying with high temperature resulted to a characterize the morphology of the mPCM using Scanning Electron
low enthalpy. Likewise, at constant drying time, increasing the Microscopy (SEM); identify the functional groups present in the
drying temperature resulted to an increased in latent heat of melting mPCM using Fourier Transform Infrared (FTIR) Spectroscopy; and
and a less steep melting temperature. In addition, in using SEM it determine the thermal properties of the mPCM using Differential
was seen that mPCM produced from low temperature and slower Scanning Calorimetry (DSC).
drying condition resulted to longest diameter due to longer time for
agglomeration. The significance of the drying time and drying
temperature in the product weight, diameter, and encapsulation ratio II. METHODOLOGY
was studied. It was seen that the drying time is significant in the
product weight and diameter. However, no conclusion was made on A. Materials and Methods
its significance in the encapsulation ratio and latent heat of melting. The methods included in the study were the treatment of
Moreover, the drying temperature was seen as significant in all four the waste coconut oil, the extraction of lauric acid, the
parameters. microencapsulation of lauric acid, and the characterizations of the
Keywords—lauric acid, phase change material, coconut mPCM. The reagents used were obtained from Sigma Aldrich and
oil, sol-gel, waste-to-energy the analyses were done at the Department of Science and
Technology – Advanced Device and Materials Testing Laboratory
(DOST-ADMATEL) and the UP Diliman Department of Chemical
I. INTRODUCTION Engineering. For the treatment of waste coconut oil, filtration, water
removal, acid-catalyzed esterification, and alkali-catalyzed
Various studies have already been done to determine the transesterification were done [4]. For the acid-catalyzed
capability of forming phase change material (PCM) from lauric acid. esterification, 2 mL 95% sulfuric acid, 6:1 methanol-to-oil molar
Variations include the combination with other fatty acids or ratio, 120-min reaction time, and 60 °C reaction temperature were
catalysts. The unique characteristics of lauric acid that makes it applied. Moreover, for the alkali-catalyzed transesterification, a 5.5
viable for further studies are its low vapor pressure, low degree of g/L KOH-to-oil ratio, 5:1 methanol-to-oil molar ratio, 60-min
supercooling, high heat of fusion, and compatibility with reaction time, and 60 °C reaction temperature were used. The
encapsulating material. Hence, it can prevent, or at least delay, the

20
forming glycerol was removed using a separatory funnel to avoid
unwanted products in the succeeding reactions [5][6].
For the extraction of lauric acid, a 10 M solution of NaOH
was prepared. In a heated water bath, the treated coconut oil and
NaOH were mixed. Afterwards, the coconut oil was converted into
glycerine and lauric acid and a distinct two-layer immiscible mixture
was observed. The lauric acid was separated using a separatory
funnel. The glycerine by-product was removed from time-to-time.
To confirm the presence of lauric acid, a two-g sample was placed
in a beaker and acidified KMnO4 was added. The lauric acid
decolorized the acidified KMnO4 completely; thus, confirming the
presence of lauric acid [7].
The microencapsulation method used was Sol-Gel
Process. In a beaker containing 50 mL distilled water, a 43 mL
tetraethyl orthosilicate (TEOS) and 50 mL absolute ethanol were
mixed. The beaker was placed in a hot plate with magnetic stirrer.
The alkoxide solution was agitated at 300 rpm. A 30-mL distilled
water and sulfuric acid solution was prepared on a separate beaker
before being mixed with the alkoxide solution. To ensure
miscibility, the mixture was heated to 65 °C for 90 min and agitation
speed was increased to 600 rpm to prevent agglomeration. The sol
mixture was separated after the completion of the hydrolysis and
condensation reactions of tetraethyl silicate. A 24-g lauric acid was
added to the sol and stirring at 800 rpm for 40 min was needed to
ensure uniform dispersion within the mixture. The mixture
temperature was maintained at 80 °C. The sol was poured into a
slightly uncovered glass Petri dish and was placed in a gravity
convection oven using a combination of drying temperature and
drying time [8][9][10]. After the drying process, the glass Petri dish
was removed from the oven and stored in a desiccator. Gel time, or
the storage time of the glass Petri dish in the desiccator, was 5 days
since the acid catalyst used was sulfuric acid. The product was the
PCM which was microencapsulated.
B. Characterization of mPCM
The In determining the effect of drying temperature and
drying time in the characteristics of the microencapsulated lauric
acid-based PCM, 7 runs were designed. Sample A did not undergo
drying, while Samples B, C, D, E, F, and G underwent drying with
varying drying time, drying temperature combination, and lauric
acid source. The mPCMs were formed using either the waste
coconut oil (WCO), technical grade coconut oil (TGCO), or
technical grade lauric acid (TGLA). For the characterization, DSC,
FTIR Spectroscopy, and SEM were applied. The response variable
that was monitored in each run was the encapsulation ratio, product
weight, latent heat of melting, and diameter. The encapsulation ratio
is the ratio of the latent heat of meting of lauric acid and the
microencapsulated PCM [8][11].
∆𝐻𝑚,𝑚𝑃𝐶𝑀
𝑅= 𝑥 100 (1)
∆𝐻𝑚,𝐿𝐴
where:
R = encapsulation ratio (%)
ΔHm,LA = latent heat of melting of the lauric acid (J/g)
ΔHm,mPCM = latent heat of melting of the mPCM (J/g)
TABLE 1. SAMPLES WITH THE CORRESPONDING
PREPARATION PARAMETER
Sample Sample Drying Time Drying Temperature
Code Source (h) (°C)
A WCO - -
B WCO 5 100
C WCO 7 100
D WCO 7 90
E WCO 12 75
F TGCO 12 75
G TGLA 12 75
III. RESULTS AND DISCUSSION
A. Treatment of Waste Coconut Oil
Pretreatment of the waste coconut oil was needed to
prevent the formation of unwanted by-products due to the reduction
of transesterification productivity caused by the impaired oil quality
[12]. The importance of transesterification in the production of fatty
acid methyl ester (FAME) involves the reaction between short-chain
alcohol (e.g. methanol or ethanol) and triglycerides (oil), using an
appropriate catalyst, to produce fatty acid alkyl esters (biodiesel) and
glycerine. Transesterifications may be differentiated in terms of the
catalyst used as either acid-catalyzed, alkali-catalyzed, or enzyme-
catalyzed. Enzyme-catalyzed transesterification is impractical due
to its slow reaction and high cost. The acid-catalyzed esterification,
which used sulfuric acid as catalyst, was needed since WCO has a
high composition of FFA. The additional acid treatment converted
the FFA to ester [4][6].
B. Physical Properties of mPCM
Removal of water prior to the microencapsulation process
ensures a higher product weight. Likewise, a lower yield is obtained
because the high water component in the mixture increases the sol-
gel condensation. At low pH level, hydrolysis occurs slowly. Linear
molecules that are cross-linked are formed from the silica.
Moreover, the entanglement of the molecular chains result to the
additional branching [8]. Consequently, gelation occurs. Of the
samples which source is the treated waste coconut oil, the one with
fast drying and high temperature got the highest yield. On the other
hand, comparing the samples with the same drying conditions but
different sources, the one from the technical grade lauric acid had
the highest weight. This can be attributed to the purity and
abundance of the lauric acid due to its grade.
The morphology of the mPCM was studied using a
Hitachi S-3400 Scanning Electron Microscope. The SEM images or
micrographs of the samples were exposed in a 15 kV electron beam
with a magnification of x50, x500, x1000, and x5000. The
agglomeration of non-homogenous microparticles or ‘multicore
structure’. The microencapsulated lauric acid can be assumed by
searching for the lumping mass in the middle of each particle. From
the mPCM products, the one with low temperature and longer drying
time, had the highest diameter. This may be attributed to the longer
agglomeration period. On the other hand, the samples with faster
drying time and medium to high drying temperatures got a lower
diameter.
Fig.1. SEM image of Sample B at x50, x500, x1000, and x5000
21
C. Chemical Properties of mPCM
The chemical compounds present in the mPCM was
investigated by applying FTIR. This was needed in confirming the
encapsulation of the lauric acid core material by the silica-based
shells. The spectroscopy was conducted from 0 to 4000 cm-1
wavenumber. From the IR spectra, there is a consistency in the dips
found from 0 to 2500 cm-1 from the various microencapsulated
samples. However, beyond 2500 cm-1, deviations were observed.
This may be attributed to the varying concentration of lauric acid in
the samples. Transmittance is the ratio of the light energy entering
on a body to that transmitted through the body. In the IR spectra, a
greater detection of lauric acid may be inferred from the higher
transmittance. Likewise, a lower detection of lauric acid in the
microencapsulated PCM may be attributed to the ineffective
encapsulation which may result to leakage. The tetraethyl
orthosilicate peaks are at 463 cm-1, 960 cm-1, and 1083 cm-1. The
hydrolysis-polycondensation reactions were catalyzed by acid
nucleophilic substitution reactions. There is no chemical interaction
between the silicate and lauric acid. Likewise, surface tension of the
silica-based shell was retained which prevents leakage of the lauric
acid core material. Moreover, silanol (Si-OH) and siloxane (Si-O-
Si), both basic groups, are responsible for hydrogen bonding which
produces a negatively-charged surface.
Fig.2. Infrared spectra of the microencapsulated PCMs
D. Thermal Properties of mPCM
Thermal properties of the mPCM was investigated using
a Perkin Elmer DSC 4000. These thermal properties include glass
transition temperature (Tg), melting point (Tm), freezing point (Tc),
and latent heats. In using the DSC, the difference in the heat flowrate
between the inert reference and the sample as a function of
temperature and time.
Fig.3. DSC curve of the mPCM from Sample B
A fast drying with high temperature resulted to a low
latent heat of melting. Hence, drying using the 7 hours drying time
is more practical. Moreover, the melting temperature of the product
was affected by the drying temperature. For the latent heat of
melting of the mPCM, at constant drying temperature, an increased
in the drying time resulted in an increased in the latent heat of
melting of the mPCM; while at constant drying time, an increased
in the drying temperature also resulted in an increased in the latent
heat of melting of the mPCM.
E. Statistical Analysis
The significance of the drying time and drying
temperature in product weight, diameter, encapsulation ratio, and
latent heat of melting of the microencapsulated PCM was analyzed
using Two-Sample T-Test in Minitab. A 95% confidence interval
was applied; hence a calculated p < 0.05 infers a significance.
Applying the analysis, it was found that the drying time of the
mPCM is significant for the product weight and diameter. Likewise,
the drying temperature is significant for the product weight,
diameter, encapsulation ratio, and latent heat of melting of the
mPCM. Additional runs and specific statistical tools are needed to
determine if drying time is not significant in the encapsulation ratio
and latent heat of melting of the mPCM.
TABLE 2: Physical, chemical, and thermal properties of the
microencapsulated PCMs.
Latent Heat Encapsulation
Weight Diameter
Code of Melting Ratio
(g) (μm)
(J/g) (%)
B 35.94 1.86 31.82 17.27
C 31.75 2.16 44.74 24.28
D 31.82 1.36 5.47 2.97
E 33.57 9.72 47.05 25.53
F 34.42 - 27.31 14.82
G 40.32 - 29.47 15.99
IV. CONCLUSION
After Waste coconut oil was physically and chemically
treated for the extraction of lauric acid. Sol-Gel Process was used to
encapsulate the extracted lauric acid. The encapsulation procedure
was also applied extraction lauric acid from technical grade coconut
oil and technical grade lauric acid for comparison purposes. This
paper aims to produce and characterize microencapsulated PCM
(mPCM) from waste coconut oil. In the preparation of the mPCM,
the effect of drying time and drying temperature on the product
weight, diameter, encapsulation ratio, and latent heat of melting was
determined.
For the characterization, the chemical, physical, and
thermal properties were investigated. Using FTIR, the presence of
silicate and lauric acid were determine. A greater detection of lauric
acid may be inferred from the higher transmittance. Likewise, a
lower detection of lauric acid in the microencapsulated PCM may be
attributed to the ineffective encapsulation which may result to
leakage. The peaks for the functional groups of lauric acid and
silicate were found. On the other hand, a faster drying with high
temperature result to a high yield. Among the lauric acid sources,
the one from technical grade lauric acid obtained the highest due to
purity. In the DSC, a fast drying with high temperature resulted to a
low latent heat of melting. Likewise, at constant drying time,
increasing the drying temperature resulted to an increased in the
latent heat of melting and a less steep melting temperature. In
22
addition, in using SEM it was seen that mPCM produced from low
temperature and slower drying condition resulted to longest
diameter due to longer time for agglomeration.
Using T-Test, it was seen that the drying temperature has
significant effect on product weight, diameter, encapsulation ratio,
and latent heat of melting of the mPCM. Moreover, it was seen that
drying time is significant on the product weight and diameter.
However, no conclusion was made for its effect on the encapsulation
ratio and latent heat of melting of the mPCM.
An optimization study on the effect of drying time and
drying temperature on the production of microencapsulated PCM
from waste coconut oil is recommended. Moreover, a change in the
waste feedstock is suggested due to its similar abundance in Metro
Manila and nearby provinces. Likewise, the effect of varying pH
level, water content, and oil ratio may also be looked in to.
ACKNOWLEDGEMENT
Utmost appreciation is given to the Department of Science
and Technology – Engineering Research and Development for
Technology (DOST-ERDT) for funding the research study.
REFERENCES
[1] Alva, G., Lin, Y., Liu, L., & Fang, G. (2017). Synthesis,
characterization and applications of microencapsulated phase
change materials in thermal energy storage: a review. Energy
and Buildings, 144, 276-294
[2] Dimaano, M. N. R., & Watanabe, T. (2009). The capric-lauric
acid and pentadecane combination as phase change material for
low-temperature thermal energy storage
[3] Pielichowska, K., & Pielichowski, K. (2014). Phase change
materials for thermal energy storage. Progress in materials
science, 65, 67-123
[4] Duti, I. J., Maliha, M., & Ahmed, S. (2016). Biodiesel production
from waste frying oil and its process simulation. J. Mod. Sci.
Technol, 4(1), 50-62
[5] Ullah, Z., Bustam, M. A., & Man, Z. (2014). Characterization of
waste palm cooking oil for biodiesel production. International
Journal of Chemical Engineering and Applications, 5(2), 134
[6] Adepoju, T. F. & Olawale, O. (2014). Acid-catalyzed
esterification of waste cooking oil with high FFA for biodiesel
production. Chemical and Process Engineering Research, 21,
80-85.
[7] Ezigbo, V. & Mbaegbu, E. (2018). Extraction of Lauric Acid
from Coconut Oil, Its Applications and Health Implications on
Some Microorganisms. African Journal of Education, Science
and Technology, 3(3), PP 59-62
[8] Gecain, M. K. R. (2015). Preparation and properties of
microencapsulated lauric acid phase change material
incorporated in glass sheet for thermal regulation (Unpublished
master’s thesis). University of the Philippines - Diliman, Quezon
City, Philippines
[9] Sciortino, M., Alonzo, G., Ciriminna, R., & Pagliaro, M. (2011).
Sol–Gel microcapsulation in silica-based particles: A
comparative study. Silicon, 3(2), 77-83
[10] Brodhacker, K. L. (2011). The sol-gel process [Powerpoint
slides]. Lander University, South Carolina, USA
[11] Fang, G., Li, H., Chen, Z., & Liu, X. (2011). Preparation and
properties of palmitic acid/SiO2 composites with flame retardant
as thermal energy storage materials. Solar Energy Materials and
Solar Cells, 95(7), 1875-1881.
[12] Guerrero-Romero, A., & Sierra, F. E. (2011). Biodiesel
production from waste cooking oil. In Biodiesel-Feedstocks and
Processing Technologies. IntechOpen.
23
Investigation of Heat Transfer Efficiency of Tungsten
Carbide and Cobalt Oxide Nanoparticles Dispersed on
Distilled Water as Base Fluid
Jhone Mychale I. Mariquit1, a *, Edgar Alan A. Donasco2, b *
1
Luinab, Iligan City, Lanao del Norte 9200, Philippines
2
Xavier Heights, Cagayan de Oro City, Misamis Oriental 9000, Philippines
a
jhonemychale.mariquit@g.msuiit.edu.ph, bedgaralan.donasco.g.msuiit.edu.ph,
Abstract—Dealing with the inquiries on the effects on
heat transfer efficiency of nanofluids with water as base fluid at
laminar and transient regimes, Tungsten Carbide (WC), and Cobalt
(II, III) Oxide (Co3O4), were tested in a shell and tube type crossflow
heat exchanger with 0.25-inch inner diameter. Percent volume
concentrations of nanoparticles ranges from 0.5% to 2.0% in
increments of 0.5%. Impacts on the thermo – physical properties
were investigated. Considered also was the comparison of Nusselt
number between theoretical, experimental, and computer simulated
results. The study showed that adding nanoparticles had substantial
impacts. Decrease in heat capacity can be perceived for all
nanofluids investigated. While increase in thermal conductivity,
viscosity, and density was recorded. Theoretical computations
compared with computer simulated results had the highest
discrepancy. While experimental data and simulated results had
smaller difference. Regarding the effects on heat transfer efficiency,
WC nanofluid presented deteriorations in all particle loading even
though it exhibited boost in thermal conductivity. While cobalt
oxide nanofluid displayed substantial improvements, with the
highest augmentation recorded at 13.27% for 1.5% volume
concentration. Based on the experimental data, correlations are
proposed in approximating the Nu of the nanofluids investigated
with ±10% consistency between predicted and experimental values.
Keywords—heat transfer, laminar flow, nanofluids,
transient condition
I. INTRODUCTION
The development of convective heat transfers and the
associated investigational as well as experimental and theoretical
research turn out to be at present a self-regulating, significant and
rapidly developing field of heat transfer theory.
High heat flux elimination or removal is a key attention in
the design of many machineries, apparatus and technologies, and
can be achieved using several types of heat transfer equipment. High
heat-flux cooling methods have obtained great consideration from
researchers due to its standing in thermal management of both
commercial and defense high-power electronic devices. While
remarkable development has been made in the past, high heat-flux
removal still principally remains as a stimulating topic that needs
additional investigation and study [1]. One of the several answers to
the problem of heat transfer performance improvement could be the
use of the so called nanofluids, which are basically fluids containing
nanoparticles of countless composition. Nanofluids can be deemed
as the new heat transfer fluids for they propose stimulating potentials
to improve heat transfer performance against pure liquids. They are
expected to have superior properties compared to conventional heat
transfer fluids [2]. Carbon nanotubes, copper nanoparticles and gold
nanoparticles, along with hybrid nanoparticles such as carbon
nanotube–copper nanoparticles or carbon nanotube–gold
nanoparticles were used in the study [3] to improve the thermal
conductivity of fluids. Single type nanoparticle suspensions
displayed the highest improvement in thermal conductivity at
around 74% increase compared to that of the base fluid. Between
nanoparticles considered, copper nanoparticle showed the greatest
improvement. An environment - friendly process utilizing intensive
solar electromagnetic radiation is used to synthesize graphene and
graphene–multi walled nanotube (MWNT) hybrid nanoparticle [4].
For 0.04% volume concentration, thermal conductivity
improvement of 9.2% with graphene nanofluids was observed and
10.5% increase for graphene–MWNT nanofluids dispersed in de-
ionized water considered at room temperature. Additionally, at
Reynolds number of 2000, an improvement in heat-transfer
coefficient of 193% was observed for 0.02% volume fraction of
graphene–MWNT hybrid nanofluids. A study [5] which defines an
innovative synthesis method for a hybrid nanostructure comprising
of multi walled nanotubes, hydrogen exfoliated graphene and silver
nanoparticles measured the thermal conductivity and convective
heat transfer characteristics at varied volume concentrations. An
improvement of around 8% in thermal conductivity is attained for a
volume concentration of 0.04% while improvement in heat transfer
coefficient of nearly 570% for 0.005% volume fraction was
reported. Friction factor as well as convective heat transfer
characteristics of Al2O3–water nanofluid in a round tube with
uniform wall temperature on turbulent flow settings were considered
in an experiment [6]. An average of 25% rise in heat transfer
coefficient was detected but with a 20% drawback in pressure drop.
As can be perceived, the intensification in heat transfer coefficient
in nanofluids paralleled to base fluids can vary from a few percent
up to several hundreds of percent. Nonetheless, there are studies
where the heat transfer efficiency decreases or there is no significant
improvement observed with the supplement of nanoparticles to the
base fluids.
An experimental research [7] considered alumina and
zirconia nanoparticles. Alumina nanoparticle concentration was
varied from 0.9 – 3.6% while zirconia nanoparticle was varied from
0.2 – 0.9%. The test revealed no unusual heat transfer improvement.
A different set – up [8] which considers natural convection of
nanofluids inside horizontal cylinder reported a seemingly
contradictory behavior of heat transfer deterioration. Nanoparticles
utilized in the study were aluminum oxide and copper oxide with
variation of percent volume concentration of 1% and 4%. According
to the report, the decline was dependent on particle density,
concentration and the aspect ratio of the cylinder. Titanium oxide –
water nanofluid with particle sizes ranging from 30 – 40 nm with
volumetric concentration of up to 3% was investigated
experimentally [9]. The findings also revealed that the
supplementation or presence of nanoparticles in the base fluid
decreases the natural convective heat transfer coefficient within the
conditions of the investigation. Considering hybrid nanoparticles, an
experimental investigation [10] utilizing silver – aluminum oxide
hybrid nanoparticle was carried out. Varying the percent volume
24
concentration from 0.005% to 0.1%. The study reported that a
grooved heat pipe system with nanofluids and hybrid nanofluids led
to worse system performance. According to a research review [11],
there is an incongruity among experimental data of various authors
and the issue of heat transfer enhancement or deterioration with
nanoparticle utilization is still unsolved and it requires follow up
investigations. Conferring to the review, different data, sometimes
contradictory ones, are obtained even for the same compositions of
nanofluids. Numerous available studies and related literature can be
analyzed and cited that shows that there is no quantifiable nor
qualitative unanimity with regards to the heat transfer performance
of nanofluids. Thus, such circumstance requires additional
experimental study of the heat transfer of nanofluids. The primary
purpose of this paper – the study on the effect on heat transfer
efficiency when adding tungsten carbide (WC) nanoparticles and
cobalt oxide (Co3O4) nanoparticles to distilled water as the base fluid
tested in a shell and tube type crossflow heat exchanger with 0.25-
inch inner diameter at laminar and transient regimes. This study also
intends to present a computer simulation of the experiment utilizing
SOLIDWORKS Flow Simulation program and assess whether such
method would prove to be beneficial in finding out the effects of the
nanofluids investigated.
II. METHODOLOGY
A. Nanofluid Preparation
Nanoparticles investigated in the study were produced by
SAT nano Technology Material Co., Limited. WC nanoparticles
were less than 100 nm at 99.95% purity. Co3O4 nanoparticles were
less than 40 nm at 99.5% purity. WC – water and Co3O4 – water
nanofluids were prepared via two-step method. With 0.5% to 2.0%
volume concentration in increments of 0.5%. The mixtures were
then placed on a H3770-HS digital magnetic stirrer for 30 minutes
at stirring speed of 1000 rpm. After that, the solution was subjected
to ultrasonic vibration with a Cole – Parmer EW-08895-04 for 1.5
hours at 40kHz. The produced nanofluids were stable for at least 4
hours without any visible sedimentation.
B. Test Rig
The test rig is outlined with the main objective of
investigating the effects or impacts on the convective heat transfer
capability of the nanofluid as volume concentration is increased. The
employed test rig consisted of a fabricated heat exchanger, pump,
flow meter, thermocouples, and water tank. Nanofluid from the
vessel was transported via the pump. Passing through the flow
sensor and in to the shell and tube, cross flow with multiple pass heat
exchanger unit. Temperature of the incoming hot water will be
maintained via an electric heater. Inlet and outlet temperatures of the
nanofluid will be verified with the use of type T and type J
thermocouples with an accuracy of ±0.1 ̊C, considering also other
points in the heat exchanger set up. 16 thermocouples are connected
to the National Instruments data acquisition system. Similarly, the
incoming and outgoing hot water temperature will be measured. The
flowrate of nanofluid will be measured by the use of a flow sensor.
While that of the hot water will be measured by using the bucket
method. Investigation were done in varying Reynolds number of
1900 to 3020. Fig. 1 presents the schematic illustration of the
experimental set – up utilized in the study.
C. SOLIDWORKS Flow Simulation
The simulation aspect of the study will be carried out
through the utilization of SOLIDWORKS Flow Simulation tool. In
order to come up with the different nanofluids to simulate in the
program, the researcher used the application’s Engineering Database
that allow users to create various types of items with defined
properties. To create a new type of fluid, properties such density,
dynamic viscosity, specific heat, and thermal conductivity are
needed at least to define a fluid. Such properties were taken from the
calculations made using the equations and correlations which will
be described in the succeeding sections. Incoming hot water
temperature, nanofluid inlet temperature, hot water mass flow rate
and nanofluid mass flow rates will be taken as input parameters in
the simulation. Such parameters will be based on the results taken
from the experimental aspect of the study.
D. Data Reduction
a. Thermo-physical properties
As dispersions of solid particles in a continuous liquid
matrix, nanofluids are expected to have a thermal conductivity that
obeys the effective medium theory developed by Maxwell [12] as:
Knf Knp + 2Kbf + 2Φ(Knp − Kbf )
= (1)
Kbf Knp + 2Kbf − Φ(Knp − Kbf )
where Knf is the nanofluid thermal conductivity, K bf is the thermal
conductivity of the base fluid, Knp is the nanoparticle thermal
conductivity and Φ is the particle volume fraction.
Nanofluids, which are commonly undiluted dispersions of
nanoparticles, are deliberated to be better suitable utilizing the
power law-based representations proposed by Krieger and
Dougherty [13] for the estimation of the viscosity of a nanofluid
paralleled to other representations like the one credited to Einstein
[14]. Einstein’s correlation is defined as:
μnf / μbf = 1 + KμΦ (2)
where μnf is the viscosity of the nanofluid while μbf is the viscosity
of the base fluid. Kμ = 2.5 as Einstein suggested for dilute systems
of spherical particles. Krieger – Dougherty correlation is expressed
as:
Φ - [μ]Φm
μnf = μbf (1- ) (3)
Φm
where where [μ] = 2.5 for hard spheres. Φm ≈ 0.605, which is the
max packing friction. This values ranges from 0.495 to 0.54 for inert
conditions and 0.64 with random suspended spheres.
Density and specific heat capacity for nanofluids are
projected based on the mixture rule [15, 16] as:
ρnf = Φ ρnp + (1 – Φ) ρbf (4)
Cpnf ρnf = ΦCpnp ρnp + (1 – Φ) Cpbf ρbf (5)
where ρnf, ρnp, ρbf, are density of nanofluid, nanoparticle and base
fluid, respectively. Cpnf, Cpnp, Cpbf are heat capacities of nanofluid,
nanoparticle and base fluid, respectively.
b. Heat Transfer
Basically, heat transfer is the transmission of thermal
energy from one object to another of different temperature. Heat
transfer can be defined as:
Q = ṁCpΔT (6)
where ṁ the mass flow rate, Cp is the specific heat capacity, and ∆T
25
is the change in temperature. With heat exchangers, the correlation where f is the friction factor relating to friction losses in pipe flow
concerning the overall heat transfer coefficient and the rate of heat and flow in open – channel [24],
transfer can be expressed by the subsequent equation,
f = (0.79ln(Re) – 1.64)-2 (14)
Q = UA (LMTD) (7)
Pe is the nanofluid Peclet number, which is a quantity of
where Q is the heat transfer rate, A is the heat transfer surface area, comparative significance of advection contrasted with diffusion
U is the overall heat transfer coefficient, and LMTD is the [25]. Peclet number can be defined as:
logarithmic mean temperature difference.
For identified values of overall heat transfer coefficient, Pe = Re * Pr (15)
heat transfer coefficient of the nanofluid, hnf, can be determined by
where Re is the Reynolds Number and Pr is the Prandtl Number as
1 defined previously.
hnf = 1 As t A 1 A (8)
( )( )-( )( s )-( )( s )
U Am k Am hs Anf

where As, Am, Anf are shell side surface area, logarithmic mean of
the surface areas, and nanofluid surface area, respectively. t
represents the thickness of the tube, k is the thermal conductivity of
the tube used and hs is the heat transfer coefficient at the shell side,
which can be identified accordingly through [17]

ℎ𝑠 (𝑑𝑜− 𝑑𝑖 ) 𝑑
= 0.020𝑅𝑒 0.8 𝑃𝑟 0.33 ( 𝑜 )0.53 (9)
𝑘 𝑑𝑖

where do and di are outside and inside diameters of the tube,

respectively. Re and Pr are Reynolds number and Prandtl number
which will be defined in the subsequent sections.

c. Dimensionless Number

Reynolds number typifies the relative strength of inertial

forces to viscous forces. The relative strength of these two actions - Fig. 1 Schematic diagram of Test Rig
their ratio - does have a lot of influence on how the fluid flow acts
[18, 19]. The Reynolds number of the nanofluid can be calculated
considering the nanofluid properties as:
III. RESULTS AND DISCUSSION
Renf = ρnfvnfD / µnf (10)
Thermal conductivity of nanofluids has an exceptionally
where ρnf is the density of the nanofluid, vnf is the flow prominent part during the heat transfer process. One of the main
velocity of nanofluid, μnf is the viscosity of the nanofluid and D is criteria that was considered in selecting of nanoparticles for the
the tube diameter. study was its thermal conductivity, that is considerably better than
The ratio of momentum diffusivity to thermal diffusivity that of the base fluid. With the hypothesis that the addition of
is known as Prandtl Number and can be defined considering nanoparticles leads to liquids with higher thermal conductivity.
nanofluid properties as [20]: Fig. 2 and 3 shows the graphed results of the thermal conductivity
of the nanofluids at different percent volume concentrations against
Prnf = Cpnf µnf / Knf (11) different correlations [26, 27] for thermal conductivity
approximations.
Heat transfer function is considered to be a better gauge
than the effective thermal conductivity for nanofluids used as
coolants in different applications, the attention for modeling of
nanofluid heat transfer coefficients is expanding and continues to
grow. However, it is still on its initial phase, and the theoretical
correlations for nanofluid heat transfer coefficients are moderately
partial. Most of the correlations, if not all, are revised from
conventional equations such as the Dittus-Boelter equation or the
Gnielinski equation with experimental considerations added [21–
23]. Gnielinski correlation is defined as:

Nunf = 0.012(Renf0.87 – 280) Prnf0.4 (12)

valid for Re = 2300 to 5 x 106, 1.5 ≤ Pr ≤ 500

𝑓
( )(𝑅𝑒−1000)𝑃𝑟
8
𝑁𝑢𝑛𝑓 = 𝑓 0.5 2 (13) Fig. 2. WC – Water Nanofluid Thermal Conductivity
1+1.27( ) (𝑃𝑟 3−1)
8

It can be perceived from the graphed results that thermal

valid for Re = 2300 to 5 x 106, 0.5 ≤ Pr ≤ 2000 conductivities of the nanofluids have increased considerably with

26
increasing particle concentration compared to that of the base fluid. introduction of solid nanoparticles which have densities that are
Highest increase recorded at 6.02% for WC-water nanofluid and greater than the base fluid, this would in turn lead to specific heat
5.55% increase for Co3O4-water nanofluid both at 2.0% volume capacity that is lower than water.
concentration. The considered correlation implores the Effective
Medium Approximations (EMA) or Effective Medium Theory
(EMT). Such approximations or estimates are depictions of a
medium centered on its properties as well as the properties of the
qualified fractions of its components. Considering component level
of a certain material, values differ and are diverse. Particular
computation of the several component values is almost irresolvable
but theory as such can come up with satisfactory estimates which
can define valuable parameters and characteristics of the combined
material in its entirety [28, 29]. Also, in a benchmark study [30], it
affirmed that the effective medium approximations or effective
medium theory for well-dispersed particles can accurately
reproduced experimental data in the limited set of nanofluids in their
exercise. With regards to viscosity, as expected viscosities of the
nanofluids are higher than that of the base fluid. Since nanoparticles
are dispersed in the base fluid, it developed its resistance to flow. Fig. 5a. Nanofluid Density
Thus, increasing its viscosity. Fig. 4 shows the graphed results of the
viscosities of the nanofluids compared to approximated results using
different correlations [31 – 33].

Fig. 5b. Nanofluid Heat Capacity

To ensure accuracy and repeatability, water was the first

Fig. 3. Co3O4 – Water Nanofluid Thermal Conductivity
and last working fluid to be tested in the set – up in between the
nanofluids. Results were then evaluated and a 0.149% difference
was recorded between the initial and final water test.
Figs. 6a, 6b, 6c, and 6d depicts the comparison of Nu
number results taken from the different aspects considered in the
study.

Fig. 4. Nanofluid Viscosity

Concerning materials science, mixture rule is a weighted mean

applied to calculate several properties of a compound matter
comprising of continuous and simplex fibers [34, 35]. Heat capacity
and density of the nanofluids investigated in the study are calculated
precisely based upon this rule. Figs. 5a and 5b illustrates the heat
capacity and density of the nanofluids considered in the study.
The heat capacity of nanoparticles considered in the study
are lower compared to that of the base fluid. By definition, heat
capacity is the quantity of heat delivered to a certain mass of an
object to create a unit change in its temperature [36]. With the
Fig. 6a. WC – Water Nanofluid Nu at Laminar Conditions
As observed, the correlation employed to calculate the
Nusselt number for the theoretical side (i.e. Gnielinski) does not
correspond well with the experimental (i.e. Present study) and
simulated (i.e. SOLIDWORKS) aspect. It either undershoots the
calculation for both nanofluids with Reynolds number within the
laminar condition (i.e. Re < 2300) and overshoots the computation

27
with Reynolds number in the transient condition (i.e. 2300 < Re < Figures 8a and 8b illustrates the graphed results of the percent
4000). Discrepancy between the different Nusselt number enhancement or detraction of nanofluid with distilled water as the
calculations revealed that the average percent difference between base line.
experimental – theoretical, experimental – simulated, and theoretical
– simulated are 11.793%, 2.775%, and 12.901%, respectively for
WC-water nanofluid. While Co3O4-water nanofluid showed an
average percent difference of 11.364%, 4.937%, and 10.702% for
experimental – theoretical results, experimental – simulated results,
and theoretical – simulated results, respectively. Such observations
of discrepancy between theoretical calculation and experimental
data can also be perceived in a certain study [37]. Wherein
theoretical calculations undershoot experimental results by 50%. A
correlation is then proposed for this study to approximate the results
of the experimental Nu where coefficients of the correlation are
based on percent volume concentration, Reynolds number, Prandtl
Number, and friction factor. The general assessment for the
coefficient of determination on how sound the approximated results
are in the proposed correlation is 97%. Proposed correlations are
described in the concluding section.
Fig. 6d. Co3O4 – Water Nanofluid Nu at Transient Regime

Fig. 6b. WC – Water Nanofluid Nu at Transient Conditions

Fig. 7a. Distilled Water as coolant Simulation

Fig. 6c. Co3O4 – Water Nanofluid Nu at Laminar Conditions

Figures 7a, 7b, and 7c depicts some of the completed

simulation using define nanofluids and water as the cooling agents.
To fully assess if whether the nanofluids investigated in
the present study provided proved to perform better as the cooling
agent compared with water, percent enhancement in heat transfer
efficiency is employed. This is outlined as overall heat transfer Fig. 7b. WC – Water Nanofluid Simulation
coefficient of nanofluid, Unf, divided by the overall heat transfer
coefficient of the base fluid, Ubf As per evaluation, there was actually a detraction in the
heat transfer efficiency with the nanofluids investigated. For
% Enhancement = (µnf / µbf) *100 (16) nanofluids under laminar conditions, WC-water nanofluids showed
no indications of heat transfer efficiency enhancement at any percent
volume concentration involved in the study. The trend actually
worsens as nanoparticle loading increases.

28

Fig. 7c. Co3O4 – Water Nanofluid Simulation
With the largest reduction recorded at 4.89% for 2.0%
volume concentration. Cobalt Oxide nanofluid had better results at
laminar conditions. Recording an enhancement of 1.43% at 0.5%
particle loading, 2.19% enhancement at 1.0% and a high of 4.3%
enhancement at 1.5% volume concentration. Then at 2.0% volume
concentration, the heat transfer efficiency recorded a reduction of
2.23%. For nanofluids at transient regimes, WC-water nanofluids
still has not exhibit any significant improvement in heat transfer
efficiency. Nonetheless, the degree of reductions is considerably
lower at such regime compared to laminar conditions. For Cobalt
Oxide nanofluid however, it revealed the best outcomes among all
nanofluids in both laminar and transient states. Exhibiting a 5.76%
enhancement at 0.5% particle loading, 7.85% enhancement at 1.0%
particle loading, highest augmentation of 13.27% at 1.5% volume
concentration and 5.92% improvement at 2.0% volume
concentration.
Fig. 8a. % Enhancement / Detraction at Laminar Conditions
Fig. 8b. % Enhancement / Detraction at Transient Regime
IV. CONCLUSION
The current study dealt with inquiries on the utilization of
WC, and Co3O4 nanoparticles on water as the base fluid on its effects
on heat transfer capability. Effects of the considered nanofluids with
percent volume concentrations of 0.5%,1.0%,1.5%, and 2.0% were
analyzed with the main attention on heat transfer efficiency at
laminar to transient regimes in the particular set – up. Considering
the thermo – physical properties, adding nanoparticles to the base
fluid had substantial impacts. Decrease in heat capacity can be
perceived for all the nanofluids. While increase in thermal
conductivity, viscosity, and density was noticed with all nanofluids
investigated. Comparison between theoretical computations,
experimental data, and computer simulated results had significant
differences. Theoretical computations compared with computer
simulated data had the highest discrepancy. While experimental data
and computer simulated results had lesser discrepancy. Regarding
the effects on heat transfer efficiency, WC-water nanofluid
presented deteriorations in all particle loading even though it
exhibited boost in thermal conductivity for all percent volume
concentration. While Co3O4-water nanofluid proved to be
beneficial. With the highest augmentation in heat transfer efficiency
of 13.27% at 1.5% volume concentration. Correlations are proposed
in approximating the Nu of the nanofluids investigated with less than
10% deviation.
Nuwc = -17.394 + 30.201Φ + 0.0039Re + 1.26Pr + 284.66f
valid for 1900 < Re < 3020, 4.9 < Pr < 6.4, 0.5% - 2.0% Φ
NuCo3O4 = 73.19 – 1260.38Φ + 0.0248Re – 39.53Pr + 2733.203f
valid for 1900 < R e< 2600, 5.7 < Pr < 6.4, 0.5% - 2.0% Φ
ACKNOWLEDGEMENT
The author desires to express his gratitude to the
Department of Science and Technology – Engineering Research and
Development for Technology (DOST – ERDT) for supporting the
research.
REFERENCES
[1] Ebadian MA, Lin CX. A Review of High-Heat-Flux Heat
Removal Technologies. ASME. J. Heat Transfer.
2011;133(11):110801-110801-11. doi:10.1115/1.4004340.
[2] Xiang-QiWang, Arun S. Mujumdar. Heat transfer
characteristics of nanofluids: a review. International Journal of
Thermal Sciences Volume 46, Issue 1, January 2007, Pages 1-
19
[3] Jana S, Khojin AS, Zhong WH. Enhancement of fluid thermal
conductivity by the addition of single and hybrid nano-
additives. Thermochimica Acta 2007; 462:45–55.
[4] Aravind SSJ, Ramaprabhu S. Graphene–multiwalled carbon
nanotube-based nanofluids for improved heat dissipation. RSC
Advances 2013; 3:4199–206.
[5] Baby TT, Ramaprabhu S. Synthesis of silver nanoparticle
decorated multiwalled carbon nanotubes–graphene mixture
and its heat transfer studies in nanofluid. AIP Adv 2013;
3:012111.
[6] Fotukian SM, Esfahany M. Experimental study of turbulent
convective heat transfer and pressure drop of dilute CuO–water
nanofluid inside a circular tube. International Journal of
Communications in Heat and Mass Transfer 2010; 37:214–219.
29
[7] Williams W, Buongiorno J, Hu L. Experimental investigation Engineering: R: Reports. 63: 1–30. doi:
of turbulent convective heat transfer and pressure loss of 10.1016/j.mser.2008.07.001
alumina/water and zirconia/ water nanoparticle colloids [30] Buongiorno, Jacopo et al. “A benchmark study on the thermal
(nanofluids) in horizontal tubes. Journal of Heat Transfer 2008; conductivity of nanofluids.” Journal of Applied Physics 106
130:042412. (2009): 094312. Copyright © 2009, American Institute of
[8] Nandy Putra, Wilfried Roetzel, Sarit K. Das, Natural Physics
convection of nano-fluids, Heat and Mass Transfer 39 (2003) [31] Brinkman HC. The viscosity of concentrated suspensions and
775–784. solution. Journal of Chemical Physics 1952; 20:571–81.
[9] Dongsheng Wen; Yulong Ding, Natural convective heat [32] Krieger IM, Dougherty TJ. A mechanism for non-Newtonian
transfer of suspensions of titanium dioxide nanoparticles flow in suspensions of rigid spheres. Transactions of the
(nanofluids), IEEE Transactions on nanotechnology, vol. 5, no. Society of Rheology 1959;3: 137–52
3, May 2006. [33] Batchelor GK. The effect of Brownian motion on the bulk
[10] Han WS, Rhi SH. Thermal characteristics of grooved heat pipe stress in a suspension of spherical particles. Journal of Fluid
with hybrid nanofluids. Therm Sci 2011; 15:195–206. Mechanics 1977; 83:97–117
[11] Terekhov V I, Kalinina S V and Lemanov V V. The mechanism [34] Askeland, Donald R.; Fulay, Pradeep P.; Wright, Wendelin J.
of heat transfer in nanofluids: state of the art (review). Part 2. (2010-06-21). The Science and Engineering of Materials (6th
Convective heat transfer. Thermophysics and Aeromechanics, ed.). Cengage Learning. ISBN 9780495296027
2010, Vol. 17, No. 2. [35] "Stiffness of long fibre composites". University of Cambridge.
[12] J.C. Maxwell, A Treatise on Electricity and Magnetism, second Retrieved 1 January 2013.
ed., Clarendon Press, Oxford, UK, 1881. [36] Halliday, David; Resnick, Robert (2013). Fundamentals of
[13] Krieger IM, Dougherty TJ. A mechanism for non-Newtonian Physics. Wiley. p. 524
flow in suspensions of rigid spheres. Transactions of the [37] Lazarus Godson Asirvatham, Balakrishnan Raja, Dhasan
Society of Rheology 1959; 3:137–52 Mohan Lal, Somchai Wongwises, Convective heat transfer of
[14] Einstein A. Eine neue Bestimmung der Moleküldimensionen. nanofluids with correlations, Particuology 9 (2011) 626–631
Annalen der Physik 1906; 19:289–306.
[15] Pak, B. C. & Cho, Y. I. (1998). Hydrodynamic and heat transfer
study of dispersed fluids with sub-micron metallic oxide
particles. Experimental Heat Transfer, 11, 151–170
[16] Xuan,Y. and Roetzel,W.(2000).Conceptions for heat transfer
correlation of nanofluids. International Journal of Heat and
Mass Transfer, 43, 3701–3707.
[17] Yoo S.J., France D.M., Tarshish T.M.: Post-critical heat flux
swirl flow heat transfer with two refrigerants and water. J.
Thermophys. Heat Tr. 11(1997), 189–195.
[18] Falkovich, Gregory (2018). Fluid Mechanics. Cambridge
University Press. ISBN 978-1-107-12956-6.
[19] Zohuri B., McDaniel P. Thermodynamics in Nuclear Power
Plant Systems. Springer; 2015, ISBN: 978-3-319-13419-2
[20] Coulson, J. M.; Richardson, J. F. (1999). Chemical Engineering
Volume 1 (6th ed.). Elsevier. ISBN 978-0-7506-4444-0.
[21] Wang, Xiang-Qi, & Mujumdar, Arun S.. (2008). A review on
nanofluids - part I: theoretical and numerical investigations.
Brazilian Journal of Chemical Engineering, 25(4), 613-630.
[22] Fang, XD and Bullard, C.W. & Hrnjak, Pega. (2001). Heat
Transfer and Pressure Drop of Gas Coolers. ASHRAE
Transactions. 107. 255-266
[23] Gnielinski, V. (1976). New equations for heat and mass transfer
in turbulent pipe and channel flow. International Chemical
Engineering, 16, 359–368
[24] Munson, B.R. (2006). Fundamentals of Fluid Mechanics (5
ed.). Hoboken, NJ: Wiley & Sons.
[25] Coulson and Richardson's Chemical Engineering (Seventh
Edition) Volume 1b: Heat and Mass Transfer: Fundamentals
and Applications 2018, Pages 529-546 - Transport Processes in
Microfluidic Applications
[26] Hamilton RL, Crosser OK. Thermal conductivity of
heterogeneous two-component systems. Ind Eng Chem
Fundam 1962;1(3):187–91
[27] Maiga SEB, Palm SJ, Nguyen CT, Roy G, Galanis N. Heat
transfer enhancement by using nanofluids in forced convection
flows. Int J Heat Fluid Flow. 2005; 26:530–46
[28] Wenshan, Cai; Shalaev, Vladimir, Optical Metamaterials:
Fundamentals and Applications. November 2009, Springer. pp.
Chapter 2.4. ISBN 978-1-4419-1150-6.
[29] Wang, M; Pan, N. "Predictions of effective physical properties
of complex multiphase materials". 2008, Materials Science and

30
Development of Hybrid Convection Solar Dryer for
Seaweeds (Kappaphycus alvarezii S.)
Reniel Z. Rocaberte1, a *, Marvin M. Cinense2,b, Ruel G. Peneyra3,c
and Jeffrey A. Lavarias4,d
1
Puerto Princesa City, Palawan
234
Central Luzon State University, Science City of Muñoz, Nueva Ecija
a
rocaberte.reniel01@gmail.com, bmarvin_cinense@clsu.edu.ph, crgpeneyra@clsu.edu.ph,
c
jeffreyalavarias@clsu.edu.ph
Abstract—The seaweeds solar drying experiment was
conducted under metrological condition of Philippines using
hybrid convection solar dryer with solar thermal air collector and
solar-powered auxiliary heater. This study addressed the problem
encountered by local seaweed farmers in dealing with seaweed
production likewise it based on the need for more efficient and
practically cost-effective drying system during cloudy and rainy
season. Seaweeds drying sample was subjected at different drying
temperature of 40C, 50C and 60C. Under these drying
conditions the moisture content reduced from 93.70% to 35.31%
moisture in 440 minutes. It showed that temperature played
significantly to the mass transfer of seaweeds. The drying kinetic
were fitted to exponential models using the coefficient of
determination (R2) and mean bias error (MBE). Results showed
that the henderson and pabis model best described the
characteristics of seaweeds which it was dominated by diffusion.
In addition, effect of drying temperature on the color of dried
seaweeds was evaluated by measuring the RGB value. The image
that was taken during the study was analyzed in capture image
software. It reveals that the resulting RGB mean values of dried
seaweed at different drying condition were significantly affected
by drying temperature. Overall, this report indicates that drying at
60C is most effective with good color appearance attributes with
uniform color. Further, the change of RGB value and moisture at
different drying condition can be used to predict the seaweeds
drying process.
Keywords—color analysis, solar drying, kinetic modeling, ,
seaweed
I. INTRODUCTION
Seaweed is vital for human life because of its nutritional
benefits and multiple uses. The Kappaphycus alvarezii S. is
one of the major variables that used to make carrageenan.
This carrageenan known as hydrocolloid which is a natural
extract from seaweeds as gelling agent, thickener, and
stabilizer in a wide variety of food applications, non-food
applications and cosmetics, pharmaceutical products. In the
export market, seaweeds are traded as dry to facilitate
efficient transport and product logistic (Phang et al, 2014).
Consequently, seaweeds must be dried before they can be
utilized for agricultural, industrial, and other purposes.
Seaweeds are reported as an extremely perishable
commodity which in most cases losses its nutritive value
within hours at a tropical temperature and recommended that
quick processing must be undertaken to prevent the losses of
the product. Effective preservation techniques, therefore, are
needed because inadequate processing and handling of
products can lead to quality loss.
Drying is an essential agricultural technique used to
preserve food products by reducing the moisture of the
product. Locally, sun drying is one of the conventional
drying methods that is used by seaweeds farmers, the
seaweeds are placed in an open space or a wooden platform
and or hanged outside. However, this method is extremely
weather dependent and the material are exposed to foreign
matters such as soil, sand which susceptible to contamination
infestation, microbial attacks, thus affecting product quality
(Seveda, 2013). Since the exportation of seaweeds started,
the drying using the conventional method have been used but
due to intensive utilization in the global market the
requirement of quality is one of the primary concerns and it
is now requiring improvement (Phang et al, 2014).
Being a weather dependent methods and with the given
limitations in open solar drying, there is a clear need to
develop an innovative dryer addresses the defined issues.
Furthermore, such intervention must take advantage of the
abundant solar energy to ensure low production cost coupled
with its capacity to dry during short and reduced sunshine
and even at night time to ensure the production of good
quality dried seaweeds. Though there were several drying
systems have been designed, developed and tested to address
the issue such that the seaweeds product will be more
competitive and economically viable to venture on. But
expected this innovation requires a lot of energy hence an
added cost to farmers, reducing their net potential income.
The lack of proper drying facilities and techniques in the
intensive quality requirement of dried seaweeds for the
global demand drives the effort to an innovative solution
through enclosed drying system with solar thermal air heat
collector and solar-powered backup heater
II. METHODOLOGY
A. Preperation of Sample
The seaweeds sample was obtained from the nearest
seaweeds production area of BFAR- Masinloc, Zambales.
The harvested sample were wash thoroughly with seawater
to removed epiphytes and other foreign matters. The
seaweeds sample were cut into smaller pieces (4 inches)
before soaking into fresh water. The sample was soaked for
31
about two hours. Then, then initial moisture content of the
seaweeds was determined by oven drying method. The
average initial moisture content obtained is 93.70%. The
dryer was pre-heated for 20 minutes in order to establish
steady state condition to the desire temperature of 40°C,
50°C and 60°C using the temperature controller while the
samples of the same mass (1700kg) were evenly arrange in
the three (3) drying trays. The reduction of weight was
determined from each tray at the top, middle and bottom
every two hours. The air temperature and relative humidity
inside the dryer and the surroundings were recorded every
ten minutes using a data logger.
B. Hybrid Convection Solar Dryer
The drying system was fabricated using locally and
commercially available materials with an approved design
plan. The unit and component are in accordance with
requirement of the seaweed. The dryer has an outer
dimension of 150cm x 150cm x 120cm (length x width x
height) and it was installed in Land and Water Resource
Management Center (LWRMC) Farm. Figure 1 shows the
schematic diagram of hybrid convection solar drying
system. The fans and auxiliary heater were powered by solar
PV system. The auxiliary heater served as back-up
mechanism to augment the needed energy for the air to
reach its desired temperature for drying. The air temperature
was controlled by a microcontroller-based temperature
controller operating on a feedback mechanism. This was
installed to actively regulate the temperature inside by
turning ON/OFF the heater and fan simultaneously during
the drying operation. The dried maximum capacity is 30kg
per batch.
Figure 1. Schematic diagram of HCSD of seaweed
C. Drying Analysis
Only good quality seaweeds were used in the experiment
5.1kg of fresh seaweeds were dried until the required final
moisture content was attained. Moisture content was
expressed as percent wet basis and then converted to gram
of water per gram of dry matter. The experimental drying
data of seaweed were fitted to the exponential model using
non-linear regression analysis. The section of best fit was
made using the highest coefficient of determination (R2) and
lowest mean bias error (MBE), root mean square error
(RMSE), and reduced chi-square (X2) (Ibrahim et al, 2009;
Hao et al, 2020).
1 2
MBE= ∑N
i=1(MRpre.i -MRexp.i ) (1)
N
1
1 2 2
RMSE= [ ∑N
i=1(MRpre.i -MRexp.i ) ] (2)
N
D. Color Analysis
The dried seaweeds at different temperatures were
evaluated using color analyzer. The color analyzer was used
to measure color value of the certain part of an image by
zooming into the region of interest and to obtain the RGB
value. The color change measure in RGB color can be in
color channels with intensity value for red, green and blue
color from 0-255. The results are presented as the mean
value of RGB (Mendoza et al, 2006; Lukinac et al, 2009).
E. Statistics Analysis
The data collected in this study were analyzed and
interpret using STAR program for agriculture research
version 2.0 to evaluate the effects of temperature to the
resulting performance and quality parameter of dried
seaweeds. The research design used single factorial
experiment arrange in CRD with three replications and
different drying temperature 40°C, 50°C and 60°C as
treatments. The test of comparison among means was made
using the Tukey’s Honest Significant Difference (HSD) test.
III. RESULTS AND DISCUSSION
The hybrid convection solar dryer was designed and
fabricated to address the need of the industry for hygienic
and cost-effective solution to seaweeds processing. the
hybrid dryer consisted of a solar thermal air collector as main
heating system with three (3) auxiliary heaters with fans.
That ensure suction and circulation of air from the solar
collector to the drying chamber. This served as the roof of
the dryer and it was installed at 16-degree inclination at the
top of the drying chamber. Both fans and heaters were
powered by solar PV system. The drying chamber was
thermally insulated with polycarbonate sheet while the floor
area was painted with black acting as great heat absorber. A
non-contact temperature sensor made of type-K
thermocouple feeds the temperature level directly to the
controller that triggers the relays to switch the heater
ON/OFF based on the desired set-point.
A. Drying Performance
The temperature inside the dryer shows significant result
with the ambient temperature as shown in the Figure 3. The
temperature inside the dryer shows significant result with the
ambient temperature, where the inside temperature is higher
than the outside temperature. This might be associated with
absorptive effect of the solar thermal air collector and the
auxiliary heater that was properly maintained the heating
requirement of the dryer. The average relative humidity
inside the dryer at different drying conditions of 40°C, 50°C
and 60°C were 40.17%, 34.07%, and 34.75% respectively.
Lower relative humidity was observed at higher temperature
due to the effect of heating to the vapor pressure exerted by
the vapor in the air (Rathore and Panwar, 2010).
32
 Figure 4. Krischer curve-drying rate versus moisture content
The quality of fit was expressed by the highest
coefficient of determination and lowest and the lowest Mean
Bias Error (MBE), Root Mean Square Error (RMSE), and
reduced chi-square (X2) (Fudholi et al 2014). Along with
Figure 2. Hybrid convection solar dryer for seaweed
criteria and the goodness of fit was statistically showed in
Henderson and Pabis model. For all treatments, the 60°C
drying temperature gave lower MBE, Reduce Chi-Square
which corresponds to higher results of R-square value. The
70
results showed a very high percentage of the possibility that
60
Temperature,°C

the model could predict the drying rate behavior at any given
50
condition of drying temperature. This means that Henderson
40
and Pabis model best describes the characteristics of
30 seaweeds in a convection solar dryer where it was dominated
20 by diffusion.
0 50 100 150 200 250 300 350 400 450 500 550 600
Time, min
40°C 50°C 60°C
40°C Amb 50°C Amb 60°C Amb TABLE I. VALUES OF MODEL CONSTANT AND STATISTICAL
PARAMETERS
Figure 3. Variation of drying temperature and ambient temperature
Temperature R2 (%) MBE RMSE X2
Model
Lewis 92.32147 0.40700 1.01736 1.03503
The seaweeds dried at 40°C reached 35% moisture 40°C
content in 590 minutes while seaweeds dried at 50°C and Henderson
86.35699 0.00000 0.25182 0.05888
and Pabis
60°C reached their desired moisture content of 35.78% in
453 minutes and 35.31% within 440 minutes, respectively. Lewis 91.26493 0.48814 1.19031 1.41685
Drying temperature had a great influence to the removal of 50°C
Henderson
surface moisture of seaweed as it was attributed to the 96.22279 0.00000 0.11693 0.01269
and Pabis
highest reduction rate (Yahya 2007). It was observed that
Lewis 91.47816 0.48563 1.18032 1.39316
the increase in drying temperature consequently decreased 60°C
the drying time (Fudholi et_al, 2011). Henderson
99.01455 0.00000 0.06031 0.00337
and Pabis
Furthermore, the external moisture and internal moisture
evaporates at different rate. The external surface of seaweed
had slight changes during the initial adjustment period
wherein the surface moisture was exposed to the drying air
evaporating at constant time, and on the later period, it has
undergone moderate changes in moisture reduction.
Figure 4 shows the relationship of moisture content and
drying rate as represented Krischer curve of dried seaweeds.
It was found that the drying rate initially increased and
reached its maximum value in about two hours from the
beginning of drying operation. This suggests that the rate of
movement and removal of moisture by diffusion and
evaporation in a convection drying process were energy- Figure 4. Plot of experimental and predicted value of 60°C
dependent and requires constant temperature regulation. - Henderson and Pabis model

B. Color Quality of dried Seaweed

The drying efficiency was computed based on the ratio
of the actual heat utilized to evaporate the moisture of the

33
seaweeds to the energy input. The electrical energy input
was measured using the power meter (kWh). The hybrid
convection solar dryer was evaluated considering the
parameter where the heat energy of the solar thermal air
collector was supplied by solar radiation and the
supplemental heating system was driven by the power of
PV/T solar panel. %. During the initial drying process, the
heater and the fan worked simultaneously to augment the
heat requirement for drying conditions thus, consumed more
energy when set on high drying temperature. At a
temperature of 60°C, both the auxiliary heater and fan had
longer ON time with a total of seven hours as compared to
drying temperature of 40°C that had shorter ON time during
the drying process with a total of four hours. The functional
combination of auxiliary heater and fan was proven to be
effective as it gave the best results for the dried seaweeds.
At 60°C, the seaweeds dried better with lesser drying time
of about eight hours but higher rate of energy consumption
which further resulted to a reduced efficiency with respect
to the time of operation of the heater and the fan, likewise,
drying temperature corresponds to the energy utilized.
C. Color Quality of dried Seaweed
Color is one of the quality attributes in dried products.
This simple color analysis is to compare the difference in
color of the dry seaweeds as affected by the drying
treatments. The results suggest that dried seaweeds at higher
temperature had better appearance in terms of color. It is
worth mentioning that the temperature at 60°C indicated a
remarkable uniform lightness during drying. The increase of
value of blue signifies blacken reaction. Temperature at
40°C obtained the higher blue value that signifies the
darkening of color on dried seaweeds. This can be supported
by the observation on dried commodity that lower
temperature setting results to darker colored surface. Similar
results were observed in the study of Mina, (2018) on which
the browning of the surface of the dried product is due to
effect of lower drying temperature and thus significantly
affects the final product.
IV. CONCLUSION
In this study, to help seaweed farmers produce
white dried seaweeds anytime, a hybrid convection solar
dryer was designed and fabricated using locally available
materials. It is equipped with heater and solar thermal air
collector, and it utilizes heat from the sun. The solar thermal
air collector combined with solar-powered auxiliary heater
provide enough hot air to dry the seaweeds. Likewise, the
use of temperature controller maximized the utilization of
heat in the system and it gave a satisfactory result in
maintaining and regulating the temperature of air inside the
drying chamber between temperature levels of 40°C, 50°C
and 60°C.The moisture content of seaweeds in the hybrid
convection solar dryer was reduced from an initial moisture
value of 93.70% (w.b) to a final value of 35% (w.b) within
ten hours of drying. The drying temperatures of 60°C
significantly reduced the moisture and increased drying rate
thus, shorter drying time compared to 40°C. However, lower
temperature level resulted to lower energy consumption.
Consequently, The Henderson and Pabis model showed the
best fit with a high value of 99.01% for the coefficient of
correlation and lower values of mean bias error of 0.00005,
a root mean square error of 0.00108, and reduced chi-square
of 0.000001 at a temperature of 60°C. Thus, Henderson and
Pabis best described the drying characteristics of seaweed
drying in a hybrid convection solar dryer. Results of color
determination showed that seaweed samples dried at 40°C
had the darkest appearance which was attributed to an
increase of oxidation reaction.
ACKNOWLEDGMENT
The author would like to thank the Department of
Science and Technology through Engineering Research and
Development for Technology (DOST-ERDT) for the
financial support through the graduate scholarship program,
Central Luzon State University grant scheme. The author is
grateful to Dr. Marvin M. Cinense for the guidance
throughout the experiment. To Prof Ruel G. Peneyra, his
second mentor, for overflowing intelligence, knowledge and
effort in providing help to completely attain the good
research result. Prof Jeffrey A. Lavarias, for his advice and
fruitful discussions. To Dr. Majid Khan Majahar Ali for the
guidance and support. The Bureau of Fisheries and Aquatic
Resources, Zambales Station Mr. Marlon Alejandro for their
support in sending the commodity to the Central Luzon
State University in this hard time of pandemic.
REFERENCES
[1] A. Fudholi, M.Y. Othaman, M.H. Ruslan, M. Yahya, A.
Zaharim and K. Sopian “The effects of drying air
temperature and humidity on drying kinetic of
seaweed,” International conference on energy and
development-environment-biomedicine. 2011, pp. 129-
133.
[2] A. Mina. 2018. Convection drying of mushroom
(Pleurotus florida) using solar thermal heatpipe water
heater. Master Degree Thesis. Central Luzon State
University.
[3] A. U. Sundaria, P. Neelamegamb and C.V.
Subramanianc, “Drying kinetics of muscat grapes in a
solar dryer with evacuated tube collector. International
journal of Engineering,” 2014, 27(5):811-818.
[4] BFAR, “Fisheries Commodities Road Map,” Seaweeds:
Fisheries Policy and Economic Diversion; Bureau of
Fisheries and Aquatic Resources. 2010.
[5] D. G. Mercer. 2008. Solar drying in developing
countries: possibilities and pitfalls. International Union
of Food Science and Technology. 4:1-11.
[6] D. J. McHugh, “A guide to the seaweed industry,” FAO
Fisheries Technical Paper 441. Rome: Food and
Agriculture Organization, 2003.
[7] D.Y. Dasin, A.I. Zakari and I. Yahuza, “Development
of solar energy dryer for fish and meat tested in
Modibbo Adama University of Technology Yola,”
International Journal of Trend in Scientific Research
and Development. 2019, 3(2):pp. 774-777.
[8] F. Mendoza, P. Dejmek and J. Aguiler, Calibrated color
measurements of agricultural foods using image
analysis. Postharvest Biology and Technology. 2006,
41(3):285-295.
[9] H. K. Phang, C. M. Chu, S. Kumaresan, M. Mizanur
Rahman and S. M. Yasir. “Preliminary study of
seaweed drying under a shade and in a natural draft
solar dryer,” International Journal of Science and
Engineering, 2014, 8(1):10-14.
[10] J. Lukinac, S. Jokic, M. Planinic, D. Magdic, M. Bilic,
S. Tomas, D. Velic and A. Bucic-Kojic, “ An
application of image analysis and colorimetric methods
on color change of dehydrated asparagus (Asparagus
34
 maritimus L.),” Agriculturae Conspectus Scientificus.
2009, 74(3): pp. 233-237
[11] M. K. M. Ali, A. Fudholi, J. Salaima, M.S. Muthuyala,
S. M. Yasir, A. Q. Hortado, “Post-harvest handling of
Eucheumatoid Seaweeds: Tropical Seaweed farming
Trends, Problems and Opportunity: Focus on
Kappaphycus and Eucheuma of Commerce,” 2017, pp.
131-145.
[12] N. Rathore and N. Panwar, “Design and development of
energy efficient solar tunnel dryer for industrial drying
Clean Technology,” 2010, Environment. 13 pp 125-32
[13] S. Suherman, M. Djaeni, A. Kumoro, A. Prabowo, S.
Rahayu and S. Khasanah. “Comparison drying behavior
of seaweed in solar, sun and oven tray dryers,” 2018,
MATEC conference 156.
[14] S. Abasi,S.M. Mousavi, M. Mohebi and S. Kiani,
“Effect of time and temperature on the moisture
content, shrinkage, and rehydration of dried onion,”
Iranian Journal of chemical Engineering. 6(3): 2009, pp.
58-70.
[15] P. Sain, V. Songara, H. Balan and R. Karir. “Natural
convection type solar dryer with latent heat storage,”
2016,
https://www.researchgate.net/publication/286662466.
[16] W. Hao, S. Liu, B. Mi and Y. Lai, “Mathematical
modeling and performance Analysis of new hybrid
solar dryer of lemon slice for controlling drying
temperature,” Energies, 2020, pp. 13:350
[17] N. Rathore and N. Panwar, “Design and development of
energy efficient solar tunnel
[18] dryer for industrial drying Clean Technology,” 2010,
Environment. 13 pp 125-32

35
GIS-based Site Suitability Analysis for Biomass
Energy Plant in Bohol’s Rice and Coconut
Producer Municipalities
Wenyville Nabor Galang1,2, a *, Ian Dominic Tabañag3,b and Michael Loretero4,c
1
School of Engineering, University of San Carlos, Cebu City, Cebu
2
College of Engineering and Computer Studies, Holy Name University, Tagbilaran City, Bohol
3
Department of Chemical Engineering, University of San Carlos, Cebu City, Cebu
4
Department of Mechanical and Manufacturing Engineering, University of San Carlos, Cebu City, Cebu
a
wgalang@hnu.edu.ph, biftabanag@usc.edu.ph, cmeloretero@usc.edu.ph
Abstract—Site suitability is an essential process in determining
potential sites for power plant development with respect to some
limiting factors. This study utilized a Geographic Information
System (GIS)-based land suitability analysis model considering
accessibility for locating optimal sites to the biomass power plant
development against environmental and economic pull factor. In
order to determine suitable site, the important factors are
incorporated, namely accessibility and topography, land cover, and
economic by using multi-criteria decision analysis (MCDA)
method in decision making process. The integration of MCDA and
a GIS model, allowed the identification and analysis of optimal
suitable locations that balanced economic and environmental
criteria within the biomass supply. The study produced a suitability
map showing the potential sites for biomass power plant
development in the area of consideration. It was found that among
the top rice and coconut areas in the province of Bohol, the most
suitable site is the municipality of Ubay.
Keywords—Biomass, Waste to Energy, Energy Potential, Site
Suitability, GIS-MCDA
I. INTRODUCTION
Globally, the use of renewable energy as an alternative
to fossil fuel is highly considered to counter the increasing
fuel cost and environmental concerns. Biomass is
considered as one of the main option for renewable energy
sources given its abundance to nature [1].
The term biomass refers to any organic material derived
from plants. Biomass is produced by green plants that
convert sunlight into plant material through photosynthesis.
Biomass resource can be considered as an organic material,
in which the solar energy stored in chemical bonds. When
the bonds between adjacent carbon, hydrogen and oxygen
molecules are broken by digestion, combustion, or
decomposition, then it releases their stored chemical energy
[2]. Biomass basically can be divided into three major
groups, wood biomass, non-wood biomass and secondary
fuels [3].
In the Philippines, biomass as a renewable energy
resources have great energy potential considering its
abundance in the sorrounding of an agricultural country. It
represents 25.5% of the indigenous energy supply as of
2016. Bulk of the demand for biomass was consumed by the
household sector. It is the most preferred fuel among
households because of its abundance, accessibility and
affordability, particularly in the rural areas [4]. Biomass
gradually stepped up its contribution to the power sector as
its level of fuel input to electricity generation increased by
as much twice its 2015-2016 level.
The province of Bohol in Central Visayas is a main
producer of agricultural products in the region. It is called
the “Rice Granary of Central Visayas” after it was cited for
surpassing other provinces in the region in terms of rice
output [5]. The rice production in the province hiked by
54.2 percent from 161,003 metric tons in 2016 to 248,314
metric tons in 2017, representing the bulk of the rice
produced in the region. The province took the bigger share
of the rice production in Central Visayas, improving by 45
percent from 231,982 metric tons in 2016 to 335,420 metric
tons in 2017. In addition, the province is abundance in
coconut plantation. Bohol Island has around 10 million
coconut trees producing an annual harvest of more than 343
million coconuts [6].
Bohol is primarily supplied by power plants located in
Leyte and Cebu through the Leyte-Bohol Transmission
Interconnection. Power plants outside Bohol supply about
68.86 MW or 89% of the peak demand of the province
through the Leyte-Bohol Interconnection. The remaining 8.4
MW (11%) is collectively supplied by the three mini-hydro
plants inside Bohol [7]. The vulnerability of its electric
power system and the need to make it resilient to the effects
of natural disasters lead to recognize the need to put in place
measures that can ensure sufficient power is supplied to the
whole island-province in the event that it is disconnected
from the rest of the Visayas Grid. Given the abundance of
agricultural sources, the province has the means of potential
energy from agricultural products waste such as rice and
coconut, respectively.
However, the meaning of potential is not just defined by
energy transfer, but defined by economic and environmental
factors as well. The location of a bioenergy plant is affected
by a wide variety of factors and criteria. This study seek to
identify the most suitable locations for a renewable energy
plant in the identified top rice and coconut municipalities in
the province. To find the best suitable land for bioenergy
36
facility based on multiple criteria decion analysis (MCDA) municipalities of Bien Unido, Buenavista, Clarin, Dagohoy,
encorporated in a Geographical Informational System (GIS). Danao, Getafe, Inabanga, Pres. Carlos P. Garcia, Sagbayan,
San Isidro, San Miguel, Talibon, Trinidad and Ubay; and
The GIS support the solution of complex spatial
the rest of the municipalities particularly Alicia, Anda,
problems, providing the decision-maker with a flexible
Batuan, Bilar, Candijay, Carmen, Dimiao, Duero, Garcia
environment in the process of the decision research and in
Hernandez, Guidulman, Jagna, Lila, Loboc, Mabini, Pilar,
the solution of the problem. To create visualized suitability
Sevilla, Sierra Bullones and Valencia belong to the 3rd
map for users and decision makers, the integration of
district.
MCDA and GIS has been widely promoted for solving
spatial problems in urban assessment and planning [8].
The availability of biomass residues varies spatially in the
Another study was conducted to introduce the Multi-criteria
province of Bohol, mostly due to land cultivation area and
decision analysis into the land use suitability analysis along
land topography. First district of Bohol have 15
with the existing perspective evolving the role of GIS [9].
municipalities including the city of Tagbilaran. The second
Meanwhile, some case studies were performed using the
district is abundant in rice and coconut production. It
method of integrating GIS and multi-criteria analysis to
includes the municipalities of Bien Unido, Buenavista,
create an output with a set of land use suitability maps
Clarin, Dagohoy, Danao, Getafe, Inabanga, Pres. Carlos
incorporating complex criteria integrating several stake-
Garcia, Sagbayan, San Isidro, San Miguel, Talibon,
holders' points of view. In the Philippine setting, a study
Trinidad and Ubay. The rest of the 19 municipalities belong
using Multi Criteria Decision Analysis and GIS processes
to the third district which is abundant in corn production.
were employed in the province of Nueva Ecija. The land
cover was classified from LandSAT 8 image using Support
Vector Machine classification technique. The resulting
classified map was used to generate the biomass resource
map along with expert consultation for a three level
hierarchy of factors used in the site suitability analysis [9].
However, in the region of Visayas, there is limited study
done for site suitability particularly in the province of Bohol
given its biomass potential for renewable energy sources.
This study utilizes the integration of GIS-MCDA to support
decision-makers in providing them with ways to evaluate
several alternatives, based on multiple and conflicting
criteria.
II. METHODOLOGY

The study was conducted within the scope of Bohol Fig. 2. Bohol’s Rice and Coconut Producer Municipalities
province in the country’s Central Visayas region. There are
3 districts composed by 47 municipalities and a city in To determine a suitable land for a new plant, the
Bohol with a total land area of 411,726 ha and arable area of following major approach was developed: (1) selection of
256,400 ha., see Fig. 1 cultivation area (2) ranking and (3) determine the most
suitable land.
The initial step where cultivation areas were selected based
on multiple criteria:
1. road accessibility,
2. slope of the area,
3. large cultivation area;
4. populated area.

Using the given criteria with weightage given 20%, 20%,

40% and 20% respectively was utilized to find the most
suitable location. The particular location of the land which
satisfies these multiple criteria shall be the suitable area to
put up a biomass power plant.

The ranking and suitability determination involves a

Fig. 1. Area of study, the Province of Bohol.
The municipalities of Alburquerque, Antequera,
Baclayon. Balilihan, Calape, Catigbian, Corella, Cortes,
Dauis, Loon, Maribojoc, Panglao, Sikatuna, Tagbilaran and
Tubigon composed the 1st district; the 2nd district are
general process to convert and obtain a raster file for road
access, slope, cultivation area and population. Then identify
the location through calculation in a raster calculator using
the weightage.

37
 The input data are vectors in shape file processed using
QuantumGIS, a free and open-source cross-platform
desktop application that supports viewing, editing, and
analysis of geospatial data [10]. The QGIS steps involve
creating a new field in the attribute table of each layer added
for a particular vector and classify them according to
importance. The new field shall be used as an input to
rasterize the respective layers for major road, slope, land
area and population.

Fig. 6. Population

In the first step, croplands were selected based on

multiple criteria. First, they must near the major roads and
easily accessible. The primary road in the identified area of
study was given the top priority. Secondly, slopes of the
land near the cropland. A higher slope will increase the
difficulty of construction, so lower slope area is more
Fig. 3. Major Roads suitable for the new plant. Moreover, the croplands must be
big enough in order to supply plenty of crop residues.
Lastly, there must have large population in some distance
from the croplands, so the power can be used efficiency.
Attributes were ranked on a scale of 1 and 0 with 1(in
black colored) being the unrestricted area for development
(shows in Figure 3, Figure 4, Figure 5 and Figure 6). Then
the selection of the suitable land for development and
combined it with ranking.

III. RESULTS AND DISCUSSION

Cultivation area was the most valued layer (Figure 5)

because of having the most abundance of agricultural
Fig. 4. Slope residue guaranties a sufficient energy source for biomass
plant development. The slopes from 0% to 3% provided by
the slope layer (Figure4) were the most suitable areas in
order to place a biomass plant due to ease of construction
given its description as level to nearly level. The
municipality of Ubay possesses 50% larger area and level
slopes as compared to the rest of the rice and coconut
producing municipalities of the province. With regards to
the accessibility of roads (Figure 4), the “primary road” was
the most valued. These primary roads are the ones that
directly connect to major cities. In the case of energy
demand, the population layer (Figure 6) of high numbered
household was most valued.
The combined layers for road, slope, land area and
population lead to the identification of exact location were
to best station a biomass power plant. The spatial
distribution of various criteria influence the site analysis. In
Fig. 5. Land Area
rasterized suitability map layer (Figure 7), it is in the
municipality of Ubay that serve as best location for biomass
plant. This map was calculated utilizing raster calculator in

38
QGIS following the waightage given for each criterion. The
satellite view of the specific location is given in Figure 8.
Fig. 7. Rasterized Suitability Map
Fig. 8. Biomass Plant Suitability Map in Satellite View
IV. CONCLUSION
Suitability maps as methods for visualization problem
provided by the combination of MCDA and GIS methods
can therefore be seen as a powerful tool for solving power
planning problems, such as the location of biomass plants.
GIS-MCDA techniques can used to obtain territorial
information for planning power supplies. In addition it can
provide the necessary tools to integrate this knowledge into
the development of the project to support decision making
and guarantee sustainable activities.
GIS-MCDA model as a decision support system, allows
efficient resolution of complex problem such is plant site
selection. The development of the method lead to the
selection of the most suitable site for the biomass plant
location within the areas with higher value with respect to
evaluation criteria (weighted criteria).
A. Authors and Affiliations
1.Wenyville Nabor Galang
School of Engineering, University of San Carlos, Cebu City,
Cebu, Philippines;
College of Engineering and Computer Studies, Holy Name
University, Tagbilaran City, Bohol. Philippines
2.Dr. Ian Dominic Tabañag
Department of Chemical Engineering, University of San
Carlos, Cebu City, Cebu, Philippines
3.Dr. Michael Loretero
Department of Mechanical and Manufacturing Engineering,
University of San Carlos, Cebu City, Cebu, Philippines
ACKNOWLEDGMENT
This study was supported by the Department of Science
and Technology, Science Education Institute under
Engineering Research and Development for Technology
program. We are grateful to DOST for the financial support
given and University of San Carlos, School of Engineering
for the accessibility and use of available facilities.
REFERENCES
[1] Woo, H., Acuna, W., Moroni, M., Taskhiri, M.S., and Turner,
P., 2018. Optimizing the Location of Biomass Energy
Facilities by Integrating Multi-Criteria Analysis (MCA) and
Geographical Information Systems (GIS). MDPI Forest
[2] McKendry P. , 2002. Energy production from biomass (part
1): overview of biomass. J. Biores Tech, 83, 37–46
[3] Calle F, Rosillo P, Groot S L, Hemstock and J Woods, 2007.
The Biomass Assessment Handbook: Bioenergy for a
sustainable environment. London: Earthscan
[4] Department of Energy, 2017. Energy Demand and Supply
Outlook 2017- 2040
[5] Philippine News Agency. Available from:
https://www.pna.gov.ph/articles/1024700 [Accessed April 7,
2020].
[6] Biological Homeopathic Medical Corp.. Available
from:
https://bhmcproducts.com/p5-m-coconut-hub-to-rise-in-
bohol/?v=a25496ebf095 [Accessed April 7, 2020].
[7] Province of Bohol, 2015. Bohol Island Power Supply Plan
(2016-2045)
[8] Phua M., Minowa, M., 2005. A GIS-based multi-
criteria decision making approach to forest conservation
planning at a landscape scale: a case study in the
Kinabalu Area, Sabah, Malaysia. Land Scape Urban
Plan 71(2), 207–222.
[9] Saladaga, I., Remolador, M., Sevilla,K., Baltazar, B.,
Inocencio, L.C., Ang, M.R. 2015. Site suitability analysis for
biomass power plant development in nueva ecija, philippines
using landsat based biomass resource map. Environmental
Science
[10] QGIS Official Website. QGIS. Retrieved 16 October 2013
39
[11] Chen, J.. 2014. GIS-based multi-criteria analysis for [13] Department of Energy, 2018. Available from:
land use suitability assessment in City of Regina. https://www.doe.gov.ph/sites/default/files/pdf/e_ipo/02_eib_r
Environmental Systems Research egion_vi_visayas_power_situation_and_outlook.pdf
[Accessed August 19, 2019]

[12] Department of Energy, 2018. Available from:

https://www.doe.gov.ph/electric-power/2018-power-supply-
and-demand-highlights [Accessed August 10, 2019]

40
Environment and
Infrastructure
Session Chair : Dr. Sheila K. Ramos
Moderator : Dr. Bryan G. Alamani
Panelists : Dr. Mary Donnabelle L. Balela
Dr. Candy C. Mercado
Dr. Annaliza Rollon
Dr. Maria Antonia Tanchuling
Colorimetric Detection of Copper Ions in Water using
Polymer-decorated Silver Nanoparticles
Edgar Clyde R. Lopez1, a, *, Jeanette Vivien R. Jalalon1, b, Mary Faith C. Delas Alas1, c, Cheska Mae P. Aldana1, d,
Marlon L. Mopon11, e, Jem Valerie D. Perez1, f, *
1
Nanotechnology Research Laboratory, Department of Chemical Engineering, University of the Philippines Diliman
a
edgarclydelopez09@gmail.com, bjrjalalon2@up.edu.ph, cmcdelasalas1@up.edu.ph, dcpaldana@up.edu.ph,
e
mlmopon@up.edu.ph, f jdperez@up.edu.ph
Abstract— Polymer decorated silver nanoparticles (PD-
AgNPs) were used as an assay-based nanosensor to detect copper
ions in water. Here, we decorated AgNPs with chitosan crosslinked
with polyethyleneimine using glutaraldehyde. The composition of
the PD-AgNPs was optimized by varying the CS, GLA, and PEI
concentrations using a Box-Behnken design. Response surface
analyses showed that a reduced cubic model is sufficient to model
and predict the Φ and LOD of PD-AgNPs with 95% confidence.
Numerical optimization revealed that the optimum formulation
comprises of 1.1642 wt.% CS, 0.8298 wt.% GLA, and 1.3687 wt.%
PEI. The predicted Φ = 82.5559 and LOD = 2.1566 mg L-1 agrees
well with our validation experiments (actual Φ = 82.9540 and LOD
= 2.2498 mg L-1). The calibration curve exhibits a dynamic linear
range between 0.10 to 4.00 mM of copper with an R2 value of
0.9930. Selectivity studies showed that the PD-AgNPs were
selective against common metal ions found in water. Overall, we
demonstrated that PD-AgNPs could be used for the rapid and facile
colorimetric detection of copper ions in aqueous media, making on-
site Cu (II) detection possible.
Keywords— Silver nanoparticles, Nanosensor,
Colorimetric sensing, Response surface methodology, Surface
plasmon resonance
I. INTRODUCTION
Heavy metals continue to be a significant problem
globally due to their tendency to bioaccumulate, which pose a
danger to humans, aquatic organisms, and the environment [1].
Conventional analytical techniques used in detecting heavy metals
such as atomic absorption spectroscopy (AAS) and inductively-
coupled mass spectroscopy (ICP-MS) have high accuracy and
precision, high sensitivity, and good selectivity even at very low
heavy metal concentration. However, these require the use of costly
and complex equipment requiring professional assistance for
operation [2]. This hinders the widespread monitoring of heavy
metal pollution at the source. A reliable, fast, portable, and
inexpensive method must be developed to enable on-site detection
of heavy metals in water.
Here, we report the synthesis of polymer-decorated silver
nanoparticles (PD-AgNPs) and its application to detect Cu (II) ions
in water. Chitosan and polyethyleneimine were used to decorate the
synthesized AgNPs, acting as functionalizing and stabilizing agents.
To achieve maximum sensing performance, we optimized the
composition of the PD-AgNPs using response surface methodology.
The concentration of chitosan (CS), glutaraldehyde (GLA), and
polyethyleneimine (PEI) was varied according to a three-factor
three-level Box-Behnken design of experiments. The optimized PD-
AgNPs were used for the colorimetric detection of copper ions in
water based on the nanoparticles' optical absorbance profile. To the
best of our knowledge, this is the first time that silver nanoparticles
have been decorated with both chitosan and polyethyleneimine and
used for copper sensing in aqueous solutions.
II. METHODOLOGY
A. Synthesis and Characterization of PD-AgNPs
18.0 mL of 2.0 mM AgNO3 was added to 18.0 mL of 2.0
mM trisodium citrate. This solution was placed in a 30 mL syringe,
and then it was added to 36.0 mL of freshly-prepared and ice-cold
2.0 mM NaBH4 solution at a rate of 50 mL min-1 using an infusion
pump. The solution was stirred continuously in an ice bath until the
reaction is complete. 6.0 mL of CS was then added to the
synthesized AgNPs, and the mixture was stirred continuously for 30
minutes. Afterward, 6.0 mL of the GLA solution was added, and
after 30 minutes, 6.0 mL of PEI solution was added. The final
solution was allowed to stabilize for 60 minutes before use. The
optimized PD-AgNPs were characterized for their optical absorption
spectra using a Vis-NIR spectrophotometer (Go Direct SpectroVis
Plus, Vernier, USA). Meanwhile, their surface morphology was
examined using a Field Emission Transmission Electron
Microscope (FE-TEM, JEM-2100F, JEOL, Japan).
B. Colorimetric Detection of Copper Ions
In a typical assay, 3.00 mL of the optimized PD-AgNPs
were mixed with 500 µL of Cu (II) solutions with varying
concentrations. Their corresponding absorption spectra were
measured using a Vis-NIR spectrophotometer. A calibration curve
was generated, and then statistical analyses were performed to
validate the obtained linear calibration curve. For the selectivity
studies, aqueous metal solutions containing 2.00 mM of Na+, K+,
Ba2+, Ca2+, Cu2+, Cd2+, Fe2+, Mg2+, Mn2+, Ni2+, Pb2+, Sn2+, Zn2+,
Al3+, Fe3+, and Cr6+ were used.
III. RESULTS AND DISCUSSION
A. Response Surface Modeling
A cubic polynomial model was fitted to develop a
correlation between the components of PD-AgNPs and the selected
responses, i.e., 𝛷 and LOD. The response contour plots for the
reduced cubic models for both 𝛷 and LOD are shown in Figure 1.
There is a high degree of curvature involved between the model's
parameters indicating the high degree of model dependence to the
interaction terms. Since the chemistry behind chemical crosslinking
is quite complicated and depends on the reactants' concentration, it
is expected that the interaction terms of the two models are
significant. Specifically, strong C=N bonds are formed between CS-
GLA and PEI-GLA during the GLA-aided chemical crosslinking
process produces a dense and rigid polymer decoration around the
surface of the silver nanoparticles. Numerical optimization was
performed based on the reduced cubic models to determine the
optimum composition of the PD-AgNPs that will provide the
maximum optical response (maximum 𝛷) and highest sensitivity
42

Figure 1. Response contour plots of PD-AgNPs showing: (a) effect
of CS and GLA (PEI = 1.75 wt.%) and (b) effect of CS and PEI
(GLA = 1.75 wt.%) on 𝛷; (c) effect of CS and GLA (PEI = 1.75
wt.%), (d) effect of CS and PEI (GLA = 1.75 wt.%) and (e) effect of
GLA and PEI (CS = 1.00 wt.%) on LOD.
(minimum LOD). The following formulation of PD-AgNPs is
deemed optimum: 1.1642 wt.% CS, 0.8298 wt.% GLA, and 1.3687
wt.% PEI. Our models predicted that the optimized PD-AgNPs has
𝛷 value of 82.5589 ± 0.6907 and LOD at 2.1566 ± 0.2243 mg L-1.
These values agree with our validation experiments (actual 𝛷 =
82.9540 ± 0.1772 and LOD = 2.2498 ± 0.0597 mg L-1) at a 95%
confidence level. This shows that our models are robust and can
generally be applied for predicting 𝛷 and LOD values of other
concentration combinations within the design range investigated.
The optimized PD-AgNPs show a deep-yellow color and are by an
optical spectrum exhibiting an LSPR peak at λ = 419.8 nm as shown
in Figure 2(a) and an average FWHM of 63.3641 ± 0.1134 nm,
indicating a narrow particle size distribution. The PD-AgNPs have
an average diameter of 15.6617 ± 1.3418 nm, which is typical of
pseudo-spherical silver nanoparticles [3].
B. PD-AgNPs as a Colorimetric Sensor for Cu (II) Detection
Figure 2. (a) Vis-NIR spectra of bare PD-AgNPs; (b) TEM image
of bare PD-AgNPs; (c) TEM image of PD-AgNPs after addition of
2.0 mM Cu2+; (d) Vis-NIR absorption spectra of the optimized PD-
AgNPs at varying Cu2+ concentration; (e) calibration curve for Cu2+
sensing using PD-AgNPs; (f) selectivity of PD-AgNPs against other
metal ions in water.
Upon addition of Cu (II) to PD-AgNPs, the AgNPs
changed in color from golden-yellow to green, as seen in Figure
2(a). This may be due to the complexation reaction between PEI and
Cu (II). This caused the aggregation of the individual PD-AgNPs, as
evidenced by the TEM images in Figure 2(b-c). Strong changes in
the optical absorbance spectra of the PD-AgNPs at varying Cu (II)
concentrations were observed in Figure 2(d). In particular, the
absorbance intensity of the peak at λ = 419.8 nm and λ = 617.7 nm
increases as the Cu (II) concentration increases. As such, the ratio
between these two peaks (617.7 nm / 419.8 nm) can be used as a
signal for detecting Cu (II) concentrations. A calibration curve was
generated based on the absorbance ratio and the Cu (II)
concentration and is shown in Figure 2(e). A strong linear
correlation was established between the absorbance and Cu (II)
concentrations within the range of 0.10 – 4.00 mM Cu (II). Our
generated calibration curve can be used for determining the
concentration of Cu (II) in an unknown water sample at a 95%
confidence level.
We also investigated the selectivity of the PD-AgNPs
against highly-valent cations such as Cr6+, Fe3+, and Al3+, divalent
cations such as Sn2+, Fe2+, Ni2+, Pb2+, Mn2+, Cd2+, Zn2+, Ca2+, Ba2+,
and Ca2+, and monovalent cations such as K+ and Na+. Common
metal cations present in various water bodies show only a tiny
response relative to Cu (II), as shown in Figure 2(e). As such, these
metals will not interfere with sensing Cu (II) ions in water.
IV. CONCLUSION
In this study, a reduced quadratic model was shown to fit
the experimental values accurately obtained using both the 𝛷 and
LOD as responses with 95% confidence. Our optimized nanosensor
composition is comprised of 1.1642 wt.% CS, 0.8298 wt.% GLA,
and 1.3687 wt.% PEI. The optimized PD-AgNPs are monodisperse
with an average diameter of 15.66 nm and peak absorbance at 419.8
nm. CS and PEI coordinate well with copper ions, which cause
distinct changes in the optical spectra of AgNPs. A strong linear
correlation was established between the absorbance ratio and the Cu
(II) concentrations within the range of 0.10 – 4.00 mM Cu (II) with
a detection limit of 2.2498 mg L-1. PD-AgNPs were found to be
selective towards copper compared to other metals commonly found
in water. Here we showed that our optimized PD-AgNPs could be
used for the rapid and facile colorimetric detection of Cu (II) in
water, thus making on-site copper detection a reality.
ACKNOWLEDGEMENT
The authors would like to acknowledge the Philippine
Council for Industry, Energy, and Emerging Research and
Technology for Development (DOST-PCIEERD), the Advanced
Materials Laboratory (ADMATEL), the University of the
Philippines Office of the Vice-Chancellor for Research and
Development (OVCRD), and the Engineering Research and
Development for Technology (ERDT) for funding this work.
REFERENCES
[1] B. O. Anyanwu, A. N. Ezejiofor, Z. N. Igweze, and O. E.
Orisakwe, "Heavy metal mixture exposure and effects in developing
nations: An update," Toxics, vol. 6, no. 4, p. 65, 2018.
[2] M. Mahmoudpour, J. Ezzati Nazhad Dolatabadi, M. Torbati, and
A. Homayouni-Rad, "Nanomaterials based surface plasmon
resonance signal enhancement for detection of environmental
pollutions," Biosens. Bioelectron., vol. 127, pp. 72–84, 2019.
[3] B. Khodashenas and H. R. Ghorbani, "Synthesis of silver
nanoparticles with different shapes," Arab. J. Chem., vol. 12, no. 8,
pp. 1823–1838, 2019.
[4] L. Tang and J. Li, "Plasmon-based colorimetric nanosensors for
ultrasensitive molecular diagnostics," ACS Sensors, vol. 2, no. 7, pp.
857–875, 2017.
43
 Geotechnical Characterization of Quarry Waste
Limestone as an Alternative Road Base Course
Material
Emerzon S. Torres1, a *and Mary Ann Q. Adajar2,b
1
Graduate Student, De La Salle University, Taft Avenue, Manila
2
Professor, De La Salle University, Taft Avenue, Manila
a
emerzon_torres@dlsu.edu.ph, bmary.ann.adajar@dlsu.edu.ph

Abstract — The rapid infrastructure development in the
Philippines over the past years poses problem on construction cost
due to the limited supply of suitable road foundation materials
within economic haul distances. An economic and sustainable
solution is to use locally available but suitable materials. This study
focused on geotechnical characterization of quarry waste Bulacan
limestone as an alternative road base course material. It includes
physical characterization using ASTM Standards, material
composition using petrographic analysis and SEM-EDX image
identification and bearing strength through California Bearing Ratio
(CBR) test. The results of the study showed that the quarry waste
Bulacan limestone is a potential road base course material and pilot
testing is highly recommended to perform.
Keywords—limestone, road base course, CBR, petrographic
analysis, SEM-EDX
I. INTRODUCTION
The "Build, Build, Build" program of the Department of
Public Works and Highways (DPWH) of the Philippines
promises a significant construction development. In the recent
Philippine Infrastructure Report of 2017, strong growth in the
construction and infrastructure industry was predicted to
occur over the next five years [1]. It follows that there is a
greater need for construction materials such as embankment
and road foundation aggregates. Searching for alternative
construction materials is considered a sustainable way of
development if secondary material is free from hazardous
components and does not have any adverse long-term effects.
Road construction costs can be very high because of the
limited opportunity of locating suitable road base material
within economic haul distances. It is preferable and
economical to use and to stabilize, if necessary, locally
available but abundant materials than to import conventional
aggregates [2].
The 2017 Minerals Yearbook shows that limestone quarry
production in 2014 reached 77, 665 thousand metric tons, and
it is the third top quarried industrial mineral commodity in the
country [3]. Its production rate doubled since 2010. The trend
in quarrying increased the production of limestone waste
material as well. The abundance of these crushed limestone
wastes creates an opportunity to use it as an embankment
material or aggregate replacement.
Sustainability calls on the use of alternative locally-
available materials and waste reduction in our quarry sites.
The utilization of waste limestone can be a big help in
providing more sustainable road construction materials. The
objective of this study is to characterize the quarry waste
limestone in a quarry site in Norzagaray, Bulacan, Philippines.
This study aims contribute to answering the problems of the
scarcity of locally-available construction materials and quarry
waste management and disposal.
II. METHODOLOGY
A. Source of quarry waste limestone
The quarry waste limestone was obtained from a quarry
site in Norzagaray, Bulacan as shown in the figure below. A
very wide range of particles was found in the site. Washing,
drying, and sieving were performed before the start of each
experiment. Manual crushing was done to meet the gradation
needed for some tests.
a. quarry site in Bulacan b. washed limestone
Figure 1. Limestone used in this study
B. Experimental program
The laboratory tests were performed on the samples. These
procedures were based from either ASTM or AASHTO
standards to establish a reference in evaluating the index
properties and strength of the alluvial soil. Morphology
includes the SEM-EDX (Scanning Electron Microscope –
Energy Dispersive X-ray) analyses, and petrographic analysis.
These material analyses supported the results of laboratory
tests and explained the soil behavior that is important for long-
term applications.
CBR tests in accordance with AASHTO T193 were
performed for the limestone aggregates compacted at the
maximum dry density and under soaked conditions to simulate
the worst possible scenario in the field. Moreover, specimens
were soaked for 96 hours (4 days). However, a shorter
immersion period of not less than 24 hours was applied, as
suggested by the AASHTO standards, if the shorter period did
not affect the test results.

44
 III. RESULTS AND DISCUSSION the petrographic analysis that calcite with chemical formula
CaCO3 comprises the majority material.
A. Geotechnical laboratory test results
Using the Grading A of the DPWH Blue Book [4] the
results of the experiments are summarized in Table 1. The
value of 34% shows a very strong resistance to abrasion of the
Norzagaray limestone. The range of values for the Los
Angeles abrasion value of Cebu limestone is 42 – 61% [2].
The ratio of particles passing No. 200 and particles passing
No. 40 or the dust ratio and the Atterberg limits were within (a) (b)
the range of values for road base course application. For the
Figure 2. SEM micrographs of alluvial soil finer than .075mm at (a) x5000
bearing strength, the CBR values of 225 and 340 for 2.54 and (5𝜇), and (b) x7500 (1𝜇) magnification.
5.08mm penetrations surpassed the stipulated 80% soaked
CBR value for road base course materials.
IV. CONCLUSION
The strong resistance of limestone aggregates to bearing
After laboratory tests and analyses on the suitability of
are attributed to the angular shape of the particles, the
Norzagaray quarry waste limestone as a road base course
cementitious property of limestone, and its high resistance to
material were performed, the following conclusions were
abrasion. The results of the morphology and petrography of
made:
the limestone explain this further.
1. the coarse aggregates of limestone has an abrasion value
TABLE I. MATERIAL CONSTANTS
of 34% and the fine aggregates has no plasticity;

DPWH Specs 2. the bearing strength in terms of CBR are 225% and
Soil Property This study
Grading A or B
340% for 2.54mm and 5.08mm penetrations, respectively;
Particle Size Distribution Grading A
Abrasion Value Not greater than 50% 34% 3. the limestone is a biomicrite with 65% of
Dust ratio ≤ 66% 0.5% microcrystalline calcite matrix; the results of SEM-EDX and
Apparent Specific Gravity - petrographic analysis supported the positive findings of the
(coarse) 2.6609 geotechnical laboratory tests;
Specific Gravity (fine) - 2.5375
Liquid Limit, LL % Not greater than 25 Non-plastic 4. the Norzagaray quarry waste limestone is a suitable
Plastic Limit, PL % Not greater than 6 Non-plastic roadbase course material as per the DPWH specifications.
Min. Void Ratio, emin - 0.28806
Max. Void Ratio, emax - 0.51798 Pilot testing is warranted to determine the long-term
Max. Dry Unit Weight,γdmax - performance of the road base course limestone aggregates. It
kN/m3 20.25386 is also recommended to observe the weathering and possible
Optimum Moisture Content, - dissolution that may happen when calcium carbonate reacts
ωopt % 7.07014 with carbon dioxide and water. The effects on the wear of tires
CBR at 2.54mm penetration, % 80 225.29
is suggested for the application of limestone to pavements.
CBR at 5.08mm penetration, % 80 340.39

ACKNOWLEDGMENT
B. SEM-EDX and Petrographic analysis The authors would like to acknowledge the support of
Petrographic analysis was performed in this study to DOST-ERDT for providing financial assistance for this
determine the chemical and physical characteristics of a rock research, the laboratory testing assistance of DPWH-BRS, and
sample. It was employed to identify the rock name as well. the help of Mr. Michael Torres for the material acquisition.
The analysis was also conducted by the geologists of the Also, the study would not be possible without the help of the
DENR-MGB in their geological laboratory services section. DLSU Civil Engineering Faculty.
The limestone is composed of 30% carbonate fragments,
REFERENCES
65% of microcrystalline calcite, and 5% microsparite which
acts as a cement. It was named biomicrite after Folk’s
Limestone Classification [5]. This limestone composition [1] Philippines Infrastructure Report Q3, 2017,
supports the positive results of the laboratory tests on abrasion https://www.giiresearch.com/report/bmi177867-
infrastructure.html.
and CBR.
[2] DPWH, “Use of weathered limestone (Anapog) as blending
The micrographs of the limestone sample in Fig. 2 materials for base course (Item 201). Manila: Department of
revealed clues that it is composed of grains of different shapes Public Works and Highways”, 1991.
and sizes from different materials. Highly visible are the [3] Fong-Sam, Y., “The mineral industry of the Philippines.
Minerals Yearbook 2013: Area Reports: International, Asia
particles that are intact glued together in the matrix. and the Pacific, 2”, 2017.
The EDX revealed that the limestone consists of 52.96% [4] Department of Public Works and Highways, “Standard
oxygen, 38.79% carbon, and 7.21% calcium with traces of specification for Highways, Bridges and Airports”, 2013.
0.62% silicon and 0.42% magnesium. This is consistent with [5] Folk, R.L., “Stages of textural maturity in sedimentary rocks.
Journal of Sedimentary Research”, 21 (3), 127 – 130, 1951.

45
A nonlinear optimization model for subsidy allocation
to disrupted infrastructures and its application in the
Philippines during the COVID-19 pandemic
Egberto F. Selerio Jr.1, a * Renan S. Maglasang1, b
1
Department of Industrial Engineering, University of San Carlos, Talamban, Cebu City
a
20102931@usc.edu.ph
b
rmaglasang@usc.edu.ph
Abstract— There have been several extensions to the
classical Leontief Input-Output model applied in infrastructure risk
research. Among the said extensions is the Inventory Dynamic
Inoperability Input-Output Model (I-DIIM). The I-DIIM illustrated
the importance of inventory to the preparedness and resilience of
infrastructures in a production environment, e.g., a macroeconomic
system, to disruptive events. Although the economic implications of
such an event to a macroeconomic system are primarily the
government's concern, no work has yet attempted to provide a way
to empirically determine the optimal intervention required from
governments to facilitate the preparedness and resilience of
economic sectors. This work addresses this significant gap in the
literature by developing a Subsidy Optimization Model (SOM)
based on the I-DIIM. This work also proposes a simulation-based
subsidy optimization procedure with the Generalized Reduced
Gradient (GRG) with a Multi-Start algorithm for solving the SOM.
The SOM's applicability is tested through an illustrative case study
in the Philippines during the COVID-19 pandemic. This work
revealed that the SOM differs from conventional methods for
allocating divisible resources at low budgets. The analysis also
highlighted the SOM's ability to exploit and leverage economic
sectors' interdependency to develop an allocation scheme that
minimizes total economic loss, which would otherwise be an
incredibly difficult task to fulfill without it. The robustness of such
results is tested through a sensitivity analysis. Some practical
implications of the findings in the Philippines are detailed in this
work.
Keywords— Infrastructure Risk Analysis, Input-Output
Economics, Optimization, COVID-19, Disaster Preparedness
I. INTRODUCTION
A. The Role of Inventory in Infrastructure Risk Management
The focus of infrastructure risk management and decision-
making has recently shifted from preventing and protecting
infrastructures from calamities to recovery and response [1].
Inventory availability contributes significantly to infrastructures'
preparedness and resilience [1-3]. The efficacy of inventory as a
mitigation strategy to delay the impact of disruptive events has been
widely explored in previous literature, e.g., [1], [3-4]. For instance,
[3] showed that an optimal sectoral inventory configuration could
minimize economic loss for a given disruption scenario. Some
works also explored the impact of inventory in real cases. For
example, [4] illustrated the importance of inventory in reducing
Philippine economic sectors' inoperability during the typhoon
Yolanda (internationally known as Haiyan). These works provided
strong empirical results supporting the need to configure inventory
The present work was part of the master's thesis of the author funded by Engineering Research and Development for Technology (ERDT) scholarship of the Department of Science and Technology,
Republic of the Philippines.
1
levels to increase infrastructure resilience and speed up their
recovery from disruptions – leading to a reduction in economic loss.
While the macroeconomic contributions of inventory have
been widely recognized in recent literature, it should be noted that
inventory is something to be minimized for for-profit institutions
rather than promote because of its associated costs. Furthermore, in
regular times, the likelihood of calamities (e.g., pandemic) to occur
is very low, which further demotivates businesses to keep inventory
just to prepare for such unlikely events. However, the significance
of inventory may be of more value to decision-makers concerned
with the national economy. Since a government aims to maintain a
sustainable national economy, it is, therefore, in its best interest to
support economic sectors in crisis. While governments' role in
supporting enterprises during crises is an obvious fact, no attempt
has yet been made in previous literature to explicitly measure the
optimal intervention required from governments to minimize
economic loss consequent to calamities. Furthermore, such analysis
is also not yet explored while considering the complex
interdependencies of these sectors. Such analysis is vital in making
these decisions because changes made in one of the macroeconomic
system's elements, may it be through some policy, may cause a
discreet yet critical impact to the entire system because of their
interconnectedness.
A financial subsidy is among the most common
government intervention to support economic sectors in crisis [5].
The financial subsidy helps the sectors sustain their operations while
in a disrupted state. In line with this, it can be assumed that financial
subsidy converts into some inventory upon allocating these sectors.
Inventory is defined herewith as an asset that can be kept and used
to maintain a sector's ability to provide goods or services to other
infrastructures, sectors, and final consumers while it is in a disrupted
state. For instance, inventory in the oil and gas sector could include
excess stored gasoline ready for use [2]. In the electric power sector,
backup generators could provide some total output to be maintained
[1]. Note that these sectors are interdependent – a sectors' output is
an input to another – which has several economic implications [4],
as previously discussed. The intrinsic interdependencies among
these sectors can trigger the propagation of disaster consequences
that often result in a broader range of inoperability and amplified
losses [3]. Thus, assessing these economic sectors' ability to
overcome shocks is an essential factor in allocating financial subsidy
to configure inventory levels and optimize their preparedness for
critical events; to reduce damage, ensure resilience, and mitigate
monetary losses [6-7].
46
 A critical tool that can be employed to address the problem [b.1.] Integration of local heuristic model with a stochastic
espoused in preceding discussions is the input-output (I-O) model – model for computing a probable global solution of the
an approach proposed by Wassily Leontief, which earned him the model.
Nobel Prize in Economic Sciences in 1973. The I-O model accounts
for sector interdependencies governing the magnitude and extent of [c] To investigate the effectiveness and sensitivity of the global
ripple effects due to changes in the economic structure of a region solution and develop policy insights for the Philippine government
or nation [8]. Recent extensions to I-O modeling have enhanced the through the following:
model's capabilities to account for the impact of an economic
perturbation, e.g., the inoperability I-O model [1], [9]. These models [c.1.] Testing the robustness of the global optimal solution
introduced sector inoperability, or the inability to satisfy as-planned with the changes in the budgetary and holding capacity
production levels, into I-O modeling [8]. Among these extensions, a parameters;
notable improvement was recently proposed by [1], which
incorporates inventory in a dynamic inoperability I-O model. Their
[c.2.] Comparing the nonlinear optimization model results
model, the Inventory Dynamic Inoperability I-O Model (I-DIIM),
with the conventional resource allocation approaches
has been illustrated in various applications, e.g., [3-4]. However,
using the data from Philippine I-O tables.
several aspects of the current formulation do not account for
complexities associated with inventory capacity and budget for
inventory reconfiguration. While inventory may promote resilience
against calamities, it is limited by economic sectors' capacity to keep I. METHODOLOGY
it and governments' ability to intervene in its configuration. For
brevity, the details of the I-O model and its recent extensions leading A. Model Development
to the I-DIIM are presented in Appendices A-D.
To address the gaps espoused in previous discussions, this
B. The Research Gaps work extends the I-DIIM through a Subsidy Optimization Model
(SOM) – a nonlinear programming model based on the I-DIIM that
An issue with the present formulation of the I-DIIM lies determines the optimal allocation of subsidy to minimize calamity-
in its inability to suggest initial inventory levels, minimize induced total economic loss. The formal definition of subsidy is as
inoperability, and, in turn, economic loss. This analysis is essential follows:
in the decision to appropriate resources for the reconfiguration of
initial inventory levels as a mitigation strategy to increase economic (0)
sectors' resilience. Galbusera et al. [3] proposed an optimization Definition 1. (Subsidy) Let 𝑠̃𝑖 denote the existing
model to supplement the need for such an analysis; however, their inventory levels of sector 𝑖 at 𝑡 = 0. The interest of this work lies in
(0)
model is impractical in several aspects. For instance, their analysis reconfiguring the size of these inventories through a subsidy, 𝑠̂𝑖 ,
is built on the assumption that economic sectors are willing to invest that comes from the government or some other financing body
in the reconfiguration of their inventories in preparation for a concerned with the macroeconomic consequences of a calamity.
calamity (e.g., pandemic). However, organizations hardly ever (0)
This idea builds on the assumption that 𝑠̂𝑖 is directly converted into
prepare for such calamities because they rarely occur. Unlike these
some sort of inventory that is readily convertible to revenue upon its
organizations, a central regulating body like the government is
allocation to sector 𝑖. In line with this, the initial inventory level
concerned with this problem. Government intervention to aid (0)
economic sectors is common during calamities because it is their 𝑠𝑖 of sector 𝑖 can be represented as the sum of its existing inventory
prime goal to maintain a stable national economy. It commonly does level and corresponding government subsidy as illustrated in (1).
so by appropriating limited financial subsidies to economic sectors.
While this is an obvious fact, no work has yet been made to (0) (0) (0)
𝑠𝑖 = 𝑠̃𝑖 + 𝑠̂𝑖 , ∀𝑖 ∈ ℕ (1)
empirically determine the optimal allocation of such subsidies to
disrupted economic sectors as a divisible resource. Assuming that (0) (0)
the financial subsidy is converted into some sort of inventory upon where 𝑠̃𝑖 , 𝑠̂𝑖 ∈ ℝ+ .
allocation to the sectors, the problem is now shifted on the allocation
of subsidy to configure existing inventory levels, reduce The summary of parameters, objective function,
inoperability, and minimize the overall economic consequence of a constraints, and the SOM formulation are presented in the following
calamity. In addressing the main problem espoused in the preceding sections.
discussions, the following objectives are outlined in this work:
Summary of Parameters
[a] To develop a nonlinear optimization model for the optimal
allocation of subsidy to configure the inventory levels of ℕ, set of economic sectors
disrupted, interdependent economic sectors through the following: 𝑛, number of economic sectors
ℎ, number of time-periods
[a.1.] Identification of parameters for the model that 𝛽𝑖 , inventory capacity of sector 𝑖
sufficiently reflect a real macroeconomic system; 𝛼, total subsidy budget
(0)
𝑠̃𝑖 , existing inventory level of sector 𝑖
[a.2.] Formulation of relations to distinguish budgetary (𝑡)
𝑠𝑖 , inventory level after subsidy of sector 𝑖 at time 𝑡
and holding capacity constraints while integrating them in (𝑡)
the model and integrating the I-DIIM in the model as a 𝑝𝑖 , production inoperability of sector 𝑖 at time 𝑡
(𝑡)
component of the objective function. 𝑥𝑖 , as-planned production of sector 𝑖 at time 𝑡
∗,(𝑡)
𝑐𝑖 , demand perturbation of sector 𝑖 at time 𝑡
[b] To develop an approach for solving the global solution of the 𝑘𝑖 , resilience coefficient of sector 𝑖
nonlinear optimization model through the following: ∗
𝑎𝑖𝑗 , interdependency coefficient of sector 𝑖

47
𝑡, time or period ∑𝑛𝑖=1 𝑠̂𝑖(0) ≤ 𝛼 (7)
(𝑡)
𝑞𝑖 , general inoperability of sector 𝑖 at time 𝑡
(0)
𝑠̂𝑖 , subsidy of sector 𝑖 The Subsidy Optimization Model

Objective Function By combining (4)-(7), the I-DIIM-based SOM is

illustrated as follows:
Definition 12. (Economic loss) Let 𝑙𝑖 be the economic loss
of sector 𝑖 across all periods of the I-DIIM simulation. It is expressed Minimize
as a function of the sector's inoperability and as-planned production
output. If inoperability at 𝑡>0 is denoted as (0) (0) (0)
𝐸𝐿(𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 )
(𝑡) (𝑡) (𝑡) ∗,(𝑡) ∗ 𝑛
𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 , 𝑐𝑖 , 𝑘𝑖 , 𝑎𝑖𝑗 ), and inoperability at 𝑡 = 0 is
(0) (0) (0) (0)
(0) (0) (0) = ∑ 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ) 𝑥𝑖
denoted as 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ), then 𝑙𝑖 is expressed as follows:
𝑖=1
𝑛 ℎ
(𝑡) (𝑡) (𝑡) ∗,(𝑡) ∗ (𝑡)
(0)
𝑙𝑖 = 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ) 𝑥𝑖
(0) (0) (0)
+ + ∑ ∑ 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 , 𝑐𝑖 , 𝑘𝑖 , 𝑎𝑖𝑗 ) 𝑥𝑖
𝑖=1 𝑡=1
∑ℎ𝑡=1 𝑞𝑖 (𝑠𝑖(𝑡) , 𝑝𝑖(𝑡) , 𝑥𝑖(𝑡) , 𝑐𝑖∗,(𝑡) , 𝑘𝑖 , 𝑎𝑖𝑗
∗ (𝑡)
) 𝑥𝑖 (2)
Subject to:
where 𝑖, 𝑗 ∈ ℕ, 𝑡 = {0,1, … , ℎ}, and ℎ is the last period of (8)
the I-DIIM simulation. The derivation of
(𝑡) (𝑡) (𝑡) ∗,(𝑡) (0) (0) (0) 𝑛
∗
𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 , 𝑐𝑖 , 𝑘𝑖 , 𝑎𝑖𝑗 ) and 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ) are (0)
∑ 𝑠̂𝑖 ≤𝛼
presented in Appendix D. 𝑖=1

Definition 13. (Total economic loss) Let (0) (0)

(0) (0) (0) 0 ≤ 𝑠̃𝑖 + 𝑠̂𝑖 ≤ 𝛽𝑖 , ∀𝑖 ∈ ℕ
𝐸𝐿(𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 ) be the total economic loss dependent on the
(0) (0) (0)
subsidy to the economic sectors 𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 . The methods employed for solving (8) are presented in
(0) (0) (0)
𝐸𝐿(𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 ) is obtained by summing the losses of the following sections.
economic sectors as follows:
B. The Solver: Generalized Reduced Gradient with Multi-Start
(0) (0) (0)
𝐸𝐿(𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 ) = ∑𝑛𝑖=1 𝑙𝑖 , (3) The Generalized Reduced Gradient (GRG) algorithm is
employed to solve (8). GRG algorithms stemmed from the idea of
where 𝑖 ∈ ℕ, 𝑡 = {0,1, … , ℎ}, and ℎ is the last period of constrained derivatives [10]. It was later developed using the name
the I-DIIM simulation. reduced gradient method [11] and finally extended through
generalized reduced gradient [12]. The GRG algorithm is a local
By substituting (2) to (3), the objective function to solver, i.e., GRG's solution may only be locally optimal. To address
minimize is as follows: this inadequacy, the Multi-Start search algorithm is applied on top
of the GRG algorithm to ensure that the final optimal solution is as
(0) (0) (0) (0) (0) (0) (0) near as possible to the global solution. The computational
𝐸𝐿(𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 ) = ∑𝑛𝑖=1 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ) 𝑥𝑖 + procedures of the GRG method are detailed as follows:
∑𝑛𝑖=1 ∑ℎ𝑡=1 𝑞𝑖 (𝑠𝑖(𝑡) , 𝑝𝑖(𝑡) , 𝑥𝑖(𝑡) , 𝑐𝑖∗,(𝑡) , 𝑘𝑖 , 𝑎𝑖𝑗
∗ (𝑡)
) 𝑥𝑖 (4)
In (8), the number of independent variables (decision
(0) (0) (0)
Constraints variables) is 𝑛, which are the subsidies 𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 for each
economic sector. The independent variables are broken into basic
Definition 14. (Holding capacity) Let 𝛽𝑖 denote the variables, 𝑚, and non-basic variables (𝑛 − 𝑚). The 𝑚 constraint
holding capacity of sector 𝑖 that characterizes the amount of equations can be solved for 𝑚 basic variables in terms of the (𝑛 −
(0) 𝑚) non-basic variables. The basic variables are then substituted in
inventory the sector can keep; thus, 𝑠𝑖 ∈ [0, 𝛽𝑖 ]. The parameter 𝛽𝑖 the objective function such that the original objective function is
may be exogenously defined based on estimates or computed as only the function of (𝑛 − 𝑚) non-basic variables. Building on this,
some multiple of the sector's as-planned production. Building on this the following definitions detail the computational procedures of the
idea, the equivalence relation in (1) can be relaxed as a set of GRG algorithm:
inequalities as follows:

(0) (0)
Definition 16. Let vector 𝒘 denote the non-basic variables
𝑠̃𝑖 + 𝑠̂𝑖 ≥ 0, ∀𝑖 ∈ ℕ (5) and vector 𝒚 denote basic variables, which is a function of the basic
variables, i.e., 𝒚(𝒘). Then the GRG algorithm is illustrated as:
(0) (0)
𝑠̃𝑖 + 𝑠̂𝑖 ≤ 𝛽𝑖 , ∀𝑖 ∈ ℕ (6)
𝒁(𝒘) = 𝒛(𝒚(𝒘), 𝒘) (9)
Definition 15. (Total subsidy budget) Let 𝛼 denote the
total budget for the subsidy. This parameter characterizes the 𝑮(𝒘) = 𝒈(𝒚(𝒘), 𝒘) = 0 (10)
government's limited capacity to cushion the consequences of a
calamity through financial aid, e.g., subsidy, to its economic sectors. now,
This parameter is illustrated as follows:

48
 𝑑𝑮 𝜕𝒈 𝜕𝒚 𝜕𝒈 Step 6: (Determine the inventory parameters) Determine
=0= + (11)
𝑑𝒘 𝜕𝒚 𝜕𝒘 𝜕𝒘 (0)
the existing inventory, 𝑠̃𝑖 , and inventory capacity, 𝛽𝑖 , for all
𝑑𝒁 𝜕𝒛 𝜕𝒛 𝜕𝒚 sectors.
= + (12)
𝑑𝒘 𝜕𝒘 𝜕𝒚 𝜕𝒘
Step 7: (Determine the subsidy parameters and set the
then from (11) and (12) the following is obtained: subsidy to zero for all sectors) Determine the subsidy budget, 𝛼.
(0)
Also, set the subsidy 𝑠̂𝑖 = 0 for all sectors and decide on the last
𝑑𝒁 𝜕𝒛 𝜕𝒛 𝜕𝒈 −1 𝜕𝒈 time-period, ℎ, of the simulation.
= − ( ) (13)
𝑑𝒘 𝜕𝒘 𝜕𝒚 𝜕𝒚 𝜕𝒘

Step 8: (Determine the initial inoperability) Determine the

𝜕𝒛 𝜕𝒈 −1 (0)
where the term ( ) in (13) is the Khun-Tucker initial inoperability, 𝑞𝑖 , for all sectors using the procedures in
𝜕𝒚 𝜕𝒚
multiplier vector [13]. The nonlinear optimization problem can be Appendix D.
transformed into a reduced problem which is unconstrained in nature
with only the upper bound (𝑢𝒘 ) and the lower bound (𝑙𝒘 ) as follows: Step 9: (Calculate the inoperability for the rest of the time-
(𝑡)
periods) Calculate the inoperability 𝑞𝑖 for the rest of the time-
Minimize 𝒁(𝒘) periods using the procedures in ppendix D.

Subject to: 𝑙𝒘 ≤ 𝒘 ≤ 𝑢𝒘 (14)
For brevity, solving the reduced problem of (14) is not
explained in this work. It is mostly based on the principles and logic
developed by [13] in conjunction with the Multi-Start search routine.
The GRG with Multi-Start is executed in this work using Visual
Basic for Applications in Microsoft Excel. The execution of the
GRG with Multi-Start is detailed in the following discussions.
C. The Subsidy Optimization Procedure
Before solving the SOM, the I-DIIM is simulated for ℎ
time-periods. Also, because the model is solved multiple times using
(0)
the GRG with Multi-start, 𝑠̂𝑖 = 0 is initially set for each solution.
The following subsidy optimization procedures were carried out to
solve the SOM:
Step 1: (Identify the system elements) Identify the
elements or sectors of the interdependent system. This system may
be in any size and form, encompassing that of an interdependent
macroeconomic system.
Step 2: (Obtain the I-O matrix) Obtain the transactional
data, 𝑥𝑖𝑗 , of the sectors organized as an I-O matrix. The data in the
I-O matrix must include total input, 𝑥𝑗 , total output, 𝑥𝑖 , and total
final demand, 𝑐𝑖 . Note that the as-planned final demand, 𝑐̂𝑖 , may be
estimated from the total final demand, 𝑐𝑖 , of the I-O matrix.
Step 3: (Determine the calamity impact parameters)
Determine the calamity situation and its impact on the sectors,
∗,(𝑡)
including the degraded demand, 𝑐̃𝑖 , demand perturbation, 𝑐𝑖 , at
(0)
period 𝑡 and the initial production inoperability, 𝑝𝑖 .
Step 4: (Calculate the interdependency parameters)
Calculate the Leontief technical coefficient of all sectors, 𝑎𝑖𝑗 ,
∗ (ECQ) – the strictest quarantine level in the country – is lifted [16].
interdependency coefficients, 𝑎𝑖𝑗 , using the procedures in
Appendices A and B.
Step 5: (Calculate the recovery and resilience
parameters) Calculate the recovery period, 𝑓𝑖 , and the production
(𝑓 )
inoperability, 𝑝𝑖 𝑖 , at the recovery period, 𝑓𝑖 , for all sectors. Also,
obtain the resilience coefficient, 𝑘𝑖 , for all sectors using the
procedures in Appendices C and D.
Step 10: (Solve the SOM using the GRG with Muti-Start)
Solve (8) using the GRG with Multi-Start. In this routine, a set of
100 different feasible starting points are generated at first. The GRG
algorithm is then applied to each of these feasible starting points turn
by turn using (9)-(14), and the local optimal solutions for each
starting point are stored. Finally, when all 100 different starting
points have been exploited, the solution that minimizes the objective
function is the best optimal solution. This optimal solution,
∗,(0) ∗,(0) ∗,(0)
𝐸𝐿∗ (𝑠̂1 , 𝑠̂2 , … , 𝑠̂𝑛 ), is considered the closest to the global
solution.
Steps 1-7 details the parameters' identification and
calculation, while Steps 8-9 details the procedures for executing the
I-DIIM. Lastly, Step 10 details the procedure for solving the SOM
using the GRG with Multi-Start. To illustrate the proposed SOM's
applicability, a case study in the Philippines during the COVID-19
pandemic is employed. The following sections detail the case study
background, datasets used in the case study, and the conventional
allocation of subsidies versus the allocation of subsidies based on
the SOM results.
D. Case Study Background: The Philippine Economy during the
COVID-19 Pandemic
Since the early stages of the COVID-19 pandemic, the
Philippine government has commenced measures to curb the spread
of the virus, which include travel bans, various levels of community
quarantines, and the T3 (test, trace, and treat) program [14], [15]. On
March 25, 2020, the Republic Act No. 11469 or the Bayanihan to
Heal as One Act (BHOA) was enacted to implement these measures
fully. A few months after its enactment, the Department of Health
(DOH) announced on May 05, 2020, in a press release that the
country is starting to achieve the "flattening of the COVID-19
curve" [16]. In the same report, the DOH warned on the possible
resurgence of cases if the people become complacent about social
distancing, mostly when the "enhanced community quarantine"
As predicted, upon the easing of lockdown protocols on June 1,
2020, the country had seen infections rising again, topping 2,000 in
recent days from about 500 when the lockdown was in effect [17].
The resurgence of cases disrupted government efforts to restart the
economy. It also prompted the government to place the country's
leading economic regions (e.g., Manila and Cebu) under strict
lockdown protocols multiple times from June to August 2020.
Despite the continuous surge of new COVID-19 cases, the country's
lockdown protocols are generally eased by the government since the
middle of August 2020, in an attempt to kick-start economic
recovery [18]. As of September 28, 2020, the country has 304,226

49
reported COVID-19 cases, ranking 8th in Asia and 21st worldwide in protocol in the Philippines. The datasets for 𝑨, 𝒙(𝑡) , and 𝒑(0) are as
terms of the highest reported cases [19]. follows:

The Philippine government's attempts at containing 𝑨 = [𝑎𝑖𝑗 ]

𝑛×𝑛
COVID-19 through quarantine and lockdowns have caused 0.12 0.01 0.12 0.03 0.00 0.01 0.02 0.00 0.02 0.01 0.03
0.00 0.00 0.05 0.05 0.02 0.00 0.00 0.00 0.00 0.00 0.00
significant disruptions in its economic activities. The private sector
0.06 0.05 0.28 0.31 0.10 0.16 0.14 0.03 0.04 0.06 0.14
investment, which drove economic growth in the past, has 0.00 0.03 0.01 0.00 0.01 0.01 0.01 0.00 0.01 0.01 0.01
significantly declined given the closure of businesses and income 0.00 0.01 0.02 0.00 0.10 0.03 0.03 0.01 0.01 0.01 0.04
loss during the lockdown [20]. The Philippine Statistics Authority = 0.01 0.01 0.03 0.01 0.00 0.08 0.03 0.02 0.01 0.04 0.02
0.07 0.09 0.11 0.11 0.06 0.11 0.15 0.08 0.11 0.09 0.12
(PSA) [21] reported that "Gross Domestic Product (GDP) growth 0.01 0.01 0.02 0.02 0.01 0.06 0.06 0.21 0.02 0.05 0.04
rate dropped by 16.5% in the second quarter of 2020." This quarterly 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
decline in GDP is the lowest recorded in 29 years. The Philippine 0.01 0.01 0.01 0.02 0.00 0.03 0.03 0.03 0.03 0.07 0.12
[0.02 0.03 0.03 0.02 0.01 0.03 0.01 0.03 0.01 0.06 0.02]
economic sectors that lead the decline were manufacturing (-21.3%),
construction (-33.5%), and transportation and storage (-59.2%) [21].
(𝑡)
In general, industry and services declined by 22.9% and 15.8%, 𝒙(𝑡) = [𝑥𝑖 ]
1×𝑛
respectively [21]. As to employment, due to the closure of several = [17.88 1.53 54.11 8.55 6.79 11.07 26.38 12.12 6.06 11.52 24.65]
establishments along with the COVID-19 pandemic, total
(0)
employment reduction was estimated to be between 0.82% to 1.69% 𝒑(𝑡) = [𝑝𝑖 ]
1×𝑛
[22]. In an attempt to alleviate the economy and mitigate the = [0.79 0.01 0.87 0.88 0.73 0.79 0.88 0.81 0.83 0.93 0.13]
pandemic's effects on the heavily hit sectors, an economic stimulus
bill – the House Bill 6815 – was proposed by the House of The value of the demand-side perturbation vector 𝒄∗,(𝑡) is
Representatives [23]. The bill seeks to protect and assist over 15.7 obtained based on the following rationalization: The prominent
million workers, create 3 million short-term jobs and 1.5 million disruption during the imposition of the 'enhanced community
infrastructure jobs, and extend help up to 5.57 million micro, small quarantine' in the Philippines originates mainly from the idea of
and medium enterprises [23]. This work applies the proposed SOM "keeping physical interaction at a minimum," which resulted in
to the Philippine case to provide a basis for developing future mandatory travel restrictions, close-down of some establishments
policies to increase the resilience of economic sectors and guide (e.g., restaurants), workforce reduction, among others, to comply
decision-makers in allocating subsidies in the proposed stimulus with distancing protocols. While this disruption affects the physical
bill. production of the economic sectors, it does not necessarily impact
demand. On these grounds, the analysis of this work is simplified by
E. Datasets fixing the value for the demand-side perturbation vector 𝒄∗,(𝑡) = 0.
The dataset for the Leontief technical coefficient matrix 𝑨 The recovery period 𝑓𝑖 was assumed to be 𝑓1 = 𝑓2 = ⋯ = 𝑓𝑛 = 24
and as-planned production vector 𝒙(𝑡) are obtained from the months (two years) based on economic recovery estimates [26]. The
(𝑓 )
periodically-released Philippine I-O Tables [4], [24]. There were physical inoperability 𝑝𝑖 𝑖 at the recovery period is also assumed to
eleven (𝑛 = 11) economic sectors identified from the Philippine I- (0)
be 0.01 for all sectors. The values for the existing inventory 𝑠̃𝑖 are
O Tables (see Table I). assumed to be 20% of as-planned production vector, 𝒙(𝑡) , for all
sectors. Furthermore, assuming that the inventory capacity, 𝛽𝑖 , of
TABLE I. ECONOMIC SECTORS OF THE PHILIPPINES the economic sectors is some multiple 𝜔 of their monthly as-planed
(𝑡) (𝑡)
output 𝑥𝑖 , hence 𝛽𝑖 (𝜔) = 𝜔𝑥𝑖 , this work assumes 𝜔 = 3.0.
Serial Sector Lastly, the subsidy budget 𝛼 = 197.0 billion Philippine pesos
Economic Sectors
Code Category (BPhp) (4.1 billion U.S. dollars) following House Bill 6815 – the
L1 Manufacturing Large bill on COVID-19 economic stimulus and subsidy to economic
L2 Trade Large sectors in the Philippines [23].
L3 Government Services Large
S4 Agriculture, Fishery, and Forestry Small
S5 Finance Small
S6 Private Services Small
III. RESULTS AND DISCUSSION
Transportation, Communication, and Small
S7 A. Baseline Results
Storage
S8 Construction Small The Unconventional Subsidy Allocation of the SOM
S9 Electricity, Gas, and Water Small
Real Estate and Ownership of Small
S10 Although the recovery period, 𝑓𝑖 , is assumed to be 24
Dwellings
months for all economic sectors; steps 1-10 are performed using the
S11 Mining and Quarrying Small datasets for 30 periods – 6 periods more than the assumed recovery
period – to consider residual inoperability that may occur as a
The sectors were categorized based on their as-planned consequence of the complex interdependencies of the sectors. The
production vector 𝒙(𝑡) . Those that comprise more than 50.0% of the results of the SOM are presented in Fig. 1.
system's total production output were categorized as "large sectors,"
while the rest were categorized as "small sectors." The monthly
value of the as-planned production vector 𝒙(𝑡) is assumed equal for
every time 𝑡. The values for the production inoperability vector 𝒑(0)
were based on the reduction in physical operations of the sectors [25]
during the 'enhanced community quarantine' – the strictest lockdown

50
 60.0 600.0
non-allocation of subsidy to the large sectors is that the SOM has
exploited the economic sectors' independencies to prioritize the
50.0 500.0 small sectors. As a result of such allocation, the inoperability of the

Economic loss per sector (in BPhp)

Subsidy per sector (in BPhp)

40.0 400.0
small sectors is delayed and reduced overtime; because they
influence the large sectors, as a consequence, the inoperability of the
30.0 300.0 large sectors is reduced as well – leading to the minimization of the
20.0 200.0
total economic loss. Therefore, this work demonstrates that to
reduce total economic loss, prioritizing the large sectors in the
10.0 100.0
allocation of subsidy, like conventional methods, is not necessarily
0.0 0.0 an optimal strategy. In line with this, it is worthwhile to investigate
L1 L2 L3 S4 S5 S6 S7 S8 S9 S10 S11
Subsidy 0.0 0.0 1.3 50.1 33.9 32.2 31.0 8.2 19.0 17.0 4.3
when would such an unconventional subsidy allocation remains an
Economic Loss 569.6 189.9 76.9 41.2 30.7 34.4 27.8 38.4 11.5 12.6 0.0 optimal strategy. In line with this inquiry, the following section
investigates the sensitivity of the optimal solution with changes in
Subsidy Economic Loss
the subsidy budget, 𝛼, and inventory capacity, 𝛽𝑖 .

Fig. 1. The optimal solution of the SOM

B. Sensitivity Analysis
Case 1: Changing the Subsidy Budget
The small sectors (S4-S11) were heavily subsidized, while
the large sectors (L1-L3) were the least subsidized. The large sectors The first case for sensitivity analysis is as follows:
are the largest contributors to the minimized total economic loss 𝐸𝐿∗ Suppose the Philippine government intends to determine the
(81.0%, 836.4/1,033.0 BPhp). As suggested by the SOM, the appropriate subsidy budget for the economic sectors. It is essential
allocation of subsidy deviates from the conventional resource to determine how the optimal solution changes with the subsidy
allocation method (i.e., proportional allocation) because the large budget for such a decision. In this regard, the SOM is solved for
sectors, which produce the majority of the system's production multiple values of the subsidy budget at a constant inventory
output (58.2%, 105.1/180.7 BPhp), were least prioritized. capacity multiplier, 𝜔 = 3.0. The results of this analysis are
presented in Fig. 2.
Conventionally, large sectors are prioritized in allocating
relief resources (e.g., subsidy) because of their large economic 5
1
1.051.1 1.15
4.95
4.85 4.9 80 1.2
value. When the large sectors become inoperable, they incur a huge 4.75
4.7
4.8 1.25
1.3
1.35
4.65 70 1.4
economic loss due to their large production output. Therefore, it is 4.6
60
1.45
4.55 1.5
imperative to prioritize the reduction of their inoperability. Subsidy L1
4.45
4.5
50
1.55
1.6
4.4 1.65
facilitates such a reduction in inoperability; however, the subsidy L2
4.35 40 1.7

budget, 𝛼, (197.0 BPhp) is relatively smaller than what is needed by L3

4.25
4.3
30
1.75
1.8
S4
these sectors as they become inoperable in multiple periods. With a 4.2
4.15
20 1.85
1.9
S5 10
limited budget, subsidizing the large sectors yields an economic 4.1 1.95
S6 4.05 0 2
disadvantage as they exhaust the budget relatively fast in the first 4 2.05
S7 3.95 2.1
instance – leaving the small sectors, which is the majority of the 3.9 2.15
S8 3.85 2.2
sectors (82.0%, 9/11), without subsidy and very inoperable. Thus, a S9 3.8 2.25
3.75 2.3
compromise is needed, which resulted in the least allocation of S10 3.7 2.35
3.65 2.4
subsidy to the large sectors in the SOM case. Such a strange result S11 3.6 2.45
3.55 2.5
may be consequent to the interdependency of the economic sectors. 3.5
3.45 2.6
2.55
3.4 2.65
3.35 2.7
3.3 2.75
3.25 3.2 2.852.8
3.15 3.13.05 3 2.952.9
The Effect of Interdependency

When it is economically disadvantageous to subsidize the Fig. 2. Subsidy of sectors (in BPhp) for different values of the subsidy budget
large sectors, those sectors that they are dependent on – the small (in BPhp)
sectors – may be subsidized instead. The small sectors provide
inputs required by the large sectors to stay operable. When the small For relatively smaller values of the subsidy budget, the
sectors become inoperable, it also affects the large sectors. While it large sectors are not prioritized. This observation is consistent with
is vital to reduce the large sectors' inoperability, reducing the small the observation in the preceding section. As the subsidy budget's
sectors' inoperability is also essential. However, when the subsidy value increases, the large sectors are prioritized except for L3
budget is small, prioritizing the large sectors would mean that the (Government Services). Note that L3 (Government Services) is
small sectors will become very inoperable, which would, in turn, among the least affected sectors during the COVID-19 pandemic,
result in the inoperability of the large sectors. A negative feedback (0)
with 𝑝11 = 0.13. This means that L3 (Government Services) is not
loop may occur in such an interdependent system, mainly because a priority for subsidy allocation because subsidizing it yields
the large sectors also provide inputs to the small sectors. Although minimal economic benefit, although it is among the large sectors.
the resilience coefficient, 𝑘𝑖 , of these sectors damps their This finding has several practical implications. For example, if the
inoperability overtime, leading to a recovery, analyzing their government decides to allocate a relatively small subsidy budget, it
interdependency to determine which sectors to prioritize in the needs to deal with political implications associated with the non-
allocation of subsidy is essential to speed up such recovery and prioritization of the large sectors. Such an unconventional approach
reduce total economic loss. However, the analysis of the complex to subsidy allocation might cause conflict between the large sectors
interdependency of the sectors in this respect is incredibly tricky. and the government.

Note that the SOM is developed within the framework of Large sectors are expected to be composed of large
the I-DIIM, which incorporates the interdependency of these sectors taxpaying institutions since their production volume is relatively
in the analysis of inoperability. Therefore, a logical reason for the

51
more extensive than in small sectors. Naturally, these sectors would The third case for sensitivity analysis intends to provide a
expect subsidy proportionate to their contribution to the national general perspective on the minimized total economic loss's
economy. The non-prioritization of large sectors can negatively sensitivity with simultaneous changes in both subsidy budget and
impact their confidence towards the government's policies and find inventory capacity values. A heat map in Fig. 4 presents the results
the country not secure for business. While the decision to least of this analysis. It can be vividly observed from the heat map that,
prioritize the large sectors in case of a small subsidy budget is for some value of the inventory capacity multiplier, there is a value
optimal, relaying the reason behind such a decision might be of the subsidy budget wherein the total economic loss can no longer
difficult. Such misunderstanding and conflict might result in large be improved. This value is indicated as the threshold line. The
businesses in these large sectors to transfer operations to other minimized total economic loss cannot be improved further than the
countries, which is terrible for the national economy. Thus, to avoid threshold subsidy that the economic sectors can utilize to
such a conflict its hefty consequences, the government may opt to reconfigure their initial inventory level due to their limited capacity
sufficiently increase its subsidy budget. For instance, when the to store inventory. Inventory capacity, therefore, limits the
subsidy budget is increased to 450.0 BPhp, the large sectors' utilization of subsidies, which can result in a surplus in the budget.
prioritization is an optimal strategy (see Fig. 2).
A surplus in the budget may be undesirable for several
Case 2: Changing the Inventory Capacity reasons. For instance, it indicates inadequate planning resulting in
inefficient utilization of the subsidy budget. For a government, the
The second case for sensitivity analysis is as follows: surplus money could have been invested or used for other profound
Suppose the Philippine government intends to impose a policy that means of public spending to serve the greater good. This finding has
requires the sectors to have a minimum inventory capacity as a several practical implications. For example, in the Philippines, this
disaster preparedness strategy. To determine the appropriate finding is of great importance, especially with the present debate on
inventory capacity to require from the sectors, it is essential to the suitability of the budget allotted for its proposed economic
determine how the optimal solution changes with the values of stimulus bill (House Bill 6815) in response to the COVID-19
parameter 𝜔 – the inventory capacity multiplier. In this regard, the pandemic. This study's findings may be utilized to adjust the debated
SOM is solved for multiple values of the inventory capacity budget to meet the government's criteria without compromising the
multiplier at a constant subsidy budget 𝛼 = 197.0 BPhp. The results efficiency of utilizing the subsidy budget. In general, this finding
of this analysis are presented in Fig. 3. suggests that, as much as possible, a decision-making body, e.g., a
government, must ensure that, for a given capacity of economic
1
sectors to keep inventory, an appropriate budget is decided such that
4.95 5 1.051.1 1.15
4.85 4.9 80 1.2 surplus is avoided.
4.8 1.25
4.75 1.3
4.7 70 1.35
4.65 1.4
4.6 1.45
4.55 60 1.5
L1 4.5 1.55 Subsidy Budget, 𝛼 (in BPhp)
4.45 50 1.6
L2 4.4 1.65
4.35 40 1.7
L3 4.3
30
1.75
4.25 1.8
S4 4.2 20 1.85
4.15 1.9
S5 4.1 10 1.95
S6 4.05 0 2
4 2.05
S7 3.95 2.1
3.9 2.15
S8 3.85 2.2
3.8 2.25
S9
3.75 2.3
S10 3.7 2.35
3.65 2.4
S11 3.6 2.45
3.55 2.5
3.5 2.55
3.45 2.6
Inventory Capacity Multiplier, 𝜔

3.4 2.65
3.35 2.7
3.3 2.75
3.25 3.2 2.852.8
3.15 3.13.05 3 2.952.9

Fig. 3. Subsidy of sectors (in BPhp) for different values of the inventory
capacity multiplier

When the inventory capacity is low, the allocation of

subsidy is focused on the large sectors. In line with this, the
allocation of subsidies to the small sectors is not economically
appealing because the government can only allocate a small amount
to them due to their limited ability to hold inventory. In this regard,
subsidizing the large sectors is a more appealing option because they
inherently have a large capacity to hold inventory. However, as the
inventory capacity increases, the large sectors are less preferred in
allocating subsidies. The focus of the allocation diverges towards
the small sectors because, in this case, the small sectors can already Economic Loss (in BPhp) Threshold Line
hold a sufficient amount of inventory to keep them less inoperable,
which, as a consequence of the interdependencies of the small and 0.0 335.3 670.7 1,006.0 1,341.3
large sectors, leads to the reduction of inoperability among large
sectors, and minimization of economic loss. Fig. 4. The minimized total economic loss (in BPhp) for different values of
both the subsidy budget (in BPhp) and inventory capacity multiplier
Case 3: Impact of Simultaneous Changes in the Values of Subsidy
Budget and Inventory Capacity to the Total Economic Loss

52
C. Comparative Case Analysis Scenario 1 Scenario 2 Scenario 3

A comparative case analysis is conducted in this work to (1)

59.0
demonstrate the importance of an inventory subsidy during

50.1
calamities, (2) show how decisions derived from the SOM can
deviate from the conventional method of appropriating resources,

33.9

32.2

31.0
and (3) contrast the results of the two methodologies. There are three

28.8

26.9
scenarios considered in this analysis, which are elaborated as

19.5

19.0

17.0
13.2

12.6

12.1
follows:

9.3
8.2

7.4

6.6

4.3
1.7
1.3
0.0
0.0
0.0
0.0
0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0
Scenario 1: No Subsidy is Allocated
L1 L2 L3 S4 S5 S6 S7 S8 S9 S10 S11

This scenario describes the case wherein the government

decides to not appropriate any budget for inventory subsidy to the Fig. 5. Subsidy allocation (in BPhp) of Scenarios 1, 2, and 3
(0)
economic sectors, thus 𝑠̂𝑖 = 0 for all sectors. There are several
reasons for such a case. A government may find such a budget 120.0
Scenario 1
worthless, or maybe the government decides to delay it too much – Scenario 2
beyond the last period of this simulation. This scenario is not an 100.0 Scenario 3

impossible phenomenon in the Philippines. Note that while the

Economic Loss (in BPhp)

80.0
discovery of the COVID-19 nears its first anniversary, and while the
rest of the world has exhausted initiatives to kick-start their 60.0
economy, the Philippines has not yet appropriated a budget to
subsidize its battered economic sectors. Therefore, it is a worthwhile 40.0
undertaking to demonstrate the consequences of such a decision
20.0
through this case, which is considered the baseline case of this
analysis. 0.0
0 5 10 15 20 25 30
Period, t (in months)
Scenario 2: Conventional Subsidy Allocation

This scenario describes the proportional allocation of

subsidies, which is conventionally applied for divisible resources. In
this case, the subsidy is proportionally allocated based on a sector's
production output. Employing this allocation scheme, the large
sectors receive more subsidies than the small sectors. The large
sectors provide more value than small sectors and contribute greatly
to total economic loss when left inoperable. This is the primary
motivation of this allocation scheme for prioritizing the large sectors
in the allocation of subsidy. However, the allocation should be
(0)
performed while conforming to the constraint in (6). Thus, 𝑠̂𝑖 =
(0)
𝑥𝑖 (0)
min (𝛼 (0) , 𝛽𝑖 − 𝑠̃𝑖 ).
∑𝑛
𝑖=1 𝑥𝑖
Scenario 3: Optimal Subsidy Allocation
The previous scenario treats the economic sectors in
isolation in the decision to allocate subsidy. However, as established
in previous discussions, the sectors are highly interdependent. Such
characteristic of the sectors is essential in this analysis because it
influences the allocation of subsidy, as detailed by the initial
findings of this work (see Section 5.3.1). This case, therefore, makes
∗,(0)
use of the results of the SOM – 𝑠̂𝑖 – as basis for allocating
Fig. 6. Economic loss (in BPhp) of Scenarios 1, 2, and 3 overtime
Subsidy allocation of Scenarios 1, 2, and 3
The large sectors were heavily subsidized in Scenario 2
(Conventional subsidy allocation), which is apparent because they
have the most extensive production output. In contrast, the subsidy
allocation in Scenario 3 (Optimal subsidy allocation), which was
obtained using the SOM, prioritized the large sectors the least.
Interestingly, the total economic loss in Scenario 3 (Optimal subsidy
allocation) (1,032.98 BPhp) is far less than that of Scenario 2
(Conventional subsidy allocation) (1,222.13 BPhp). On the other
hand, Scenario 1 (No subsidy is allocated), although it has zero-
subsidy, turned out to be the costliest scenario as it incurs the most
extensive total economic loss (1,559.46 BPhp). Thus, such a strategy
is highly unadvisable. The following discussion delves more into the
comparison of Scenario 2 (Conventional subsidy allocation) and
Scenario 3 (Optimal subsidy allocation).
Scenario 1 Scenario 2 Scenario 3
48.99%

subsidies to the economic sectors. Specifically, this case uses the

38.92%

results in Section 5.3.1 to compare with the results of the previous

cases. The comparison of subsidy allocation in each case is
presented in Fig. 5.
14.25%

11.65%
8.58%

8.52%
7.91%

7.66%
7.61%

7.45%
6.97%

4.97%

4.57%
4.50%

3.97%
3.30%

3.30%
2.41%
2.28%

2.03%

0.09%
0.06%
0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

0.00%

L1 L2 L3 S4 S5 S6 S7 S8 S9 S10 S11

Fig. 7. Contribution of economic sectors in the reduction of total economic

loss in Scenarios 1, 2, and 3

53
 merely assumes that the economic sectors are independent system
L1 L2 L3
elements. This assumption is the main reason for its wide adoption
1.00 0.80 0.25 among analysts and decision-makers in appropriating divisible
Inoperability, 𝒒

0.70
0.80 0.60 0.20 resources (e.g., subsidy), hence being conventional. While it is

Inoperability, 𝒒

Inoperability, 𝒒
0.60 0.50 0.15
0.40
simple, it doesn't mean that it is optimal. Such analysis fails to
0.40 0.30 0.10
0.20
account for and take advantage of the complex interdependencies of
0.20 0.05
0.10 the economic sectors. These interdependencies are incredibly
0.00 0.00 0.00
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30 challenging to perceive or even infer from. Thus, exploiting them is
Period, t (in months) Period, t (in months) Period, t (in months) difficult for analysts or decision-makers. However, this work
demonstrates the significance of such an analysis in reducing
calamity-induced inoperability (see Fig. 8), leading to the
S4 S5 S6 minimization of total economic loss consequent to it. Hence, the
0.70 0.80 1.00
SOM's importance is its ability to guide analysts and decision-
0.60 0.70 makers to make systemic and systematic resource allocation
0.60 0.80
Inoperability, 𝒒

Inoperability, 𝒒

Inoperability, 𝒒
0.50
0.40
0.50 0.60 strategies.
0.40
0.30 0.40
0.30
0.20 0.20 0.20
0.10
0.00
0.10
0.00 0.00
IV. CONCLUSION
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
Period, t (in months) Period, t (in months) Period, t (in months) There have been several extensions to the classical
Leontief Input-Output model applied in infrastructure risk research.
Among the said extensions is the Inventory Dynamic Inoperability
S7 S8 S9
Input-Output Model (I-DIIM). The I-DIIM illustrated the
0.70 0.80 0.70 importance of inventory to the preparedness and resilience of
0.60 0.70 0.60
0.60
infrastructures in a production environment, e.g., a macroeconomic
Inoperability, 𝒒

Inoperability, 𝒒

Inoperability, 𝒒

0.50 0.50
0.50 system, to disruptive events. Although the economic implications of
0.40 0.40
0.40
0.30 0.30
0.20
0.30
0.20
a disruptive event to a macroeconomic system are primarily the
0.20
0.10 0.10 0.10 government's concern, no work within the domains of infrastructure
0.00 0.00 0.00
0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
risk research has yet attempted to provide a way to empirically
Period, t (in months) Period, t (in months) Period, t (in months) determine the optimal intervention required from governments to
facilitate the preparedness and resilience of economic sectors. This
work addressed this major gap in the literature by developing a
S10 S11 Subsidy Optimization Model (SOM) that is based on the I-DIIM
0.01 with the following key model characteristics: (1) distinguishes
0.80
0.70 0.01 inventory constraints from budgetary constraints, (2) incorporates
Inoperability, 𝒒

0.60 0.01
Inoperability, 𝒒

0.50
0.40
0.30
0.20
0.10
0.00
0 5 10 15 20 25 30
Period, t (in months)
Fig. 8. I-DIIM simulation results on the inoperability of the economic
sectors in Scenarios 1, 2, and 3
Fig. 6 presents the economic loss in each scenario over
time. Considering Scenario 1 (No subsidy is allocated) as the
baseline case (the worst case), Scenario 2 (Conventional subsidy
allocation) reduced the total economic loss only by 21.6%, while
Scenario 3 (Optimal subsidy allocation) reduced the total economic
loss by 33.8%. The contribution of each sector in each scenario in
reducing total economic loss is presented in Fig. 7. In Scenario 3
(Optimal subsidy allocation), although little to no subsidy was
allocated for the large sectors, they were among the most significant
contributors in reducing the total economic loss. While this
observation on the contribution of the large sectors towards the
reduction of total economic loss is also true in Scenario 2
(Conventional subsidy allocation), it is worth noting that, for the rest
of the sectors, Scenario 3 (Optimal subsidy allocation) far
outperforms it in terms of total economic loss reduction. These
results vividly illustrate the strength of Scenario 3 (Optimal subsidy
allocation) over Scenario 2 (Conventional subsidy allocation),
which is primarily grounded on the ability of the SOM to exploit the
interdependency of the sectors.
The strength of Scenario 2 (Conventional subsidy
allocation) is the simplicity of its analytical procedures because it
Scenario 1
subsidy as a means to configure initial inventories, and (3) integrates
0.01
the I-DIIM in a nonlinear optimization model for minimizing total
0.00 Scenario 2
Scenario 3
0.00 economic loss within an exogenously defined simulation period.
0.00 This work also proposes a simulation-based subsidy optimization
0 5 10 15 20 25 30
Period, t (in months)
procedure with the Generalized Reduced Gradient (GRG) with the
Multi-Start algorithm for solving the SOM. The SOM's applicability
is tested through an illustrative case study in the Philippines during
the COVID-19 pandemic.
This work revealed that the SOM differs from
conventional methods for allocating divisible resources at low
subsidy budgets. For instance, at a subsidy budget of 197.0 BPhp,
the SOM's optimal solution indicated that large sectors be least
prioritized in the allocation of subsidy. Such peculiar results were
found consequent to the interdependency of the economic sectors,
which the SOM was able to exploit to minimize the total economic
loss. The robustness of such findings was tested through sensitivity
analysis by changing the subsidy budget and inventory capacity
values. Consistent with initial results, it was found that the large
sectors are least prioritized for relatively smaller values of the
subsidy budget; however, as the subsidy budget's value increases,
the priority shifts towards the large sectors except for L3
(Government Services), which is among the least affected sectors
during the COVID-19 pandemic in the Philippines. This result
somehow resembles the conventional method for the allocation of
divisible resources. It was also found that when the inventory
capacity is low, the large sectors are prioritized in the allocation of
subsidy. However, as the inventory capacity increases, the priority
diverges from the large sectors towards the small sectors, which is
still instigated by the complex interdependency of the sectors
exploited by the SOM.

54
 Furthermore, this work investigated how simultaneous
changes in the subsidy budget and inventory affects the
minimization of total economic loss. In this work, the analysis found
out that there is a corresponding value of the subsidy budget for
some inventory capacity multiplier values wherein the total
economic loss can no longer be improved. The finding is essential
for avoiding utilization inefficiencies and budget surplus, which is
essential for decision-makers and policymakers. Specifically, the
finding suggests that, as much as possible, a decision-making body,
e.g., a government, must ensure that, for a given capacity of
economic sectors to keep inventory, an appropriate budget is
decided such that surplus is avoided.
Through comparative case analysis, this work contrasted
conventional methods of subsidy allocation, e.g., proportional
allocation, and the allocation scheme proposed using the SOM. The
analysis vividly demonstrated the inadequacy of treating economic
sectors in isolation, which is employed by conventional allocation
methods. The analysis also highlighted the SOM's ability to exploit
and leverage economic sectors' interdependency to develop an
allocation scheme that minimizes total economic loss, which would
otherwise be an incredibly difficult task to fulfill without it. The
SOM, therefore, is instrumental in guiding analysts and decision-
makers in making systemic and systematic resource allocation
strategies.
For future work, researchers may explore a SOM
modification to consider allocating subsidies in multiple schedules.
In the real world, large budgets for subsidies are seldom allocated at
once. Instead, it is allocated in tranches. Scheduling each tranche
such that total economic loss is minimized within a simulation
period may be a worthwhile undertaking for a SOM extension. One
of the major limitations of this work and preceding works on I-DIIM
is the assumption that inventories are readily convertible to revenue.
In this regard, researchers may also consider developing new
constraints to consider the processing capacity and processing time
of inventories to address this limitation. The concepts mentioned
above are not addressed in the present work because it intends to
serve as a groundwork for exploring such concepts in the following
series of publications on this topic.
ACKNOWLEDGEMENT
The author expresses appreciation to Engineering
Research and Development for Technology (ERDT) program of the
Department of Science and Technology, Republic of the Philippines
who funded this work, as part of the principal author’s (EFSJ)
master’s thesis, through the Thesis and Dissertation Grant (TDG).
REFERENCES
[1] K. Barker and J. R. Santos, "Measuring the efficacy of
inventory with a dynamic input–output model," International
Journal of Production Economics, vol. 126, no. 1, pp. 130–143,
Jul. 2010, doi: 10.1016/J.IJPE.2009.08.011.
[2] O. Jonkeren and G. Giannopoulos, "Analysing Critical
Infrastructure Failure with a Resilience Inoperability Input–
output Model," Economic Systems Research, vol. 26, no. 1, pp.
39–59, Jan. 2014, doi: 10.1080/09535314.2013.872604.
[3] L. Galbusera, I. Azzini, O. Jonkeren, and G. Giannopoulos,
"Inoperability Input-Output Modeling: Inventory Optimization
and Resilience Estimation during Critical Events," ASCE-
ASME Journal of Risk and Uncertainty in Engineering
Systems, Part A: Civil Engineering, vol. 2, no. 3, p. B4016001,
Sep. 2016, doi: 10.1061/AJRUA6.0000861.
[4] E. M. Y. Jabilles, J. M. T. Cuizon, P. M. A. Tapales, R. L.
Urbano, L. A. Ocampo, and D. A. A. Kilongkilong,
"Simulating the impact of inventory on supply chain resilience
with an algorithmic process based on the supply-side dynamic
inoperability input–output model," International Journal of
Management Science and Engineering Management, vol. 14,
no. 4, pp. 253–263, Oct. 2019, doi:
10.1080/17509653.2018.1555693.
[5] L. Xu, Q. Zhang, K. Wang, and X. Shi, "Subsidies, loans, and
companies' performance: evidence from China's photovoltaic
industry," Applied Energy, vol. 260, p. 114280, Feb. 2020, doi:
10.1016/j.apenergy.2019.114280.
[6] C. M. Himang et al., "Acceptability evaluation of a developed
thermal infrared device for fire risk management using the
unified theory of acceptance and use of technology (UTAUT)
model," International Journal of Sociotechnology and
Knowledge Development, vol. 12, no. 3, pp. 1–25, Jul. 2020,
doi: 10.4018/IJSKD.2020070101.
[7] E. F. Selerio Jr., R. D. Arcadio, G. J. Medio, M. A. Nacua, and
G. A. Pedregosa, "An Application of a Hybrid Fuzzy Multi-
Criteria Decision-Making Approach in Managing Equipment
Hazards in Woodworking," International Journal of Safety and
Security Engineering, vol. 10, no. 4, pp. 517–527, Aug. 2020,
doi: 10.18280/ijsse.100411.
[8] J. R. Santos, M. J. Orsi, and E. J. Bond, "Pandemic Recovery
Analysis Using the Dynamic Inoperability Input-Output
Model," Risk Analysis, vol. 29, no. 12, pp. 1743–1758, Dec.
2009, doi: 10.1111/j.1539-6924.2009.01328.x.
[9] J. R. Santos and Y. Y. Haimes, "Modeling the demand
reduction input-output (I-O) inoperability due to terrorism of
interconnected infrastructures," Risk Analysis, vol. 24, no. 6,
pp. 1437–1451, Dec. 2004, doi: 10.1111/j.0272-
4332.2004.00540.x.
[10] R. Sharma and B. Glemmestad, "On generalized reduced
gradient method with multi-start and self-optimizing control
structure for gas lift allocation optimization," Journal of
Process Control, vol. 23, no. 8, pp. 1129–1140, Sep. 2013, doi:
10.1016/j.jprocont.2013.07.001.
[11] P. Wolfe, "Methods of Nonlinear Programming, in: Graves,
R.L. & Wolfe, P. (Eds.), Recent Advances in Mathematical
Programming, McGraw-Hill, New York, 67-86.," 1963.
[12] J. Abadie and J. Carpentie, "Generalization of the Wolfe
reduced gradient method to the case of nonlinear constraints,
in: Fletcher, R. (Eds.), Optimization, Academic Press, New
York, USA, 37–47.," 1969.
[13] L. S. Lasdon, R. L. Fox, and M. W. Ratner, “Nonlinear
Optimization Using the Generalized Reduced Gradient
Method. Revue française d’automatique, informatique,
recherche opérationnelle,” RAIRO - Operations Research -
Recherche Opérationnelle, vol. 8, no. V3, pp. 73–103, 1974,
Accessed: Sep. 17, 2020. [Online]. Available:
http://www.numdam.org/item?id=RO_1974__8_3_73_0.
[14] M. M. Ong, R. M. Ong, G. K. Reyes, and L. B. Sumpaico-
Tanchanco, "Addressing the COVID-19 Nutrition Crisis in
Vulnerable Communities: Applying a Primary Care
Perspective," Journal of Primary Care and Community Health,
vol. 11. SAGE Publications Inc., 2020, doi:
10.1177/2150132720946951.
[15] G. D. Legaspi et al., "Letter to the Editor' Service and Training
During the COVID-19 Pandemic: Perspectives from a
Neurosurgical Center in the Philippines,'" World
Neurosurgery, vol. 139. Elsevier Inc., pp. 741–743, Jul. 01,
2020, doi: 10.1016/j.wneu.2020.05.138.
[16] DOH, "PH Starts to Flatten Curve; ECQ, Physical Distancing
Pay Off," Department of Health, 2020.
55
[17] The Straits Times, "Coronavirus: Resurgence of cases threatens where
bids to kick-start Philippine economy," The Straits Times,
2020. 𝑥𝑖𝑗
[18] CNN Philippines, "Metro Manila, 4 provinces to ease into GCQ 𝑨 = [𝑎𝑖𝑗 ] =[ ] , (16)
𝑛×𝑛 𝑥𝑗
𝑛×𝑛
starting Aug. 19," CNN Philippines , 2020.
[19] ECDC, "COVID-19 situation update worldwide, as of 28 where 𝑥𝑖 is the total production output of sector 𝑖 while the
September 2020," European Centre for Disease Prevention and Leontief technical coefficient 𝑎𝑖𝑗 is the ratio of the output of sector
Control, 2020.
𝑖 that serves as an input for sector 𝑗 concerning the total production
[20] NEDA, "Statement of NEDA Acting Secretary Karl Kendrink
requirement of sector 𝑗; and 𝑐𝑖 is the final demand of the 𝑖th sector,
T. Chua On The 2020 Q2 Performance of the Philippine
Economy," National Economic and Development Authority, for 𝑖, 𝑗 ∈ ℕ.
2020.
[21] PSA, "Press Release," Philippine Statistics Authority, 2020. Appendix B. The Demand-driven Inoperability I-O Model (IIM)
[22] TESDA, "COVID-19's Impact to the Philippine
EconomSectors: A TVET Response. TVET Brief: Industry While the Classical I-O Model has gained prominence in
Trends," Technical Education and Skills Development its ability to integrate economic sectors' interdependence, its
Authority, 2020. applications are quite limited due to its inability to assess economic
[23] PNA and F. M. Cervantes, "House okays ARISE bill on 3rd risks explicitly. Based on (15) and (16), the IIM was proposed by [9]
reading | Philippine News Agency," 2020. to address such a gap. They employed the IIM in assessing the
[24] PSA, "Input Output Tables," Philippine Statistics Authority, impact of a disruption and its spread across an economic network.
2017. The capability of the IIM to model the propagation of inoperability
[25] ADB, "ADB Philippine Enterprise Survey on COVID-19 in a system makes it a relevant framework in the development of
Impact | ADB Data Library," Asian Development Bank, 2020. risk-management policies [9]. The following definitions constitute
https://data.adb.org/dataset/adb-philippine-enterprise-survey- the IIM:
covid-19-impact (accessed Sep. 18, 2020).
[26] ADB, "Philippine Economy to Decline Further in 2020 Amid Definition 2. (Demand-side perturbation) Let 𝒙 be the as-
COVID-19, With Recovery in 2021," Asian Development planned production vector that describes the as-planned production
Bank, 2020. output of each sector. Provided that each sector is defined on the
[27] W. W. Leontief, "Quantitative Input and Output Relations in premise of "economic significance" [9], then [𝑑𝑖𝑎𝑔(𝒙)]−1 always
the Economic Systems of the United States," The Review of exist since it guarantees that 𝑥𝑖 ≠ 0, for 𝑖 ∈ ℕ. The preceding
Economics and Statistics, vol. 18, no. 3, p. 105, Aug. 1936, doi: argument also suggests that 𝑑𝑖𝑎𝑔(𝒙) is always a nonsingular matrix,
10.2307/1927837. which sufficiently establishes its inverse's existence. Denoting 𝒄̂ as
[28] Y. Y. Haimes, B. M. Horowitz, J. H. Lambert, J. R. Santos, C. the as-planned final demand and 𝒄̃ as the degraded demand which
Lian, and K. G. Crowther, "Inoperability Input-Output Model may be brought about by fear and unrest due to some disaster, the
for Interdependent Infrastructure Sectors. I: Theory and demand-side perturbation vector 𝒄∗ is defined as the decrease in the
Methodology," Journal of Infrastructure Systems, vol. 11, no. final demand (𝒄̂ − 𝒄̃), normalized according to the as-planned
2, pp. 67–79, Jun. 2005, doi: 10.1061/(ASCE)1076- production 𝒙, which is illustrated as follows:
0342(2005)11:2(67).
[29] R. , Miller and P. Blair, Input output analysis foundations and 1
0 ⋯ ⋯ 0
extensions, 2nd ed. Cambridge University Press, 2009. 𝑐1∗ 𝑥1 𝑐̂1 − 𝑐̃1
⋮ 0 ⋱ ⋱ ⋮ ⋮
1
𝒄∗ = [𝑐𝑖∗ ]𝑛×1 = [𝑑𝑖𝑎𝑔(𝒙)]−1 [𝒄̂ − 𝒄̃] ⇔ 𝑐𝑖∗ = ⋮ ⋱ 𝑥𝑖
⋱ ⋮ 𝑐̂𝑖 − 𝑐̃𝑖 (17)
Appendix A. The Classical I-O Model ⋮ ⋮
⋮ ⋱ ⋱ 0
[𝑐𝑛∗ ] 1 [𝑐̂𝑛 − 𝑐̃𝑛 ]
[0 ⋯ ⋯ 0 𝑥𝑛 ]
The I-O analysis is a form of macroeconomic analysis
based on the interdependencies between economic sectors or The 𝑖th element in 𝒄∗ represents the demand-side
industries, showing how the output from one industrial sector may perturbation of the 𝑖th sector. Denoting this scalar element by 𝑐𝑖∗ , it
become an input to another industrial sector. This method is can be extracted from (17) as follows:
commonly employed for determining the impacts of economic
disruptions and their ripple effects throughout an economy [27]. The 𝑐̂𝑖 −𝑐̃𝑖
𝑐𝑖∗ = , (18)
column entries of the intersectoral matrix employed in I-O models 𝑥𝑖
represent the input to an economic sector, while the row entries
represent outputs from a given sector. The intersectoral matrix where 𝑐̂𝑖 is the as-planned demand of sector 𝑖, 𝑐̃𝑖 is the
presents each sector's dependence on every other sector, both as a 𝑐̂
corresponding degraded demand of sector 𝑖, 𝑥𝑖 ∈ ℝ+ , 𝑐𝑖∗ ∈ [0, 𝑖 ],
customer of outputs from other sectors and as a supplier of inputs. 𝑥𝑖
Each column of the matrix indicates the monetary value of inputs to and 𝑖 ∈ ℕ.
each sector, while each row represents the value of each sector's
outputs. Building on this idea, the Classical I-O Model is defined as Definition 3. (Interdependency matrix) Let 𝑨∗ be the
follows: demand-side interdependency matrix that characterizes the degree
of coupling of the sectors. The relationship of matrix 𝑨∗ to the
Definition 1. (Classical I-O model) Let ℕ = {1,2, … , 𝑛} Leontief technical coefficient matrix 𝑨 and the as-planned
be a set of economic sectors. Assuming that each sector sells some production vector 𝒙 is illustrated as follows:
of its output to other sectors (intermediate output) and some of its
output to consumers (final output, or final demand), the Classical I-
O Model [27] is expressed as follows:

𝒙 = 𝑨𝒙 + 𝒄 ⇔ {𝑥𝑖 = ∑𝑗 𝑎𝑖𝑗 𝑥𝑗 + 𝑐𝑖 }; ∀𝑖 ∈ ℕ, (15)

56
 ∗
𝑨∗ = [𝑎𝑖𝑗 ] = [[𝑑𝑖𝑎𝑔(𝒙)]−1[𝑨][𝑑𝑖𝑎𝑔(𝒙)]]𝑛×𝑛 ⇔ transforming the resulting equation into the normalized inoperability
𝑛×𝑛
1
0 ⋯ ⋯ 0 𝑎11 ⋯ 𝑎1𝑗 ⋯ 𝑎1𝑛 𝑥1 form (as in Definition 2-4) yields the discrete form of the DIIM that
𝑥1 0 ⋯ ⋯ 0
0 ⋱ ⋱ ⋮ ⋮ ⋱ ⋮ ⋮ 0 ⋱ ⋱ ⋮ presents the inoperability vector 𝒒(𝑡+1) , as illustrated in (24).
1
⋮ ⋱ ⋱ ⋮ 𝑎𝑖1 ⋯ 𝑎𝑖𝑗 ⋯ 𝑎𝑖𝑛 ⋮ ⋱ 𝑥𝑗 ⋱ ⋮ (19)
𝑥𝑖
⋮ ⋮ ⋱ 0 ⋮ ⋱ ⋱ 0 1
⋮ ⋱ ⋱ 0
1 [𝑎𝑛1 ⋯ 𝑎𝑛𝑗 ⋯ 𝑎𝑛𝑛 ] [ 0 ⋯ ⋯ 0 𝑥𝑛 ]
𝑲 = −𝑩−1 ⇔ 𝑘𝑖 = ; ∀𝑖 ∈ ℕ (23)
𝑏𝑖
[0 ⋯ ⋯ 0
𝑥𝑛 ]

(𝑡+1)
The derivation of the scalar representation of (19) is 𝒒(𝑡+1) = [𝑞𝑖 ] = 𝒒(𝑡) + 𝑲[𝑨∗ 𝒒(𝑡) + 𝒄∗,(𝒕) − 𝒒(𝑡) ], (24)
𝑛×1
omitted for brevity, but it can be found in [28]. The scalar
representation of (19) is illustrated as follows: where 𝑨∗ denotes the demand-side interdependency
(𝑡)
𝑥𝑗
matrix as defined in Definition 3, 𝑞𝑖 ∈ [0,1] ∀𝑖, 𝑡; and 𝑖 ∈ ℕ.
∗
𝑎𝑖𝑗 = 𝑎𝑖𝑗 ( ); for 𝑖, 𝑗 ∈ ℕ (20)
𝑥𝑖
Appendix D. The Inventory Dynamic Inoperability I-O Model
Definition 4. (Inoperability I-O model) Let 𝒒 be the final (I-DIIM)
inoperability vector that describes the inoperability of the different
sectors. Within the framework of the IIM, inoperability is defined as Barker and Santos [1] introduced the I-DIIM, which can
a level of a system's dysfunction expressed as a percentage of its as- model the effects of the resilience measures that delay the onset of
planned production capacity [9], [28]. Building on previous infrastructure or economic sectors' inoperability due to access to
definitions, 𝒒 is calculated as follows: some inventory. The definitions that constitute the I-DIIM are
presented as follows:
𝒒 = [𝑞𝑖 ]𝑛×1 = [𝑰 − 𝑨∗ ]−1 𝒄∗ , (21)
Definition 8. (Production inoperability) Let 𝒑(𝑡) =
(𝑡) (𝑡)
where 𝑰 is an identity matrix, 𝑞𝑖 ∈ [0,1], and 𝑖 ∈ ℕ. [𝑝𝑖 ] be the production inoperability vector, where 𝑝𝑖 denotes
𝑛×1
the production inoperability consequent only to physical disruption
The rigorous derivation of (21) is not presented herewith of processes of sector 𝑖 at the end of time 𝑡 [1]. The calculation of
for brevity, but it can be found in [9]. (𝑡)
𝑝𝑖 is illustrated as a decaying function of time 𝑡 as follows:
Appendix C. The Demand-driven Dynamic Inoperability I-O (𝑡) (0)
Model (DIIM) 𝑝𝑖 = 𝑒 −𝑘𝑖𝑡 𝑝𝑖 , (25)

(0)
The main limitation of the IIM is its inability to model the where 𝑝𝑖 is the exogenously defined initial production
concept of "dynamic economic resilience" [28]. The need to (𝑡)
inoperability of sector 𝑖, 𝑝𝑖 ∈ [0,1] ∀𝑖, 𝑡; 𝑖 ∈ ℕ, and 𝑘𝑖 is as
integrate the temporal dynamic behavior of industry recoveries in
defined in Definition 5 and is calculated using (26).
the analysis of inoperability leads to the development of the DIIM,
as an extension that supplements and complements the static IIM. In (0)
𝑝
the derivation of the DIIM, [28] introduced the concept of an ln[ 𝑖
(𝑓𝑖) ]
𝑝
industry resilience coefficient that is defined as follows: 𝑘𝑖 = 𝑖
, (26)
𝑓𝑖

Definition 5. (Industry resilience coefficient) Let 𝑲 =

[𝑘𝑖 ]1×𝑛 be the industry resilience coefficient vector wherein each where 𝑓𝑖 ∈ ℤ+ is the estimated recovery period of sector
element 𝑘𝑖 denotes the resilience of sector 𝑖 when given an 𝑖 ∈ ℕ. The derivation of (26) can be found in Barker and Santos [1]
imbalance between supply and demand some disruptive event, and is not presented herewith for brevity.
where 𝑘𝑖 ∈ [0,1) and 𝑖 ∈ ℕ. In the case of a calamity (e.g.,
(𝑡)
pandemic), 𝑘𝑖 measures the estimated recovery rate of the economic Definition 9. (Inventory level) Let 𝒔(𝑡) = [𝑠𝑖 ] be the
𝑛×1
sectors. This parameter may be exogenously defined based on expert (𝑡)
opinion or computed based on several recovery period factors. inventory level vector, where 𝑠𝑖
denotes the amount of inventory
of sector 𝑖 at the end of time 𝑡, for 𝑖 ∈ ℕ. In this work, inventory is
Definition 6. (Re-defining the Classical I-O Model) Let utilized to cushion the impact of a calamity. Thus, it depletes as it is
the Classical I-O Model take the form of (22) as in [29] and [28]. used to compensate for production inoperability as follows:

(𝑡+1) (𝑡) (𝑡) (𝑡)

𝑑𝒙(𝑡) 𝑠𝑖 = max{𝑠𝑖 − 𝑝𝑖 𝑥𝑖 , 0}, (27)
𝒙(𝑡) = 𝑨𝒙(𝑡) + 𝒄(𝑡) + 𝑩 , (22)
𝑑𝑡
(𝑡)
where 𝑥𝑖 denotes the total output anticipated to be
where matrices 𝑨, 𝒙(𝑡) , and 𝒄(𝑡) are the Leontief technical
produced by sector 𝑖 ∈ ℕ between the end of time 𝑡 − 1 and the end
coefficient matrix, production output vector at time 𝑡, and final
of time 𝑡.
demand vector at time 𝑡, respectively (as defined in Definition 1),
while matrix 𝑩 is a square matrix of capital coefficients 𝑏𝒊 that
characterizes the willingness of the economy to invest in capital Definition 10. (Inventory dynamic inoperability I-O
∗,(𝑡)
resources (see [28]). model) Denote (10) as 𝑭[𝒒(𝑡) , 𝒄∗,(𝑡) ], where 𝐹𝑖 [𝒒(𝑡) , 𝑐𝑖 ] is its 𝑖th
element at time 𝑡. Building on Definition (8, 9), the I-DIIM presents
Definition 7. (Dynamic inoperability I-O model) Let 𝑲 the inoperability vector 𝒒(𝑡+1) as follows:
and 𝑩 be 𝑑𝑖𝑎𝑔(𝑘𝑖 ) and 𝑑𝑖𝑎𝑔(𝑏𝑖 ), respectively; and their
relationship is illustrated in (23). A more rigorous definition of this
relationship can be found in [28]. Substituting (23) for (22) and

57
 𝒒(𝑡+1) = [𝑞𝑖
(𝑡+1)
] = 𝑞𝑖 (𝑠𝑖
(𝑡)
, 𝑝(𝑖𝑡) , 𝑥(𝑖𝑡) , 𝑐∗,(𝑡)
𝑖 , 𝑘𝑖 , 𝑎∗𝑖𝑗 ) =
𝑛×1
(𝑡) ∗,(𝑡) (𝑡+1) (𝑡+1) (𝑡+1)
𝐹𝑖 [𝒒 , 𝑐𝑖 ], 𝑖𝑓 𝑠𝑖 ≥ 𝑝𝑖 𝑥𝑖
(𝑡+1)
(𝑡+1) 𝑠𝑖 ∗,(𝑡) (𝑡+1) (𝑡+1) (𝑡+1)
max {[𝑝𝑖 − (𝑡+1) ] , 𝐹𝑖 [𝒒(𝑡), 𝑐𝑖 ]} , 𝑖𝑓 𝑠𝑖 ∈ (0, 𝑝𝑖 𝑥𝑖 )
𝑥𝑖
, (28)
(𝑡+1) ∗,(𝑡) (𝑡) (𝑡+1)
max {𝑝𝑖 , 𝐹𝑖 [𝒒(𝑡) , 𝑐𝑖 ]} , 𝑖𝑓 𝑠𝑖 > 0, 𝑠𝑖 =0
∗,(𝑡) (𝑡+1) (𝑡)
{ 𝐹𝑖 [𝒒(𝑡) , 𝑐𝑖 ]], 𝑖𝑓 𝑠𝑖 = 𝑠𝑖 =0

(𝑡)
where 𝑞𝑖 ∈ [0,1] ∀𝑖, 𝑡, and 𝑖 ∈ ℕ and at time 𝑡 = 0, the
(0)
initial inoperability 𝑞𝑖 varies with initial inventory scenarios as
follows:

(0) (0) (0) (0)

𝑞𝑖 = 𝑞𝑖 (𝑠𝑖 , 𝑝𝑖 , 𝑥𝑖 ) =
(0) (0) (0)
0, 𝑖𝑓 𝑠𝑖 ≥ 𝑝𝑖 𝑥𝑖
(0)
(0) 𝑠𝑖 (0) (0) (0)
𝑝𝑖 (1 − (0) (0) ) , 𝑖𝑓 𝑠𝑖 ∈ (0, 𝑝𝑖 𝑥𝑖 ) (29)
𝑝𝑖 𝑥𝑖
(0) (0)
{ 𝑝𝑖 , 𝑖𝑓 𝑠𝑖 =0

58
Earthquake Resistant Building Through Genetic
Algorithm Optimization: A Case Study Using
Ground Motion Data
Lisa Guieb1, a *, Sherwin Guirnaldo², Glenn Lorenzoᵌ
¹Department of Civil Engineering, MSU – Iligan Institute of Technology, Iligan City;
²Department of Mechanical Engineering, MSU – Iligan Institute of Technology, 9200 Iligan City;
³Department of Civil Engineering, MSU – Iligan Institute of Technology, Iligan City;
a
lisatguieb@gmail.com
Abstract— The study uses the dynamic analysis
procedure for Response Spectrum Analysis (RSA) for
buildings from the Site-specific Response Spectrum Curve
obtained on October 2019, series of earthquakes that caused
major damage to several structures in Mindanao.
Optimization techniques are used to solve structural
engineering designs considering safety, stability, and
sustainability when subjected to earthquake loadings. The
STAAD.Pro structural analysis software has been used for
the design computation. For this study, the member size of
columns and beams was checked for displacement, and
forces considered as constraints for optimizing different
cross-sections using a genetic algorithm. The design with the
highest performance, which satisfies the code parameters,
are considered the best performing design that was chosen
from the final generation out of one hundred population. The
numerical algorithm is developed in MATLAB software
incorporating the response spectrum analysis of five-story
story building and optimization routine. The most common
parameter used in this study is the minimization of inter-
story drift and dynamic base shear, which is used in seismic
design codes of the National Structural Code of the
Philippines (NSCP).
Keywords: Response spectrum analysis; site-specific ground
motion; response spectrum curve; optimization; genetic
algorithm
I. INTRODUCTION
On October 2019, series of earthquakes struck
Tulunan, Cotabato, Mindanao and vicinity. The
Philippine Institute of Volcanology & Seismology
(PHIVOLCS), reported that these earthquakes
occurred on October 16, 7:37 PM Philippine Standard
Time (PST) at magnitude 6.3, on October 29, 9:04 AM
and 10:42 AM at magnitudes 6.6 and 6.1, respectively,
and on October 31, 9:11 AM at magnitude 6.5. Several
damaged structures such as residential condo units,
school buildings, business establishments and
residential dwellings are reported after the strong
earthquakes in Mindanao need total demolition and
retrofitting.
There is an urgency to achieve an optimized design
of structure for resilient and sustainable construction.
Since the reinforced concrete is now widely used in
civil engineering projects, RC structures' economic
design is necessary. It is directed to find optimal
structural design resulting in size optimization,
minimum construction weight results in sustainable
development with lesser damage to buildings. The
optimization of building design under seismic
excitation is crucial for optimal design parameters
among considered alternatives while ensuring design
safety. It provides the satisfaction of ultimate strength
and serviceability limit states prescribed by design
codes.
2. OPTIMIZATION PROBLEM
2.1 Fundamental optimization problem
The objective function of the optimization is to
minimize the sizes of beams and columns to minimize
the weight of structure and the displacement of the
building under seismic actions. It is design to resist
lateral forces such as earthquake forces which may
cause failure to structures.
In this study, the design variables are associated
with dimensions, beam width, beam depth and column
member sizes. In the sizing optimization, ‘x’ is some
type of thickness i.e. cross sectional area of beam and
column. The objective function is expressed in
Equation 1.
𝒇(𝑿) = 𝒇([𝑪𝒐𝒍𝒖𝒎𝒏 𝒔𝒊𝒛𝒆𝒔, 𝑩𝒆𝒂𝒎 𝒔𝒊𝒛𝒆𝒔])
= 𝐦𝐢𝐧(∆𝒙𝒅𝒓𝒊𝒇𝒕 )
𝒏
The f(X) is the fitness score. Candidate based on
building description.
Find 𝑥𝑖 (i = 1,2…,n), and make objective
function 𝑓(𝑥) = 𝑥𝑡 , drift be minimum if the total
weight W is reduced. The constraint conditions of the
stress and displacement are the following.
𝜎ј ≤ [𝜎𝑗 ] (𝑗 = 1,2 … 𝑚) (Equation 1)
𝑥𝑘 ≤ [𝑥𝑘 ] (𝑘 = 1,2 … 𝑚) (Equation 2)
Where 𝒙𝒊 indicates section size, i variable
number, n the total variable, 𝝈𝒋 and [𝝈𝒋 ]the maximum
work stress of j-th beam, m the total number of beam,
59
[𝒙𝒌 ]the displacement in the node k, p the total of (a)
node, 𝒙𝒕 the total drift, W the total weight of beam.
PLAN
b=19m

x1 x2 a1 =8m

a2=7m a=23m

x1 x3
a3=8m

Figure1. The section form of beam and column

Table 1. Range of Design Variables b1=6m b2=7m b3=6m

Range of Design Variables

(b)
Figure 2. Model 5-Storey RC
(a) Two-dimensional model. (b) Top view.
Column Sizes (m) Beam Sizes (m)
Height Base
Va Fro T Va Fro T Va Fro T
In Figure 3, the building is subjected to a ground
r. m o r. m o r. m o motion. The excitation at the base of structure yields
Lev X1 0.3 1. X2 0.25 0. X3 0.25 0. displacement 𝒙 (𝐭) at the roof relative to the base.
el 1 0 8 8
&2
Lev X4 0.3 1. X5 0.25 0. X6 0.25 0.
el 3 0 8 8
&4
Lev X7 0.3 1. X8 0.25 0. X9 0.25 0.
The equation of motion is express as shown below.
el 5 0 8 8
𝒎𝒙 + 𝒄𝒙 + 𝒌𝒙̈ =̇ −𝒎𝒙𝒈̈ (Equation 3)

The fixed parameters in this structural optimization where m = mass

problem are height and number of storeys (𝒉, 𝒏𝒔 ), bay c = damping coefficient
widths (a,b), number of bays in each direction, steel k = stiffness
grades, seismic weight, seismic intensity (peak ground
acceleration, behavior factor, ground type, response The effective earthquake force is expressed in the
spectra). Equation 4 shown below. For this study, the member
size of columns and beams were checked for
The minimum member sizes is 0.25m and the displacement when the building is under seismic
minimum column member sizes is 0.30m. The optimal excitation.
solution is defined as the structure with the minimum
weight satisfying the design criteria of code. The
Figure 2.0 is the Five - storey model reinforced 𝒌𝒙 = 𝒎𝝎₂ 𝒙 = 𝒎𝝎𝑹 (Equation 4)

ELEVATION
Roof
3.5m
Floor 5
3.5m
Floor 4
3.5m h
Floor 3
3.5m
Floor 2
4m
Ground

6m 7m 6m
b

concrete building.
Figure 3. Model 5-Storey RC

60

5th Floor
4th Floor
3rd Floor
2nd Floor
1st Floor
The dynamic response of the equivalent system is
represented by the modes of vibration being equivalent
to the number of masses as shown in Figure 4.
Figure 4. Lumped mass model with mass and stiffness
More specifically, it aimed to understand the
structural behavior of buildings in Mindanao. The
purpose of this effort is to help reduce the impact of
future large quakes in the country particularly in the
seismically active areas in the Philippines. It is an
advantage for engineers to understand the recent
earthquake effects related to geology, ground motion,
earthquake intensity, and magnitude to address the
advancement in developing building codes and
designs that require site-specific analysis in the
Philippines.
II. METHODOLOGY
2.1 Structural Optimization
Structural optimization is important, Cohn and
Dinovitzer in 1994, who showed the gap between
theories and application on designs. The Structural
Engineering Institute 2002 pointed out the economical
design of RC structures.
The traditional and heuristic are two different
approaches, which was separated in two main
branches in the field of structural optimization.
Genetic algorithm is an effective, heuristic search
method (Goldberg, 1989) was used to generate high-
quality solutions to obtain best performance design
and savings generated thereafter. The heuristic
approach followed the computerized pathway with the
progress of several algorithms, bee colonies, threshold
accepting, stimulated annealing etc.
2.2 Optimization algorithm: Genetic algorithm
Genetic Algorithm (GA) is inspired by the process
of natural evolution (Goldberg 1989). In this
optimization approach, a population of individuals
(variables) is evolved to produce best performance of
designs. The evolution process in this article is one
hundred generations continue. The evolution
progresses in generations. In each generation, the
fitness of individuals are selected randomly modified
to produce the next generation; and continue until a
maximum number of generations are formed or a
satisfactory level of fitness is achieved.
MATLAB
OPTIMIZATION METHOD
Design Variables
DESIGN CANDIDATES
Constraint
s
Fitness value
EARTHQUAKE or
SIMULATOR Performance
value
(STAAD Pro
software)
EQ
Records
Figure 5. – Procedure of Design Optimization
The Figure 5 above shows the procedure for design
optimization. For a 100 design candidate, the average
performance of population test. After initial
population, reproduction occurs, wherein the
performance of best candidate may go higher among
the candidate of100. It will reproduce, thus second
generation population is obtained. The possible best
candidate in third generation can have same
performance than other generation and so on. Then, it
will go higher or maintained population. Thus
generation can reach improved and best performance
as it progresses. The performance of design candidate
goes up. The best performing design can be obtained
from final generation. These design candidates will be
an output.
The design candidates are shown in the figure
above as results of analysis of design using STAAD
software. At each step, the genetic algorithm uses the
current population to create the children that make up
the next generation. The algorithm selects a group of
61
individuals in the current population, called parents,
who contribute their genes—the entries of their
vectors—to their children. The numerical algorithm
was developed in MATLAB software to obtain the
best design performance. The drifts in x – direction is
shown below in Table 2.

Table 2. Drifts (mm) X-direction

Five-Storey Height Displace- Elastic NSCP NSCP Adequacy

Building (m) ment Interstorey Inelastic Allowable
(mm) Drifts (mm) Interstorey Storey Drifts
Drift (mm) (mm) Figure 7. Response Spectra of Different Major
Earthquakes

Level X-Dir X-Dir X-Dir X-Dir Satisfied

with Mean Response Spectrum
RoofDeck 18 14.702 3.353 19.95 70 Satisfied

5th Floor 14.5 11.349 2.446 14.55 70 Satisfied

The construction of the Response Spectrum
4th Floor 11 8.903 3.278 19.50 70 Satisfied
Analysis as defined in NSCP 2010 Section 208.5.3.4.1
3rd Floor 7.5 5.625 2.178 12.96 70 is shown in Figure 8.
Satisfied

2nd Floor 4 3.447 3.447 20.51 80 Satisfied

Grd 0 0 0 0 0

𝑥𝑡

Figure 8. NSCP 2010 Design Response Spectra

RSA is a procedure for the dynamic analysis and

structural behavior of buildings under seismic
excitation. In Figure 9, engineers use this model single
line lollipop with circles representing the building’s
(a) (b) mass. Then, the design response spectra can be
obtained by plotting the maximum response of
acceleration.
Figure 6. Inter-story drift
(a) Three-dimensional view. (b) Top view.
.

The Figure 6 above shows the three dimensional

view and top view of inter-story drift.

The study made use of the direct integration scheme

that was carried out numerically in computers for
general dynamic loading solution techniques. Thus,
conversion of the specific strong motion records was
solved using integration scheme in SEISMOSIGNAL
2016 software. A numerical method was implemented
using computer programs called STAAD.Pro for actual
building modeled in 3D for the site-specific response
Figure 9. Construction of Response Spectrum
spectrum analysis of buildings. The design spectrum is
constructed to represent the mean recorded significant The attainment of seismic safety evaluation and
earthquakes from recorded instruments as shown in building structures designs to resist future earthquakes
Figure 7.

62
would be possible if the design spectrum satisfies
specific requirements.
This also presents analysis of building structures
using Response Spectrum Method for the design
optimization of building structures. For this study, the
member size of columns and beams were checked for
displacement when the building is under seismic
excitation. The following figure shows the typical
behavior of the genetic algorithm. The whole fitness
assignment, selection, recombination and mutation
process is repeated until a stopping criterion is satisfied.
The evolution process starts with randomly created
first generation from site-specific earthquake records of
Mindanao earthquake records.
(b)
The recorded earthquakes were documented from
the “Operation and Maintenance of Strong Motion Figure 10. Best Fitness Scores per generation
Accelerograph Network of PHIVOLCS last October
2019 and November 2019.
In this study the dynamic analysis using Staad The design candidates are shown in the figure
software was communicated to MATLAB software for below as results of analysis of design using
optimization methods. Once MATLAB software gives STAAD.Pro software. At each step, the genetic
the candidate, the STAAD automatically open to algorithm uses the current population to create the
compute, then results was obtained. Then it will closed. children that make up the next generation. The
STAAD.Pro dynamic analysis is activated once called algorithm selects a group of individuals in the current
by MATLAB software. The program calls directly the population, called parents, who contribute their
results of run analysis of STAAD.Pro of software and genes—the entries of their vectors—to their children.
automatically alters the values of variables. The optimization method was used in this study
through MATLAB optimization software.
The dynamic analysis using response spectrum
III. Results and Discussion method is carried out in this five story building to assess
the seismic behavior when subjected to earthquake
The optimization process of RC beams and loadings. The result of calculated displacements or
columns were carried out in order to search for an each floor level is shown below in Figure 11.
optimum solution, the outputs of the optimization
process, minimization of the displacement versus the
number of generation presented in Figure 10. The
graph shows the plot of best fitness of the average x 5=xt
performance of population test. Result shows the best x5
performance is reached after one hundred generation. 5th Floor
x4
4th Floor
x3
3rd Floor
x2
2nd Floor
x1
1st Floor

Figure 11. Displacement for each level

The Elastic Inter-Storey Drift are the

following:
Dx1 = x1 = 3.45 mm
Dx2 = x2 - x1 = 2.18 mm
(a)
Dx3 = x3 - x2 = 3.27 mm
Dx4 = x4 – x3 = 2.45 mm
Dx5 = x5 – x4 = 3.35 mm

63
 The inelastic drift according to the NSCP code is
satisfied. The important part of the spectrum that needs
attention is the maximum peak response of strong
ground motions that might affect the structural
behavior of buildings.
The final displacement/drift response was checked
against code limits. The allowable drift limits are
satisfied. Thus, building with its structural
configuration such as member sizes, column and beam
quantities and locations are effectively proportioned.
IV. CONCLUSION
The results obtained from the final generation of
genetic algorithm progresses until it reach its final
generation. It is found that genetic algorithm is capable
for the optimization of structural sizes of reinforced
concrete beams and column wherein the best design is
chosen out of one hundred population, the design
performance goes up. Thus, the optimal building design
subjected to seismic loading can be achieved to create
resilient designs and economical designs to protect life
and property.
ACKNOWLEDGMENT
The authors wish to acknowledge ERDT, DOST,
SEAD and PHIVOLCS for allowing to use the
waveform data in this study produced under the project
entitled “Operation and Maintenance of Strong Motion
Accelerograph Network of PHIVOLCS.” MSU- Iligan
Institute of Technology and to those who in one way or
another extend their help for the success of this study.
REFERENCES
[1] Atkinson, G.A. and Boore, D.M. Emperical ground
motions relations for sub-ductions-zone earthquakes
and application to Cascadia and other regions.
Bu,93(4):1703-1729, April 2008.
[2] Boore David M, The Motion of the Ground in
Earthquakes, Scientific American, Vol. 237, No. 6. pp
68-81, December 1977.
[3] Chopra A.K. Dynamics of Structures: Theory and
Applications to Earthquake Engineering, Prentice-
Hall.1955.
[4] Clough R.W., Penzien J. Dynamics of Structures, 2nd
Edition, McGraw Hill.1994
[5] National Structural Code of the Philippines, 7th edition,
Volume 1, 2015, NSCP C101-10 ISBN 2094-5477,
Association of Structural Engineers of the Philippines
(ASEP), 2015.
[6] FEMA Publication 451, NEHRP Recommended
Provisions for New Buildings and Other Structures:
Training and Instructional Materials FEMA451B-
June2007
[7] Handbook to the Uniform Building Code: an
Illustrative Commentary. Whittier, Calif.: International
Conference of Building Officials, c1988.
[8] T. Guendelman, G.R. Saragoni, R. Verdugo.
CHILEAN EMERGENCY SEISMIC DESIGN CODE
FOR BUILDING AFTER EL MAULE 2010
EARTHQUAKE 15th World Conference on
Earthquake Engineering at Lisboa. 2012.
[9] https://openknowledge.worldbank.org/handle/10986/8
748
[10] M. Ashchheim, E. Hernandez-Montes, L.M. Gil-
Martin. Design of Optimally Reinforced RC Beam,
Column, and Wall Sections. Journal of Structural
Engineering. Vol. 134, No. 2, February 2008.
[11] I. Rahmanian, Y. Lucet, S. Tesfamariam. Optimal
Design of reinforced concrete beams: A review.
Techno-Press, Ltd. Computers and Concrete, Vol. 13,
No. 4. 2014
[12] K. Tajziehchi, A. Ghabussi, H. Alizadeh. Control and
Optimization against earthquake by using genetic
algorithm. Journal of Applied Engineering Sciences.
Vol. 8(21) Issue 1/2018/
[14] Structural Engineers Association of Davao, Inc.
64
 Forensic Wind Hazard Analysis of Typhoon Rolly
Joshua C. Agar1, a * ,Jaime Y. Hernandez Jr., 1,b ,Timothy John S. Acosta, 1,c
1
Institute of Civil Engineering, University of the Philippines, Diliman, Quezon City
a
jcagar@up.edu.ph, bjyhernandez@up.edu.ph, ctsacosta@up.edu.ph,

Abstract—Typhoon Rolly was regarded as the strongest determining the low-pressure field and the wind field of the typhoon.
tropical cyclone in 2020. At peak intensity, it made landfall to the Also, as part of the forensic analysis, the research aimed to perform
Bicol region in the morning of November 1, however the peak forensic structural analysis on documented damages to verify the
estimates from various weather agencies differed, with PAGASA magnitude of the wind hazard.
estimating 10-minute sustained winds of 225 kph and the JTWC
estimating 1-minute sustained winds of 315 kph. Differences in the II. METHODOLOGY
wind speed estimates of various weather agencies have led the
research into performing forensic analysis on both the ground A. Pressure Field Modelling
surface data and the documented damages in order to determine the
magnitude of the wind hazard brought by Typhoon Rolly. As Typhoon Rolly approached landfall, the barometric
pressure drop brought by the approaching typhoon was closely
Keywords—typhoons, forensic analysis, typhoon monitored across the Automated Weather Stations (AWS) of
modelling, severe wind, Monte Carlo simulation, Typhoon Goni, PAGASA. To capture the strength of Typhoon Rolly at the time of
failure simulation, computational fluid dynamics, structural analysis its first landfall at Catanduanes at peak intensity, the AWS on Pili,
Legazpi, Guinobatan, Juban, and Catarman were monitored.
I. INTRODUCTION
Lying on what is called the typhoon belt, the Philippines
is frequented by typhoons by about 20 typhoons a year. Out of 20
typhoons, an average of 5 typhoons make landfall in the Philippines,
bringing hazards like storm surges, floods, and severe winds.
Among the regions frequently hit by strong typhoons was the Bicol
Peninsula. The region's shared history with strong typhoons has the
residents of the region experienced devastating chapters like
Typhoon Sening (International Name: Joan) in 1970 where the
synoptic station in Virac, Catanduanes has recorded then-record
wind speeds of 277.2 kph. It would be soon supplanted by the
current record of 320 kph recorded also on Virac during Typhoon
Reming (International Name: Durian) in 2006.
Fig. 1. PAGASA Automated Weather Stations Monitored
Since Typhoon Yolanda (International Name: Haiyan),
the Philippine Atmospheric, Geophysical, and Astronomical
Services Administration (PAGASA) has introduced Public Storm
Signal #5 to accommodate the newly introduced category called
super typhoons whose 10-minute sustained wind speed was more
than 220 kph.

In November 2020, PAGASA has raised Public Storm

Signal #5 for the first time in Bicol region, particularly in
Catanduanes, as Typhoon Rolly (International Name: Goni)
approached. PAGASA estimated the 10-minute sustained winds of
Typhoon Rolly to be at 225 kph, while the Japan Meteorological
Agency (JMA) estimated 10-minute sustained winds of 220 kph.

By the time it made landfall at 5:00 am (November 1

PST), Typhoon Rolly was regarded as the strongest landfalling
tropical cyclone based from the Advanced Dvorak Technique
(ADT) satellite estimates, made by Joint Typhoon Warning Center
(JTWC), of 1-minute sustained winds of 315 kph and minimum
central pressure of 884 hPa. [1]
Evaluating the magnitude of the wind hazard is a vital
component into assessing the risk brought by a typhoon to localities,
to evaluate sufficiently the degree of priorities on disaster risk
management. To adequately perform the risk assessment, the
objectives of this research was to perform forensic analysis in
Fig. 2. Time History of the Barometric Pressure of the AWSs
(Source: PAGASA)
The distance of each of the monitored AWS with respect
to the center of the storm across time were then established through
the correlation of the position of the typhoon with respect to time –
which was obtained from the storm track of PAGASA – and the
position of the stations.

65
 As a result, a scatter plot relating barometric pressure with
respect to the distance of the center of the storm was obtained as
shown in Figure 2.
Fig. 2. Barometric Pressure with respect to distance from the
storm center.
To normalize the pressure field of Typhoon Rolly in
order to distinguish the pressure drop from the ambient pressure,
the research adapted Schloemer's hyperbolic model in 1954 [2]:
−𝐴
𝑝(𝑟) = 𝑝0 + (𝑝𝑛 − 𝑝0 )𝑒𝑥𝑝⁡( ) (1)
𝑟𝐵
Where p(r) is the barometric pressure at distance r,⁡𝑝0 is
the minimum central pressure of the storm, 𝑝𝑛 (Pa) is the ambient
pressure, and A and B are the shape parameters of the storm.
B. Regional Wind Field Modelling
The winds comprising the circulation of the tropical
cyclone can be divided into two: winds generated by the primary
circulation and winds generated by the second circulation. The
primary circulation is primarily driven by the balance between the
Coriolis forces – which is the fictitious inertial force experienced by
a moving body of fluid due to the rotation of the Earth – the
centrifugal force experienced by the rotating body of fluid, and the
force caused by the pressure gradient brought by the low pressure
center of the storm, whilst the secondary circulation is primarily
driven by the heat exchange between the lower and upper
atmospheres within a cyclone.
Focusing on the primary circulation, the regional winds of
the typhoon were then determined by achieving the balance between
the pressure gradient forces, the centrifugal force, and the Coriolis
force, which in equation form is known as the gradient wind
equation:
𝑈2 1 𝜕𝑝
𝑓|𝑈| + = (2)
𝑟 𝜌 𝜕𝑟
Where U is the gradient wind speed, r is the radius of
𝜕𝑝
curvature, 𝜌 is the air density, and is the pressure gradient.
𝜕𝑟
Substituting (1) to (2) yielded the gradient wind speed as
a function of the distance from the storm center:
𝑓𝑟 (𝑓𝑟)2 𝐴𝐵(𝑝𝑛 −𝑝0 ) −𝐴
𝑈⁡(𝑟) = ⁡ − +√ + 𝑒𝑥𝑝⁡(
2 4 𝜌𝑟 𝐵
The maximum winds 𝑈𝑚𝑎𝑥 and the location of the
maximum winds with respect to the center of the storm 𝑅𝑚𝑎𝑥 was
determined using the equation proposed by Holland in 1980 [3]:
𝑅𝑚𝑎𝑥 = 𝐴1/𝐵
𝐵(𝑝𝑛 −𝑝0 )
𝑈𝑚𝑎𝑥 = ⁡ √
𝜌𝑒
C. Wind Exposure and Local Winds
Fig. 4. Regional Wind (Green) vs the Actual Local recorded
wind speeds
Wind exposure plays a factor in either amplifying or
reducing the magnitude of the winds as the regional winds are
converted to local winds experienced by the area. Gustiness – a
fluctuating increase in magnitude due to turbulence effects,
increases as the wind transited from the sea to land. Shielding occurs
when a feature of land or a huge structure covers the areas on its
leeward side. Topographic amplification occurs when the slope of a
terrain causes contraction of the wind flow leading to the increase.
Legazpi City, for example, is only exposed to winds
coming from the east, as the Mayon Volcano shields the city from
winds coming from west. Thus, Legazpi City is only exposed to
severe winds brought if the typhoon passed over the areas south of
the city. Historically, the city experienced its record maximum wind
of 234 kph when Typhoon Sisang (International Name: Nina) passed
south of the city in 1987. Typhoon Tisoy (International Name:
Kammuri) last year, which has a regional wind of 170 kph, traversed
a similar path as Typhoon Rolly in the Lagonoy Gulf. Virac
Synoptic Station recorded peak gust of 212.2 kph during the passage
of Typhoon Tisoy, while Legazpi Synoptic Station recorded peak
gust of 151.2 kph.
As Typhoon Rolly traversed across Lagonoy Gulf, the
eastern and southern coastal towns of Catanduanes and the coastal
towns along the coast of Lagonoy Gulf experienced the full brunt of
the winds.
D. Forensic Structural Analysis
Another way to establish the magnitude of the wind hazard
brought by Typhoon Rolly was through forensic analysis of the
damage incurred by Typhoon Rolly.
) (3)
𝑟𝐵
Fig. 5. A frame from the captured footage of the entire roof
system of a school building in Catanduanes being blown away
during Typhoon Rolly.
(4) The structure was assumed to be a standard DepEd school
building. The wind flow around the structure was simulated using
(5)
Computational Fluid Dynamics (CFD) in order to obtain the wind
pressure experienced by the structure.
66
Fig. 6. Computational Domain for the CFD Analysis

Judging from the footage, the wind was assumed to strike

at an angle of 60 degrees with respect to the length of the structure.
The research has used Exposure Category V from the wind code Fig. 8. Structural Model
from the Architectural Institute of Japan (Japan) [4] in determining
the boundary layer profile over the inlet. To model the turbulence, Another structure subjected to the forensic analysis is the
the SST k-omega model was used. [5] [6] Catanduanes National High School (CNHS) situated near Virac
airport. Figure 9 shows the extent of the structural damage inflicted
The wind loads determined in the CFD analysis were then by Typhoon Rolly with purlins and roof sheet at the last two spans
passed on to the structural model where a subsequent structural of the roof structure.
analysis was performed in order to determine the roof-to-column
connection forces. These forces were then compared to the
resistance capacities of the corresponding roof-to-column
connection in order to assess its failure.

The research assumed that the roof system was mainly

supported by a protruded rebar which were bent and hooked over the
roof framing. The research adopted Doguiles (2015) resistance
capacities against unbending which has a lognormal probabilistic
distribution curve with mean of 3070 N and standard deviation of
1290 N [7]. The research also adapted the resistance capacities from
the study of Bisa (2013) for the roof fasteners [8], which have mean
of 894.5 N and standard deviation of 85.5849 N.

Fig. 9. Catanduanes National High School (From: DepEd)

A Monte Carlo scheme was implemented in the failure simulation
with 200 building cases at each increment of wind speed. The result
was the relationship between the wind speed and the probability of
the event captured in footage happening.
Unlike the previous case, no footage was at hand to ascertain the
angle of approach of winds. To simulate the wind flow around the
structure at certain angles of approach, CFD analysis was performed
on the school building, whose hallway was facing northeast.

Fig. 7. Coefficient of Pressures across the building surface Fig. 10. CFD Analysis of CNHS at 45 degrees wrt the structure (top)
and at 30 degrees wrt the length of structure (bottom) shown on the
length.

Following the component-based approach in failure simulation by

Tan (2017) on DepEd school buildings, each structural component

67
was assessed whether its resistance capacities were exceeded by the
wind load that it was subjected to. For the purlins, assuming that
they were adequate braced, two scenarios were assessed and
observed: (1) whether the yielding capacity of 248 MPa was
exceeded due to bending or (2) the purlin connection assembly with
capacity of 232.5 kN was exceeded. For the roof fasteners, which
were spaced by 0.2 m, the pull-out failure criteria was being assessed
whether each fastener exceeded the tested capacity of 538.7 N with
Coefficient of variance of 0.1. [9]
Figure 11. Details of the Component-Based Approach. Two truss
spans were analyzed, with each span having 22 purlins. Roof
fasteners represented were by dots.
Figure 12. Failure Simulation on the Purlins on the Leftmost span at
winds approaching 45 degrees wrt to the length of the building. All
the purlins at the leeward side experienced failure at 61 m/s
Both spans were subjected to the failure simulation. The lower
bound scenario indicated that all of the purlins at one side of one
span sustained bending failure and the breach in the building
envelope had led to the progression of damage to the other side and
spans whilst the upper bound scenario indicated that all purlins at
one side of both spans suffered onset failure alone.
To further eliminate the possibilities about the wind approach, the
wind pressures experienced by the eaves were evaluated also in the
CFD analysis. The failure of panels on the eaves, which were
possible only if negative pressures were experienced, were also
observed at CNHS
15 deg 30 deg
45 deg 60deg
75deg
Figure 13. Coefficient of Pressures on the Eaves at Scenarios where
the winds approaches from the north or from the east. The angles are
wrt to the length of the structure.
III. RESULTS AND DISCUSSION
The fitted values of A and B for (1) were determined to be
929028.5844 and 1.363260857 respectively with a goodness of fit
𝑟 2 to be equal to 0.952. The pressure profile fitting also determined
that the minimum central pressure of Typhoon Rolly at its first
landfall was 905.37 hPa, almost the same as the estimates of
PAGASA and JMA. Using (4), the maximum regional wind speed
of Typhoon Rolly was determined to be equal to 229.6 kph, which
is assumed to be 10-minute wind speed, closer to the estimates of
PAGASA and JMA also.
Fig.14. Probability of total roof failure of the first case
The forensic analysis determined that the structural failure
shown in Figure 5 shows probability of happening at a minimum of
68
251 kph, where the first cases of the said failure were documented ACKNOWLEDGEMENT
to have happened. However, this may reflect to possible sub- The main author thanks the PAGASA for providing the
standard construction practice since the structural code suggested historical data for the synoptic stations along with the time history
that the building should withstand wind speeds up to the 320 kph of the data from the AWS during the passage of Typhoon Rolly.
[10].
Collaborating with PAGASA in the Severe Wind Hazard
TABLE I. Failure Simulation Results - CNHS and Risk Assessment program is the UP Institute of Civil
Engineering, under the research project titled "Enhanced Severe
Wind Vulnerability Curves of Key Building Types in the
Angle wrt to Failure Simulation Philippines". Thru the project, the authors are able to collaborate
the length of under this independent research venture.
the structure
[deg] Failure Simulated Lower Upper The main author also acknowledges the Twitter user
Bound Bound @chinitsko for his footage of the failure of school building analyzed
[m/s] [m/s] in this research.

15 Purlin failure 70 81
REFERENCES

30 Purlin failure 63 71 [1] Masters, J., “Super Typhoon Goni Slams Into Philippines As
Strongest Landfalling Tropical Cyclone On Record » Yale
Climate Connections”. [online] Yale Climate Connections.
45 Purlin failure 61 72 2020 Available at:
<https://yaleclimateconnections.org/2020/11/super-typhoon-
goni-slams-into-philippines-as-strongest-landfalling-tropical-
cyclone-on-record/> [Accessed 20 November 2020].
60 Purlin failure 69 74 [2] R. W. Schloemer, “Analysis and synthesis of hurricane wind
patterns over Lake Okechobee”. FL. Hydromet Rep. 31, pp. 49.
75 Purlin failure 71 80 1954
[3] G. Holland, “An Analytic Model of the Wind and Pressure
Profiles in Hurricanes”

[4] Y. Tamura, K. Hiromasa, Y. Uematsu, H. Okada, and T.

The scenario at Catanduanes National High School, on the
other hand, could have happened if the gusts were at least 61 m/s
(219.6 kph) with a likely scenario, based from the failure sustained,
experiencing peak gust of 81 m/s (291.6 kph).
On the other hand, the evaluation of wind pressures on the
eaves, which suffered failure, suggested scenarios where the winds
were coming from the south or west when CNHS suffered the said
structural failure.
Overall, this is in good agreement to the estimates of the
regional wind field modelling, although the wind exposure also
played a vital role in the reduction of the wind speeds experienced
by the areas south of Typhoon Rolly. The areas north and west of
Typhoon Rolly, however, being exposed to the severe winds, may
have experienced, if factoring in the gustiness [11] with a gust factor
of 1.36, a peak gust of 298.48 kph, which is close to the worst
scenario for Catanduanes National High School.
IV. CONCLUSION
The forensic analysis of both the ground data and the
documented damage of a chosen structure determined that Typhoon
Rolly has 10-minute sustained winds of 229.6 kph and minimum
central pressure of 905.37 hPa, both in agreement with the analysis
and forecast of PAGASA and JMA rather than the JTWC and the
Dvorak technique estimates.
Ohkuma, "Documents for windd resistant deisgn of buildings
in Japan". 2004
[5] F. R. Menter, "Zonal Two Equation k-ω Turbulence Models for
Aerodynamic Flows", AIAA Paper 93-2906. 1993.
[6] F. R. Menter, ,"Two-Equation Eddy-Viscosity Turbulence
Models for Engineering Applications", AIAA Journal, vol. 32,
no 8. pp. 1598-1605. 1994.
[7] A. Doguiles “Failure Modeling of Roof-Truss Supports
Reinforced Concrete Frame Residential Buildings due to
Severe Wind”. Thesis. University of the Philippines – Diliman,
Institute of Civil Engineering. 2016
[8] H. Bisa, . "Determination of Empirical Threshold Values of
Roof Connections Using Pull-Out Tests of Nails and Screws. "
Thesis. University of the Philippines – Diliman, Institute of
Civil Engineering. 2013
[9] L.R.E. Tan , "Wind Vulnerability Assessment of Enhanced
Wind Resilient Design for a Standard 1-Storey School Building
in Tacloban Using Component-Based Approach in Reliability
Modeling". MS Thesis. University of the Philippines –
Diliman, Institute of Civil Engineering. 2017
[10] Association of Structural Engineers in the Philippines,
"National Structural Code of the Philippines". 2015
[11] B. A. Harper, J.D. Kepert, and J.D. Ginger, "Guidelines for
Converting Between Various Wind Averaging Periods in
Tropical Cyclone Conditions". World Meteorological
Organization. 2010.

To verify the estimated peak gust, on the other hand, the

synoptic wind data from Virac Station is needed.

69
 Environmental Analysis and Estimation of
Evaporation Rate using Pennman Numerical
Simulation in the Salt Production Industry in
Occidental Mindoro
Michelle Enriquez, a *,
michelle_d_enriquez@dlsu.edu.ph
Abstract—Salt or sodium chloride has been in use as a
flavoring substance to improve the taste; or as preservatives for
meat, fish and other food products. The salt production is
considered as one of the oldest industry in the world ranging from
the traditional solar evaporation to advanced technology-based
harvesting method. Occidental Mindoro was once known in salt
production industry contributing almost 18% of the country’s salt
production. But the industry experienced losses, as climate change
impacts continue to worsen the weather patterns. As the province
is trying to reclaim its reputation in salt industry, salt farmers are
continuously seeking ways on increasing their harvest and ensure
production year-round and even during the off-season. Historical
analysis is presented in this study to determine which among the
physical, environmental/ecological and biological system has a
great impact on the salt production, characterizing the processes in
the solar production was presented to understand how each stages
is affected by climatic patterns. Results of the study shows that
meteorological factors like temperature, wind speed, atmospheric
pressure, wind speed and humidity greatly affects the evaporation
rate. Higher temperature, minimum wind speed, lower humidity
and less rainfall increases the evaporation rate thus speeds up the
evaporation process and produces more salt).
Keywords— Environmental analysis, evaporation rate,
Pennmann- Numerical solution, salt production
I. INTRODUCTION
Salt is considered as one of the important economic
mineral that is widely distributed on all the continents occurring in
large reserves [1]. Salt production is one of the most ancient and
widely distributed industries in the world where almost 120
countries have salt producing facilities ranging from primitive solar
evaporation to advanced and multi stage [2]. The annual world
production of salt can exceed to 200 million tons with more than
one-third produced by solar evaporation of sea water and inland
brines, and the rest by mining of rock salt [3].
Salt industry in the Philippines was once a vibrant
industry. In 1990, the provinces of Bulacan, Pangasinan,
Occidental Mindoro and Cavite supplied almost the 85% of the
country’s annual salt requirement. In 1994, the country can still
produce 300,000 metric tons of salt. But in 2010, salt production
drastically dropped to less than half and significantly resulted to
the importing of salt [4].
Occidental Mindoro was once one of the biggest salt
producers in the country. In 1990, it produced and contributed
about 60,000 metric tons of the 338,000MT or 18% of the
country’s total salt production [5]. Today, while Occidental
Mindoro still supplies salt in the neighbouring provinces in
Southern Luzon, Visayas, and Mindanao, only 12 percent of the
national salt requirement comes from the province with the biggest
chunk of the demand supplied by imported salts.
The province, being an island and surrounded by water,
has one of the longest coastline but is greatly affected by the
climate change and decimated the salt industry. The traditional
method of producing salt used in the province depends primarily
on solar evaporation is affected adversely by erratic weather
patterns. The physical and chemical composition of salt produced
varies dependently in many factors: production techniques,
processes and the climatic conditions. Salt production through
solar production is mainly dependent on dry season.
As the province is trying to reclaim its reputation in salt
industry, salt farmers are continuously seeking ways on increasing
their harvest and ensure production year-round and even during the
off-season. Seeing the potential of the salt industry to drive the
local economy of the province, Occidental Mindoro Salt Council
was created in 2018 which pieced together a roadmap for the
Occidental Mindoro Salt industry with the end view of developing
a salt industry that meets the domestic and international standards
in order to provide equitable benefits by using socially responsible
and environment-friendly practices. The roadmap envisioned that
by 2022, the province of Occidental Mindoro shall be known as the
Salt Capital of the Philippines producing high quality salt products
at competitive prices for the domestic market catering to the
consumers, food processors, industrial and agricultural users [6].
As climate pattern in the province changes in time, salt
industry experiences fluctuation on the volume of salt produced
affecting salt farmers and traders. Longer dry season increases the
salt production while rainfall and continuous rain flooded the salt
beds. The objective of this paper is to understand the salt
production processes in the province through characterization of
the physical and environmental factor and determine how climatic
conditions affect the salt industry.
Historical analysis is presented in this study to determine
which among the physical, environmental/ecological and
biological system has a great impact on the salt production. This
study aims to present the existing condition of salt industry in
Occidental Mindoro using the data obtained in three weather
stations: Barrio Site, Toong and Laste. Specifically, it aims to:
characterize the physical and environment system of the salt farms;
present the processes involved in salt production; determine the
evaporation rate using the climatic characteristics identified in the
three sites; and dentify the evaporation rate in the salt production.
70
 I. METHODOLOGY
A. Study Area
In the three sites where salt farms and the weather
stations are situated, salt beds are located adjacent to the coast, and
salt beds are slight higher than the normal high water line. Sitio
Toong and Laste are located in the municipality of Magsaysay,
Occidental Mindoro, both facing the China Sea with vast plains
and valleys. Slope ranges from level to very steep sloping areas;
highest point of elevation is 543 ft. above sea level.
Figure 1. Location map of the salt farms
Laste is situated at approximately 12.253 lat and
121.1741 long with elevation of 2.7m or 8.9ft above mean sea level
[7]. Salt beds can be seen approximately from 216.17m. While
Sitio Toong is located in Brgy. Caguray, Magsaysay, Occidental
Mindoro lying at 12.2828 lat and 121.1185 long with elevation
estimated at 7.0m or 23.0 feet above mean sea level and with
distance of approximately 552.72m from the coastline (PhilAtlas,
2019)
On the other hand, Sitio Barrio site, is situated in
Labangan Poblacion, San Jose lying along at nearly 12.3582 lat
and 121.0776 long, and elevation estimated at 9.9m or 32.5feet
above mean sea level [8].
B. Pennman Evaporation rate
To characterize the physical and environment system in
thee three sites, an environmental scanning is conducted to
determine the prevailing characteristics in the salt farms.
Meteorological data from three (3) weather stations: Laste, Sition
Toong and Barrio Site from January 16, 2019 to May 29, 2019 are
used to estimate the evaporation rate on the salt farms.
A historical background in terms of annual volume of
salt produced is presented as part of the site analysis. The method
used including the preparation, harvesting and transportation of salt
is further identified to describe the processed of salt production.
The climatic factors like proximity to sea water,
elevation, soil type and other hydrological parameters were
identified to determine the evaporation rate. Numerical Simulation
using the Pennman equation as modified by [9] is used to
determine the evaporation rate for the four-month period.
𝛥 𝛾
𝜆𝐸 = 𝑅𝑛 + 𝑓(𝑢)(𝑒𝑠 − 𝑒) (1)
𝛥+ 𝛾 𝛥+ 𝛾
The evaporation rate is expressed in mm/day; λ is the
latent heat of vaporization; Δ is the gradient of the vapor pressure-
temperature curve; γ is the psychometric constant; Rn is the net
solar radiation; f(u) is the function of wind speed’ es and e are the
saturation vapor pressure of water and ambient water vapor
pressure [10].
III. RESULTS AND DISCUSSION
A. Management of Salt Production
A.1 Preparation of beds
Most of the salt farms in Occidental Mindoro are small
scale. Estimated total salt farms in San Jose and Magsaysay,
Occidental Mindoro is approximately 2,000 hectares. From this
land areas, there are 35,544 estimated salt beds own and manage
by almost 50 salt producers in the province [11].
The Tamaraw Salt Producers Cooperative [11].leads the
salt production industry in the province of Occidental Mindoro.
Salt farms are nearly located where salt water are accessible and
can easily enter the salt beds. Areas with mudflats slightly above
the high tidal range are leveled and compacted during the dry
season. Each areas is divided into salt pans arranged as reservoir,
condenser and crystallizer.
Salt farms are designed considering the natural
characteristics of the selected site and prevailing climatic
conditions. The salt farms are consists of shallow ponds ranging
from 15-60cm supplied by the sea water stored in the reservoir.
The shallow pond is divided into evaporation ponds where sea
water is concentrated up to saturation until NaCl salinity is
obtained. While, the crystallization ponds is where the salt
crystallizes by evaporation of the brine up to 28-29°Be. The
evaporation ponds constitute 90% of the production area and the
remaining 10% used as crystallization ponds, where salt is
harvested.
A.2 Stocking of Seawater
Stocking of sea water in the reservoir is done through
dikes connected to rivers or canals. The design of the salt pans
considers the level of tide to ensure that sea water will flow freely
through the gates of the reservoir. The seawater from the reservoir
flows into the condenser pan. Then water flows through the inlet
canal placed in the condenser to the crystallizer and only high
saline water is transferred from condenser to the crystallizer, then
71
the salt crystals start to form. The accumulation of salt increases
by the rate of evaporation in the environment. It is continuously
evaporated by solar heating and wind flow. As the water
evaporates, the salinity rises and salt crystals formed. When the
water dries up, the salt crystals are now ready to harvest by raking
or sweeping.
The table significantly shows that much of the salt
production is bought by big traders and buyers and only small
portion by TAMACO. As stated by the cooperative, it is because of
the higher price that the traders offered for each bag of salt. And
also one of the reason why the cooperative is appealing for
establishing a fix price of salt per bag in the province.

A.3 Harvesting and Production Rate A.4 Transportation of salt

Sea water is stored in the reservoir for almost 2 weeks,
then evaporated off from the direct sunlight and let the remaining
sea water dense with salt. The gate of the reservoir is then opened
and allowed to flow in the salt beds. It is then dried up for more
than 3 hours until salt crystals are seen in the surface. The salt
crystals are harvested by raking or sweeping. Harvesting of salt is a
difficult activity since it entails direct exposure to sun. From the
salt farmers in the province, there are almost 4,035 regular salt
farm workers harvesting the salt produced (TAMACO, 2019).
Figure 2 shows the production rate from 2008 to 2017.
The salt harvested in each salt farms brought by Traders,
other buyers and TAMACO are stored in salt warehouses and
transported outside the province. A Salt Washer located in
Mapaya, Magsaysay was owned and operated by TAMACO that
can washed the raw salt to obtain a whiter salt granules free of dirt
and other impurities. TAMACO allowed other salt farmers to use
the salt washer to ensure that a cleaner salt is transported and
delivered.
A.5 Marketing and the Price Fluctuation

Price of salt per bags is dominated by the presence of

traders and other buyers in the province and seasonality. Prices
Salt Production Volume: offered by the traders is higher compared to that of the cooperative.
2008-2017 Since most of the salt farmers are small scale, they opted to sell
their harvest to traders while, during dry season prices of salt bags
1,200,000
dropped and increases during off season. Salt producers having
PRODUCTION VOLUME (BAGS)

warehouses, stored their harvests and disposed when the price

1,000,000
escalates. Figure 3 shows the summary of variation of prices
during the 4month period of January to May, 2019
800,000

600,000
Price Variation per bag from
400,000
January-May 2019
250.00

200,000
200.00

-
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 150.00

Figure 2. Volume of Salt produced from 2008 to 2017 100.00

(TAMACO, 2018)
50.00
The figure shows that volume of salt in bags produced in
the province increases from 2008 to 2011 reaching to more than
1,100,00 bags but an almost 20% sudden decreased was -
Dec 31-Jan. Feb. 4- 10 Feb 18-24 March 4-7, March 18- May 13- 31
experienced in 2012; with a gradual increased from 2013 to 2016. 20 April 28
Moreover, a recent data on the volume of salt produced in bags Sibalat Caguray/Toong
Toong Mapaya
was obtained by the researcher during the 4-month period the same
duration of the meteorological data obtained in the three sites.
Table 1 shows the volume of salt produced based on the four-
month period bags sold Salt Traders and TAMACO. Figure 3. Variation of price of salt bags for 4 month
period (TAMACO, 2019)
Table 1. Volume of Production for 4-month period
The figure shows that as dry season is experienced in the
Month Estimated Estimated TAMACO Total province, the price of the salt per bags drastically drops as low as
Purchase by purchase actual 100 pesos. This can affects the small scale salt farmers’ operation
Big Traders by other purchase cost. The cause of price variation may be attributed to the over
during buyers/ production during dry season, dominance of middlemen or traders,
Production warehouse the lack of government policy for fixed price rating and the lack of
Season (bags) support to empower the salt farmer.
January 14,000 96,000 110,000
February 193,000 119,768 34,463 347,232
March 168,600 535,647 43,506 747,753
April 547, 950 135,940 40,220 724,110
June 126, 450 195,274 16,552 338,276
Total 1,050,000 1,082,629 134,742 2,267,371

72
B. Factors affecting Salt Production
B.1 Environmental Condition
A thorough environmental investigation is needed to
assess the acceptability of a site for salt production. Assessment is
barely based on the natural features of the surrounding
environment. The salt farm areas in Occidental Mindoro are
mainly found in the coastal wetlands. The availability and
accessibility of salt water is first to consider in determining the
viability for site production. Soil condition is also considers in the
site selection as it affects the soil permeability that determines the
capacity of the underlying soil to hold water. Seepage rate is also
consider as it is dependent on the soil type and soil particle size.
Soil type in the three sites are sandy clay and silty clay.
B.2 Seasonality
Climate plays an important role in sea salt production.
The variation in rainfall patterns as affected by the climate change
greatly affect the production rate of salt industry. The province’s
erratic rainfall patterns causes flooding in the salt beds. Salt
production process was then affected as sea beds were inundated
by water. Moreover, temperature also affects the salt production;
the higher the temperature, the more salt can be produced. Figure 4
shows the rainfall amount m average temperature and wind speed
from 2009-2018 in Occidental Mindoro. Meteorological data
during the month of April and May were obtained and considered
in the analysis since evaporation rate is significantly high during
dry seasons.
Average Annual Rainfall(mm)
temperature (C) and wind
speed (m/s)
3000.0
Rainfall amount (mm)
2500.0
2000.0
1500.0
1000.0
500.0
0.0
rainfall amount temperature (C)
wind speed
Figure 4. Average rainfall amount (mm), temperature and
average wind speed in Occidental Mindoro
(PAG-ASA, 2019)
The figure shows that the average temperature in the
province varies from 30 to 33°C from 2008. It also means that heat
index in the province is becoming warmer. While average wind
speed ranges from 2.5 to 5 m/s. The rainfall pattern from 2008 to
2018 varies significantly. This may be attributed to the erratic
changes in the weather pattern in the province. With the increasing
temperature and minimum speed, the rate of evaporation is
estimated to increase.
Moreover, the annual meteorological data obtained were
used to determine the average annual evaporation rate. Minimum
evaporation rate is during the month when the cloud cover is high
and average wind velocity is near the annual minimum. While,
maximum evaporation rate is during the dry season covering the
month of March to May when the temperature is high, skies are
clear and wind speed is high. Occidental Mindoro experiences high
temperature or dry season during the month of November to April
with high relative humidity and rainy seasons during the month of
May to October..
Pennman equation of evaporation rate is used in this
study to obtain the average annual evaporation rate in the province
and it is then compared against the salt production rate. Data from
including the wind speed, humidity and temperature were
simulated using numerical solution. Figure 5 shows the annual
evaporation rate against the production rate of salt industry in the
Evaporation rate vs. production
rate
1,200,000 9
8
1,000,000
VOLUME OF SALT PRODUCED (BAGS)
7
800,000 6
5
600,000
4
35
400,000 3
30
25 2
20 200,000
1
15
10 Annu
- 0
al
5 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Produ
0 ctio…
province from 2008-2017.
Figure 5. Evaporation rate vs. production rate from 2008-2017
The table shows a significant relationship of evaporation
rate in the salt production in the province. It can be observed when
the evaporation is high, the production rate is also high. With the
climatic characteristics obtained including the mean sea level
height, temperature, wind speed, humidity, and pressure were
simulated to estimate the evaporation rate. The results of the
simulation shows that when temperature is high, and the wind
speed is minimum, relative humidity is low, the resulting
evaporation rate is high.
Furthermore, the same analysis was obtained when the
meteorological data obtained from the three sites were simulated to
determine the recent evaporation rate and its significant impact on
production rate from January to May, 2019. Figure 6 shows the
same behavior of the estimated evaporation rate.
73
 Table 2. Summary of Evaporation rates
Location Evaporation rate: Evaporation rate:
min vale (mm/day) max value
(mm/day)
Barrio Site 3.7756 14.2650
Laste 4.9875 12.2523
Toong 9.7789 11.8849
The table significantly shows the variation of
evaporation rates in the three sites. The simulation also proved that
evaporation rates are high during the dry months with maximum
values during April and May as minimum wind speed is also
observed
The behavior of evaporation in Laste differs in Toong, it
is because the wind blows directly coming from the coast in
Toong. Laste, although small in distance with the coastline, is
surrounded by mangroves that block the wind to directly hit the
salt beds. Further, Barrio Site is situated in the inner part of the
Labangan Poblacion surrounded by almost residential and
commercial structures. Evaporation rate play an important factor in
salt production. Temperature, relative humidity, wind velocity,
atmospheric pressure can sped up or slows down the evaporation
rate.
IV. CONCLUSION
This study presented the processes in the solar salt
production existing in the province of Occidental Mindoro. Salt
farms are designed considering the ecological characteristics and
the prevailing climatic patterns in the salt farms. The solar salt
production in the province is consist of reservoir, condenser and
crystallizer. Ninety percent (90%) of the total land area is consider
as evaporation ponds and the 10% as the crystallization pond
where the salt is harvested.
Salt harvested in the province is either bought by traders,
other buyers and TAMACO and being sold or transported locally
and outside the province. Price of salt per bags varies significantly
on the volume of salt produced, dry and off-season and lack of
policy on fixed price per bags affecting the small scale salt farmers.
in this study, and DOST: PAG-ASA for the meteorological data
used in the simulation.
REFERENCES
[1] Affam M., Asamoah D.N., 2011. Economic Potential of Salt
Mining in Ghana Towards the Oil Find. Research Journal of
Environmental and Earth Sciences 3(5):448-456
[2] Hossain, Shahadat M., Hossain, Zakir M., Chowdhury
Sayedur Rahman. 2006. An Analysis of Economic and
Environmental Issues Associated with Sea Salt Production in
Bangladesh and Thailand Cost. International Journal of
ecology and Environmental Science 32: 159-172
[3] Sedivy, Vladimir M. 2009. Environmental Balance of Salt
Production Speaks in favor of Solar Saltworks. Global Nest
Journal 11 (1): 41-48
[4] Acedo, Celia E., Hontucan, Royanni Miel. 2017. PH Salt
Industry Reeling from Climate Change. Conference of the
Philippines Population Association.
[5] Business Mirror, 2017. Occidental Midnoro boosting salt
indsyutry through new technology.
https://businessmirror.com.ph/occidental-mindoro-boosting-
salt-industry-through-new-tech
[6] Department of Trade and Industry. 2018. Salt Industry
Roadmap 2017-2022. www.dti.com.ph
[7] PhilAtlas, 2019. Magsaysay, Occidental Mindoro.
http://www.philatlas.com
[8] PhilAtlas, 2019. San Jose, Occidental Mindoro.
http://www.philatlas.com
[9] Akridge, Glen D. 2008. Methods for calculating brine
evaporation rates during salt production. Journal of
Archaeological Science 35(2008) 1453-1462
[10] Penman, H.L., 1948. Natural evaporation from open water,
bare soil, and grass. Proceedings of the Royal Society, Series
A 193, 120e145.
[11] Tamaraw Salt Producers of Occidental Mindoro, 2019.
Conditions of salt farms in the province of occidental
Mindoro.

Results of the simulations show that evaporation rate is

mainly dependent on the temperature, rainfall amount, humidity,
wind speed, atmospheric pressure and elevation. The study
revealed that high temperature, minimum wind speed, lower
humidity and less rainfall amount can speed up the evaporation
rate. Moreover, evaporation rate has significant impact on the salt
production. Higher evaporation rate can speed up the
crystallization process, thus more salt can be produced.

Considering other methods in estimating the evaporation

rate can be used in the study to further proved the analysis
obtained. Investigation on the engineering factors as to the efficient
and effective area for salt beds should be done to provide a basis
on optimizing the salt production processes. Moreover, a least-cost
analysis should be done to investigate the benefits of introducing
technology-based equipment in the different stages of the salt
production

ACKNOWLEDGEMENT

The author would like to thanks the Tamaraw Salt

Cooperative (TAMACO) and JALD Incorporated for the data used

74
 Simulating the Impacts of Deficit Irrigation on Corn
Yield and Water Productivity Using Coupled DSSAT
CERES-Maize and Aquacrop Models
Marilyn S. Painagan1, a * and Victor B. Ella2,b
1
Department of Agricultural & Biosystems Engineering, College of Engineering and Information Technology, University of
Southern Mindanao, Kabacan, Cotabato.
2
Land and Water Resources Division, Institute of Agricultural and Biosystems Engineering, College of Engineering and Agro-
Industrial Technology, University of the Philippines Los Baños, Laguna.
a*
mspainagan@usm.edu.ph, bvbella@up.edu.ph
Abstract—This study aimed to simulate the effects of deficit
irrigation or intentional under-irrigation on corn yield and water
productivity using DSSAT CERES-Maize and AquaCrop Models.
The crop simulation models were calibrated and validated using
observed crop production data. Both models performed
satisfactorily during calibration and validation based on standard
statistical indices such as RMSE, NRMSE, R2, d-stat and NSE. The
calibrated and validated models were then coupled and used to
simulate the effect of various levels of deficit irrigation on corn
yield. Model simulation results showed that a 15 % level of deficit
irrigation results to maximum yield while a 60% level of deficit
irrigation leads to maximum water productivity. Results suggest that
maintenance of soil moisture very close to field capacity does not
necessarily lead to increased crop yield and water productivity. A
net application depth of 77 mm proved to be optimum when a 60%
deficit irrigation is imposed. Model simulation results also showed
that tasseling stage and silking stage until grain filling stage are the
most susceptible to water stress. This study demonstrated that deficit
irrigation is a practical water saving management strategy. The use
of coupled crop simulation model (DSSAT CERES- Maize and
AquaCrop) proved to be a viable tool to support decision making in
corn production systems in the Philippines, especially when
imposing deficit irrigation.
Keywords— AquaCrop, Crop simulation, Deficit Irrigation, DSSAT
CERES-Maize, Water productivity
I. INTRODUCTION
A global challenge for the agricultural sector is to produce
more food with less water (Toumi, 2016) to sustain the rapidly
growing world population. This problem is due to the increasing
demand of water use in both non-agricultural and agricultural
sectors. Nowadays, increasing population, increasing industries, and
the adverse effect of changing climatic condition aggravate the
problem on water scarcity.
Water scarcity leads to water stress, which in turn affects
crop growth and productivity in many ways (Ertek and Kara, 2013).
Inadequate irrigation water application leads to adverse
physiological effects like stunted growth, poor quality and reduce
quantity of crop yield. Water scarcity problems also affect water
productivity, which is basically defined as the ratio of the mass of
crop yield to the amount of irrigation water consumed. Hence,
innovations, strategies, and plans to increase water use efficiency in
crop production systems are necessary.
To address these issues of water scarcity, a number of
water saving technologies and management strategies have evolved
through the years. These include an irrigation strategy called deficit
irrigation (DI), which is basically defined as the intentional under-
irrigation of crops below full crop water requirements (crop
evapotranspiration, ETc) (Fereres and Soriano, 2007). Deficit
irrigation is a well-accepted practice to optimize and increase water
use by allowing crops to withstand mild water stress with none or
only marginal decrease in yield and quality (Costa et al., 2007;
Geerts and Raes, 2009). Yield reduction is generally expected when
crops are subjected to deficit irrigation (DI) (Lamm et al., 1995), yet
a well-designed deficit irrigation (DI) regime can minimize the
impact on yield and still lead to growers’ profitability (Fereres and
Soriano, 2007; Rodrigues et al., 2013). Deficit irrigation, coupled
with crop simulation models to investigate multiple alternatives,
have a pivotal role to play in sustainable water development (Wang
et al., 2009). In developing countries such as the Philippines,
however, no study on deficit irrigation can be found in published
peer-reviewed literature much less the application of crop models
for quantifying the effects of deficit irrigation on crop yield and
water productivity.
Application of crop simulation models is unquestionably
a sound approach in examining the effect of deficit irrigation on crop
yield and water productivity, particularly for corn production
systems in the Philippines. Corn, particularly sweet corn, is one of
the most important crops that are produced perennially in the
Philippines. However, sweet corn growers in the Philippines usually
employ surface irrigation applied on a weekly basis if rainfall does
not occur, and hence the soil moisture content is maintained near
field capacity. Consequently, farmers tend to use large amount of
irrigation water, which leads to huge water application losses and
wastage.
In recent years, simulation modelling studies for corn have
been conducted in the Philippines. For instance, Bondad et al. (2015)
simulated the impact of conservation agriculture on corn yield using
DSSAT CERES-Maize model. Lansigan and Salvacion (2015) made
assessments of the impact of climate change on rice and corn yields
using CERES-Maize and Painagan and Guiterrez (2014) conducted
a simulation study of corn yield using the Aquacrop model.
However, to date no study has been published in peer-reviewed
literature on the application of both the FAO’s AquaCrop water
productivity model and DSSAT CERES-Maize in simulating the
impact of deficit irrigation system on corn yield and water
productivity in the Philippines.
This study aimed to simulate and analyze the effect of
deficit irrigation on corn yield and water productivity by coupling
AquaCrop (4.7) and DSSAT CERES-Maize crop simulation
models.
I. METHODOLOGY
75
 To serve as basis for model calibration and validation, field
experiments were conducted at the University of the Philippines,
Central Experimental Station, Los Baños, Laguna, Philippines,
(14.1800 N 121.2500 E) with an elevation of 21 m above sea level,
from March to May 2018 and from October to December 2018 for
the first and second cropping seasons, respectively.
The climate at the experimental site falls under Type I
climate based on the Corona classification, with two pronounced
seasons, dry from November to April, and wet during the rest of the
year. It has an average annual precipitation and temperature of 1,942
mm and 27.1°C respectively. During the year 2018, the driest month
was March with a total rainfall of only 22.5 mm which occurred
during the first cropping season. Most of the precipitation during this
year fell in the month of December, averaging 450 mm due to the
occurrence of strong typhoons. The soil at the experimental site is
clay loam classified as Lipa clay loam for the upper 25 cm layer.
The site is nearly flat, well drained and has an average water content
at field capacity and permanent wilting point of 0.32 and 0.14 m3
m-3, respectively.
The climatological input data such as daily maximum and
minimum temperature, rainfall, relative humidity, dry and wet bulb
temperature, wind speed, sunshine duration, and vapor pressure
were collected from the National Agrometeorological Station,
University of the Philippines Los Baños, located at about 500 meters
from the study area (Figure 1).
Figure1. Location of Philippine Agrometeorological Station and
study area using google earth map.
Soil data such as textural class, field capacity, permanent wilting
point and bulk density were obtained from the CHED-PCARI
project (Ella, 2019). Data for soil water content were collected at the
start of the planting season, two days after every irrigation
application, and during harvest through gravimetric method, and
were counterchecked using a soil moisture meter probe.
Crop management information, such as plant population, row
spacing, seeding depth, application of fertilizer, and irrigation
application were also recorded. Plant phenological growth
parameters were also recorded. These included emergence date,
anthesis date, plant height, maximum rooting depth, observation of
maximum canopy expansion and dry above ground biomass which
were measured through destructive sampling, harvest date and fresh
sweet corn ear physical characteristics.
Five irrigation treatments were evaluated for the entire duration of
the study, using randomized complete block design replicated three
times, a total of 15 experimental plots. Summarized below are the
irrigation treatments:
Treatment 1: 40% depletion of the available water
Treatment 2: 50% depletion of the available water
Treatment 3: 60% depletion of the available water
Treatment 4: 70% depletion of the available water
Treatment 5: 80% depletion of the available water
A. Model Calibration
Calibration of the DSSAT CERES–Maize model was performed
using the observed cultivar-specific coefficients of sweet corn under
Los Baños, Laguna condition using the localized model input data
taken from the October to December 2018 cropping season, when
soil water was optimal for crop growth. It was also the season when
plants did not experience stress. For this study, the localized model
input data included the following: daily weather data, parameters of
the soil profile, local management condition, anthesis day which was
measured in days after planting (DAP), physiological maturity day
measured in DAP, stem weight at maturity (kg(dm) ha-1), leaf
weight at maturity (kg(dm) ha-1) and emergence day measured in
DAP. In order to fine-tune the crop model, manual iterative process
of adjusting crop genetic coefficient values of sweet corn was done
until simulated results corresponded to the observed data.
The AquaCrop (4.7) model was also calibrated for local conditions
using data taken from the October to December 2018 cropping.
Biomass was measured in tons/ha and the percent soil water content
for the first 0.25 m depth of the soil profile was used for calibration.
A trial and error approach was used in adjusting the non-
conservative parameters to minimize the difference between the
simulated and measured data.
For both models, the goodness of fit between simulated and
observed data during model calibration was assessed using standard
statistical indices such as root mean square error (RMSE),
Normalized root mean square error (NRMSE), coefficient of
determination R2 ,Willmott’s index of agreement (d-stat), and Nash-
Sutcliffe model efficiency coefficient (NSE).
B. Crop Model Validation
Field data sets for all treatments from the March to May 2018
cropping season were used for model validation. Fresh ear weight
(kg(fm) ha-1), leaf and stem weight (kg(dm) ha-1) were considered
to validate the DSSAT CERES-Maize model; whereas, AquaCrop
(4.7) model was validated using fresh ear weight (kg(fm) ha-1), dry
above-ground biomass (ton ha-1), and percent soil water content
data.
The goodness of fit between simulated and observed data during
model validation was similarly assessed using the same statistical
indices used during model calibration.
C. Model Simulation
Several levels of deficit irrigation were simulated using the coupled
DSSAT CERES- Maize and AquaCrop model, ranging from 10% to
100% allowable depletion of the available soil water content
(ASWC). Several irrigation depths of the optimum deficit irrigation
were also simulated using the coupled model. Climatological data
and soil data of Los Baños, Laguna were used during simulation.
The AquaCrop model was used to simulate irrigation schedules as
well as net irrigation depths. The generated irrigation depths and
irrigation schedule from AquaCrop model were then used as input
data for the DSSAT CERES-Maize model. The DSSAT CERES-
Maize model was then used to simulate the responses of a maize
crop subjected to several deficit irrigation levels. Crop responses
such as grain yield and dry above-ground biomass at several deficit
irrigation levels were then determined.
D.Water Productivity Determination
76
Water productivity is defined as the ratio of the mass of yield to the Table 2. Statistical measures of the goodness of fit between
volume of water consumed by the crop usually expressed as kg m-3 observed and simulated growth parameters during calibration of
(Geerts et al., 2009): DSSAT CERES- Maize model.
𝑌
Eq.6 𝑊𝑃 = 𝑎
𝐸𝑇𝑎
Where: No. of RMSE NRMSE R2 d-stat NSE
𝑌𝑎 = 𝑚𝑎𝑠𝑠 𝑜𝑓 𝑦𝑖𝑒𝑙𝑑 (𝑘𝑔) dataset (%)
𝐸𝑇𝑎 = 𝑣𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 𝑤𝑎𝑡𝑒𝑟 𝑐𝑜𝑛𝑠𝑢𝑚𝑒𝑑 𝑏𝑦 𝑡ℎ𝑒 𝑐𝑟𝑜𝑝 (𝑚3 )
Silking 3 0.8165 1.60
day day
In this study, water productivity was defined as the grain produced
(DAP)
in kg. per m3 of water used. The numerator was expressed as the
marketable yield, whereas, the denominator was expressed as the
Maturity 9 2.2 24.44
amount of water applied (the sum of rainfall and irrigation).
day days
(DAP)

III. RESULTS AND DISCUSSION Leaf dry 5 115.8 8.49 0.75 0.927 0.736
mass
(kg/ha)
A. DSSAT CERES- Maize model calibration Stem 5 118.61 9.245 0.97 0.992 0.97
During model calibration six genetic coefficient for maize were dry mass
parameterized for local Philippine conditions (Table 1). These six (kg/ha)
maize genetic coefficients were as follows: P1 is the thermal time
from seedling emergence to the end of the juvenile phase; P2 is the Also, the value of genetic coefficient P5 which is the thermal time
extent to which development is delayed for each hour increase in from silking to physiological maturity was estimated based on actual
photoperiod; P5 is the thermal time from silking to physiological observation of maturity during field experiments. Maturity in sweet
maturity; G2 is the maximum possible number of kernels per plant; corn was referred to as the milk stage which occurs during 70 to 80
G3 is the kernel filling rate; and PHINT is the phylochron interval, days after planting, based on sweet corn’s genetic characteristics.
the interval in thermal time between successive leaf tip appearances. During field experiments, maturity stage (milk stage) was observed
beginning 72 days after planting, as maturity progressed sugar will
Table 1. Estimated genetic coefficients of sweet corn under local be converted into starch, therefore, sweet corn was harvested at 75
Philippine conditions. days after planting. P5 values of 463 to 481 degree days with an
average of 472 degree days were the predicted physiological
Variab Unit GSS09 Satur BSS09 Calibrate maturity with a minimum RMSE value of 2.2 days and NRMSE
le 66 sh2 n sh2 77 sh2 d value of 24.44% (Table 2). As a result, a P5 value of 472.0 was used.
Coefficie Mean leaf and stem dry weight obtained from the 15 experimental
nt plots were 1,364 kg ha-1 and 1,283 kg ha-1, respectively. In the
Default sweet corn genetic process of iteration, it was observed that a G2 (maximum possible
coefficient number of kernels per plant) value of 550 and G3 (the kernel filling
P1 Growing 150 180 175 164.0 rate during linear fill stage under optimal conditions) value of 3 gave
degree the minimum value of RMSE which is 115.8 and 118.61 kg ha-1 for
days leaf and stem dry weight, respectively (Table 2). Phylochron interval
(GDD) (PHINT) as well as P2 (extent to which development is delayed for
P2 expressed 0.30 0.30 0.30 0.30 each hour increase in photoperiod) were kept constant, since
as days photoperiodism does not affect the crop’s development.
Calibrated genetic coefficients for sweet corn were compared with
P5 expressed the genetic coefficients of GSS0966 sh2, Saturn sh2 and BSS0977
in degree 680 738 700 472.0 sh2. These sweet corn varieties were pre- calibrated by DSSAT and
days used as default sweet corn genetic coefficients of crop simulation
above a model CERES- Maize. The genetic coefficient values of sweet corn
base calibrated under local conditions are within the range of the default
temperatu values as shown in Table 1.
re of 80C Results showed a high degree of goodness of fit between simulated
G2 number 600 850 500 550.0 and observed data. Dry weight of leaf and stem were predicted with
G3 mg/day 5.50 15 5.0 3.0 R2 of 0.75 and 0.97, d value of 0.927 and 0.992 and NSE of 0.736
PHINT degree 43.0 35 50 43.0 and 0.97, respectively, as shown in Table 2 and Figure 2.
days

During field experiments, occurrence of silk was observed during

50-52 days after planting, wherein during 52 DAP was the maximum
occurrence of silk. P1 values of 154 to 174.59 with an average of
164.3 predicted anthesis day (52 DAP) yielded a minimum RMSE
value of 0.8165 day and NRMSE value of 1.60% (Table 2). As a
result, a P1 value of 164.0 GDD was used as the genetic coefficient
of sweet corn.

77
Figure 2. Simulated and observed dry mass of leaf and stem during Number of plants 185000 25000
calibration of DSSAT CERES- Maize model. per hectare

B. Calibration of AquaCrop (4.7) model Minimum effective 0.30 0.30

rooting depth (m)
The main parameters used to calibrate AquaCrop (4.7) model under Maximum 1 1
local conditions are shown in Table 3. It also shows the default effective rooting
values in the AquaCrop program files. The non- conservative depth (m)
parameters of the AquaCrop (4.7) were calibrated using the data
collected during the October to December 2018 cropping season.
Calibrated parameters were based on observations during the actual
Calibrated values are mostly lower than default values since sweet
field experiments. Maximum canopy cover was set at 70%,
corn variety of maize is a short season crop.
maximum rooting depth was set at 53 DAP, days from sowing to
flowering was set at 52 DAP, building up of harvest index starting
Table 3.Default and calibrated values of non-conservative
parameters used during the calibration of AquaCrop model. from flowering was set at 22 DAP as per observation when plant
started to bear fruit. Number of plants per hectare was set at 25,000
based on planting density, days from sowing until the start of
AquaCrop non- Default values Calibrated values senescence was set at 70 DAP and physiological maturity of sweet
conservative corn was set at 80 DAP.
parameters Results of model calibration are shown in Table 4 and Figure 3.
Maximum canopy 0.80 0.70 Biomass observed values were taken during 50 and 70 DAP of
cover (CCx) in October to December 2018 cropping season. A high goodness of fit
fraction soil cover between simulated and observed dry above-ground biomass and soil
water content was obtained. Calibration based on dry above-ground
Calendar Days: 5 5 biomass showed an R2 value of 0.901, d-stat value of 0.895, RMSE
from sowing to of 0.226 ton/ha, NRMSE of 5.1% and NSE value of 0.98. On the
emergence other hand, calibration based on observed values of soil water
content which were taken during 40, 50, 60 and 70 DAP of the same
Calendar Days: 100 53 cropping season, showed an R2 value of 0.897, RMSE of 5.2mm,
from sowing to NRMSE of 1.9%, d-stat value of 0.845 and NSE value of 0.69.
maximum rooting
depth Table 4. Statistical measures of the goodness of fit between observed
and simulated dry above-ground biomass and soil water content
Calendar Days: 110 70 during calibration of AquaCrop model.
from sowing to
start senescence
Variab Mean Mean RM NRM R2 d- NS
Calendar Days: 125 80 les obser simula SE SE stat E
from sowing to ved ted val
maturity (length of ue
crop cycle) Above 4.437 4.439 0.22 5.1% 0.9 0.8 0.9
- 6 01 95 8
Calendar Days: 70 52 groun ton
from sowing to d ha-1
flowering bioma
ss
Length of the 10 10 (ton
flowering stage ha-1)
(days)
Soil 275.4 280.1 5.2 1.9% 0.8 0.8 0.6
Building up of 50 22 water mm 97 45 9
Harvest Index conten
starting at t (mm)
flowering (days)

Water Productivity 17.0 33.7

normalized for
ETo and CO2
(WP*) (gram/m2)

Crop performance 50 50
under elevated
atmospheric CO2
concentration (%)
Figure 3. Simulated and observed dry above-ground biomass and
Reference Harvest 50 45 soil water content during calibration of AquaCrop model.
Index (HIo) (%)

78
C. Validation of Calibrated Models
Models were validated by simulating the biomass and fresh ear yield
of sweet corn under different levels of deficit irrigation. The
observed field experiment data from March to May 2018 cropping
season was used to validate the crop simulation models. Table 5
shows the simulation result of fresh fruit yield using DSSAT
CERES- Maize models and AquaCrop. Both DSSAT CERES-
Maize and AquaCrop exhibited good performance in simulating the
fresh fruit yield of sweet corn with R2 values of 0.9529 and 0.8707,
respectively (Figure 4 & 5).
Table 5. Comparison of simulated and observed fresh ear mass
during validation of DSSAT & AquaCrop models during model
validation.
Figure 4. Observed and simulated values of fresh sweet corn ear
mass using DSSAT CERES-Maize model during model validation
Figure 5. Observed and simulated values of fresh sweet corn ear
mass using AquaCrop model during model validation.
DSSAT CERES- Maize over predicted the fresh ear weight at high
irrigation application (40% depletion of the Available Water). This
result is also similar to the reported performance of CERES- Maize
by Lizaso et al., (2007). The study reported that “the model
consistently over-predicted fresh ear weight across the range of
values examined”. However, AquaCrop also over-predicted fresh
ear weight at low irrigation application (80% depletion of the
Available Water). This performance of AquaCrop is comparable to
the study of Greaves and Wang, (2016) when they assess AquaCrop
in simulating maize growth and productivity under tropical
environment, they reported that the model tended to overestimate
the biomass accumulated, with the deviation becoming more distinct
in the DI treatments especially in water stress treatment, and
therefore, they concluded that as water stress intensifies model
reliability lessens (Greaves and Wang, 2017). Nevertheless, the
models still generally performed satisfactorily based on the standard
statistical indices as shown in Table 6 and 7.
Table 6. Statistical indicators of the goodness of fit between
observed and simulated dry above-ground biomass using AquaCrop
model under different levels of deficit irrigation during model
validation.
Table 7. Statistical indicators of the goodness of fit between
observed and simulated leaf dry mass during model calibration of
DSSAT-CERES Maize.
C. Model Simulation
Nineteen levels of deficit irrigation, or percent depletion between
field capacity and permanent wilting point, with an interval of 5%
were simulated using the calibrated and validated models. The
depths of irrigation used in the simulations were generated using the
AquaCrop model, which in turn, were used to run the simulation
using DSSAT CERES-Maize model. In the simulation process
deficit irrigation was applied throughout the growing season.
Simulation results for the various levels of DI are shown in Figure
6. Results showed that 15% DI and 20% DI gave the highest grain
yield at an average yield of 5,267 kg/ha and 5,216 kg/ha,
respectively. These results are plausible since crops under 15% DI
and 20% DI did not demonstrate any observable water stress during
the field experiments. Similarly, 10% DI did not show any sign of
water stress, and yet gave relatively lower yield compared to 15%
and 20% DI.
79
Figure 6. Simulated grain yield at different levels of deficit irrigation
using DSSAT CERES-Maize.
Simulation results also showed that 60% DI gave relatively higher
yield than 40% DI, 45% DI, 50% DI, 60% DI and 70% DI. On the
basis of crop yield effects alone, model simulation results suggest
that while increasing levels of deficit irrigation or levels of depletion
generally lead to yield reduction, maintenance of soil moisture very
close to field capacity does not necessarily lead to increased crop
yield.
In terms of water productivity, model simulation results showed that
60% DI obtained the highest irrigation water productivity with 3.24
kg m-3. In this study, water productivity was measured as the ratio
of marketable yield and the amount of water used by the crop
throughout its growing season. Among all DI treatments, the 60%
DI utilized the least amount of water and still obtained a high yield,
making it the most productive in terms of water consumption.
Moreover, 50% DI and 70% DI obtained the next highest irrigation
water productivity with 2.51 kg m-3 and 2.50 kg m-3, respectively.
On the other hand, 40% DI obtained the lowest irrigation water
productivity with 1.50 kg m-3. These results showed that a small
amount of irrigation water is capable of producing high yield.
Results also showed that there are phenological stages of crops
wherein deficit irrigation has less impact on crop growth and yield.
It was found that V3 to V6 stages of corn were not sensitive to water
deficit. Furthermore, after the grain filling stage until harvest, corn
crop is no longer sensitive to water deficit. These results are
generally consistent with the findings of other studies (e.g. Denmead
and Shaw, 1960).
Under 60% deficit irrigation, different irrigation depths were applied
at different phenological stage of corn. (Table 8). Results showed
that a net irrigation of 77.3 mm that was scheduled during start of
tasseling to grain filling stage produced the highest grain yield of
5009 kg ha-1. Furthermore, a 69.9 mm of net irrigation that was
applied during tasseling to grain filling stage produced a grain yield
of 5006 kg ha-1. In terms of water productivity, an irrigation depth
of 6.3mm that was applied at the end of tasseling stage to grain
filling stage obtained the highest water productivity of 8.2 kg m-3.
These results highlight the importance of considering the crop’s
phenological stage when scheduling an irrigation. Results also
showed that tasseling stage, silking stage until grain filling stage are
stages that are susceptible to water stress. Stress can lower the crop’s
yield at high level if it occurs during these stages.
Table 8. Irrigation depths, grain yield, water productivity and
irrigation schedule under 60% deficit irrigation.
CONCLUSION
Successful imposition of deficit irrigation strategy can be achieved
if supported with a decision support system like crop simulation
models. In terms of yield, model simulation results showed that 15%
DI and 20% DI provide the highest grain yield at an average of 5,267
kg/ha and 5,216 kg/ha, respectively. The 10% DI did not show any
sign of water stress, and yet gave a relatively lower yield compared
to 15% and 20% DI. Results suggest that maintenance of soil
moisture very close to field capacity does not necessarily lead to
increased crop yield. In terms of water productivity, model
simulation results showed that 60% DI generate the highest water
productivity at 3.24 kg/m3. Among all DI treatments, the 60% DI
utilized the least amount of water and still obtained a high yield.
Therefore, a small amount of irrigation application can still lead to
increased water use efficiency and crop productivity, as long as it is
being applied at the right phenological stage of a crop. Results also
showed that there are phenological stages of crops wherein deficit
irrigation has less impact on crop growth and yield. It was found out
that V3 to V6 stages of corn were not sensitive to water deficit.
Furthermore, results showed that growth stages after the grain filling
stage until harvest corn crop are no longer sensitive to water deficit.
Model simulation results also showed that tasseling stage and silking
stage until grain filling stage are stages that are most susceptible to
water stress, which consequently affects the quality and quantity of
crop yield.
This study demonstrated that deficit irrigation is a practical water
saving management strategy. Furthermore, the use coupled crop
simulation model (DSSAT CERES- Maize and AquaCrop) proved
to be a viable tool to support decision making in corn production
systems in the Philippines, especially when imposing deficit
irrigation.
It is recommended to have further data collection and consider
external parameters to further verify the model.
ACKNOWLEDGEMENT
This study was supported by the CHED-PCARI project
entitled “Development of wireless sensor network-based water
information system for efficient irrigation water management in the
Philippines” and the Engineering Research and Development
Technology (ERDT) under DOST-SEI. Special acknowledgement
to Dr. Aurelio A. delos Reyes, Dr. Ronaldo B. Saludes and Dr.
Delfin Suministrado for your support and comments in this study.
REFERENCES
[1] BONDAD, R. M., ELLA, V. B., SALUDES, R. B., REYES,
M. R., & MERCADO, A. R. 2015. Simulating the impact of
conservation agriculture on corn yield in the Philippines using
the DSSAT CERES-Maize model. The Philippine Agricultural
Scientist, 98(2): 190-201.
[2] COSTA, J.M., ORTUÑO, M.F., CHAVES, M.M. 2007. Deficit
irrigation as a strategy to save water: physiology and potential
application to horticulture. J. Integr. Plant Biol. 49: 1421–1434.
[3] DENMEAD O.T. AND R.H. SHAW. 1960. The effects of Soil
Moisture Stress at Different Stages of Growth on the
Development and Yield of Corn. Agronomy Journal. J-3064:
272-274.
[4] ELLA, V.B. 2019. Annual report for the CHED-PCARI project
entitled "Development of Wireless Sensor Network-Based
Water Information System for Efficient Irrigation Water
Management in the Philippines".
[5] ERTEK A., B. KARA. 2013. Yield and quality of sweet corn
under deficit irrigation. Agricultural Water Management 129:
138– 144. http://dx.doi.org/10.1016/j.agwat.2013.07.012.
[6] FAO Reference Manual, Chapter 2. June 2012 – AquaCrop,
Version 4.0
[7] FERERES, E., SORIANO, M.A.2007. Deficit irrigation for
reducing agricultural water use. Special issue on ‘Integrated
approaches to sustain and improve plant production under
drought stress’ J. Exp. Bot. 58: 147–159.
[8] GEERTS, S., RAES, D., GARCIA, M., MIRANDA, R.,
CUSICANQUI, J.A., TABOADA, C., MENDOZA, J.,
80
 HUANCA, R., MAMANI, A., CONDORI, O. 2009.
Simulating yield response of quinoa to water availability with
AquaCrop. Agron. J. 101: 499–508.
[9] GEERTS, S., DIRK RAES. 2009. Deficit Irrigation as an on-
farm strategy to maximize crop water productivity in dry areas.
Agrl. Water management. 96: 1275-1284.
[10] GREAVES, G.E., & YU-MIN WANG. 2017. Effect of
regulated deficit irrigation scheduling on water use of corn in
southern Taiwan tropical environment. Agricultural Water
Management. 188: 115–125.
[11] GREAVES, G. E., & YU-MIN WANG. (2016). Assessment of
FAO AquaCrop model for simulating maize growth and
productivity under deficit irrigation in a tropical environment.
Water. 8(12): 557. doi:10.3390/w8120557
www.mdpi.com/journal/water
[12] HOOGENBOOM, G., JONES, J.W., PORTER, G.H.,
WILKENS, P.W., BOOTE, K.J., BATCHELOR,W.D.,
HUNT, L.A., TSUJI, G.Y. 2004. DSSAT v.4.0, vol. 1,
Overview. ICASA, Univ. of Hawaii, Honolulu, USA.
[13] HOOGENBOOM, G. 2000. Contribution of agrometeorology
to the simulation of crop production and its applications. Agric.
For. Meteorol. 103: 137–157.
[14] LAMM, F. R., MANGES, H. L., STONE, L. R., KHAN, A. H.,
& ROGERS, D. H. 1995. Water requirement of subsurface
drip-irrigated corn in Northwest Kansas. Transactions of the
ASAE. 38: 441–448.
[15] LANSIGAN, F. P., & SALVACION, A. R. 2015. Assessing
the Effect of Climate Change on Rice and Corn Yields in
Selected Provinces in the Philippines.
[16] LIZASO JI, BOOTE KJ, CHERR CM, SCHOLBERG JMS,
CASANOVA JJ, JUDGE J, JONES JW, HOOGENBOOM G.
2007. Developing a sweet corn simulation model to predict
fresh market yield and quality of ears. J Am Soc Hortic Sci.
132: 415–422.
[17] MAGAIA, E., S. FAMBA, I. WESSTRÖM, R. BRITO, A.
JOEL. (2017). Modelling maize yield response to plant density
and water and nitrogen supply in a semi-arid region. Field
Crops Research. 205: 170–181.
http://dx.doi.org/10.1016/j.fcr.2016.12.025
[18] PAINAGAN, M. S., GUTIERREZ, H. M. 2014. Simulation of
Corn Yield in Carmen, North Cotabato Philippines, using
AquaCrop Model. USM R&D journal. 22(1): 1-8.
[19] RODRIGUES, G.C., PAREDES, P., GONCALVES, J.M.,
ALVES, I., PEREIRA, L.S.2013. Comparing sprinkler and drip
irrigation systems for full and deficit irrigated maize using
multi criteria analysis and simulation modelling: ranking for
water saving vs. farm economic returns. Agric. Water Manag.
126: 85–96, http://dx.doi.org/10.1016/j.agwat.2013.05.005.
[20] Singh, J., Knapp, H. V., & Demissie, M. 2004. Hydrologic
modeling of the Iroquois river watershed using HSPF and
SWAT. Journal of the American Water Resources Association,
Illinois State Water Survey.
[21] TSUJI, G.Y., UEHARA, G., BALAS, S. 1994. (Eds.), Decision
Support System for Agrotechnology Transfer (DSSAT)
Version 3. University of Hawaii, Honolulu, Hawaii.
[22] WANG, Y.-M., NAMAONA, W., TRAORE, S., ZHANG, Z.
2009b. Seasonal temperature-based models for
evapotranspiration estimation under semi-arid conditions of
Malawi. Afr. J. Agric. Res. 4: 878–886.
[23] WILLMOTT, C. J. 1981. On the validation of models. Physical
Geography 2: 184–194.

81
 Chi and hypsometric analysis for watershed
prioritization and conservation
Imelida G. Torrefranca1, a *, Roland Emerito S. Otadoy2,3,b, and Alejandro F. Tongco 1,c
1
Engineering Graduate Program, School of Engineering, University of San Carlos – TC, Talamban, Cebu City
2
Department of Physics, University of San Carlos – TC, Talamban, Cebu City
3
Center for Geoinformatics and Environmental Solutions, University of San Carlos – TC, Talamban, Cebu City
a*
imetorrefranca@gmail.com, brsotadoy@usc.edu.ph, caftongco@usc.edu.ph

Abstract— The study presents a rapid assessment tool in
landform characterization. The method integrates the proxy for
steady-state river channel elevation called chi (χ) and the
hypsometric analysis using a 30-m SRTM DEM determined using
TopoToolbox (MATLAB) and further processed in GIS platforms.
χ gages the stability of drainage divides while hypsometric analysis
quantifies the stages of basin geological development. A quantity
called mean chi difference, 𝜒𝑚𝑑 , is proposed as a measure of the
degree of potential divide mobility. The disaggregation of large
basin into small hydrologic units enables the identification of areas
of different erosional stages. Highly erosional areas with highly
mobile divides are identified as areas of high priority for
conservation.
Keywords— chi, hypsometry, divide mobility, 𝜒𝑚𝑑 , basin
geological development, TopoToolbox, GIS
I. INTRODUCTION
River basins are hydrologically-defined landforms comprising
of drainage divides and stream channel networks. These landforms
are previously assumed to be static, however, observations of stream
capture [1, 2, 3], long-term erosion studies [4] and numerical
modelling [5, 6] demonstrate that basin shape and network topology
adjust in response to any perturbations to boundary conditions.
Geomorphic analysis would give us a glimpse of the dynamic nature
of these landforms and further provide us guidance on catchment
management process.
The 𝜒 – metric is a measure of drainage divide stability
fundamentally based on the ‘differential rates of river channel
erosion on opposite sides of a divide’ [7]. 𝜒 is the integral solution
to the stream power model, and is expressed as:
𝑚 ⁄𝑛
𝑥 𝐴
𝜒 = ∫𝑥 ( 𝑜 ) 𝑑𝑥 (1)
𝐴(𝑥)
𝑏
where 𝐴 is the upstream drainage area, 𝐴𝑜 is an arbitrary scaling
area, 𝑥 is the location of a point along the channel, 𝑥𝑏 is the base
level, and 𝑚 and 𝑛 are empirical constants. When disequilibrium
exists between opposing river channels, the shared drainage divide
is set in motion moving from lower to higher 𝜒 area until equilibrium
is reached. 𝜒 - maps provide a rapid visual assessment of divide
stability over a large area and also characterizes the variability of
river network’s topology and geometry in a river basin [8] as they
transmit tectonic and/or climatic signals throughout the landscape
[9].
Hypsometry quantifies the erosion proneness and stage of
geologic development of a basin [10]. Hypsometric analysis studies
the distribution of ground surface area, or horizontal cross-sectional
area, of a landmass with respect to elevation [11]. The hypsometric
curve and hypsometric integral (HI) of drainage basins are common
tectonic geomorphologic indices [12]. The geometry of the
hypsometric curve classifies a basin into different stages of geologic
development while HI represents the area under this curve and
expresses the volume of a basin that has not been eroded [13].
The study explores the integration of the 𝜒 -metric and the
commonly used hypsometric analysis as rapid assessment tools in
river basin management and prioritization as applied in the island of
Bohol, Central Philippines.
I. METHODOLOGY
The study comprises of two major analysis: the divide stabilty
analysis with χ and the hyspometric analysis. Both analysis use a
30-m Shuttle Radar Topography Mission (SRTM) digital elevation
model (https://earthexplorer.usgs.gov).
χ is computed using the DivideTools [14], a series of Matlab-
based tools for analyzing drainage divide stability built on top of
TopoToolbox [15]. Inputs to run the functions are a DEM and a flow
direction grid, and the parameter values for minimum accumulation
area and theta. As described by [14], the stability of a divide is based
on a criterion wherein the ‘mean of one side of the divide is within
one standard deviation of the mean of the other side’, otherwise a
divide is unstable. The mean chi difference, 𝜒𝑚𝑑 , is proposed to
represent the degree of potential divide mobility. It is computed as
|𝜒𝑚𝑒𝑎𝑛,1 − 𝜒𝑚𝑒𝑎𝑛,2 | where, 𝜒𝑚𝑒𝑎𝑛 is the mean values, while
subscripts 1 and 2 refer to the sides of the divide section. The 𝜒𝑚𝑑
values are classified into low, moderate and high with values of
<135, 135 – 255 and >255, respectively.
The hypsometric analysis starts with the delineation and
parameterization of basins and subbasins using a DEM with a
hydrologic model SWAT (www.swat.tamu.edu). Basins/subbasin
attributes are used to compute the hypsometric curve and
hypsometric integral (HI). The hypsometric curve is a plot of the
relative area (a/A) versus the relative altitude (h/H) where a, A, h
and H refer to area within the two adjacent contour lines, total area
of the basin or subbasin, contour interval, and relief of the basin or
subbasin, respectively. The value of a/A is 1.0 at the lowest point in
the basin where h/H=0, and it is 0.0 at the highest point of the basin
where h/H=1. HI is computed as (Zmean- Zmin )/(Zmax- Zmin) where
Zmean, Zmin, and Zmax refers to mean, minimum and maximum
elevations, respectively.
III. RESULTS AND DISCUSSION
A. 𝜒𝑚𝑑 and divide migration
The χmd of a divide is a quantitative anomaly between
opposing channel heads that share that divide. It is proposed to show
the degree of χ differences which can be used as an indicator of the
potential extent of divide mobility: the higher the χmd, the higher the

82
Figure 1. Divide stability analysis applied in Bohol Island, Philippines. A)
Drainage divide sections colored according to 𝜒𝑚𝑑 values. B) Direction of
potential migration direction and their 𝜒𝑚𝑑 classification.
degree of divide migration. Selected divides are shown in Figure 1A
where mobile divides are classified based on χmd. Divides with high
χmd are located at the headwaters of the major basins that border
smaller basins draining towards the southern coast of the island.
These divides extend in the SW-NE direction shown in Figure 1B.
Other classes of χmd are dispersed and found to be between the major
and smaller basins that faced east and northwest coastal areas.
Between major basins, divide is stable specially in the interior part
of the island. However, in the eastern side of the island, divides
between major basins are predicted to be migrating at a low to
moderate levels. Though χmd criteria presented in this study is
arbitrary, however, it is useful in illustrating the purpose of the
study.
B. Hypsometric analysis at different scales
Hypsometric analysis provides information on the erosion
status of basins which is used as the basis for prioritizing areas for
conservation measures [10]. This is indicated by HI values which
refer to the remaining landmass after erosion. Figure 2 (A) shows
the spatial pattern of HI in the island which ranges from 10 to 60
percent. High values are mostly found in the headwater areas of
major basins that border south-facing smaller catchments. Patches
of high HI values are also located in the western part of the island
and in the lower portion of the Wahig-Inabanga River Basin – a
primary agriculture basin (Figure 2B). Within this basin three
randomly selected subbasins show different HI values and
hypsometric curves (Figure 2C). Since the curves use
nondimensional parameters, they are described and compared
irrespective of true scale [11].
The Wahig-Inabanga River Basin drains towards the north of
the island and has an HI value of 0.19. Subbasin 431, draining
towards the main river from the south towards the north, is located
at the headwater of the basin indicates an HI value of 0.69. In the
middle of the basin, subbasin 260 drains from the west to the east
and gives an HI of 0.28. Subbasin 103 drains from the north to the
south in the lower part of the basin and demonstrates an HI of 0.58.
The Wahig-Inabanga River Basin is generally in its old stages and
Figure 3 Spatial variation of HI in (A) whole island and (B) major basin.
C) Hypsometric curves of the basin and selected subbasins. Reference
hypsometric curves adopted from [16].
Figure 3 Map of the subbasins that share with high 𝜒𝑚𝑑 divide sections.
The color indicates HI values (light to dark means low to high HI). The
dark-colored subbasins have HI of 0.5 and up. In one of these subbasins,
the 2005 Mayana landslide had occurred. Inset is the latest satellite image
of the landslide area extracted from Google Earth. Below are the
hypsometric curves of the nine subbasins arranged from low to high HI
values.
Figure 3. χ - map and the major basins in the island. A) highlights the
is graded to base level so that only 19% of the its total landmass has
remained after erosion. Subbasin 260 has undergo erosion that only
28% of its original landmass is left. Subbasins 431 and 103 are
highly erosional areas where 69% and 58% of their total land mass,
respectively, have remained which may be subject to further erosion.
C. Landform dynamics prioritization
Since basins and subbasins are associated with divides and their
catchment areas, we identify and prioritize them based on these
indicators. There are six divide sections with high χmd as shown in
Figure 1A. These divide sections are shared with 24 subbasins
whose HI values range from 0.24 – 0.69, as shown in Figure 3, with
catchment areas ranging from 382 ha to 1233 ha. By considering
only subbasins with HI values of 0.5 and up, the list reduces to nine
subbasins. These subbasins have catchment areas where more than
50% of their total landmass remaining after erosion so are
considered high erosion-prone areas and their drainage divides are
classified with high χmd. The hypsometric curves of these nine
subbasins exhibit different patterns which are associated with
different erosion processes that are happening in this individual
subbasins.
IV. CONCLUSION
The topographic metric, χ, is integrated successfully with
hypsometric analysis, and a new metric is introduced, χmd, to
develop criteria that allow for the selection of target areas for
conservation to arrest erosion process and for allocating areas for
development planning purposes as these areas are also associated
with geologic hazards. The intersection of highly unstable divides
and the highly erosional areas help us to narrow down further target
areas which facilitate prioritization process. The method presented
could be used as a rapid assessment tool in the characterization phase
of developing river basin management program.
.
ACKNOWLEDGEMENT
The study is supported by the Engineering Research and
Development Technology Program of the Department of Science
and Technology (ERDT-DOST).
83
REFERENCES
[1] D. H. Shugar et al., “River piracy and drainage basin
reorganization led by climate-driven glacier retreat,” Nat.
Geosci., vol. 10, no. 5, pp. 370–375, 2017.
[2] L. F. Bloxom and T. J. Burbey, “Determination of the
location of the groundwater divide and nature of
groundwater flow paths within a region of active stream
capture; the New River watershed, Virginia, USA,”
Environ. Earth Sci., vol. 74, no. 3, pp. 2687–2699, 2015.
[3] R. Linkevičiene, “Impact of river capture on hydrography
and water resources: Case study of Ula and Katra
catchments, south Lithuania,” Holocene, vol. 19, no. 8, pp.
1233–1240, 2009.
[4] H. W. Beeson, S. W. McCoy, and A. Keen-Zebert,
“Geometric disequilibrium of river basins produces long-
lived transient landscapes,” Earth Planet. Sci. Lett., vol.
475, pp. 34–43, 2017.
[5] L. Guerit, L. Goren, S. Dominguez, J. Malavieille, and S.
Castelltort, “Landscape ‘stress’ and reorganization from χ-
maps: Insights from experimental drainage networks in
oblique collision setting,” Earth Surf. Process. Landforms,
vol. 43, no. 15, pp. 3152–3163, 2018.
[6] L. Goren, S. D. Willett, F. Herman, and J. Braun, “Coupled
numerical-analytical approach to landscape evolution
modeling,” Earth Surf. Process. Landforms, vol. 39, no. 4,
pp. 522–545, 2014.
[7] J. T. Perron and L. Royden, “An integral approach to
bedrock river profile analysis,” Earth Surf. Process.
Landforms, vol. 38, no. 6, pp. 570–576, 2013.
[8] S. D. Willett, S. W. McCoy, T. J. Perron, L. Goren, and C.
Y. Chen, “Dynamic reorganization of River Basins,”
Science (80-. )., vol. 343, no. 6175, 2014.
[9] K. X. Whipple and G. E. Tucker, “Dynamics of stream-
power river incision model: Implications for height limits
of mountain ranges,landscape response timescales, and
research needs,” J. Geophys. Res., vol. 104, no. B8, pp.
17,661-17,674, 1999.
[10] O. Singh, “Hypsometry and erosion proneness: a case study
in the lesser Himalayan Watersheds,” J. Soil Water
Conserv., no. 2, pp. 53–59, 2009.
[11] A. Strahler, “Hypsometric (Area-Altitude) Analysis of
Erosional Topography,” Bull. Geol. Soc. Am., vol. 63, no.
2, pp. 1117–1142, 1952.
[12] Y. C. Chen, K. Y. Cheng, W. S. Huang, Q. C. Sung, and H.
Tsai, “The relationship between basin hypsometric integral
scale dependence and rock uplift rate in a range front area:
A case study from the coastal range, Taiwan,” J. Geol., vol.
127, no. 2, pp. 223–239, 2019.
[13] R. El Hamdouni, C. Irigaray, T. Fernández, J. Chacón, and
E. A. Keller, “Assessment of relative active tectonics,
southwest border of the Sierra Nevada (southern Spain),”
Geomorphology, vol. 96, no. 1–2, pp. 150–173, 2008.
[14] A. M. Forte and K. X. Whipple, “Criteria and tools for
determining drainage divide stability,” Earth Planet. Sci.
Lett., vol. 493, pp. 102–117, 2018.
[15] W. Schwanghart and D. Scherler, “Short Communication:
TopoToolbox 2 - MATLAB-based software for
topographic analysis and modeling in Earth surface
sciences,” Earth Surf. Dyn., vol. 2, no. 1, pp. 1–7, 2014.
[16] F. Pavano, G. Romagnoli, G. Tortorici, and S. Catalano,
“Morphometric evidences of recent tectonic deformation
along the southeastern margin of the Hyblean Plateau (SE-
Sicily, Italy),” Geomorphology, vol. 342, pp. 1–19, 2019.

84
 Automated Furrow Irrigation System for Maize (Zea
mays L. var. saccharata) Production
Sheila G. Griengo1, a *, Armando N. espino Jr. 2, b, Marvin M. Cinense3, c, Romeo B. Gavino 4, d, Vitaliana U.
Malamug5, d, Theody B. Sayco6, d
1
Mindanao State University -Main Campus, Marawi City, Lanao del Sur
23456
Central Luzon State University, Science City of Munoz, Nueva Ecija
a
sittieayeesha@yahoo.com
Abstract - This study aimed to develop an automated
furrow irrigation system to improve irrigation water management in
maize (Zea mays L. var. saccharata) production. An Automated
Furrow Irrigation System (AFIS) was developed and installed in the
field during maize production with the following components:
reservoir with motorized offtakes, AFIS microcontroller, soil
moisture sensor, and field advance sensor which was solar-powered
and with wireless data transmission. The field advance sensor was
located at 60 m from the furrow head for an optimized irrigation
process. A total of 2508 out of 2880 or 87.08 % of the expected data
points from the start of operation in automated blocks were
transmitted from the soil moisture sensors to the AFIS
microcontroller. Moreover, the soil moisture sensors and field
advance sensors successfully transmitted all requests for opening
and closing the gates, respectively, during the irrigation process.
Grain yield (14.1 tons/ha), and water use efficiency (43.25 kg/ha-
mm) from the produced maize using the automated system were
significantly different compared to those planted using the
conventional system. Cost analysis of the system resulted in a BCR
of 2.37 which infers that the system was viable for maize production.
The developed AFIS allowed irrigation processes to proceed with
minimum human intervention except for periodic inspections and
routine maintenance. The AFIS also improved water management
through efficient irrigation scheduling which improved the yield of
maize by 30% compared to the conventional system.
Keywords—furrow irrigation, soil moisture sensor, field
advance sensor, microcontroller, water management
I. INTRODUCTION
In the Philippines, the most commonly used irrigation
system is surface irrigation which is considered the oldest and most
common method of applying water to croplands. One of the cheap
and probably low-technique methods of surface irrigation is the
furrow irrigation. This irrigation scheme was characterized by low
application efficiency (45–60%) and caused significant water losses,
mainly due to excess deep percolation from the irrigated fields [1].
Rising labor costs, the competition for water and the need to
conserve the environment has intensified the move to more
advanced systems. An irrigation system should be designed
efficiently to uniformly apply irrigation water and maintain adequate
soil water for plant growth and production without causing
excessive water loss, erosion, or water quality impairment [2]. The
modern challenge for improving plant growth and reducing costs
justifies the development of an automated irrigation system that will
minimize the waste of water and reduce labor and monitoring
overhead [3]. Automation is one means of ameliorating the problems
of high labor requirements and low water use efficiency of surface
systems. Automation of the irrigation system will enable the farmer
to make smart decisions based on accurate and real‐time data (flow
rate, advance rate, automated gates, and other variables). Thus, this
study aimed to develop an automated furrow irrigation system to
minimize the water input and human intervention and improve
application efficiency while satisfying the plant’s needs.
I. METHODOLOGY
A. Design of the Major Components of AFIS.
The AFIS has several components such as concrete
reservoir with motorized offtake and distribution lines as a source
of irrigation water; soil moisture sensor (SMS) to monitor the
real-time soil moisture status of the field which initiate the
opening of the offtake when the reading stretches below 50%
moisture allowable deficit (MAD); field advance sensor (FAS)
which monitors the advancing water from the head furrow until it
reaches the sensor located before the end of the furrow and
triggers the closing of the offtake ending the irrigation process;
and, microcontroller which run the embedded program and
control the actions and features of the system.
B. Test Performance of AFIS
The sensors were tested through its capability to transmit
the necessary data to the AFIS microcontroller. Lag time and speed
of transmission were also observed. Transmitted data of the soil
moisture sensors were also subjected to reliability tests. It is the
probability that the software will not fail for a specified time under
specified conditions.
C. Evaluation of Field Performance of AFIS in Maize Production
F1 hybrid sweetcorn (Sweet Victory) was planted in all
the assigned blocks with 75-day maturity. An area of 8 m x 65 m
plot was utilized for both the automated system and the
conventional system. Each block contains 10 furrows each spaced
70 cm apart. All agronomic and plant protection practices were
applied uniformly to both treatments (automated and conventional)
except for the timing of irrigation. Automated block was irrigated
based on the actual soil moisture sensor readings while the
conventional block was irrigated based on farmer’s practice.
II. RESULTS AND DISCUSSION
85
A. Concrete Reservoir with Automated Offtake
The installed reservoir was constructed using three 24 in.
diameter x 39.37 in. long, made of concrete circular culvert. Three
(3) culverts were stacked together at 2.5-m high to store water from
the pump for irrigation. The reservoir was equipped with a 3-in
automated offtake for controlling the water flow. The offtake was
made of a 3/16-in. thick flanged steel plate body with 4 holes (Fig.
1). A disc valve with steel base rubber seal ring and a 12-V linear
actuator were connected at the top of the offtake to mechanically
operate the device. The offtake was further connected to the
microcontroller to remotely close and open the offtake during the
irrigation process.
Fig. 1. Concrete reservoir with automated offtake
B. Soil Moisture Sensor (SMS) Assembly
Fig. 2 shows the soil moisture sensor (SMS) assembly.
The Arduino Nano 3.0 board base was used and mounted to Nano
3.0 expansion board adapter. It served as an adapter to easily hook
up for Arduino Nano to an outside world via hook-up wires and has
a re-adjust fuse that protects the board from short circuits and over
current. A capacitive analog soil moisture sensor was used to
measure the soil moisture content by capacitive sensing. This is a
corrosion-resistant and transmits data through analog signals. The
sensor measured the real-time soil moisture content of the active root
zone of the plant which monitored and scheduled irrigation events
when the soil moisture was within the user-defined set point. A 5-V
relay module was then installed for remote device switching
specifically after data had been relayed to the main controller. A
three-pin IC LM 7805 voltage regulator was connected to the relay
module to maintain the output voltage in the soil moisture sensor at
a constant value. The system used a wireless transmission of data
using a transceiver which is a combination of transmitter and
receiver in a single package. This was used to receive a command
from the microcontroller during data collection and transmit signals
about the status of the soil moisture content. The sensor utilized solar
power as power source. The sensor was set to transmit soil moisture
data at a 3-hour interval. Six (3) sensors were installed in the study
site. Three were placed in Block 1 and the other three were placed
in Block 2 arranged in the front, middle, and end part of each block.
Fig. 2. Soil moisture sensor (SMS) assembly
C. Field Advance Sensor Assembly
Fig. 3 shows the components of the field advance sensor
(FAS) which includes the following: Arduino Nano 3.0, Nano 3.0
expansion board adapter, a water level sensor, a relay module, a
voltage regulator and a transceiver. These were all connected to a
solar power system composed of solar panels with charge controllers
and battery.
Almost all the components for FAS resembled that of the
soil moisture sensor assembly except for the sensor used. The FAS
assembly utilized a DC3-5V liquid water level module sensor as a
field advance sensor. The advancing water in the furrow was sensed
before it reached the end of the furrow. This also triggered the
termination of irrigation when the water reached the predetermined
location of the sensor. This location was identified through the
furrow performance evaluation before the start of planting. The FAS
was placed 60 meters from the furrow inlet with an approximate cut-
off time of 75 minutes.
Fig. 3. Field advance sensor (FAS) assembly
D. AFIS Microcontroller Assembly
The AFIS microcontroller controls all the activities of the
system. It sends and receives a request from the sensors and is
connected to the motorized gate of the reservoir for opening and
closing. Fig. 4 shows the components of the AFIS microcontroller
assembly.
The Arduino Mega 2560 has an 8-bit board with 54 digital
pins, 16 analog inputs, and 4 serial ports which serves as the brain
of the microcontroller that can sense the environment by receiving
input from a variety of sensors and can affect its surroundings by
controlling lights, motors, and other actuators. It was attached to a
prototype screw terminal block shield to eliminate loose connections
when connecting to digital and analog signals. The XD-05 was
attached on top of the terminal block which is a data logger module
shield with real-time clock and SD card interface where the card was
inserted for data logging purposes. A 16 x 2 LCD screen with two
horizontal lines comprising a space of sixteen display characters was
fitted above the XD-05. It displays the real-time soil moisture
reading of the soil. A dual motor driver module board H-bridge was
also connected to the Arduino Mega 2560 to drive the actuator
installed in the gate. It drives the actuator to open or close the gate
depending on the signal received by the microcontroller. The
assembly also used a transceiver which was utilized to send and
receive signals from the soil moisture sensor and water advance
sensor connected to the microcontroller. The assembly was
connected to a solar panel as the power source. A toggle switch was
provided to manually restart the microcontroller.
86

Figure 4. AFIS microcontroller assembly
The program for the microcontroller and soil moisture and
field advance sensors were written in C language. Sketches were
written in an Arduino Integrated Development Environment (IDE)
and uploaded in a microcontroller that executes the code. The soil
moisture sensor reads an analog signal which was converted into a
volumetric soil moisture reading. For field advance sensors, if the
analog was higher than 101, it would signal the controller to execute
the closing of the gates. If the reading was between 0 to 100, this
means that the gates should remain open for a certain amount of
time.
E. System Performance of AFIS
The speed of data transmission in SMS and FAS to the
AFIS controller and vice versa was between 100-300 milliseconds
which varied depending on the distance of the sensors to the
controller. A total of 2508 out of 2880 or 87.08 % of the expected
data points from the start of operation for both blocks were
transmitted from the soil moisture sensors to the AFIS
microcontroller. A total of 372 data points or 12.92% failed to be
transmitted. The field advance sensor successfully transmitted all of
the signals needed to close the gates. A recurrence was observed
during the conduct of the study. This means that the soil moisture
sensor gave the signal to open the gate ahead of its scheduled time
of opening. Readings were sent by the sensors to the AFIS
microcontroller and it took 23 seconds for Gate 1 and 15 seconds for
Gate 2 to open and close the offtakes. The opening of the gate was
initiated if the soil moisture was below 50% moisture allowable
deficit (MAD). The soil moisture sensor with readings below 50%
MAD triggered the opening of the gate to commence irrigation
regardless of location and block. The closing of the gate was
initiated when the advancing water had reached the location of the
field advance sensor
F. Grain Yield
Table 1 shows that minimum grain yield of 9.8 t/ha was
obtained from the conventional system which differed statistically
from the automated system with maximum grain yield of 14.1 t/ha.
Water availability is usually the most important crop production
factor which limits yield and yield characteristics of maize. Water
deficit remarkably influenced productivity and quality in maize [4].
Water stress in maize also reduced the grain yield by 33% as
compared to non-stress conditions [5].
G. Water Use Efficiency (WUE)
Crop water use efficiency (WUE) is the ratio of the grain
yield (kg) and the volume of water consumed by the crop during the
growing season. Crop WUE was highly influenced by irrigation
scheduling by AFIS (Table 1). The highest crop WUE (43.25 kg/ha-
mm) was observed from the automated system and the least value of
WUE (30.20 kg/ha-mm) was observed in the conventional system.
Results showed that irrigation scheduling with the help of an
automated system significantly increased the WUE. The automated
system helped reduce unproductive losses while maintaining a
Table 1. Mean comparison of different variables using t-test
at 5% level of significance.
VARIABLES TREATMENT
Conventional Automated
Grain yield ( tons/ha) 9.80a 14.10b
water use efficiency 30.06a 43.25b
(kg/ha-mm)
*Means with the same letter are not significantly different.
healthy and productive crop. Providing optimized water supply
when crops need it most and promptly terminating the irrigation
process on time prevented additional losses from excess irrigation.
H. System Viability
Simple cost analysis of the AFIS showed annual cost of
production for maize in the automated system was Php 337,551.04
while total cost in the conventional system was Php 321,629.52 The
annual cost of production in the automated system was higher due
to the initial investment of the system. The annual net return in the
automated system was Php 800,448.96 and in the conventional
system was Php 465, 970.48. Though the automated system has
higher annual cost of production it still has a higher annual net return
than the conventional since the cost in investing in the automated
system was offset by a higher yield from the said system.
Benefit-Cost Ratio (BCR) with AFIS was 2.37 while
without AFIS the value was 1.45. BCR with AFIS was higher
compared to the BCR without AFIS. This indicates that the use of
AFIS was feasible with a higher benefit-cost ratio than maize
production without AFIS.
Irrigation labor cost consisted of 45.76 % of the total labor
cost in the conventional system. This was reduced to 21.95 % in the
automated system since it only requires a minimum manual
intervention during the system operation. It did not only save labor
cost in the irrigation process but also made process more efficient.
Reduction in the fuel cost was also observed since the automated
system optimized the irrigation process thereby reducing the time of
operation during irrigation thus reducing the fuel cost.
CONCLUSION
The AFIS was successfully installed using available
materials for optimizing the irrigation process and proved that real-
time collection of soil moisture readings and water scheduling and
water-efficient technology was potentially a win-win strategy with
environmental and social benefits. This study also proved that
automation in the irrigation process demonstrates that efficient
irrigation management can significantly improve not only grain
yield in maize production but also increase water use efficiency and
that the system is economically viable to the farmers making the
maize production industry more competitive and sustainable.
ACKNOWLEDGEMENT
The author would like to thank the Department of Science
and Technology (DOST)-Engineering Research and Development
for Technology (ERDT) for the scholarship program. The author
was also equally indebted to her adviser Dr. Armando N. Espino Jr.
87
and the rest of the committee members for their invaluable ideas,
time, and patience accorded during the conduct and in doing this
paper.

REFERENCES

[1] Mitchell, A.R., Shock, C.C. & Perry, G.M., (1995). Alternating-
furrow irrigation to minimize nitrate leaching to groundwater.
Conference Proc ‘Clean Water Clean Environment – 21st
Century’, Kansas City, Missouri, ASAE, 1995
[2] Cinense, M. (2010). Modification, Fabrication and Evaluation
of Low Energy Precision Application System for Furrow
Irrigation. Unpublished Master’s thesis. Central Luzon State
University, Munoz, Nueva Ecija, Philippines
[3] Leroux, M. F. And Raghavan, G. V. (2005). Design of an
Automated Irrigation System
[4] Barutcular C, Yildirim M, Koç M, Akinci C, Toptai I, Albayrak
O, Tanrikulu A, & El Sabagh A (2016). Evaluation of SPAD
chlorophyll in spring wheat genotypes under different
environments. Fresenius Environmental Bulletin 25:1258-1266
[5] Alemi, M. H. (1981). The effect of water deficit on yield and
agronomic characters of corn. Field Crop Research. 37(7):
5955

88
 Physical and Mechanical Properties Assessment of
Lignocellulosic Rice Straw-based Rigid Polyurethane
Foam for Structural Sandwich Panel Cores
Mary Ann N. Ahalajal1, a *, Arnold A. Lubguban2,b, Roberto M. Malaluan3,c
1
DOE Student, Department of Civil Engineering and Technology, COET, MSU-IIT
2
Professor, Department of Chemical Engineering and Technology, COET, MSU-IIT
3
Professor, Department of Chemical Engineering and Technology, COET, MSU-IIT
a
ann.ahalajal@gmail.com, baalubg@gmail.com, crmmalaluan@yahoo.com

Abstract—This study investigates the physical properties
such as density and thermal conductivity and the mechanical
properties such as compressive strength and shear strength of
lignocellulosic rice straw-based rigid polyurethane foam. These
properties were further assessed using ASTM E 1730-02 Standard
Specification for Structural Sandwich Panel Core. Further, the
results were compared to petroleum-based rigid polyurethane foam.
Keywords—rice straw-based rigid polyurethane foam,
bio-based rigid polyurethane foam, structural sandwich panel core
rigid polyurethane foam first, such as its physical and mechanical
properties to perform a complete structural analysis and design. This
investigation was made to determine the properties of the
lignocellulosic rice straw-based rigid polyurethane foam such as its
density, thermal conductivity, compressive strength, and shear
strength based on American Standard Testing Method (ASTM) and
to assess if the properties meet the requirements for structural
sandwich panel cores as specified in ASTM E 1730-02. The results
were compared to the properties of petroleum-based rigid
polyurethane foams.

I. INTRODUCTION II. METHODOLOGY

One type of insulation is rigid polyurethane foams. They A. Foaming Formulation and Raw Materials Preparation
create one of the world's most popular, energy-efficient and versatile
insulations. These foams can significantly cut energy costs while Based on the results of a preliminary experiment [3], rigid
making commercial and residential properties more efficient and polyurethane foam with 15% rice straw-based polyol resulted in an
comfortable. To maintain uniform temperature and lower noise optimum density-compressive strength property. Accordingly, 15%
levels in homes and commercial properties, builders turn to rigid rice straw-based polyol was used in the final foaming formulation.
The lignocellulosic rice straw-based polyols were produced in large
polyurethane foam [1]. batches in MSU-IIT, Philippines. Voranol® 490 is a petroleum-
based polyether polyol with a hydroxyl number of 490 mg KOH/g
Polyurethane is a versatile polymer traditionally prepared for rigid foam applications. PAPI 135C SH, a polymeric
using a petroleum-based raw material. Petroleum, however, is a non- diphenylmethane diisocyanate (MDI) with 2.7 NCO functionality,
renewable material and polyurethane produced was found to be non- has 340 average molecular weight, and 31.4% part by weight of
biodegradable. In addition, petroleum is not environmentally NCO content. Voranol® SC 490 and PAPI 135 SH were obtained
friendly and requires high production costs with high technology from Chemrez Philippines, Inc (gratis). The catalyst for rigid
processing system [2]. polyurethane foaming was POLYCAT 8
(Dimethylcyclohexylamine, gel biased amine catalyst) while the
The BioProducts Research Laboratories (BPRL) of surfactant is INV® 690. POLYCAT® 8 and INV® 690 were obtained
Mindanao State University - Iligan Institute of Technology, from Sigma-Aldrich Chemicals, Philippines. Distilled water was
Philippines focuses on the use of natural, renewable resources to used as the blowing agent. The foaming formulation is shown in
produce innovative materials and bio-energy in a sustainable Table 1.
manner. The research group addresses compelling global interest to
make a transition to a bio-based industry where petroleum-derived
products can be replaced by better-performing products based on TABLE I. FORMULATIONS FOR LIGNOCELLULOSIC RICE
renewable materials. One of the main research products of BPRL is STRAW-BASED RIGID POLYURETHANE FOAM (LRS PU
lignocellulosic rice straw-based rigid polyurethane foam. Local FOAMS) AND CONTROL PU FOAMS
renewable material such as rice straw is utilized as raw material Concentration, phpa
which is chemically processed to produce a polyol for the production Ingredients LRS PU Control PU
of rigid polyurethane foams. foams foams
B-side materials
In home construction, rigid polyurethane foams have been VORANOL® 490 85 100
widely used as a nonstructural component such as thermal insulation (Petroleum-based polyol)
and sound absorber material, and as structural components such as Rice straw-based polyol 15 0
core of a sandwich panel. In designing a structural component such Polycat® 8 0.5 0.5
as sandwich wall panel, it is necessary to study the behavior of the (gelling catalyst)

89
 INV® 690 (surfactant) 1.5 1.5
Distilled water 1.0 1.0
(blowing agent)
A-side materials
Isocyanate indexb of PAPI® 27 110 110
a
The concentrations of all ingredients are expressed in parts per hundred part of polyol,
which conventionally dictates that the sum of all polyols adds up to 100 parts.
b
The amount of isocyanate index used relative to the theoretical equivalent amount.
The isocyanate index is the ratio of the actual amount of isocyanate used over the
theoretical amount of isocyanate required times 100.
B. Rigid Polyurethane Foam Production
Standard laboratory mixing and pouring procedures for
making water-blown polyurethane foams were performed for this
study as used by Tu, et. al (2007). The petroleum-based polyol
(VORANOL 490), rice straw-based polyol, catalysts, surfactant, and
blowing agent (B-side materials) were added into a 3-liter plastic
container and mixed at 2000 rpm for 10-15 seconds. The mixture
was then allowed to degas for 120 seconds. Methylene diphenyl
diisocyanate was then added rapidly into the mixture. Stirring
resumed at the same speed for another 15 seconds. The mixture was
then poured immediately into a wooden mold with dimensions 18
cm x 18 cm x 30 cm covered in aluminum foil lining and allowed to
rise freely at ambient conditions. The foam was then demolded and
allowed to cure for 7 days. After curing, the foam block was cut into
appropriate sizes. Fig. 1 illustrates the foaming procedures.
accordance with the ASTM C 518-15, by using samples of 200 x
200 x 35 mm.
C.2 Mechanical Properties
C.2.1 Compressive Strength
The compressive strength of the foam was determined in
accordance with ASTM D1621 by a Shimadzu Universal Testing
Machine with Trapezium X material testing software. The test
specimens were square of cross-section 50 mm x 50 mm and a
minimum height of 25.4 mm and loaded at a constant rate of 2.5
mm/min for every 25.4 mm height. The loaded ends of the specimen
are parallel to each other and perpendicular to the sides.
C.2.2 Shear Strength
The shear strength of the foam was determined in
accordance with ASTM C 273 and ISO 1922 by a Shimadzu
Universal Testing Machine with Trapezium X material testing
software. The test specimen had a thickness equal to 25 mm, a width
of 50 mm, and a length of 250 mm and loaded at such a constant rate
that the maximum load will occur within 3-6 minutes. A constant
rate of 0.7 mm/min was applied.
D. ASTM E 1730-02 Standard Specification for Rigid Foam for
Use in Structural Sandwich Panel Cores

Petroleum-based Table 2 shows the standard specification for rigid foam for
polyol use in structural sandwich panel cores as listed in ASTM E 1730-02.
This specification is used as a reference in assessing the physical and
Blowing agent (H2O)
mechanical properties of the lignocellulosic rice straw-based rigid
Catalyst
polyurethane foam.
Rice straw-based Surfactant
polyol
TABLE II. PHYSICAL AND MECHANICAL PROPERTIES OF
RIGID FOAM FOR USE IN STRUCTURAL SANDWICH
PANEL CORE
Requirement Type 1 Type 2 Type 3
Procedure
Density, kg/m3, 41.6 55.7 72.0
Polymeric max
Transfer to mold
diisocyanate Thermal 0.036 0.037 0.037
conductivity,
W/mK, max
Fig. 1 Procedure for preparing rigid PU foams
Compressive 172.4 275.8 448.2
strength, kPa,
C. Foam Characterization min
Shear strength, 172.4 241.2 275.8
C.1 Physical Properties kPa, min
C.1.1 Density
The density of the foam was measured in accordance with III. RESULTS AND DISCUSSION
ASTM D 1622-08. Density was determined by dividing the weight
The following presents and discusses the results obtained
of each sample by its corresponding volume. The samples were in performing the methodology in this study.
square in cross-section with an area of 25 cm2 and a height of 30
mm.
A. Physical Properties
A.1 Density
C.1.2 Thermal Conductivity
Table 2 shows the density test results of PU foams in this
The thermal conductivity measurements of the foams study. It can be observed that the average density of control PU
were performed using a heat flow meter. The measurements were foams is higher than the average density of the LRS PU foam
made under steady heat flow conditions through the test samples, in samples. Also, the density results of LRS PU foam have no
significant difference from each other.

90
 TABLE II DENSITY OF PU FOAM SAMPLES
Density of PU Foam Samples
Sample Density, kg/m3
LRS PU foam Control PU foam 90
1 55.53 48.88
80
2 55.47 48.76

Density (kg/m3)
3 53.56 50.00 70

4 55.98 80.39 60
5 56.05 77.85
50
6 54.86 75.02
7 56.76 85.36 40
8 56.92 85.23
30
9 56.73 84.93
Standard Type 1 PU Foam, max Standard Type 2 PU Foam, max
Average 55.54 70.71 Standard Type 3 PU Foam, max LRS PU foam
Conrol Pu foam

Fig. 2 illustrates the density results in Table 2. The solid

lines show the standard density values for rigid foams for use in Fig. 2 Density of PU foam samples in comparison with the
structural sandwich panel cores as indicated in ASTM E 1730-02. It standard density specifications
can be seen that the density of lignocellulosic rice straw-based rigid
polyurethane foam in this study is classified under Type 2 PU foam.
Thermal Conductivity of PU Foams
0.050
A.2 Thermal Conductivity Thermal Conductivity (W/mK)

Table 3 shows the thermal conductivity of the PU foam 0.040

samples. It can be seen that the thermal conductivity of LRS PU
foam is superior to that of control PU foam. 0.030

0.020
TABLE III THERMAL CONDUCTIVITY OF PU FOAM
SAMPLES
0.010
Thermal Conductivity, W/mK Standard Type 1 PU Foam, max Standard Type 2 and 3 PU Foam, max
LRS PU foam Control PU foam LRS: Perp. to Rise Direction LRS: Par. to Rise Direction
Control: Perp. to Rise Direction Control: Par. to Rise Direction
Sample Perp. to Par. rise Perp. to Par. to rise
rise direction rise direction Fig. 3 Thermal conductivity of PU foam samples in comparison
direction direction with the standard thermal conductivity specifications
1 0.03121 0.02786 0.03341 0.04339
2 0.03992 0.03526 0.03986 0.03385 B. Mechanical Properties
3 0.03952 0.03531 0.04095 0.03786 B.1 Compressive Strength
Ave. 0.03688 0.03281 0.03807 0.03837
Table 4 shows the compressive strength results of the PU
foam samples. It is noted that PU foams made with 100% Voranol ®
490 (control PU foam) have higher average compressive strength
Fig. 3 illustrates the thermal conductivity results found in compared to LRS PU foam.
Table 3. These values were compared with the maximum thermal
conductivity values for all types (Type 1, Type 2 and Type 3) of PU TABLE 4 COMPRESSIVE STRENGTH OF PU FOAM
foams used in structural sandwich panel cores as indicated in ASTM SAMPLES
E 1730-02, which is shown in solid line in Fig 2. It can be seen that
the LRS PU foam passed the maximum standard value of 0.037 Sample Compressive Strength, kPa
W/mK. LRS PU foam Control PU foam
1 495.471 414.491
2 544.642 433.798
3 529.328 436.515
4 422.609 638.281
5 461.992 824.466
6 429.083 749.656
7 468.014 893.057
8 476.633 829.393
9 497.645 875.609
Average 480.601 677.252

91
 Fig. 4 illustrates the compressive strength test results
found in Table 4. It can be seen that the LRS PU foam passed all
minimum standard compressive strength values specified in ASTM
E 1730-02.
Compressive Strength of PU Foams
1000
900
Compressive Strength (kPa)
800
700
600
500
400
300
200
100
0
Standard Type 1 PU Foam, min
Standard Type 3 PU Foam, min
Conrol Pu foam
Fig. 4 Compressive strength of PU foam samples in comparison
with the standard compressive strength specifications
B.2 Shear Strength
Table 5 shows the shear strength test results of the PU
foam samples. Shear strength test of the control PU foam yet to be
completed. The average value is calculated from the initial test
results. It can be seen that the average shear strength value is higher
than the LRS PU foam.
TABLE 5 SHEAR STRENGTH OF PU FOAM
SAMPLES
Sample Shear Strength, MPa3
LRS PU foam
1 -
2 2.640
3 2.116
4 2.152
5 2.327
6 2.352
7 3.175
8 3.061
9 3.097
Average 2.615
Figure 5 illustrates the shear strength test results found in
Table 5 in comparison with the standard minimum shear strength
values specified in ASTM E 1730-02. It can be observed that all PU
foams passed the minimum standard shear strength values for all
types (Type 1, Type 2 and Type 3) of PU foams used in structural
sandwich panel cores.
Shear Strength of PU Foams
5.0
Shear Strength (MPa)
4.0
3.0
2.0
1.0
0.0
Standard Type 1 PU Foam, min Standard Type 2 PU Foam, min
Standard Type 3 PU Foam, min LRS PU foam
Conrol Pu foam
Fig. 5 Shear strength of PU foam samples in comparison with the
standard shear strength specifications
IV. CONCLUSION
Standard Type 2 PU Foam, min
LRS PU foam The results of the study clearly show that the physical
properties and mechanical properties, such as density, thermal
conductivity, compressive strength and shear strength, of the
lignocellulosic rice straw-based rigid polyurethane foam passed the
specifications for structural sandwich panel core listed in ASTM E
1730-02. Further, it can be classified as a Type 3 structural sandwich
panel core.
This study shows that the lignocellulosic rice straw-based
polyurethane foam can be used in actual applications such as
sandwich wall panels, roof panels, and others.
ACKNOWLEDGEMENT
This study was made possible by BioProducts Research
Laboratory, MSU-IIT, Philippines.
REFERENCES
Control PU foam [1] “Introduction to Polyurethanes: Polyurethane Applications”,
2.298 https://polyurethane.americanchemistry.com/Applications/.
2.188 [2] Enderus N.F. and Tahir, S.M. (2017), “Green Waste Cooking
- Oil-based Rigid Polyurethane Foam,” IOP Conf. Series:
3.37 Materials Science and Engineering.
3.292 [3] Icalina, K.M. and Lubguban, A.A. (2018), “Development and
4.453 Optimization of the Physical Properties of Water-Blown Rigid
- Polyurethane Foams Produced from Rice Straw-Based
Polyols,” Material Science and Engineering, MSU-IIT, Iligan
-
City, Philippines.
-
3.120
92
 Silanated Coconut Oil-Based 2K Waterborne
Polyurethane Coating for Corrosion Protection
Dave Joseph E. Estrada1,2 a *, Amierson C. Tilendo2,b, Roberto M. Malaluan2,3,c Arnold C. Alguno4,d,
and Arnold A. Lubguban2,3,e
1
College of Engineering Graduate School, MSU – Iligan Institute of Technology
2
Bio Products Research Laboratory, MSU – Iligan Institute of Technology
3
Department of Chemical Engineering and Technology, MSU – Iligan Institute of Technology
4
Department of Physics, MSU – Iligan Institute of Technology
a
davejoseph.estrada@g.msuiit.edu.ph, bamierson.tilendo@g.msuiit.edu.ph, croberto.malaluan@g.msuiit.edu.ph
d
arnold.alguno@g.msuiit.edu.ph, arnold.lubguban@g.msuiit.edu.ph

Abstract— This study aims to develop novel techniques
on the synthesis of an effective two-component (2K) coconut oil-
based waterborne polyurethane with varying silane loadings applied
on a carbon steel substrate as protective coating against metal
corrosion. Electrochemical analysis shows a significant decrease on
corrosion rate as silane loading ratio increases. Moreover, Tape Test
method is utilized to determine the adhesion strength of the
synthesized coating material. Results revealed that the adhesion
strength of the synthesized WBPU material is enhanced as silane
loading increases.
Keywords—Waterborne, coconut oil, silane, adhesion
strength, electrochemical analysis.
I. INTRODUCTION
Metallic substrates are susceptible to corrosion attacks in
natural environments and there are so many ways on preventing it
but generally, the use of coating is the most common one [1], [2].
Organic coating such as the use of environmentally friendly
polyurethane (PU) (e.g., waterborne polyurethane, WBPU) coatings
for corrosion protection have garnered attention over the past couple
of years, as certain pioneering researchers have improved the PU
coating. PU is mainly composed of polyol (called the soft segment)
and diisocyanate (called the hard segment) [3].
The major components for the synthesis of PUs are
produced from petroleum oil as the feedstock [4]. With the
fluctuation of prices for petroleum feedstock, this dangerously
affects the price of raw materials for industries [5]. Moreover,
Desroches et al. (2012) emphasized that the instability regarding the
depletion of fossil based assets and worldwide temperature
alteration issues were identified [6], [7]. Renewable resource such
as natural oil-based polyols are widely used to prepare PUs. For
example, soybean oil, palm oil, castor oil, and rapeseed oil have
higher unsaturated fatty acid content, hence they can be hydroxyl
functionalized easily using various methods. However, some of
these nature oils can be produced in small scale and will compete
between food industrials and petrochemical industrials [4], [8].
Coconut oils are abundantly available plant oils and can
produce an appreciable amount of coconut oil up to 60-65 % of its
actual weight [9]. But in comparison to other nature oils, it has a low
unsaturated fatty acid content making it difficult to use in PU
synthesis. Fortunately, this problem was addressed by Ampon et al.
(2019), they have successfully developed a highly functionalized
coconut oil-based polyester polyol (CCP) via polycondensation
process [10]. The appreciable change in the properties of the coconut
oil after the polycondensation process provided baseline information
that coconut oil can also be utilized in synthesizing polymeric
materials despite the challenges in its molecular structure.
But, as stated earlier that due to the high hydrophobicity
of coconut oil, water-borne approach is not applicable because it will
lead to the phase separation between oil and water. Several
researches were conducted to address this concern, and the use of
silane is quite conventional. Silane is a silicon containing
compound. One silicon molecule can have both organic and
inorganic reactivity, thus silane is widely used as adhesion
promoters and coupling agents. According to Lyu et al. (2019),
inorganic Si-O bond in silane results in low surface energy, good
thermal stability and excellent flexibility [11].
In this study, the coconut oil will be utilized as a partial polyol
replacement for petroleum based polyol for the synthesis of
silanated coconut oil-based 2K waterborne polyurethane coating for
corrosion protection. Furthermore, the addition of silane at different
loadings will be investigated if it contributes to its corrosion rate
adhesion strength properties.
II. METHODOLOGY
A. Materials
The coconut polyester polyol (CCP) was supplied by the
Bio Products Research Laboratory, MSU – IIT. Additives such as
polyether-modified siloxane (BYK 349), polyether block copolymer
(DISPERBYK-190), foam-destroying polysiloxane (BYK 022),
non-ionic urethane rheology modifier (ACRYSOL RM-8W), and
titanium dioxide (Ti-Pure R960) were supplied by supplied by
Chemrez Technologies. The tetraethoxysilane used in this
experiment was reagent grade.
B. Methods
The NCO:OH ratio was kept constant at 1.2:1. The
additives and pigments, including the isocyanate and polyol will be
measured in terms of grams that would account to a 15 g WBPU
sample. The first step is to disperse the distilled water, defoamer,
dispersing agent, rheolate, and pigment using and overhead stirrer
for 20 minutes at 3000 rpm. Add distilled water and continuously
stir for 15 minutes at 1500 rpm. Add the bayhydrol, CCP, distilled
water, defoamer, wetting agent, and rheolate and stir for 20 minutes
at 2000 rpm. After 20 minutes, add an additional distilled water and
continuously stir for 3 minutes at 2000 rpm. Lastly, add the HDI and
continuously stir at 2000 rpm until a thick consistency is achieved.

93
Add silane to the synthesized WBPU with 0-5wt% concentration potential (Ecorr), corrosion current density (icorr), cathodic slope
and stir for 2-3 minutes at 2000 rpm. The parts of all the reagents for (βc), anodic slope (βa) and corrosion rate (CR) which are calculated
each steps are indicated in the Table 1 below: and tabulated in Table 2. The corrosion rate is calculated by using
(1):
0.13 𝐼𝑐𝑜𝑟𝑟 (𝐸. 𝑊. )
𝐶𝑅 (𝑚𝑚𝑝𝑦) =
𝑑
TABLE 1. FORMULATION AND PARTS FOR THE It can be observed in Fig. 1 that tafel plots shifted to lower
SYNTHESIS OF PU COATING current density as the silane loading increases. It indicates that the
Steps Function/supplier Parts incorporation of silane on the improvement of corrosion resistance
Local 5-6 with Si-O-Si act as an inhibitory effect additive with WBPU coating.
Defoamer/BYK 1-2 In potentiodynamic polarization curve analysis, the increase in Rp
Dispersing and decrease in Icorr represented the better protection of corrosion.
1 1-2
agent/BYK
Rheolate/DOW 0.1-1.0
Pigment/Dupont 20-25
2 Local 1.0-2.0
Resin/Bayer 26.0-27.0
Polyol/BPRL 26.0-27.0
Local 0.9-1.0
3
Defoamer/BYK 0.1-0.5
Wetting agent/BYK 0.1-0.5
Rheolate/DOW 0.1-0.5
4 Local 3.0-5.0 Fig. 1. Electrochemical polarization curve test results.
5 Isocyanate 32.0-35.0
The results in Table 2 showed that increasing of silane
loading up to 1 wt% increased the Ecorr and decreased the Icorr values.
C. Application of the synthesized WBPU Since corrosion rate is directly proportional to Icorr value [12], it can
Place the metal substrate on the spin coater. On top of the be deduced that higher silane loading improved the anticorrosive
metal substrate, place the synthesized PU and let the spin coater to properties of the coatings. To further support this finding, the
coat the metal substrate at 1000 rpm for 30 seconds. Repeat calculated corrosion rate (CR) of the WBPU coatings also showed a
application until the coating is uniformly distributed on the metal declining trend upon increasing the loading of silane. The improved
substrate. Curing time must take for at least 1 week. corrosion rate could be due to the fact that silane-based coatings are
chemically inert and can provide corrosion protection because of
D. Characterizations and specifications their good barrier properties that delay the entry of aggressive
species to the metallic surface due to its hydrophobic property as
The corrosion resistance of waterborne PU coatings is
evaluated using potentiodynamic polarization using stated by Asadi et al. (2014) [13].
Potentiostat/Galvanostat (SP150, Bio-Logic). To examine the
adhesion of coatings with the substrate, cross cut adhesion test of TABLE 2. POLARIZATION PARAMETERSFOR THE
prepared coating panels was carried out using the ASTM D-3359- WBPU COATING WITH DIFFERENT SILANE LOADINGS
02 standards. Thermal degradation properties of the polyurethane
foam samples were studied between -30°C to 400°C using DSC Ecorr Icorr CR
Samples βa βc
(DSC 4000, Perkin Elmer) under nitrogen atmosphere at 20 L/min (mV) µA (mmpy)
with a ramp rate of 10°C /min. Chemical compositions of the 0% -589.79 13.765 193.2 382.4 0.029
synthesized PU coating will be evaluated by Fourier Transform 0.2% -468.02 0.466 231.3 262.7 0.98e-3
Infrared (FTIR) using Shimadzu IRTracer-100 ATR mode at 0.4% -496.84 0.371 150.0 776.2 0.78e-3
wavelengths between 4000 and 400cm−1. 0.6% -745.39 0.063 68.9 201.2 0.15 e-3
1% -575.01 0.071 141.4 82.9 0.13 e-3
E. Statistical analysis
B. Adhesion strength test
All experiments are repeated thrice. Therefore, the
experimental results will be on average values from two replications The photographs of all samples that have undergone tape
with standard deviations on the error bars. The imprecision, test method can be seen in Fig. 2. It can be clearly seen that more
coefficient of variation (CV) will be calculated to be < 5 %. In areas are removed during the tape test at 0% loading of silane.
addition, analysis of variance (one-way ANOVA) will be carried out However, as the silane loading increases, the area removed tends to
by Microsoft Excel tool package within 95% confidence interval. decrease.

III. RESULTS AND DISCUSSION 0%

A. Corrosion test
The corrosion performances of WBPU coatings are also
evaluated by Tafel extrapolating method. Fig. 1 shows the Tafel
polarisation curves for WBPU hybrid coatings. The plots are
extrapolated in order to derive the parameters such as corrosion

94

0.2%%
0.4%
0.6%
1%
Fig. 2. Photographs of all samples after tape test.
Figure 3 shows the average remained area after the tape
test is conducted. An increasing trend for its adhesion strength can
be observed as the silane loading ratio increases. The average area
remained with standard deviation for samples 0%, 0.2%, 0.4%,
0.6%, and 1% silane loadings are 48±23.25mm2, 62±10.19 mm2,
74.66±27.76 mm2, 94±3.74 mm2, and 98.33±0.47 mm2, respectively.
The 1-way annova is calculated and the p value is 0.047 which is a
bit lower than the threshold value of 0.05. This indicates that there
is a significant difference in terms of its adhesion strength as the
silane loading increases.
Fig. 3. Average remained area of all samples.
IV. CONCLUSION
It can be concluded that it is indeed possible to
synthesized coconut oil-based waterborne polyurethane as
protective coating against metallic corrosion. In agreement with
the corrosion current densities obtained from polarization
curves, the 1 wt% tetraethoxysilane loading exhibits the best
result in terms of lower cower corrosion rate. Furthermore,
the adhesion strength of the 1 wt% tetraethoxysilane loading
WBPU coating shows a significant increase.
ACKNOWLEDGEMENT
The researchers would like to thank the following
sponsors: Chemrez as the official supplier of all reagents used, and
Bio Products Research Laboratory – MSUIIT for allowing the
student to conduct its experiment.
REFERENCES
[1] P. Y. Tsai, T. E. Chen, and Y. L. Lee, “Development and
characterization of anticorrosion and antifriction properties
for high performance polyurethane/graphene composite
coatings,” Coatings, vol. 8, no. 7, 2018, doi:
10.3390/coatings8070250.
[2] O. S. I. Fayomi, I. G. Akande, and S. Odigie, “Economic
Impact of Corrosion in Oil Sectors and Prevention: An
Overview,” J. Phys. Conf. Ser., vol. 1378, no. 2, 2019, doi:
10.1088/1742-6596/1378/2/022037.
[3] M. M. Rahman, M. H. Zahir, M. B. Haq, D. A. Al Shehri,
and A. M. Kumar, “Corrosion inhibition properties of
waterborne polyurethane/cerium nitrate coatings on mild
steel,” Coatings, vol. 8, no. 1, 2018, doi:
10.3390/coatings8010034.
[4] T. Kattiyaboot and C. Thongpin, “Effect of Natural Oil
Based Polyols on the Properties of Flexible Polyurethane
Foams Blown by Distilled Water,” Energy Procedia, vol.
89, pp. 177–185, 2016, doi: 10.1016/j.egypro.2016.05.024.
[5] B. J. Rashmi, D. Rusu, K. Prashantha, M. F. Lacrampe, and
P. Krawczak, “Development of water blown bio-based
thermoplastic polyurethane foams,” Adv. Mater. Res., vol.
584, pp. 361–365, 2012, doi:
10.4028/www.scientific.net/AMR.584.361.
[6] M. Desroches, M. Escouvois, R. Auvergne, S. Caillol, and
B. Boutevin, “From vegetable oils to polyurethanes:
Synthetic routes to polyols and main industrial products,”
Polym. Rev., vol. 52, no. 1, pp. 38–79, 2012, doi:
10.1080/15583724.2011.640443.
[7] S. Sahoo, S. Mohanty, and S. K. Nayak, “Biobased
polyurethane adhesive over petroleum based adhesive: Use
of renewable resource,” J. Macromol. Sci. Part A Pure
Appl. Chem., vol. 55, no. 1, pp. 36–48, 2018, doi:
10.1080/10601325.2017.1387486.
[8] F. Carré, “1. Introduction,” L’archéologie en Haute-
Normandie, Bilan des connaissances, pp. 7–12, 2018, doi:
10.4000/books.purh.5104.
[9] M. Kumar and R. Kaur, “Effect of Different Formulations
of MDI on Rigid Polyurethane Foams based on Castor Oil,”
vol. 2, no. May, pp. 29–42, 2013.
[10] B. S. Ampon, J. J. C. Budiongan, J. D. Mangubat, and B.
A. N. Semense, “Optimum Functionalization of Coconut
Oil-Based Polyester Polyol (CPP) via Polycondensation
with Phthalic Anhydride,” Mindanao State University -
Iligan Institute of Technology, 2019.
[11] J. Lyu et al., “In situ incorporation of diamino silane group
into waterborne polyurethane for enhancing surface
hydrophobicity of coating,” Molecules, vol. 24, no. 9, 2019,
doi: 10.3390/molecules24091667.
[12] M. M. Ariffin, M. M. Aung, L. C. Abdullah, and M. Z.
Salleh, “Assessment of corrosion protection and
performance of bio-based polyurethane acrylate
incorporated with nano zinc oxide coating,” Polym. Test.,
vol. 87, no. April, p. 106526, 2020, doi:
10.1016/j.polymertesting.2020.106526.
[13] N. Asadi, R. Naderi, M. Saremi, S. Y. Arman, M. Fedel,
and F. Deflorian, “Study of corrosion protection of mild
steel by eco-friendly silane sol-gel coating,” J. Sol-Gel Sci.
Technol., vol. 70, no. 3, pp. 329–338, 2014, doi:
10.1007/s10971-014-3286-8.
95
 Qualitative risk assessment of tailings dam
Cielo D. Frianeza1, a* and Andres Winston C. Oreta1, b
1
De La Salle University Manila, 2401 Taft Ave, Malate, Manila, 1004 Metro Manila
a
cielo_frianeza@dlsu.edu.ph, bandres.oreta@dlsu.edu.ph
Abstract—By-products from the mining process, called
tailings, are composed of uneconomical ore, process water, and
hazardous substances that are usually stored in structures made of
soil called tailings dams. Despite the importance of tailings dam
stability because of the hazardous materials it stores, failure often
occurs, which severely impacts the environment, specifically
groundwater, surface water, vegetation, and soil. Thus, the study
aims to assess risk qualitatively using the preliminary qualitative
checklist rating and failure modes and effects analysis (FMEA).
Thus, the hazard occurrence, vulnerabilities of the tailings dam, and
the factors that contribute to the vulnerabilities were identified to
accurately conduct the qualitative risk assessments. Prior to a
detailed assessment, a preliminary qualitative checklist rating was
conducted wherein ratings were assigned to the vulnerabilities,
hazards, and occurrence of hazards to determine whether a detailed
analysis of the tailings dam is required based on the level of risk.
Since Padcal Mine Site Tailings Pond 3 is a high-risk facility, a more
detailed risk assessment, FMEA, was applied wherein the
vulnerabilities of the said tailings dam were evaluated to determine
which component will cause failure. Based on the analysis,
overtopping, inadequate foundation design, inadequate structural
design, and mine subsidence are the probable cause of failure of
Padcal Mine Site Tailings Pond 3 due to the frequent typhoon and
earthquake incidence, as well as the existing condition of the facility.
Keywords—Tailings dam failure, Qualitative risk
assessment, Failure modes and effects analysis, Environment
I. INTRODUCTION
Tailings is a slurry composed of the uneconomical part of
the ore, excess water used during the process of separating the
valuable minerals, and hazardous substances like heavy metals,
sulphide minerals, and cyanide. An estimate of 14 billion tons of
tailings were produced in 2010; this waste material is stored in
tailings dams and other storage facilities [15]. An example of a cross
section of a tailings dam is shown in Fig. 1 that illustrates the
important components in determining the condition of a tailings
dam. Tailings dams are constructed in sequential “lifts” over time;
hence, the tailings dam partially depends on the tailings themselves
for support [15]. Usually, a single tailings dam facility is used for
the entire life of the mine, which is approximately 5 to 40 years [18].
Inactive tailings dams are facilities that have been filled or when
tailings production ceases; while active dams are facilities that are
still used to store produced tailings. One of the ways to close a
tailings dam is by enforcing a wet cover, a common closure strategy,
which requires the surface of tailings dams to be submerged to
prevent acidic drainage by restricting oxidation of sulphides [18].
Fig. 1. Tailings dam [18]
There is no publicly accessible inventory of tailings dams
[18]. Hence, most of the data from studies involving tailings dam
come from recorded tailings dam failure. Tailings dam failure
impacts the environment severely due to the heavy debris and
hazardous substances that are present in the tailings. Groundwater,
surface water, vegetation, and soil are at risk of contamination and
destruction if tailings are discharged onto them. Thus, the study aims
to assess the risk to protect tailings dams from failure. However, data
about tailings dams is limited and the only known tailings dams in
the Philippines are those that have already failed. Research has
shown that there is a lack of publicly accessible qualitative or
quantitative risk assessments for active and inactive tailings dams in
the Philippines. To show the application of qualitative risk
assessment, the study applied the preliminary qualitative checklist
rating and failure modes and effects analysis (FMEA) on an active
tailings dam that have failed in the past. The study will also
determine the hazard occurrence in areas wherein operational mines
are situated, assuming that the tailings dam of the respective
operational mine is nearby as a reference for future risk assessments.
II. CONCEPTUAL FRAMEWORK
The conceptual framework of the study is shown in Fig. 2.
The study will focus on risk assessment rather than the other
components of disaster preparedness. Furthermore, to assess the
risk, the hazard occurrence, element at risk, and vulnerability of the
element at risk must be determined. Since the common natural
hazards that trigger tailings dam failures in the Philippines are
rainfall and earthquakes, the occurrence of these hazards on the
operational metallic and non-metallic mines in the Philippines will
be determined using hazard maps. On the other hand, the study will
focus on common vulnerabilities of tailings dams, such as
overtopping, external erosion, internal erosion, slope instability,
seismic vulnerability, inadequate foundation design, inadequate
structural design, and mine subsidence.
Fig. 2. Conceptual framework of the study
III. METHODOLOGY
FMEA is a qualitative risk assessment tool used in
mapping out the impact of all harmful events that can occur during
construction or operation of the tailings dam [19]. As shown in Fig.
3, to conduct FMEA, the system must be defined in which the
objective of the project and other pertinent information is
established to set the scope of the study. The first step of FMEA is
to establish the hierarchy of levels. A system is the set of all
components that affect the failure of the structure studied; it is
systematically divided into subsystems down to the basic component
level. The functions of the basic components are then defined, which
are used to describe the role of the component in the system. Failure
modes are the description of how failure occurs in a component.
Failure is defined as the cessation of the ability of a component,
96
subsystem, or system to perform one of its functions [19]. Effect is
the impact of a failure mode in the performance of the system and
components; furthermore, it consists of outcomes associated with
the loss of ability to accomplish a function [19]. The effect on the
component is the direct effect, while the intermediate effect is the
effect on the subsystem, and the end effect is for the whole system.
The effects are used to formulate measures to prevent failure in each
level. Detection measures are methods a failure mode can be
detected under normal conditions. The control measures carry out
remedial work to control sequence of its effects by delaying or
stopping them; while, mitigation measures reduce end effects and
their consequences [19].
Fig. 3. Flowchart of the summary of procedures
IV. COMPONENTS OF ASSESSING RISK
A. Factors contributing to vulnerability of element at risk
a) Construction method
The upstream method is the oldest and most commonly
used construction method. The upstream method starts with a starter
dyke in the extreme downstream point then progressively raised
towards the upstream side. Upstream constructed dams require
fewer construction materials, but the dam has a high phreatic line
and poor stability; furthermore, the dam wall is built on top of
previously deposited unconsolidated tailings slurry retained behind
the wall [16]. Thus, the upstream method is more susceptible to
instability from seismic loading compared to other types of
construction methods, such as downstream and centerline.
b) Deposition of tailings
The general methods of tailings deposition are single point
discharge, spigotting, and cycloning. Single point discharge is used
to discharge tailings away from the crest; furthermore, the discharge
point must be periodically moved to another location to prevent
unequal raising of the dam section. Spigotting discharges tailings
through small pipes or spigots that originate from multiple points at
regular intervals along the tailings header line [21]. Uniform flow of
tailings, resulting in uniform beaches, are achieved using spigotting.
The distribution characteristics of spigotting is effective in reducing
the water level across the dam. Furthermore, uniform beaches are
necessary to maintain stability for dams constructed using upstream
method. Cycloning is used to separate the fines from the coarse
fraction in the tailings slurry [21].
c) Embankment material of tailings dam and properties of tailings
Tailings sands are often used as an inexpensive source of
material for construction of tailings dams that can have both
effective strength and high permeability [21]. Native soil and
tailings sand used as embankment are appropriate for tailings dams
constructed using upstream, downstream, or centerline method. The
disadvantage of using tailings as embankment include high
susceptibility to piping and high erodible surfaces; moreover, loose
and saturated tailings are subject to earthquake-induced liquefaction.
d) Tailings dam height
The stability analysis carried out for tailings dams under
different heights shows that the design height of the tailings dam
should be reduced to ensure safety [12]. A higher tailings dam would
require more stability due to the accumulated wastewater that the
dam needs to retain. According to U.S. Environmental Protection
Agency [21], tailings dams that are constructed using upstream
method have a height increase restriction of 15 to 30 feet per year,
while a height increase that would exceed 50 feet per year are
considered hazardous. Generally, tailings dam heights are kept
below 200 feet because high embankments, which are greater than
400 feet, pose design and construction problems [21].
e) Steepness of tailings dam slope
Steep slopes are prone to external erosion and slope
instability, which could collapse a portion of the dam embankment
if no protections are in place. Furthermore, steep slopes are more
likely to become unstable during seismic events, such as ground
shaking. Steep-sided sand has a slope of 3H:1V to 4H:1V [21].
f) Drainage system
Tailings are discharged into the dam as slurry, wherein the
solid material separates as it settles into the dam. A suitable drainage
system must be used to avoid dam failure due to overtopping in
countries that experience heavy rains [12]. Since the outer slope of
the dam is very sensitive to a high phreatic line, the dam is more
susceptible to failure due to seepage and internal erosion if proper
drainage is not in place.
g) Demand of mine products
Tailings dam failures in developed countries happened
before the 21st century; while in developing countries, failures
happened after the 21st century during their economic growth [12].
Thus, as the mining industry begins to flourish in the period of
economic growth and development, the frequency of failure
increases, since developing countries often overlook the proper
design and management of tailings dams when there is an influx of
tailings from mining.
h) Date of construction
Most of the tailings dam failures are attributed to tailings
dams that were constructed before the 21st century. According to
Larrauri [10], prior to the 1970s the construction method used was
the upstream method regardless of local condition and there was no
real stability analysis. During the 1970s to 1980s a few
advancements and techniques were utilized to improve the stability
of tailings dams and avoid failure, wherein deterministic approach
was used for seismic hazards, centerline and downstream method
were used in seismic zones, and usage of safety factors were
predominant [10]. During the 1990s to 2000s, seismic stability
calculations, deterministic and probabilistic assessment of seismic
hazards, probable maximum flood determination, seepage
modelling and detection, and limit equilibrium methods for slope
protection were utilized [10].
i) Tailings dam activity
Active tailings dams are more likely to fail than inactive
tailings dams. The increase of water level, wherein the phreatic line
is close to the crest of the downstream slope, is a common cause of
inactive dam failures, which leads to slips or overtopping [22].
B. Vulnerability of element at risk
a) Overtopping
Overtopping, or the flooding of the top of the tailings dam,
make the slope unstable that causes tailings dam failure [12]. The
initially constructed tailings dam should have a good permeability
to ensure drainage of water and prevent overtopping. According to
Lyu et al. [12], the three stages of the tailings dam break process,
which may occur multiple times, are the beginning of destabilization
of tailings dam under adverse condition; interaction of tailings sand
97
with water and formation of debris flow with high energy; and debris g) Inadequate structural design
flow moving downstream. Structural design errors are often caused by failure of
drainage systems, such as the decant facilities. Decant facilities are
b) External erosion used to remove excess water from the tailings dam to manage the
Erosion of tailings dams due to wind and rain can affect water level and phreatic line. Structural failure, such as breakage of
the stability and create environmental problems. Since larger pipes or a destruction of the dam wall caused by the failure of a
particles are near the crest, local small rotational slips may occur component of the decant facility, can cause tailings dam failure.
where the water level meets the downstream slope. Downstream Other forms of surface water control systems aside from decant
slopes suffer slight gullying caused by rainfall; the first sign of a facilities are diversion channels, closed channels, and spillways.
close water level is given by water issuing into one these gullies Spillways are designed as temporary structures constructed with
wherein small slips may develop deepening the gully, which will impervious material to withstand rapid flow velocities [21].
move the surface of that part of the downstream slope further back
into the water level [22]. This leads to larger slips that eat away the h) Mine subsidence
base of the slope until the whole downstream face becomes unstable. Mine subsidence occurs when the tailings dam is built on
The erosion of the embankment can also lead to overtopping [18]. top or near an underground mine [18]. Hence, collapse of the
underground mine can lead to a breach in the foundation, which will
c) Seepage and internal erosion lead to collapse of tailings dam due to the pressure from the flow of
The stability of the tailings dam is greatly influenced by tailings into the underground mine.
the seepage field since tailings dams are permeable [12]. According
to Lyu et al. [12], the factors that increase the susceptibility of
seepage and affect the level of the phreatic line are the level of water C. Hazard occurrence
in the dam, permeability of the initial dam before tailings were Heavy rain, such as typhoons, trigger overtopping and
stored, and rainfall. As the seepage increases, the hydrostatic internal erosion; while, earthquakes trigger seismic instability such
pressure increases, reducing the shear strength of the embankment, as liquefaction of tailings. External erosion, slope instability,
which will cause the failure of the tailings dam. Moreover, structural and foundation failure, and settlement of tailings dams
components of the structure with high permeability are susceptible near underground mines are triggered by rainfall and earthquakes.
to piping or the internal erosion due to seepage [12]. The operational metallic and non-metallic mines were obtained from
the directory of Mines and Geosciences Bureau as of August 2019
d) Slope instability [13, 14]. The occurrence of typhoons and earthquakes in the mines
Rotational sliding results in slope failure ranging from were mapped using Quantum Geographic Information System
local sloughing of tailings at random areas along the face of an (QGIS); the base maps used were from the Typhoon incidence map
embankment to massive circular arc slides that extend over the entire and Earthquake-prone areas map of Center for Environmental
structure [21]. The causes of rotational sliding include changes in Geomatics, & Manila Observatory [3].
water table, changes in permeability of the foundation materials,
disturbance to the embankment caused by impact or vibration, and TABLE I. Operating metallic mines in the Philippines [13]
settlement of foundation materials. Slope instability may also be ID Contractor Location
caused by deformation from static load and partial saturation of Itogon-Suyoc Resources,
areas in the tailings dam designed to remain dry [18]. 1 Sangilo, Itogon, Benguet
Inc.
Benguet Corporation –
e) Seismic vulnerability
2 Acupan Contract Mining Acupan, Itogon, Benguet
Earthquake-induced liquefaction of tailings is the main
Project
cause of damage of high tailings dams, while the influence of
Lepanto Consolidated
seismic inertial force on the seismic stability of the dam is 3 Mankayan, Benguet
Mining Company
secondary. According to Lyu et al. [12], as the height of the dam
4 FCF Minerals Corporation Quezon, Nueva Vizcaya
increases, the seismic performance is reduced. Liquefaction of
tailings are affected by the composition, size, gradation, Filminera Resources
5 Aroroy, Masbate
arrangement, compactness, depth of wetting line, and seismic Corporation
intensity [12]. Studies show that the tailings stored in the dam are 6 Apex Mining Company Inc. Maco, Compostela Valley
saturated and relatively loose [12]. Thus, under seismic loading, the Philsaga Mining Bunawan and Rosario,
7
tailings dam body is susceptible to liquefaction, which makes it Corporation Agusan del Sur
easier for seepage channels to form that can cause collapse. Tmc-Tribal Mining
8 T’boli, South Cotabato
Corporation
f) Inadequate foundation design Alegria, Mainit, Tubod
Greenstone Resources
Weak material beneath the slope of the embankment, low 9 and Bacuag, Surigao Del
Corporation
shear strength of foundation strata, and fissured clay, when not Norte
addressed, are some of the causes of foundation failure [21]. Abra Mining & Industrial
10 Bucay, Abra
Furthermore, the permeability of the foundation affects the stability Corporation
of the embankment; if the foundation material is impervious, the Carmen Copper
11 Biga, Toledo City, Cebu
load of the embankment will generate excess porewater pressure, Corporation
which will reduce the shear strength of the foundation. On the other 12 Philex Mining Corporation Tuba, Benguet
hand, if the foundation material is pervious, excessive seepage can Oceanagold (Philippines),
13 Kasibu, Nueva Vizcaya
lead to piping. Hence, the properties of the tailings beach that will Inc.
become the foundation of the next dyke must comply with the 14 Techiron Resources, Inc. Guiuan, Eastern Samar
requirements for a stable foundation. [21]. Loreto, Dinagat Islands
15 Krominco Inc. (within Parcel III Surigao
Mineral Reservation)

98
TABLE II. Continuation of operating metallic mines [13] TABLE III. Continuation of operating metallic mines [13]
ID Contractor Location ID Contractor Location
Mt. Sinai Mining Oriental Synergy Mining
Homonhon Island, 46 Loreto, Dinagat Islands
16 Exploration and Corporation
Guiuan, Eastern Samar
Development Corporation Citinickel Mines and Narra and Sofronio
47
Rio Tuba Nickel Mining Development Corporation Espanola, Palawan
17 Bataraza, Palawan
Corporation MacArthur and Javier,
48 Nicua Corporation
Tubay, Santiago, and Leyte
18 Agata Processing, Inc.
Jabonga, Agusan del Norte Dulag and Mayorga,
Strongbuilt (Mining)
19 SR Metals, Incorporated Tubay, Agusan del Norte 49 MacArthur, Javier and
Development Corp.
Surigao Integrated Abuyog, Leyte
20 Claver, Surigao del Norte
Resources Corporation 50 Atro Mining-Vitali, Inc. Vitali, Zamboanga City
21 Berong Nickel Corporation Quezon, Palawan
Hinatuan Mining Tagana-an, Surigao del TABLE IV. Operating non-metallic mines in the Philippines [14]
22
Corporation Norte ID Contractor Location
Adnama Mining Resources Apo Land and Quarry
23 Claver, Surigao del Norte 1 Naga, Cebu
Incorporated Corporation
Taganito Mining Republic Cement &
24 Claver, Surigao del Norte 2 Norzagaray, Bulacan
Corporation Building Materials, Inc.
Marcventures Mining and Holcim Mining and
25 Carrascal, Surigao del Sur 3 Norzagaray, Bulacan
Development Development Corporation
CTP Construction and Adlay, Carrascal, Surigao Republic Cement &
26 4 Taysan, Batangas
Mining Corporation del Sur Building Materials, Inc.
CTP Construction and Adlay, Carrascal, Surigao Lazi Bay Resources
27 5 Lazi and Maria, Siquijor
Mining Corporation del Sur Development, Inc.
Carrascal Nickel Republic Cement Iligan City, Lanao del
28 Carrascal, Surigao del Sur 6
Corporation Mindanao, Inc. Norte
Cagdianao Mining Holcim Resources and Iligan City and Lugait,
29 Cagdianao, Dinagat Island 7
Corporation Development Corporation Misamis Oriental
30 Libjo Mining Corporation Libjo, Dinagat Islands Republic Cement Land and Norzagaray and San Jose
8
Oriental Vision Mining Resources, Inc. del Monte, Bulacan
31 Tubajon, Dinagat Island
Philippines Corporation Solid Earth Development
9 San Fernando, Cebu
Loreto and Libjo, Dinagat Corp.
32 Century Peak Corporation
Islands Holcim Mining and
Bunawan District, Davao
Benguetcorp Nickel Mines 10 Development Corporation –
33 Sta. Cruz, Zambales City
Inc. HMDC
34 Eramen Minerals, Inc Sta. Cruz, Zambales Republic Cement Iligan, Kiwalan and Iligan City,
11
Zambales Diversified Sta. Cruz and Candelaria, Inc. Lanao del Norte
35
Metals Corporation Zambales Maria Cristina Chemical
LNL Archipelago Minerals, 12 Industries (MCCI) Kiwalan, Iligan City
36 Sta. Cruz, Zambales Corporation
Inc.
Claver, Surigao del Norte Northern Cement
13 Sison, Pangasinan
(Within Parcel I of Corporation
37 Claver Mineral Dev't. Corp.
Surigao Mineral 14 Teresa Marble Corporation Antipolo City, Rizal
Reservation) Island Quarry and
15 Antipolo City, Rizal
Sinosteel Phils. H. Y. Aggregates Corp.
38 Loreto, Dinagat Islands
Mining Corporation Republic Cement &
16 Teresa, Rizal
Libjo and Tubajon, Building Materials, Inc.
39 Wellex Mining Corporation
Dinagat Islands Holcim Mining and
Pacific Nickel Philippines, Cagdianao, Surigao Del Development Corporation Doña Remedios Trinidad
40 17
Inc Norte and Doric Marble and Norzagaray, Bulacan
Dinapigue Mining Corporation
41 Dinapigue, Isabela
Corporation Ibalong Resources and Camalig and Guinobatan,
18
Basilisa and San Jose, Development. Corporation Albay
AAM-Phil Natural
Dinagat Islands (Parcel II Bohol Limestone
42 Resources Exploration and 19 Garcia Hernandez, Bohol
of Surigao Mineral Corporation
Development Corporation
Reservation) Solid North Mineral
20 San Ildefonso, Bulacan
Basilisa and San Jose, Corporation
AAM-Phil Natural
Dinagat Islands (Parcel II 21 Eagle Cement Corporation San Ildefonso, Bulacan
43 Resources Exploration and
of Surigao Mineral Bigte and Norzagaray,
Development Corporation 22 Spar Development Co., Inc.
Reservation) Bulacan
Mil-Oro Mining Mati and Gov. Generoso, Solid Earth Development
44 23 San Fernando, Cebu
Corporation Davao Oriental Corporation
Panamaon, Loreto, Rio Tuba Nickel Mining
45 Century Peak Corporation 24 Bataraza, Palawan
Dinagat Island Corporation

99
TABLE V. Continuation of operating non-metallic mines [14] upon by extreme weather events, such as typhoons. Using the spot
ID Contractor Location map in Fig. 6 and Fig. 7, it was determined that mines located in
Holcim Mining and South Cotabato, Surigao Del Norte, Dinagat Island, Agusan del
Bacnotan and Balaoan, La Norte, and Sarangani experience approximately 48 to 75 hits per
25 Development Corporation-
Union year, while mines in Agusan del Sur, Surigao del Norte, and Surigao
HMDC
Holcim Mining and del Sur experience 76 to 100 earthquake hits per year. Thus, tailings
Davao City and Maasim, dams near the mines located in these areas are susceptible to failure
26 Development Corporation-
Sarangani triggered by earthquake events. In addition, mines located in
HMDC
Holcim Resources and Iligan City and Lugait, Benguet, Nueva Vizcaya, Pangasinan, La Union, and Bataan has a
27 high or very high typhoon incidence, and experience 48 to 100
Development Corporation Misamis Oriental
Holcim Mining and earthquake hits per year; thus, tailings dam near the mines in these
28 Development Corporation- Bacnotan, La Union areas are susceptible to failure triggered by earthquakes and rainfall.
HMDC
29 Teresa Marble Corporation Antipolo City, Rizal
30 Teresa Marble Corporation Antipolo City, Rizal
Naga and Pinamungahan,
31 Quarry Ventures Phils., Inc.
Cebu and Toledo City
Rapid City Realty and
32 Teresa, Rizal
Development Corp.
Rapid City Realty and
33 Antipolo City, Rizal
Development Corporation
34 Apo Land and Quarry Corp. Naga, Cebu
Concrete Aggregates
35 Angono, Rizal
Corporation
Concrete Aggregates
36 Angono, Rizal
Corporation
Rapid City Realty and
37 Antipolo City, Rizal
Development Corporation
Asensio Pinzon Aggregate
38 Rodriguez, Rizal
Corp.
Rapid City Realty and
39 Antipolo City, Rizal
Development Corporation
Rapid City Realty and
40 Antipolo City, Rizal
Development Corporation
Island Quarry and
41 Antipolo City, Rizal
Aggregates Corporation
Rapid City Realty and
42 Antipolo City, Rizal
Development Corporation
Fig. 4. Typhoon incidence in operating metallic mines
Rapid City Realty and
43 Baras, Rizal
Development Corporation
JLR Construction and Naga and Minglanilla,
44
Aggregates, Inc. Cebu
45 Hardrock Aggregates, Inc. Antipolo City, Rizal
Rolando B. Gimeno and La
Concepcion Construction
46 Antipolo City, Rizal
and Development
Corporation
Gozon Development
47 Antipolo City, Rizal
Corporation
Dolomite Mining Alcoy and Dalaguete,
48
Corporation Cebu
UBS Marketing
49 Legazpi City, Albay
Corporation
50 Heirs of Elias E. Olegario Mangatarem, Pangasinan
51 Orophil Stonecraft, Inc. Baao, Camarines Sur
52 Rockmix, Inc. Balanga, Bataan
Island Quarry and Palangoy, Binangonan,
53
Aggregates Corporation. Rizal

Using the spot maps in Fig. 4 and Fig. 5, it was determined

that the operating mines located in Zambales, Dinagat Island, Leyte,
Bulacan, Batangas, Rizal, Albay, and Camarines Sur have high
typhoon incidence, while mines in Abra and Isabela have very high
typhoon incidence. Thus, tailings dams near the location of the said
mines are susceptible to failure triggered by heavy rainfall brought Fig. 5. Typhoon incidence in operating non-metallic mines

100
 TABLE VII. Typhoon incidence in operating non-metallic mines
Typhoon
ID Location
Incidence
2 Norzagaray, Bulacan High
3 Norzagaray, Bulacan High
4 Taysan, Batangas High
8 San Jose del Monte, Bulacan High
14 Antipolo City, Rizal High
15 Antipolo City, Rizal High
16 Teresa, Rizal High
Doña Remedios Trinidad and
17 High
Norzagaray, Bulacan
18 Camalig and Guinobatan, Albay High
20 San Ildefonso, Bulacan High
21 San Ildefonso, Bulacan High
22 Bigte and Norzagaray, Bulacan High
29 Antipolo City, Rizal High
30 Antipolo City, Rizal High
32 Teresa, Rizal High
33 Antipolo City, Rizal High
35 Angono, Rizal High
36 Angono, Rizal High
37 Antipolo City, Rizal High
38 Rodriguez, Rizal High
39 Antipolo City, Rizal High
Fig. 6. Earthquake-prone operating metallic mines
40 Antipolo City, Rizal High
41 Antipolo City, Rizal High
42 Antipolo City, Rizal High
43 Baras, Rizal High
45 Antipolo City, Rizal High
46 Antipolo City, Rizal High
47 Antipolo City, Rizal High
49 Legazpi City, Albay High
51 Baao, Camarines Sur High
53 Binangonan, Rizal High

TABLE VIII. Earthquake-prone operating metallic mines

ID Location Earthquake Hits
Bunawan and Rosario, Agusan del
7 76 to 100
Sur
8 T’boli, South Cotabato 48 to 75
Alegria, Mainit, Tubod and Bacuag,
9 48 to 75
Surigao del Norte
Loreto, Dinagat Islands (within
15 Parcel III Surigao Mineral 48 to 75
Reservation)
Tubay, Santiago, and Jabonga,
18 48 to 75
Agusan del Norte
20 Claver, Surigao del Norte 76 to 100
23 Claver, Surigao del Norte 76 to 100
Fig. 7. Earthquake-prone operating non-metallic mines
24 Claver, Surigao del Norte 76 to 100
25 Carrascal, Surigao del Sur 76 to 100
TABLE VI. Typhoon incidence in operating metallic mines
26 Adlay, Carrascal, Surigao del Sur 76 to 100
ID Location Typhoon Incidence
27 Adlay, Carrascal, Surigao del Sur 76 to 100
10 Bucay, Abra Very High
28 Carrascal, Surigao del Sur 76 to 100
33 Sta. Cruz, Zambales High
Claver, Surigao del Norte (Within
34 Sta. Cruz, Zambales High
37 Parcel I of Surigao Mineral 76 to 100
35 Sta. Cruz & Candelaria, Zambales High
Reservation)
36 Sta. Cruz, Zambales High
41 Dinapigue, Isabela Very High TABLE IX. Earthquake-prone operating non-metallic mines
45 Panamaon, Loreto, Dinagat Island High ID Location Earthquake Hits
48 MacArthur and Javier, Leyte High 26 Davao City and Maasim, Sarangani 48 to 75
Dulag and Mayorga, MacArthur,
49 High
Javier and Abuyog, Leyte

101
TABLE X. Typhoon incidence in earthquake-prone metallic mines TABLE XII. Tailings dam failure in the Philippines
Typhoon Earthquake ID Date Mine/Project and Location
ID Location
Incidence Hits A 28/Oct/16 Antamok, Baguio (Benguet Corp.)
1 Sangilo, Itogon, Benguet High 76 to 100 B 2/Aug/12 Padcal No 3, Benguet (Philex)
2 Acupan, Itogon, Benguet High 76 to 100 Bayarong dam, San Marcelino, Zambales
C 11/Sep/02
3 Mankayan, Benguet Very High 48 to 75 (Benguet Corp-Dizon Copper-Silver Mines)
4 Quezon, Nueva Vizcaya High 48 to 75 Camalca dam, San Marcelino, Zambales
D 27/Aug/02
13 Kasibu, Nueva Vizcaya High 48 to 75 (Benguet Corp-Dizon Copper-Silver Mines)
12 Tuba, Benguet High 76 to 100 E 9/Aug/99 Toledo City (Atlas Con Mining Corp)
TP7, Placer, Surigao Del Norte (Manila
F 26/Apr/99
TABLE XI. Typhoon incidence in earthquake-prone non-metallic Mining Corp)
mines Sibutad Gold Project, Zamboanga Del Norte
G 27/Jun/98
Typhoon Earthquake (Philex Mining Corp)
ID Location
Incidence Hits Sibutad Gold Project, Zamboanga Del Norte
H 6/Nov/97
13 Sison, Pangasinan High 76 to 100 (Philex Mining Corp)
Bacnotan and Balaoan, La Marcopper Mine,Tapian Pit, Marinduque
25 Very High 76 to 100 I 24/Mar/96
Union Island (Placer Dome & President Marcos)
28 Bacnotan, La Union Very High 76 to 100 Bulawan Mine, Sipalay River, Negros
J 8/Dec/95
50 Mangatarem, Pangasinan High 48 to 75 Occidental (Philex Mining Corp)
52 Balanga, Bataan High 48 to 75 Bulawan Mine, Sipalay River, Negros
K Apr/95
Occidental (Philex Mining Corp)
V. TAILINGS DAM FAILURE IN THE TP5, Placer, Surigao del Norte (Manila
L 2/Sep/95
Mining Corp) 2nd event
PHILIPPINES Marcopper, Mogpog, Marinduque Island
M 6/Dec/93
(Placer Dome-President Marcos)
A. Local tailings dam failure Baguio gold district, Itogon-Suyoc (Benguet
N 26/Jun/93
Corp)
O 2/Jan/92 TP2, Tailings Pond, Tubu, Benguet
Placer, Surigao Del Norte (Manila Mining
P 9/Jul/87
Corp)
TP3, Mankayan District (Benguet Corp
Q 17/Oct/86
subsidiary Lepanto Con Mining Co)
R 8/Nov/82 TP3, Sipalay (Maricalum Mining Corp)

TABLE XIII. Continuation of tailings dam failure in the Philippines

Ore Fill Dam
ID Dam Type Status
Type Material Height
A Au Inactive
Fig. 8. Activity of failed tailings dam
B Au Cu Upstream Active
C Cu Au Earth fill Active
D Cu Au Earth fill Inactive
E Cu Inactive
F Au 50.43 m Active
G Au Active
H Au Active
I Cu Active
J Au Active
K Au Upstream Earth fill 17 m Inactive
L Au Upstream Earth fill 17 m Inactive
M Cu Active
N Au Ag Active
Fig. 9. Triggers of tailings dam failure
O Cu Active
P Au Active
Q Cu Au Earth fill Active
Mine
R Cu Active
Waste

TABLE XIV. Continuation of tailings dam failure in the Philippines

ID Natural Hazard Cause Release (cu.m.)
A Super Typhoon Lawin ST 50,000
B Typhoon Ferdie and Gener OT 13,000,000
C Heavy rain OT 1,000,000
D Heavy rain OT Undetermined
E Undetermined ST 5,700,000
Fig. 10. Causes of tailings dam failure
F Heavy rain ST 2,032,000

102
TABLE XV. Continuation of tailings dam failure in the Philippines Heavy rain or typhoons trigger 56% of the recorded
ID Natural Hazard Cause Release (cu.m.) tailings dam failures listed in TABLE XIV and TABLE XV; in
G Undetermined OT Undetermined addition, the majority of tailings dam failures were caused by
H Heavy rain OT Undetermined overtopping. On the other hand, 33% of tailings dam failures are
I Earthquake ST 1,600,000 caused by structural failure, which include the failure of drainage
J Undetermined ST Undetermined systems. During an intense rainfall, the increase in volume of water
K Heavy rain U 50,000 may exceed the capacity of the drainage system which may lead to
L Heavy rain FN 50,000 failure or a breach in pipes. Failure of the foundation to support the
M Undetermined U Undetermined load of the tailings dam, is also 17% of the cause of tailings dam
failure, wherein the excess volume brought by heavy rain may
N Undetermined OT Undetermined
exceed the designed load that the structure is able to support. Since
O Earthquake FN 80,000,000
major weather events are more frequent than major seismic
P Undetermined U Undetermined
activities, only 11% of the failed tailings dams in the Philippines
Q Heavy rain ST 100,000 were triggered by earthquakes, wherein the cause of failures were
R Heavy rain FN 25,000,000 largely attributed to structural failure and foundation failure.
Earthquake events may cause a rupture in the pipes underground
TABLE XVI. Continuation of tailings dam failure in the Philippines designed for the drainage system. Furthermore, seismic events may
ID Effects of Disaster on Elements Source trigger foundation failure by rupturing the foundation strata.
Contaminated Liang River, Ambalanga
A [24]
River, and Agno River Surface water was mostly affected by the tailings dam
Contaminated Balog River, Agno River, failure, while a few reports indicate the effects of acid mine
B [1, 24]
and San Roque Dam drainage. Nearby croplands were also affected by the sediments
Contaminated Mapanuepe Lake and from the tailings dam that contaminated fertile croplands by burying
C [23, 24]
Santo Tomas River it. Most of the tailings that were stored in the failed tailings dam
Contaminated Mapanuepe Lake and were from the production of gold and copper, which usually contains
D [23, 24]
Santo Tomas River hazardous substances, such as heavy metal waste and cyanide waste
E Contaminated Sapangdaku River [9, 24] [18]. From the reports, the tailings spill is composed of acidic mine
Contaminated 51 hectare of cropland and [2, 8, 20, 23, water and sediments that contain hazardous substances; thus, bodies
F
Batayakan area 24] of water, such as Boac River and Mogpog River were declared
G Undetermined [24] biologically dead after the tailings spill.
H Fish kills in nearby bodies of water [20, 24]
Contaminated Makulaquit River and Boac
River System, and Barangay Hinapulan B. Padcal Mine Site Tailings Pond 3 in Padcal, Tuba, Benguet
[4, 6, 11, 17,
I cropland; Boac River Biologically Dead; Tailings Pond 3 was used to store tailings from Padcal
24]
Seepage in Hinapulan creek; Acid Rock Mine Site, wherein the main products obtained are gold and copper.
Drainage in Groundwater According to AGHAM [1], Padcal Mine pioneered the underground
J Contaminated Sipalay River [20, 24] block cave; a type of mining that creates a cave and extracts a large
K Coastal Area Pollution in Placer Bay [23, 24] volume of rock from the formation of surface depression. Tailings
L Coastal Area Pollution in Placer Bay [8, 20, 24] Pond 3 has a catchment area of 350 hectares wherein the south end
Contaminated Mogpog Town and of the main dyke consists of two main penstock, A and B. Tailings
Mogpog River; Mogpog River [6, 7, 17, 24] Pond 3 started construction in 1983 and commissioned in 1992, with
M
Biologically Dead; Acid Rock Drainage a lifespan of 18 to 20 years [1]. Moreover, the dam crest went
in Groundwater beyond the permitted elevation of 600 meters above sea level. Also,
N Undetermined [24] fault lines, such as Albian and Santa Fe, traverse Tailings Pond 3.
Heavy siltation in Irrigation systems
O [20, 23, 24]
downstream
P Fish kills in Placer, Surigao del Norte [20, 24]
Q Contaminated Abra River [20, 24]
Laid 1.5 meters thick of sediment on
R [23, 24]
agricultural land

There are 18 tailings dam disaster events and 16 failed

tailings dams in the Philippines tabulated in TABLE XII, TABLE
XIII, TABLE XIV, TABLE XV, and TABLE XVI. Surigao del
Norte and Benguet are areas that experience high seismic activity; Fig. 11. Tailings Pond 3 before the incident [5]
thus, the dams in TABLE XIII that are constructed using upstream
method should have proper designs to avoid failure from
destabilization due to seismic activity. The height of the dam is an
important factor that contributes to the likelihood of failure. Tailings
Pond 7 in Placer, Surigao del Norte exceeded the design height of
50 meters by 0.43 meters, which may have contributed to the failure
of the tailings dam in 1999 among other factors. The activity of the
tailings dam also contributes to the likelihood of its failure since
active tailings dams are more likely to fail than inactive ones. Most
of the incidents occurred on active tailings dams, while only 28% of
Fig. 12. Tailings Pond 3 after the incident [5]
the incidents occurred on inactive tailings dams.

103
 Penstock A was breached that resulted in the discharge of TABLE XX. Risk scoring
tailings, which contaminated Balog River, Agno River, and San Risk class Description Risk Number (RN)
Roque Dam, and polluted nearby fertile lands and vegetation. The I Low 1 to 15
incident was triggered by heavy rain brought by typhoon Ferdie and II Medium 16 to 30
Gener. There were five major incidents of discharge reported on III High 31 to 45
August 1, 4, 11, and 30, and September 13, 2012. Fig. 11 shows the
idealized cross section of Tailings Pond 3 prior to the incident and
Fig. 12 shows the cross section after the incident, both illustrations
VII. QUALITATIVE RISK ASSESSMENT
from Mines and Geoscience Bureau. Sinkholes, due to the OF PADCAL MINE SITE TAILINGS
underground block cave, were also found in the tailings pond, which POND 3
was plugged using concrete spheres following the discharge to also
prevent tailings from leaking out [5]. The tunnels were sealed by
concrete to prevent leakage; thus, spillways were built to drain water A. Preliminary qualitative checklist rating
from the tailings dam instead of using penstocks and underground
drain tunnels, since the Department of Environment and Natural Padcal Mine Site Tailings Pond 3 is located in Tuba,
Benguet which is an earthquake-prone area, hit by approximately 76
Resources approved the life extension of Tailings Pond 3 to store
to 100 hits per year, and has a very high typhoon incidence. TABLE
waste from the operating Padcal Mine Site [5].
XXI shows the preliminary qualitative risk assessment for Padcal
Mine Site Tailings Pond 3. Each vulnerability of the tailings dam
VI. PRELIMINARY QUALITATIVE was evaluated by multiplying the corresponding hazard and
CHECKLIST RATING probability of the hazard. Padcal Mine Site Tailings Pond 3 is
classified a Class III risk since the likelihood that a tailings dam
TABLE XVII, TABLE XVIII, and TABLE XIX shows breach may occur due to overtopping, inadequate foundation design,
the rating of the vulnerabilities, hazards, and hazard occurrence for and inadequate structural design is high. Thus, further analysis to
tailings dams in the Philippines. The ratings are based on the studies determine which component failure will lead to tailings dam breach
and literature that were reviewed, which showed the critical factors is necessary to properly assess the risk.
that affect tailings dam failure in the local and global setting.
Equation 1 is used to obtain the risk class of a tailings dam to TABLE XXI. Padcal Mine Site Tailings Pond 3 risk class
determine which tailings dam should have a focused risk reduction Risk number computation Risk class
measure. Since tailings dam failures in the Philippines are mostly V1 × H3 × P5 45 III (High)
caused by typhoons, typhoons were given a high hazard rating. V2 × H3 × P5 15 I (Low)
Moreover, the common vulnerability of tailings dam failure in the V3 × H3 × P5 15 I (Low)
Philippines are overtopping, inadequate foundation design, and V4 × H3 × P5 30 II (Medium)
inadequate structural design; thus, the said vulnerabilities were V5 × H2 × P5 20 II (Medium)
assigned a rating of 3. V6 × H3 × P5 45 III (High)
V7 × H3 × P5 45 III (High)
Equation 1 V8 × H3 × P5 30 II (Medium)

TABLE XVII. Rating for tailings dam vulnerability

Vulnerability Rating B. Failure modes and effects analysis
V1 Overtopping High 3 To obtain a more detailed qualitative analysis, FMEA was
V2 External erosion Low 1 utilized to determine the failure that will lead to a dam breach.
V3 Seepage and internal erosion Low 1
V4 Slope instability Medium 2 Main System: Padcal Mine Site Tailings Pond 3 in Tuba, Benguet
V5 Seismic vulnerability Medium 2
V6 Inadequate foundation design High 3
V7 Inadequate structural design High 3
V8 Mine subsidence Medium 2
Fig. 13. Tailings Pond 3 subsystems and basic components
TABLE XVIII. Rating for tailings dam hazards
Hazard Rating TABLE XXII. Functions of the components of Tailings Pond 3
H1 Light rain Low 1 Basic
H2 Earthquake Medium 2 Function
Component
H3 Typhoon High 3 To contain the tailings in case of overflow
Main Dyke
from offset dyke
TABLE XIX. Rating for probability of tailings dam hazard To contain the tailings before it reaches the
occurrence Offset Dyke
main dyke
Probability Rating Typhoon incidence Earthquake hits Penstock A Used to control water level; however, it is
P1 1 Very low 0 to 11 and B already sealed by concrete
P2 2 Low 12 to 28 Drain Tunnel Used to connect the tailings dam to Balog
P3 3 Medium 29 to 47 A and B River; however, it is sealed by concrete
P4 4 High 48 to 75 Spillway To regulate the water level in the tailings dam
P5 5 Very high 76 to 100 Foundation
To support the load of the tailings dam
Strata
Mountain
To serve as one of the tailings dam’s wall
Slope

104
TABLE XXIII. Failure modes of components and its root causes TABLE XXIV. Continuation of failure modes of components and
Component Failure Mode Root Causes its root causes
Heavy rainfall; excessive Component Failure Mode Root Causes
External
wind action; weathering; Changes in water level;
Erosion (EE)
vibration changes in permeability;
Slope Instability
Changes in water level; disturbance of due to
(SLE)
changes in permeability; vibration; deformation due
Mountain
disturbance due to vibration; to static load
Slope
Slope Instability settlement of foundation; Seismic activity; mechanical
Seismic
(SLI) partial saturation of areas properties of mountain slope
Vulnerability
designed to remain dry; material; uncompacted
Main Dyke (SV)
deformation due to static mountain slope soil particles
load
Seismic Seismic activity; low shear TABLE XXV. Effects analysis for Tailings Pond 3
Vulnerability strength of embankment and Failure Intermediate
Component Direct Effect End Effect
(SV) tailings Mode Effect
Heavy rainfall; inadequacy Collapse at Reduction
Deformation
Overtopping of drainage system; increase downstream in stability
SV of
(OT) in sediments due to eroded face of of tailings
embankment
materials tailings dam dam
Heavy rainfall; excessive Local Eroded
External
wind action; weathering; rotational materials at Reduction
Erosion (EE) EE
vibration slips on the the bottom of of dam
Seepage and Permeable embankment; rise crest the dyke height that
Main Dyke
Internal Erosion of water table and phreatic Circular arc Collapsed prevents
(SIE) line slide at materials at overflow
SLE
Offset Dyke Seismic Seismic activity; low shear downstream the bottom of of tailings
Vulnerability strength of embankment and face the dyke
(SV) tailings Overflow of
Destabilizatio Tailings
Heavy rainfall; inadequacy tailings over
OT n of dyke dam
Overtopping of drainage system; increase crest of main
system breach
(OT) in sediments due to eroded dyke
materials Eroded
Inadequate material
Penstock A Seismic activity; external
Structural Local will
and B forces from tailings Reduction in
Design (ST) rotational increase
EE compactness
Drain Inadequate Seismic activity; external slips near volume
and stability
Tunnel A Structural forces from tailings and water level retained in
and B Design (ST) embankment tailings
Inadequate Seismic activity; increase dam
Spillway Structural water flow from extreme Weaker
Design (ST) weather events Seepage of dam wall
Inadequate Weak material beneath slope tailings Piping or due to
Foundation of embankment; low shear SIE water; internal voids or
Design (FN) strength; fissured clay layer reduction of erosion presence
Offset
Seepage and Permeable foundation shear strength of seepage
Dyke
Internal Erosion material; rise of water table channels
(SIE) and phreatic line Collapsed
Seismic activity; static load material
Foundation Deformation; will
Mine from weight of tailings; rise
Strata liquefaction Collapse at increase
subsidence of water table and phreatic SV
(MS) line, which also increases the of tailings upstream face volume
load sand retained in
Seismic activity; low shear tailings
Seismic dam
strength of foundation
Vulnerability Overflow of Mobilization Offset
material and tailings;
(SV) tailings over of tailings dyke is
uncompacted tailings OT
Mountain External Heavy rainfall; excessive crest of offset onto main unable to
Slope Erosion (EE) wind action; weathering; dyke dyke function
vibration Damage on
Penstock A Leakage of
Seepage and Permeable mountain slope ST the concrete
and B tailings on Tailings
Internal Erosion soil particles; rise of water plugs
mountain dam
(SIE) table and phreatic line Drain Damage on
slope and breach
Tunnel A ST the concrete
ground
and B plugs

105
TABLE XXVI. Continuation of effects analysis for Tailings Pond 3 The structural component of the drainage system must be
Failure Intermediate protected because it is prone to failure that will likely lead to tailings
Component Direct Effect End Effect
Mode Effect spill. In TABLE XXVI, it can be observed that the end effect of the
Overflow failure mode for penstocks and drain tunnels is tailings dam breach.
Inability to Inadequacy to of tailings Based on the incident of Tapian Pit in Marinduque, Philippines in
regulate flow remove dam; March 1996, discontinuities in the structure, such as plugged tunnels
Spillway ST are susceptible to failure since the pressure from the tailings is
of water; excess water tailings
damage on dam dam considerably large [18]. In addition, a breach already occurred in
breach penstock A that mobilized a large volume of tailings into Balog
Reduction of River; thus, the plug in the drain tunnels must be effective and
Excess regularly monitored to assure that the same penstock will not be
shear Tailings
porewater breached again. In addition, the end effect of the failure for spillways
FN strength; dam
pressure; is also tailings dam breach because spillways are temporary
internal breach
seepage structures that are removed before the tailings dam is raised and will
erosion
Weaker be reconstructed again after.
Seepage of dam wall
Instability; Failure of tailings dam foundation due to mine subsidence,
tailings due to
internal seismic vulnerability, and inadequate foundation design will likely
SIE water; voids or
Foundation erosion or lead to mobilization of a large volume of tailings. The Padcal Mine
reduction of presence
Strata piping Site pioneered underground block cave mining, in which the
shear strength of seepage
channels incidence of tailings dam breach in 2012 exposed sinkholes due to
Tailings underground mining, which was then filled using concrete spheres.
MS Settlement Instability dam However, sinkholes may still be present near the tailings dam, which
breach will increase the likelihood of settlement since the dam is still active
Settlement; and more tailings are acquired that will increase the load that the
Tailings foundation must support. Furthermore, the location of Padcal Mine
liquefaction Permanent
SV dam Site Tailings Pond 3, is prone to earthquakes; thus, the likelihood of
of tailings deformation
breach settlement triggered by earthquake is also significant. Since the
sand; rupture
Local Eroded tailings dam has an existing plugged underground drain tunnel
Reduction in illustrated in Fig. 11 and Fig. 12, the likelihood of foundation failure
rotational and
EE compactness may increase due to the discontinuity of the component. Fissures in
slips near collapsed
and stability the foundation strata may be present near the drain tunnels, which
water level material
Deformation Collapse at will may exacerbate tailings dam failure if not regularly monitored.
SV
of slope upstream face increase
volume VIII. CONCLUSION
Circular arc Reduction in
retained in
Mountain SLE slide at compactness
tailings Storage facilities, such as tailings dams, must be protected
Slope upstream face and stability
dam to avoid failure that will lead to discharge of a large volume of
Weaker tailings that will directly affect the groundwater, surface water, soil,
Seepage of dam wall and vegetation. Thus, qualitative risk assessment, such as FMEA, is
Instability;
tailings due to conducted to mitigate said disaster. To conduct the risk assessment,
internal
SIE water; voids or the vulnerability of element at risk and hazard occurrence must be
erosion or
reduction of presence identified. Since there are no available directories of active and
piping effect
shear strength of seepage inactive tailings dams, the hazard occurrence for the mines was
channels determined with the assumption that its designated tailings dam is
nearby. Mines in Zambales, Dinagat Island, Leyte, Bulacan,
The effects analysis of Padcal Mine Site Tailings Pond 3 Batangas, Rizal, Albay, and Camarines Sur have high typhoon
shows that overtopping will likely lead to tailings dam spillage incidence, while mines in Abra and Isabela have very high typhoon
among any other failure modes for the main dyke. Overtopping is incidence. On the other hand, mines in South Cotabato, Surigao Del
mostly caused by heavy rainfall and mismanagement of water level. Norte, Dinagat Island, Agusan del Norte, and Sarangani experience
Since there are several strong typhoons in the Philippines, the 48 to 75 hits per year, while mines in Agusan del Sur, Surigao del
likelihood of rain-induced failure is high as observed from past Norte, and Surigao del Sur experience 76 to 100 earthquake hits per
tailings dam incidents in the country. Furthermore, once the tailings year. Furthermore, Mines in Benguet, Nueva Vizcaya, Pangasinan,
overflow from the dam wall, it will likely lead to tailings spill due La Union, and Bataan have a high or very high typhoon incidence,
to the high energy flow of debris that can also mobilize part of the and experience 48 to 100 earthquake hits per year; thus, risk
embankment, which will lead to the collapse of the entire tailings assessment should be conducted.
dam wall. Overtopping in the offset dyke will not immediately lead
to the collapse of the dam wall since the main dyke can still prevent Since FMEA is an intensive analysis that requires the
the tailings from spilling onto the environment; however, once the investigation of every component of the tailings dam, a preliminary
offset dyke is unable to function, the embankment material of the qualitative checklist rating was conducted in which the factors that
offset dyke will serve as an additional external force that the main increased the risk of tailings dam breach were evaluated and
dyke has to resist aside from the external force from the flow of categorized as low, medium, or high risk. In applying FMEA, the
debris and tailings water. Hence, external erosion, slope instability, factors that contribute to the vulnerability of the element at risk is
and seismic vulnerability of the main dyke must be avoided to ensure related to the root cause of failure modes; on the other hand, the
that it is strong enough to resist external forces from embankment vulnerability of the element at risk are related to the failure modes.
material of offset dyke and tailings. The factors that contribute to the vulnerability of the tailings dams
are the construction method, deposition of tailings, embankment

106
material of tailings dam and properties of tailings, tailings dam
height, steepness of tailings dam slope, drainage system, demand of
mining products, date of construction, and tailings dam activity. The
factor that contributes the most to the vulnerability is the
construction method; using an upstream method increases the
likelihood of failure caused by its poor stability and high phreatic
line. The vulnerability of the tailings dams are overtopping, seepage
and internal erosion, external erosion, slope instability, seismic
vulnerability, inadequate foundation design, inadequate structural
design, and mine subsidence. Since Padcal Mine Site Tailings Pond
3 was classified as a Class III risk, it was further analyzed using
FMEA. The effects analysis of Padcal Mine Site Tailings Pond 3
shows that overtopping, inadequate foundation design, inadequate
structural design, and mine subsidence will cause tailings dam
breach. Furthermore, the discontinuities due to underground
structures will lead to tailings dam breach, which is exacerbated by
the additional forces from rainfall and earthquake.
IX. RECOMMENDATION
The accuracy of the rapid risk assessment using the
preliminary qualitative risk checklist may be improved by
incorporating weights for each factor based on quantitative data that
may be obtained once the tailings dams are monitored using
measurement devices such as piezometers and sensors to monitor
the condition of the tailings dam. Since data is not sufficient to
conduct a more detailed FMEA, the FMEA of the Padcal Mine Site
Tailings Pond 3 may still be improved with additional data. FMEA
for tailings dams in other sites in the Philippines can be conducted
properly with sufficient data to analyze the failure modes based on
the actual design, which can be compared to the industry standard.
Detection, control, and mitigation measures may also be formulated
with the proper data from monitoring instruments in the dam so that
the impending disaster can be prevented using the detection
measures and controlled using the control measures. In addition, the
relation between failure modes should be considered in the effects
analysis to provide a more accurate qualitative risk assessment that
can also be used as a basis for quantitative risk assessment.
ACKNOWLEDGEMENT
Engr. Cielo Frianeza wishes to thank Dr. Andres Oreta for
the guidance with regards to disaster risk reduction.
REFERENCES
[1] AGHAM, “Environmental Investigation Mission on the
Impacts of the Philex Mining Corporation (PMC) Mine
Tailings Pond 3 Failure,” 2013.
[2] Carating, R. B., and Galanta, R. G., “The Soils of the
Philippines,” Springer Science Business Media Dordrecht
Heidelberg, 2014.
[3] Center for Environmental Geomatics, & Manila Observatory,
Mapping Philippine Vulnerability to Environmental Disasters,
2005.
[4] de la Cruz, G., “Look Back: The 1996 Marcopper mining
disaster,” retrieved from https://www.rappler.com/move-
ph/issues/disasters/165051-look-back-1996-marcopper-
mining-disaster, 2017.
[5] Dinglasan, R. R., “No stranger to mine spills, Philex still wants
rehab of aging tailings pond,” retrieved from
https://www.gmanetwork.com/news/news/specialreports/2940
83/no-stranger-to-mine-spills-philex-still-wants-rehab-of-
aging-tailings-pond/story/, 2013.
[6] Dizon, N., “The Marcopper mine spill and the unending wait
for justice,” retrieved from
https://www.philstar.com/headlines/2019/04/11/1909180/mar
copper-mine-spill-and-unending-wait-justice, 2019.
[7] Dizon, N., and Vera Files, “The Marcopper disaster: A tragedy
that continues in people's veins,” retrieved from
https://news.abs-cbn.com/spotlight/04/04/19/the-marcopper-
disaster-a-tragedy-that-continues-in-peoples-veins, 2019.
[8] Environmental Science for Social Change, “Manila Mining
Corporation in the Surigao del Norte Mining District,”
Philippine Working Group on Mining, 2008.
[9] Landingin, R., and Aguilar, J., “Dirty Past,” retrieved from
https://www.rappler.com/business/special-
report/whymining/whymining-latest-stories/7195-dirty-past,
2008.
[10] Larrauri, P., “Assessing Risks of Mine Tailing Dam Failures,”
Norges Bank Investment Management, 2017.
[11] “Latest mining disaster further signal to junk mining law,”
retrieved from https://www.ibon.org/latest-mining-disaster-
further-signal-to-junk-mining-law/, 2015.
[12] Lyu, Z., Chai, J., Xu, Z., Qin, Y., and Cao, J., “A
comprehensive review on reasons for tailings dam failures
based on case history,” Advances in Civil Engineering, 2019.
[13] Mines and Geosciences Bureau, “Directory of Metallic
Operating Mines in the Philippines,” 2019.
[14] Mines and Geosciences Bureau, “Directory of Non-Metallic
Operating Mines in the Philippines,” 2019.
[15] Owen, J. R., Kemp, D., Lebre, E., Svobodova, K., and Perez
Murillo, G., “Catastrophic tailings dam failures and disaster
risk disclosure,” International Journal of Disaster Risk
Reduction, 2020.
[16] Penman, A. M., Williamson, J. G., Phillips, J., Pimenta de
Avila, j, Abadjiev, C. B., Klohn, E. J., … Noguera, G.,
“Tailings Dams Risk of Dangerous Occurrences,” Commission
Internationale des Grands Barrages, 2001.
[17] Plumlee, G. S., Morton, R. A., Boyle, T. P., Medlin, J. H., and
Centeno, J. A., “An Overview of Mining-Related
Environmental and Human Health Issues,” Marinduque Island,
Philippines. U. S. Geological Survey, 2000.
[18] Roche, C., Thygesen, K., and Baker, E., “Mine Tailings
Storage: Safety Is No Accident,” Nairobi and Arendal: United
Nations Environment Programme and GRID-Arendal, 2017.
[19] Santos, R., Caldeira, L., and Serra, J., “FMEA of a tailings
dam,” Georisk: Assessment and Management of Risk for
Engineered Systems and Geohazards, 2012.
[20] Stark, J., Li, J., and Terasawa, K., “Environmental Safeguards
and Community Benefits in Mining: Recent Lessons from the
Philippines,” In Environmental Safeguards and Community
Benefits in Mining: Recent Lessons from the Philippines, 2006.
[21] U.S. Environmental Protection Agency, “Design and
Evaluation of Tailings Dams,” Washington, DC: Office of
Solid Waste Special Waste Branch, 1994.
[22] Wills, B. A., and Finch, J. A., “Tailings Disposal” in Wills'
Mineral Processing Technology: An Introduction to the
Practical Aspects of Ore Treatment and Mineral Recovery (8th
ed., pp. 439–448), Waltham, MA: Butterworth-Heinemann,
2016.
[23] Wood, H., “Disasters and Minewater,” London: IWA
Publishing, 2012.
[24] “World Mine Tailings Failures from 1915”, Retrieved from
https://worldminetailingsfailures.org/, 2019.
107
Information and
Communications
Technology
Session Chair : Dr. Edwin Sybingco
Moderator : Engr. Jose Martin Maningo
Panelists : Dr. Kathleen Aviso
Dr. Angela Bandala
Engr. John Anthony Jose
 Predicting Stag and Hare Hunting Behaviors Using
Hidden Markov Model
Rex Bringula1,a* and Ma. Mercedes T. Rodrigo2,b
1,2
Ateneo de Manila University
a
rexbringula@gmail.com, bmrodrigo@ateneo.edu
Abstract— In this paper, we used Hidden Markov Model (HMM)
to describe the gaming behaviors of students and whether they will
exhibit “stag” or “hare” hunting behavior in a mobile game for
mathematics learning. We found that there is a 99% probability
that the students will stay either as stag or hare hunters. Our results
also suggest that they would choose arithmetic problems involving
addition. These game behaviors are not beneficial to learning
because they are only exhibiting mathematical skills they already
know. The results of the study show that stag and hare hunters
have unique traits that separate the one from the other.
Keywords—collaborative learning, game behavior,
hidden Markov model, Viterbi algorithm
I. INTRODUCTION
A collaborative learning environment posits that students
learn while they are in a social group [4]. As social learners,
students may display varying degrees of participation [1]. Stag and
hare hunting behaviors are the tendency of a leaner to contribute to
the scores of the group by choosing either a high-risk mode but
with higher points or a low-risk mode but with lesser points [2].
This game behavior is relatively unknown in the field of a
collaborative mobile learning environment. In this study, we
attempted to describe the game behaviors of the students and
predict whether students will exhibit the stag and hunting
behaviors. The output of the study will serve as a basis in the
development of adapted mobile-based learning for mathematics.
II. METHODOLOGY
This study used the dataset collected by Bringula and
Rodrigo [2]. The dataset is composed of interaction log files that
were generated through the use of a mobile collaborative game
named Ibigkas!Math. It is a mobile-based learning application for
grades 1 to 6 students. It is a collaborative game that covers
arithmetic problems (addition, subtraction, multiplication, and
division of whole numbers and fractions). The students utilized the
mobile game for 15 minutes in an experiment conducted during
their class session. The application generates arithmetic problems
and it displays on one of the team members' mobile devices. The
player has to read aloud the arithmetic problems. The answers are
presented in multiple-choice. The correct answer will appear on
one of the team members' devices. The complete description of this
game is discussed extensively in Bringula and Rodrigo [2].
The dataset contains 4,628 solved problems. The log files
have six features, such as difficulty level, type of problem solved,
speed, time spent, and the number of correct attempts. The
difficulty level and type of problem solved are game modes. In the
difficulty level, students may choose a very easy (ve), easy (ea),
medium (md), hard (ha), or very hard (vh) game setting. The type
of problem solved is the type of arithmetic problems (i.e., addition
(add), subtraction (sub), multiplication (mul), division (div)) that
the student attempted to solve. Students were informed that prizes
would be awarded to three groups of students with the highest
game scores. Twenty-five boys and 12 girls with an average of 11
years old participated in the study. They had varying degrees of
mathematics abilities – eight were low-performing, 13 were
average-performing, and 16 were high-performing.
Speed is a game setting that determines the pace of the
game. Students may choose a very slow (2 points), slow (5 points),
medium (10 points), fast (15 points), or very fast (20 points) mode.
Students were informed that there were no deductions of points for
wrong answers but on the time limit. Students that chose the first-
three settings were labeled as hare hunters; otherwise, they were
labeled as stag hunters. We used the hidden Markov model (HMM)
to predict whether students would exhibit a stag or hunting
behavior given the sequence of difficulty level and type of problem
solved. These variables were used because we found in our
separate study that these were the significant variables of a
decision tree model that could classify stag and hare hunter
behaviors [3]. There are 20 possible game settings (or states) (5
difficulty problems times 4 types of problem) that a participant
may choose from. Furthermore, there are 400 possible transitional
states (20 states multiplied by 20 states). Initial probabilities for the
HMM were determined based on the log files. HMM using the
Viterbi algorithm was implemented in Jupyter Notebook. Thirty
random samples with ten random observations were fed into the
HMM model to determine the possible game behavior of the
students. The random samples were generated through a
spreadsheet. The number of random observations is based on the
average number of questions a student can solve per game session.
On average, one student solved nine questions per game session.
III. RESULTS AND DISCUSSION
We found that there is a very high probability (99%) that
the students will stay at their current game behavior (Figure 1). The
result suggests that the students will persistently choose a game
mode based on the speed setting. This means that they will choose
a game setting that they are comfortable with and are less likely to
explore new game speed settings. This game behavior may not be
beneficial to mathematics learning since students are only
exhibiting the skills that they already know.
The graph in Fig 1 also shows that they choose only 10
states out of the 20 possible states. These states are easy and
addition (eaadd), easy and multiplication (eamul), easy and
division (eadiv), very easy and subtraction (vesub), very easy and
addition (veadd), very hard and addition (vhadd), medium and
addition (mdadd), medium and multiplication (mdmul), medium
and subtraction (mdsub), and hard and addition (haadd). It is worth
noting that they are more likely to choose problems involving
addition. These game behaviors can be attributed to the fact that
the goal of the students is to achieve high scores. It can be also
observed from the figure that the distribution probabilities in hare
hunters are scattered among the 10 states. The finding implies that
hare hunters are solving diverse problems than stag hunters.
109
 This study aims to describe the game behavior of
students who utilized a mobile game learning application in
mathematics and predict whether they will exhibit a stag or a hare
hunter game behavior. It is found that there is a very high
probability that students will stay in the game settings they choose
until the end of the game. Students that display a hare hunting
behavior solve a variety of problems in an easy and slow pace
manner. This is different from the students who are stag hunters.
Fig 1. A hidden Markov model for relating the game mode settings Stag hunters are risk-takers – solving more difficult problems also
to the game behavior (stag and hare) with a faster game setting. Thus, we found evidence that stag and
hare hunters of a mobile game learning application have unique
Tables I and II show the three possible cases the students traits that separate the one from the other.
may exhibit during game sessions. The other twenty-seven possible
cases have the same output as the first-three observations (Obs). It The results of this study will be implemented in the
indicates that students are more inclined to be hare hunters – a development of an adaptive version of Ibigkas! Math. The adaptive
"slowly but surely" game behavior. The results shown in Table 1 version will be then tested to determine the changes in the gaming
further confirm that hare hunters solve a variety of problems. The behaviors of the students. Its influence on students’ mathematics
hare hunters will attempt all arithmetic problems with varying learning will also be investigated.
levels of difficulties.
ACKNOWLEDGEMENT
TABLE I. Predicted Game Behaviors Given the Game Settings
N Obs1 Behavior n Obs2 Behavior The researchers are indebted to Dr. Nieva Discipulo,
1 haadd Hare 1 vhadd Stag Ellen Tabayan, UE Elementary and Senior High School Laboratory
2 mdmul Hare 2 vhadd Stag School staff, my research assistants, and to all participants of the
study. This paper is accepted for oral presentation at the
3 mdsub Hare 3 vhadd Stag International Conference on Computers in Education 2020.
4 mdadd Hare 4 vhadd Stag
5 vesub Hare 5 vhadd Stag REFERENCES
6 eadiv Hare 6 vhadd Stag
7 vesub Hare 7 vhadd Stag [1] R. P. Bringula, M. M. T. Rodrigo, J. Ocumpaugh, K.
8 eamul Hare 8 vhadd Stag Porayska-Pomsta, I. Olatunji, and R. Luckin, “Towards the
development of a computer-based game for phonemic
9 vesub Hare 9 vhadd Stag
awareness”, In J. C. Yang et al. (Eds.), Proceedings of the
10 veadd Hare 10 vhadd Stag 26th International Conference on Computers in Education.
Paper presented at the International Conference on Computers
TABLE II. Predicted Game Behaviors Given the Game Settings in Education (pp 657-662). Philippines: Asia-Pacific Society
(continuation) for Computers in Education, 2018.
N Obs2 Behavior [2] R. P. Bringula, M. M. T. Rodrigo, “Can "stag-and-hare hunt”
1 vhadd Stag behavior be modeled using interaction data from a mobile-
2 vhadd Stag supported collaborative learning application?”. In M. Chang
3 vhadd Stag et al. (Eds.), Proceedings of the 27th International Conference
4 veadd Stag on Computers in Education. Paper presented at the
5 vhadd Stag International Conference on Computers in Education (pp 761-
6 eadiv Hare 766). Taiwan: Asia-Pacific Society for Computers in
7 eamul Hare Education, 2019.
8 mdadd Hare [3] R. P. Bringula, M. M. T. Rodrigo, “Modeling stag and hare
9 veadd Hare hunting behaviors” (unpublished manuscript). Ateneo de
10 vesub Hare Manila University, Philippines, 2020.
[4] R. Gillies, “Cooperative learning: Review of research and
practice” Aust. J. of Teach. Edu., vol. 41, no. 3, pp. 39–51,
Meanwhile, a stag hunter tends to solve very hard 2016. doi:10.14221/ajte.2016v41n3.3
addition problems. Perhaps, they believe that they will gain more
points by solving more difficult problems. Consistent with the
definition of stag behavior, the students who chose to set the game
mode in a faster setting are risk-takers. A faster game setting may
incur higher points but the game imposes higher time deductions
for every wrong answer. The HMM model can also predict that
students may also shift from being a stag to a hare hunter (and vice
versa). As previously shown, the shift in a game behavior is very
minimal. Nonetheless, it provides an opportunity for learners to
explore more game settings.

IV. CONCLUSION

110
 Marketing and Engagement Tool for Educational
Institutions
John Cristian Barrameda1, a , Eizer Edzon Relayson1,b, Richelle Ann B. Juayong1,c, Jasmine A. Malinao2,d
,
Jaime D.L. Caro1,e, and Jozelle C.Addawe*1,f
1
Service Science and Software Engineering Lab
Department of Computer Science
University of the Philippines Diliman
Quezon City, Metro Manila
2
Headstart Business Solutions, Inc.
Quezon City, Metro Manila
a
jmbarrameda@up.edu.ph, begrelayson@up.edu.ph, crbjuayong@up.edu.ph, djas_malinao@yahoo.com,
e
jdlcaro@up.edu.ph, fjcaddawe@up.edu.ph*

Abstract— Educational institutions find it necessary to make
efforts to promote high student recruitment, minimize student
dropouts and encourage continuous student engagement and
alumni engagement. However, in the Philippines, there is currently
a lack of research that aims to develop a system which focuses to
address student recruitment, student engagement, and alumni
engagement. In this paper, we present a web application system
design that can be used by institutions as a marketing and
engagement tool designed to promote student recruitment, student
engagement, and alumni engagement of an institution. The features
of the application include an interactive survival guide, a chatbot, a
Social listener and an account management system. Altogether, the
system is a timely and cost efficient way to evaluate and maintain
the engagement level of students to an institution even in a remote
manner.
Keywords—education, marketing tool, engagement tool,
engagement tool, student recruitment, student engagement
of marketing and communication actions [2]. Making sure that
students do not suffer a tendency to get bored, dispassionate,
disaffected, or otherwise “disengaged” fosters student engagement
[3]. Finally, an institution may offer personally relevant benefits
that attract the attention of the alumni, and incentivize them to
respond in order to encourage alumni engagement [1, 4].
Some of the factors that influence student recruitment are school
visibility, school prestige, and suitable programs for the students
while student retention may be influenced by an institution’s ser-
vices and policies, buildings and infrastructures or concerns
regarding school fees.
School visibility covers the marketing strategies of the institution
[5] while school prestige includes faculties’ reputation, government
and private agencies’ recognition, and student’s proximity and
accessibility to the educational institute [6, 7]. Suitable program is
related to course compatibility, tuition fee affordability, and an
institution committed to quality [8, 9].

I. INTRODUCTION Services and policies include orientation, tutoring, advisement, and

Student recruitment refers to the process of finding and recruiting
the best-qualified students, from within or outside an institution, in
a timely and cost effective manner. Student engagement, on the
other hand, is the degree of attention, curiosity, interest, optimism,
and passion that students show when they are learning or being
taught. It extends to the level of motivation they have to learn and
continue to progress in their education. Moreover, alumni
engagement is the level of attraction, connection, affection, and
influence an alumnus has with their alma mater over time [1].
It is important for educational institutions to identify factors that
influence student recruitment and student retention so that they can
recruit the best students, make sure they graduate on time, and
make students continue to come back and be engaged in the
institute through different means.
progress monitoring [10]. Buildings and infrastructures cover the
facilities, library literature, and information science [11,12].
Concerns regarding school fees are related to inability or perceived
inability to afford the total costs of the education institution and the
perception that the cost of an educational institution outweighs its
benefits [13].
In order to recruit more students, keep students from dropping out
and transferring, and to make them continue to come back and be
engaged in the institute even after they graduate, previous studies
[14, 15] have shown that schools are finding it necessary to
prioritize efforts which aim to achieve the goals stated above. But
to the best of our knowledge, there is a lack of studies in the
Philippines which aim to develop a system which focuses to
address student recruitment, student engagement, and alumni
engagement.

Promoting and fostering each of the key aspects mentioned above In this paper, we propose a design of a web application system for
require some effort to be carried out by an institution. In order to institutions to promote and foster student recruitment, student
attract, engage and convert individuals into new and consistent engagement and alumni engagement of an institution.
enrollments, an institution, university or school, may carry out a set

111
The outline of the paper is as follows: Section II presents our ob-
jectives and considerations in designing the application, in Section
III we present the application design, initial visualizations and
system testing plans. Conclusions are then summarized in Section
IV.

II. OBJECTIVES AND

CONSIDERATIONS
This study aims to address student recruitment, student engage-
ment, and alumni engagement in the Philippines by developing a
web application system that consists of the following features: (1)
An Interactive survival guide which may help orient the students
with the institutional buildings and keep them updated with events
within the institution. (2) A Social listener to track the engagement
of people through social media posts. (3) A Chatbot to allow
students or guests to ask questions or clarify details remotely
through the system. (4) An Account management system that will
implement access control across the application.

These features were identified by the proponents so that it can Fig. 1. Login Interface
cover important considerations that might influence a student’s
decision making in entering a school and continuously engaging
with the institution.

Open source libraries are used in the creation of the said features.
The map used in creating the Interactive Survival Guide is limited
to the vicinity of the school and the Chatbot is trained using
English phrases from frequently asked questions about the school.
The social listener is limited to Twitter posts.

The web application system is intended solely to encourage student

recruitment, student engagement and alumni engagement in high
school institutions and not for promotional efforts. It is accessible
through a web browser and also for mobile viewing.
Furthermore, to demonstrate that the system fulfills its intended use
upon deployment to an appropriate environment, validation testing
is implemented.
III. APPLICATION DESIGN
The application system has 3 main components: an interactive sur-
vival guide, a social listener and a chatbot. An account
management system is incorporated to allow the system contents to
be managed and customized using an administrator account.
A. Account Management
The users of the system can either be an administrator or a guest.
Administrators are authorized persons within the institution such as
IT Administrators or Admissions Office members which are given
username and password credentials in order to manage the contents
of the application. A Viewer/Guest user does not need login cre-
Fig. 2. Main Menu / Homepage
dentials to view the system contents and may include prospective
students, current students, and alumni. Upon opening the system
application, the user first selects a user type then login to the
system if he/she has an authorized credential, this is shown in
Fig.1. The user is then directed to the homepage where he/she may
choose to open the main features of the system with the interface
shown in Fig. 2.
B. Interactive Survival Guide
The interactive map menu shown in Fig. 3 has two links, one for
the registration tutorial (New Game) module and another for the
event/place markers (Markers) module. Configure buttons are only
accessible when an administrator is signed in and leads to the
dataset editor in the mapbox studio where the map markers are
added, updated, or deleted.
The New Game module in Fig. 4 displays the map of the institution
and a text area on the right containing instructions for the
registration procedures of the institution. The map has a search bar,
geolocator button, toggle 3D buildings, and basic navigation
controls. The text area is only editable when an administrator is
signed in.

The Markers module in Fig. 5 has a similar UI to the New Game

112
Fig. 3. Interactive Map Menu
Fig. 4. Interactive Map New Game Module
Fig. 5. Interactive Map Markers Module
module. It contains a list of marker suggestions made by guests and
other administrators with its description, name, and coordinates.
The list is used by administrators as a guide in updating the dataset
that stores the map markers and details of the Interactive Map.
Administrators of the system can submit a form containing the
coordinates and the label name to suggest a mark for the system
administrator to add, edit, or delete map markers. Guests/Viewers
may only view the list of requests of the users to add, edit, or delete
a mark.
Guests can only view the interactive map and the details associ-
ated to the marked locations and events or view the animation of
the institution’s registration process created by the administrator.
Fig. 6. Institution Contact Details and Chatbot Options
C. Chatbot
A chatbot is created within the system to allow conversation be-
tween the user and the application. It acts as an on demand source
of common knowledge regarding student recruitment, student en-
gagement, and alumni engagement. The chatbox was built using
Snatachbot and using Natural Language Processing techniques to
learn from frequently asked questions (FAQs) and basic
information inputted by the institutions and communicate it to the
user.
In case that user queries cannot be resolved through the chatbot,
the user has the option to contact the institution or initiate a live
chat with an online system administrator as shown in Fig. 6.
D. Social Listening
Social listening is the process of listening to and analyzing conver-
sations happening online about specific topics and keywords. This
feature allows software administrators to:
• Capture Twitter posts concerning the institution via hash- tags
and keywords, and classifying them regarding their sentiment
(positive, negative or neutral).
• Display positive feedback and trending topics about the
institution from Twitter posts on the Application so that guests to
the application system can view it.
• Assess data through the analytics that will be provided by the
application
Control management of the social listening feature involves a
configure dropdown accessible by the administrator. There will be
two options in the dropdown menu:
The Guest View page in Fig. 7 displays feedback regarding the
institution through an embedded link that will be set up by an
administrator. Twitter Publish will be used to get the collection of
twitter posts from the institution’s page. The proponents suggest to
use the institution page’s likes collection to be displayed. Em-
bedded twitter links were used in the system due to automation rule
113

Fig. 7. Social Listening Guest View Sample Layout
Fig 8. Social Listening Admin View Sample Layout
constraints.
The Admin View page in Fig. 8 displays a collection of tracked
conversations about a specific phrase, word, or the institution name
in social media posts. The purpose of this page is to provide some
degree of insight through the analysis of said conversations to be
able to devise a strategy that will better influence targeted
audiences. This page allows the admin to view Graphs and
Analytics, access Keyword Settings and use the Keyword Search.
The Graph and Analytics allow the administrator to monitor social
media interactions that include keyword and institution mentions
through summary charts and graphs containing the following
information:
(1) Number of institution/keyword mentions, to be able to visu-
alize the volume of messages, and peaks in the conversations, as
to predict the potential impact digital marketing plans have on
specific institutional campaigns.
(2) Positive-Negative feedback ratio, to be able to distinguish
the general opinions regarding the keyword or the brand
 the school’s office hours. All these fields are editable by the
(3) Mentions over time by day, to provide a timeframe of when
Fig 9. System User Flow
the specific keyword is being mentioned. 

(4) User post with the highest retweet and followers, to identify
people who greatly impact the brand or keyword.
(5) Trending Topics about the institution.
Within the Keyword Settings, the administrator has the option to
narrow down which social media posts will be extracted by the
social media listener by indicating a list of the following keywords:
(1) Required Keywords - The listener will strictly collect
mentions that are added in this list
(2) Additional Keywords - The listener will collect any mentions
that contains keywords which are added in this list plus the
required keyword
(3) Inessential Keywords - The listener will not collect any of the
mentions that contains keywords added in this list
The administrator can use the keyword search to indicate which
analytics result will be displayed on the page. Same rules from
Keyword Settings can be applied to Keyword Search.
After the administrator provides a keyword to the social listening
process, guests are granted access to feedback, mentions, and list
of trending topics about the institution in the Guest View page.
E. System User Flow
Fig. 9 summarizes the user flow of the system and includes the
main components used behind the creation of the system features.
F. Updating Institution Details
Details of the institution may be incorporated into the system using
an administrator account. Fig. 10 shows the interface design for
filling in the information regarding the institution. Details included
are school name, contact person, school email, mobile number and
administrator.
114

Fig 10. School Details Interface Design 1
E. Testing Plan
Validation testing shall be done to verify that the application
system correctly executes the system features and meets its
purpose. The proponents recommend manual testers from faculty
members and the institution’s registrar while automated testing can
be done by the IT specialists. All testers will be provided a set of
questions and test cases available at
https://tinyurl.com/srseaTestCases. Usability and user satisfaction
will also be tested by interviewing manual testers about the
graphical user interface (GUI) of the system and their overall
experience.
IV. CONCLUSION
In this paper, we presented a design of a web application system
for institutions to promote student recruitment, student
engagement, and alumni engagement of its current, alumni and
prospective students. Its main features include an interactive
survival map, a chatbox and a social listener which can be
managed and customized using administrator accounts. The
Interactive Survival Map allows the students to be continually
engaged by being remotely updated with the current events or
activities within the institution while the registration guide
provides an interactive way to go through the registration process
of the school. The Chatbot and Social Listener together provides
continuous engagement by allowing its users to see how others
perceive the institution as they are mentioned in social media.
Altogether, the features of the application serves as a timely, cost
efficient, web-based marketing and engagement tool as it adds to
the level of attraction, connection, affection and influence of the
institution to its users and viewers. Student engagement is
encouraged through the system so that it allows them to interact
with the institution and be updated of events remotely through their
personal devices over the internet.
The repository link of the application being developed is at
https://github.com/linocrusher/SrSeA while the web application is
currently deployed and can be accessed at
https://srsea.herokuapp.com/. Currently, the social listening
module is still for implementation. Moreover, the proponents
recommend further optimizations to the user interface and leverage
other integration tools to enhance the system’s user experience and
system usability.
REFERENCES
[1] Student engagement. https://www.edglossary.org/student-
engagement/, Feb 2016.
[2] J. Camba, R. E. David, A. Betan, A. Lagman, and J. D. L.
Caro. Student analytics using support vector machines. 2016
7th International Conference on Information, Intelligence,
Systems Applications (IISA), pages 1–6, 2016.
[3] M. M. Courage and K. L. Godbey. Student retention: policies
and services to enhance persistence to graduation. Nurse
educator, 17 2:29–32, 1992.
[4] J. Cuseo. Student retention: The big picture. page 1–10.
[5] 
K. K. Esteves. Exploring facebook to enhance learning and
student engagement: A case from the university of philippines
(up) open university. 2012.

[6] R.Ewing, W.Schroeer, and W.Greene. School location and
student travel analysis of factors affecting mode choice.
Transportation Research Record, 1895:55–63, 01
2004.

[7] A.N.Hess,K.Greer,S.V.Lombardo,andA.Lim.Books,bytes, and
buildings:The academic library’s unique role in improving
student success. Journal of Library Administration,
55(8):622–638, 2015.

[8] F. Maringe. University and course choice: Implications for
positioning, recruitment and marketing. 2006.

[9] C. S. Olson. Is your institution user-friendly? essential
elements of successful graduate student recruitment. 1992.
[10] 
I. Oplatka and J. Hemsley-Brown. The research on school
marketing. 2004.
[11] M.V.Patti.Therelationshipofcollegefacilitiesandservicestostud
entretention. 1993.

[12] R. A. Sevier. How students choose a college. 1987.
[13] 
G. Toyn. Download the 2015 alumni relations survey
results
now!http://blog.alumniaccess.com/webinar_to_release_vaese_
alumni_ benchmarking_survey-results- 0- 0.

[14] G. Toyn. Alumni advancement: Agree on a definition of
engagement.http://blog.alumniaccess.com/alumni_engagemen
t_definition_attraction_ connection_affection_influence, Sep
2016.
[15] 
C. Vargas. What is student recruitment? student recruitment
is... https://www. web2present.com/what-is-student-
recruitment/, Apr 2019.
115
 Semi-immersive Virtual Reality Game for Upper Body
Rehabilitation of Pediatric Cerebral Palsy Patients
Jesrelle Anne S. Abenojar, Kyle Steffi D. Bonifacio, Maria C. Torres
Jaime Caro, Ph.D., Richelle Ann Juayong, Ph.D., Josiah Cyrus Boque*
Service Science and Software Engineering Laboratory
Department of Computer Science
University of the Philippines Diliman
Quezon City, Metro Manila
*jrboque@up.edu.ph

Abstract—Cerebral palsy is a group of permanent and
nonprogressive disorders that can cause problems in movement, motor
coordination, and posture. There are traditional rehabilitation methods
for patients with cerebral palsy, but pediatric patients are often
unmotivated in engaging in these methods since they are arduous. This
research explores the use of virtual reality as a rehabilitation tool for
CP patients as it allows simulation of the real world and gamification,
which can give motivation and engagement to the patients. The
researchers of this study proposed a game that uses the Kinect sensor
for rehabilitating the upper body of pediatric cerebral palsy patients.
On this research, a semi-immersive VR learning and
rehabili-tation system is created for pediatric patients with cerebral
palsy.This VR shall provide educational activities through
gamification of proposed system’s set of activities for physical and
occupational therapy. Also, it will provide visual feedback to the
patients’ perfor-mance and improvements. A questionnaire based on
ISO standard25022 and 25023 will be used to analyse the usability on
patients,engagement and enjoyment factor of this VR

Keywords—Cerebral Palsy, Virtual Reality, Therapy II. REVIEW OF RELATED

I. INTRODUCTION
Cerebral Palsy (CP) is a group of permanent and non-
progressivedisorders that affects movement, motor coordination and
posture[11,19]. According to Rosenbaum et al. [19], this is due to
malfor-mation or damage to one or more parts of the developing fetal
or infant brain. Furthermore, Rosenbaum and his colleagues explain
that cerebral palsy may be associated with problems in
cognition,sensation, communication, and behavior; while it may
sometimes be co-occurring with epileptic disorders and autism.
Usually, patients with Cerebral Palsy undergo therapy and
rehabilitation, and physical therapy and occupational therapy are some
that are mostly used for patients. Physical Therapy involves physical
manipulation of the body to restore, maintain, and im-prove physical
body function[1]. While for occupational therapy,Patel [14] describes
this as the utilization of assessment tools and intervention techniques
to develop, enact, and maintain meaning-ful activities in one’s daily
life for improvement(e.g. washing and dishes, taking out the trash etc.)
In the Philippines set-up, data from the profile of cerebral palsy by
Hebreo et al. [11] shows that 64%resorted to physical therapy, 29%
were occupational therapy, and7% were unreported.
However these therapies are often long and require a
rigorous amount of work [21]. Thus, children oftentimes would not be
mo-tivated in performing rigorous and repetitive exercises [20]. So to
address the identified concerns, virtual reality(VR) is being con-
sidered to be as a therapy tool, because it can simulate real-life
environment with varying the activities[16,18] through gamifica-tion
of the rehabilitation process [22].
LITERATURE
Virtual Reality (VR) is an environment that emulates the real world or
simulates a world to seem real [18]. Furthermore, this emulation of the
real world and simulation of another is experienced in vari-ous levels
of immersion created through perception and interaction.Perception is
the degree in which one perceives the world with the basic senses
(sight, hearing, touch, smell, taste). Interaction is the degree to which
one can affect the virtual environment with one’sbody, such as with
moving one’s arms or walking in place.
Cipresso et al. [6] define three levels of immersion based
on their degree of perception and immersion. The following levels
are:
● Non-immersive - This level is described as having a limited
degree of perception and interaction. The setup for this level
is usually a desktop virtual environment (DVE) where the
user uses a keyboard or mouse input device.
● Semi-immersive - This level is described as having a limited
perception but wide or full use of interaction. The setup
found is the user using semi or full-body movement for input
without any head-mounted devices (HMDs) for vision.
● Fully Immersive - This level is described as utilizing full use
of both perception and interaction. The setup uses HMDs
forvision and body motions for interaction. Some setups can
have treadmills that provide altered elevation or
simulatescents for added immersion.
With the ability of VR to emulate the real world while also
cre-ating new environments not found, it provides us with a way to
efficiently gamify the rehabilitation process. The use of games and
game elements has risen to occupy many fields in our real
world,including health, fitness, marketing, and politics [10]. Games
are used to promote fitness by giving badges corresponding to different

116
achievements. This gives a sense of achievement to users as VRallows
them to perform tasks they cannot do in the real world [5].Reward
systems are used by businesses to create rapport and give exclusivity
to their best customer.
Lee and Hammer [10] state that the goal of gamification is
touse the motivation and engagement found in game to apply to real
problems, such as health. Additionally, they state that games allow
repeated failure. It invites a safer and more welcoming environment
for trial and error, which helps in maintaining the engagement of the
user.
Several studies explore the application of gamification and
VRto CP rehabilitation. These studies offer different perspectives and
data in developing this research.
One research uses a semi-immersive multi-touch display
screen to simulate the virtual environment in different games [8]. A
motion sensor ball and pen are used for user interaction. The
researchers develop a series of games that are aimed to be used by
occupational therapists. One of the games created is called Catching
Butterflies,which requires the use and coordination of both hands.
The research states their design process which includes the
study of current related literature, visit of cerebral palsy clinics and
ther-apy sessions, and consultation with medical experts [8]. These are
considered in the development of the proposed system, especially with
the consultation with medical people.
Another research uses the sit-down setup but instead of
motion sensor balls, the researchers use the Leap Motion device, which
isa sensor that accepts fine motor input [7]. The system features a
multiple game setup that allows the patient to pick and drop objects in
the VR world. The game aims to improve joint flexion in the hands
which is related to fine motor skills.
Said games feature colorful environments and animation
mod-els. It also heavily focuses on feedback and positive
reinforcement as the study aims to maintain user engagement. To
measure user engagement, the researchers use a device called
MindWave that detects the brain waves of the person wearing the head
device [7].
While they found that the MindWave provides accurate
results in brain wave mapping, they could not make a general claim for
its efficiency as some of the patients rejected the MindWave [7].The
researchers proceed to suggest the use of Kinect sensor for future
iterations of the study to increase the scope of movement captured by
the system [7].
The previous studies utilize a sit-down setup in the game
process.Other researches have a standing setup, most likely to create a
more flexible space for the patient to interact in.The next research
decides to integrate wearable haptic device to simulate wrist and finger
movement. These devices are paired with an HMD in a fully
immersive VR setup [3].
In this paper, level progression is being included as part of
the games. The results (time completed, failed attempts, range of mo-
tion) are computed by the system and are then used to generate the next
level. This is done to avoid stagnation and motivation loss because of
the repetitive game. Nah et al. [13] identifies competition as one of the
principles of gamification. This means that games are aimed to provide
a challenge for users to maintain engagement[13]. A form of
“difficulty progression” is needed to maintain en-gagement and
motivation.
An issue raised by this research is the use of HMDs in the
VRsimulation. Some patients reported nausea and eye-irritation upon
using the HMDs within the length of testing [3]. Pereira [16] also
discusses nausea due to prolonged exposure to HMDs. This provides
us the idea of removing the HMD from the proposed game's hardware
as per the comfort of the patients.
Subsequently, there is a study that incorporates all of this
using a Kinect sensor to create a ludic game system for mobility
disorders[17]. The system consists of six games developed using
Unity. Their Use of Kinect provides us the idea to use the
commercially available device for research. The use of Kinect also
saves the users from wearing body sensors that can cause discomfort
[5].
The study also provides the details of the technical part of
developing their system, which includes the data layers. This can be a
guide in designing the architecture of the proposed system.
Another factor that is being improved upon is the level of
progression or the difficulty curve of the activities. The researchers
develop the games wherein it can be customized by the administer-ing
therapist. Each game has a set of controls to alter the number of
obstacles, distance of each obstacle from each other, and more. This
Approach aims to have experience for each patient and therapist more
personalized.
III. METHODOLOGY
A. Theoretical Framework
Figure 1 shows the details of the theoretical framework that dis-cusses
the bulk of theories and concepts used to create and support the system.
As seen in figure, the patient undergoes pre-assessment to
deter-mine their GMFCS and MACS levels. These levels will
determine if they are included in the scope of the project [9,15]. After
which, the upper body motor proficiency of the patient will undergo
measure-ment based on the BOT2 guidelines on manual coordination
and body coordination [4]. Exercises for manual coordination consist
of reaching, wrist movement, and hand-eye coordination. While he
exercises for the body coordination part consists of alternating
movements of the left and right side of the body. In some
cases,occupational therapy is used instead of physical therapy. The
researchers have listed down some activities that can serve as occu-
pational therapy for the patients [14].
117
 Fig 1. Theoretical Network

Figure 1 also illustrates the different researches discussed in
the literature review and the different approaches they used to
integrateVR to the rehabilitation process [3,7,8,17]. The proposed
game will incorporate the exercises from the manual coordination
inthe controls and objectives of the game. It will use Kinect as the input
and output device that will simulate the VR environment. There-
assessment of the patient will be inputted into the system to keep track
of records. After that, a testing period will occur where testers will
answer two questionnaires; one for usability, which is based on ISO
standards, and one for engagement, which is based on
GAMEFULQUEST.
The therapist can adjust the game parameters based on the
mo-tor skills of the player. The increase in difficulty will then be based
on the set parameters. For now, the identified modifiable game
parameters are the number of ingredients and the hands to be used(left,
right, or both). These are subject to change based on future
consultations with medical experts.
The setup of the proposed system is shown in Figure 2. It is
a sit-down setup; the player is seated facing the sensor and themonitor.
Fig 2. Proposed Game Set-up

B. Therapy Environment and Game Concept Design

The proposed game aims to highlight reaching and grasping
exercises for upper limb coordination. The game will feature a
cauldron surrounded by different ingredients. The player will be asked
to reach and drag the glowing ingredient towards the cauldron. The
hand to be used, whether left or right hand, will depend on the color of
the glowing light. The game will end as soon as the players able to put
all the ingredients as the game is instructed based on the set game
parameters of the therapist. When the game is finished, the system will
generate a report to help the therapist track the progress of the patient.
This report will be viewed as the player's record.

118
 Fig. 3. Wireframe for Therapist
Figure 3 shows the wireframe for the therapist. The system
will first display an option for viewing the player list, creating a
player,or closing the application. When the first option is chosen, it
will lead to the display of all created players. The therapist will then
choose a player from the list. This will lead to the display of the
player’s details with options of editing the details, viewing therecords,
or starting the game.
The edit details option will lead to the scene where the
therapist can change the player’s details. The view records option will
lead to the display of player’s records. The play option will lead to the
scene of setting the game parameters. Then an option for starting the
game is displayed.
The create player option will lead to a form for creating a
newplayer. After it is successfully created, it will display the details of
the player, an option to go back to the main menu, and a playoption,
which will lead to the set game parameters scene.
When the therapist chooses to start the game, they will be
firstled to an interface where they can set the game parameters that will
suit the motor skills of the player. After setting the game pa-rameters,
they can continue to the game. The game will be both displayed on the
screens of the player and therapist.
Figure 4 shows the wire The game will display a cauldron
and different ingredients. An Ingredient will glow with a specific color
that indicates which hand will be used. The player is then asked to drag
the glowing ingredient towards the cauldron. The dragging of the
ingredients simulates the reaching exercises and elbow flexion done in
traditional therapy settings. The therapist can apply pressure onto the
arm if they want and assist the patient in how long they hold the hand
out. An added difficulty to be inserted if the therapist wishes is the
stirring of thepot. The player uses the right and left flicks of the wrist
to move the ladle. The wrist to be used depends on the color of the
ladle orto be indicated by the therapist.
The game is finished when all the ingredients are
successfully dragged towards the cauldron. It then proceeds to print a
compre-hensive report to the therapist, which is also saved as the
player's record.
C. Hardware
The proposed system would use Kinect 2 for motion capture.
Itallows full free body motion detection, supports Unity game
development, and has available Blender toolkits for motion capture
modeling.
119
 that the therapist will use will have at least the following specifications
so that the Kinect 2 will properly function:
● 64-bit (x64) processor
● Physical dual-core 3.1 GHz (2 logical cores per physical
sys-tem) or faster processor
● 4 GB of RAM
● Graphics card that supports DirectX 11
● Windows 8 or 8.1, Windows Embedded 8, or Windows 10
Additional hardware necessary for the system:
● USB 3.0 controller dedicated to the Kinect for Windows v2
sensor or the Kinect Adapter for Windows for use with
theKinect for Xbox One sensor

D. Software
The system will be using Blender and Unity for software
development. Blender is a free and open source 3D creation suite used
for modeling, rigging, animation, simulation, rendering, compositing,
motion tracking, video editing and game creation. Unity is a cross-
platform game engine that is compatible with multimedia or game
development software such as Blender, and uses C# and JavaScript.

E. Architectural System
The proposed system consists of three layers.

a. Data Layer.This layer stores and allows the retrieving of

data that consist and are generated by the system, which
includes patient records, environment models, therapy
results and gameresults. SQLite database will be used.

b. Smart System.This retrieves the data provided by Kinect

for user detection and interprets the data for the system.

c. User Interface.This layer allows the interaction of the user

with the system as it provides the game interface, patient
profiles,level modification, and reports viewing.

Fig. 4. Wireframe for Players

The Kinect 2 Sensor has the following specifications:

● Field of View: 70 degree horizontal by 60 degree vertical
● Resolvable Depth: 0.8 m -> 4.0 m
● Colour Stream: 1920 x 1080 x 16 bpp 16:9 YUY2 @ 30 fps
● Depth Stream: 512 x 424 x 16 bpp, 13-bit depth
● Infrared Stream: 512 x 424, 11-bit dynamic range
● Registration: Colour <-> depth and active IR
● Audio Capture: 4-mic array returning 48K Hz audio
● Data Path: USB 3.0
● Latency: 60 ms with processing
● Tilt Motor: No tilt motor

The system will also be using a monitor for displaying the

game,and a desktop or laptop for the therapist. The researchers have Fig. 5. Architectural System
estimated a television screen with 32 - 43" will be suitable for visi-
bility and efficiency of use of the system. While the computer or laptop F. Standards and Guidelines for MeasuringUsability and
Engagement

120
 ISO 25022 and ISO 25023 are standards and guidelines to
be used for measuring usability.
ISO 25022 focuses on the quality of use of the software.
These Include measures in user efficiency and risk management. The
focus is on how well the system executes the functionalities, responds
to errors, and the overall user satisfaction upon using the system [2].
ISO 25023 focuses on quality as a product of the software.
These include measures in the maintainability of the system as a
product.The focus is on how easy the product is to maintain after the
initial release and how intuitive the product is to the intended users [2].
These standards will be used to create a questionnaire to be
dis-tributed to the testers during the project’s testing period.
For user engagement, the researchers have decided to create
a questionnaire that emulates the structure of the GAMEFULQUEST.
This is a questionnaire that measures the gamefulness of a game system
based on seven parameters: accomplishment, challenge,guided,
playfulness, competition, immersion, and social [12].
IV. CONCLUSION
With the power of VR, therapy for Cerebral Palsy has given a new
dimension for patients to enjoy. Using this new way of rehabilitation,
patients will be encouraged to do their therapy process, and at the same
time, enjoy the use of the technology. With some evidence of
improvement using similar devices for therapy of Cerebral Palsy or
other mental diseases, it is also expected that this type of VR Therapy
for Cerebral Palsy patients will give the same improvement for patients
mobility especially on their limbs. However, this has not been tested
to show results of improvement of the patients.
To further improve this research, testing for patients is
recommended. Consultation with medical people and therapy centers
is highly recommended too to prove its conceptual framework with
medical evidence and theories. Observing a therapy session can also
be helpful in gaining knowledge and insights to improve the system
and thus creating more gamified activities that patients can enjoy, and
at the same time, improve their situation and way of life.
REFERENCES
[1] 2019. Role of a Physical Therapist.
https://www.apta.org/PTCareers/RoleofaPT/
[2] Nigel Bevan, Jim Carter, Jonathan Earthy, Thomas Geis, and
Susan Harker. 2016.New ISO standards for usability, usability
reports and usability measures. InInternational Conference on
Human-Computer Interaction. Springer, 268–278.
[3] Ilaria Bortone, Daniele Leonardis, Massimiliano Solazzi,
Caterina Procopio,Alessandra Crecchi, Luca Bonfiglio, and
Antonio Frisoli. 2017. Integration of serious games and wearable
haptic interfaces for Neuro Rehabilitation of children with
movement disorders: A feasibility study. In2017 International
Conference onRehabilitation Robotics (ICORR). IEEE, 1094–
1099.
[4] Robert H Bruininks et al.1978.Bruininks-Oseretsky test of motor
proficiency.American Guidance Service Circle Pines, MN.
[5] Yao-Jen Chang, Wen-Ying Han, and Yu-Chi Tsai. 2013. A
Kinect-based upper limb rehabilitation system to assist people
with cerebral palsy.Research in developmental disabilities 34, 11
(2013), 3654–3659.
[6] Pietro Cipresso, Irene Alice Chicchi Giglioli, Mariano Alcañiz
Raya, and GiuseppeRiva. 2018. The past, present, and future of
virtual and augmented reality research:a network and cluster
analysis of the literature.Frontiers in psychology 9 (2018),2086.
[7] Juliana M de Oliveira, Rafael Carneiro G Fernandes, Cristtiano S
Pinto, Plácido RPinheiro, Sidarta Ribeiro, and Victor Hugo C de
Albuquerque. 2016. Novel Virtual environment for alternative
treatment of children with cerebral palsy.Computational
intelligence and neuroscience 2016 (2016).
[8] Alan Dunne, Son Do-Lenh, Gearóid Ó’Laighin, Chia Shen, and
Paolo Bonato. 2010.Upper extremity rehabilitation of children
with cerebral palsy using accelerome-ter feedback on a
multitouch display. In2010 Annual International Conference of
the IEEE Engineering in Medicine and Biology. IEEE, 1751–
1754.
[9] Ann-Christin Eliasson, Lena Krumlinde-Sundholm, Birgit
Rösblad, Eva Beckung,Marianne Arner, Ann-Marie Öhrvall, and
Peter Rosenbaum. 2006. The ManualAbility Classification
System (MACS) for children with cerebral palsy: scale
development and evidence of validity and
reliability.Developmental medicine and child neurology 48, 7
(2006), 549–554.
[10] JESSICA Hammer and J Lee. 2011. Gamification in Education:
What, How, WhyBother.Academic Exchange Quarterly 15, 2
(2011).
[11] Angelita Roma P Hebreo, Cynthia D Ang-Muñoz, Joycie Eulah
E Abiera, Mon-alisa L Dungca, and Betty D Mancao. 2017.
Profile of Pediatric Patients withCerebral Palsy at the Department
of Rehabilitation Medicine, Philippine GeneralHospital. (2017).
[12] Johan Högberg, Juho Hamari, and Erik Wästlund. 2019. Gameful
ExperienceQuestionnaire (GAMEFULQUEST): an instrument
for measuring the perceived gamefulness of system use.User
Modeling and User-Adapted Interaction(2019),1–42.
[13] Fiona Fui-Hoon Nah, Venkata Rajasekhar Telaprolu, Shashank
Rallapalli, andPavani Rallapalli Venkata. 2013. Gamification of
Education Using Computer Games. InHuman Interface and the
Management of Information. Information andInteraction for
Learning, Culture, Collaboration and Business, Sakae
Yamamoto(Ed.). Springer Berlin Heidelberg, Berlin, Heidelberg,
99–107.
[14] Dilip R Patel. 2005. Therapeutic interventions in cerebral
palsy.The IndianJournal of Pediatrics 72, 11 (2005), 979–983.
[15] Andrea Paulson and Jilda Vargus-Adams. 2017. Overview of
four functional classification systems commonly used in cerebral
palsy.Children4, 4 (2017), 30.
[16] E Monge Pereira, F Molina Rueda, IM Alguacil Diego, R Cano
De La Cuerda,Alessandro De Mauro, and JC Miangolarra Page.
2014. Use of virtual reality systems as a proprioception method
in cerebral palsy: clinical practice guideline.Neurología (english
edition)29, 9 (2014), 550–559.
[17] A Pérez-Muñoz, P Ingavélez-Guerra, and Y Robles-Bykbaev.
2018. New approach of serious games in ludic complements
created for rehabilitation therapies in children with disabilities
using Kinect. In2018 IEEE XXV International Conferenceon
Electronics, Electrical Engineering and Computing
(INTERCON). IEEE, 1–4.
[18] Albert A Rizzo and J Galen Buckwalter. 1997. Virtual reality
and cognitive assessment.Virtual Reality in Neuro-Psycho-
Physiology: Cognitive, Clinical andMethodological Issues in
Assessment and Rehabilitation 44 (1997), 123.
[19] Peter Rosenbaum, Nigel Paneth, Alan Leviton, Murray Goldstein,
Martin Bax,Diane Damiano, Bernard Dan, and Bo Jacobsson.
121
 2007. A report: the definition and classification of cerebral palsy rehabilitation tool.Journal of neuroengineering and rehabilitation
April 2006.Developmental medicine and child neurology. 1, 1 (2004), 12.
Supplement 109 (2007), 8–14. [22] David M Whittinghill. 2014. Gamification of physical therapy for
[20] Andrew Patrick Hayes Weightman, Nick Preston, Raymond Holt, the treatment of pediatric cerebral palsy: a pilot study examining
Matthew All-sop, Martin Levesley, and Bipinchandra Bhakta. player preferences.age 24(2014), 1.
2010. Engaging children in healthcare technology design:
developing rehabilitation technology for children with cerebral
palsy.Journal of Engineering Design21, 5 (2010), 579–600.
[21] Patrice L Weiss, Debbie Rand, Noomi Katz, and Rachel Kizony.
2004. Videocapture virtual reality as a flexible and effective

122
 The Effects of Quality of Service Mechanisms
in a Software-Defined Networking Environment
on HTTP Requests
Josiah Eleazar T. Regencia1, a *, William Emmanuel S. Yu, Ph.D2,b
1
Ateneo de Manila University, Loyola Heights, Quezon City
2
Ateneo de Manila University, Loyola Heights, Quezon City
a
josiah.regencia@obf.ateneo.edu, bwyu@ateneo.edu
Abstract—This study explores how Quality of Service
(QoS) mechanisms using the Software-Defined Networking
distributed approach, Leaf-enforced, affect HTTP requests across
the network on multiple layers. It is important to note that traditional
approaches commonly enforced QoS in a core switch, Core-
enforced since this is assumed to perform better because it is only
enforced at a single node. Results in this study have shown that for
2 layers and 3 layers of client leaf switches Leaf-enforced QoS
performed over 30% better than the Core-enforced. Although Core-
enforced QoS performed better by at least 2.2% in 4 layers of client
leaf switches, 5 layers and 6 layers of client leaf switches resulted in
Leaf-enforced having better performance by at least 1% and 2.8%,
respectively.
Keywords—Software-Defined Networks, Quality of
Service, Class-based Queueing, HTTP Traffic
I. INTRODUCTION
The rapid increase in networking applications and
devices connected to the internet has opened opportunities to
network researchers to explore the field of Quality of Service (QoS)
inside a Software-Defined Networking (SDN) environment, a
relatively new technology that decouples the control plane and the
data plane. SDN has been regarded as a potential solution to deal
with the challenges traditional networking architecture have with
regards to QoS. A previous study by the authors introduced a testing
framework for QoS mechanisms and network topologies inside a
SDN environment [1]. Using the test framework, the authors
explored how layers in the network topology affect QoS
performance in the network inside an SDN environment [2] and how
additional network traffic flows affect the performance of the SDN
environment enforced with QoS across multiple hops [3]. All studies
involve simulating synthetic HTTP and streaming traffic along with
ICMP Ping packets. The study involving additional network traffic
flows [3] also included generation of raw IP data traffic with the use
of the iperf3 tool in order to flood the network with large traffic
flows. This study extends the previous study [3] by examining how
QoS mechanisms in the SDN environment affect HTTP requests. In
this study, the SDN approach is referred to as Leaf-enforced QoS
since QoS is enforced distributedly in client leaf switches. The
traditional centralized approach will be referred to as Core- enforced
QoS as QoS will only be enforced in the core switch.
II. RELATED LITERATURE
A. Software-Defined Networking and OpenFlow
Software-Defined Networking (SDN) is a networking
paradigm that introduces a centralized approach to network
management where the control logic of the network is dictated by a
controller located inside the control plane which is directly
connected to all packet forwarding networking devices, the data
plane [4]. The centralized controller is less prone to error when
modifying network policies through high-level languages and
software components. Furthermore, SDN eases into Network
Function Virtualization (NFV) which potentially gives major
advantages in network flexibility, scalability, and also reduction of
infrastructure costs. Lastly, more fine-grained policies can be
developed and deployed for the purposes of traffic engineering,
Quality of Service (QoS) provisioning, security and other network
management essentials [5]. Multiple studies have been explored on
taking advantage of the distributed nature of SDN to address security
issues. Peña and Yu developed the use of a distributed firewall for
SDN [6].
OpenFlow is the protocol that enables network operators
to perform fine-grained implementations of network flows inside the
SDN architecture [7]. Most common SDN controllers use OpenFlow
as the backbone of their framework. Inside every OpenFlow enabled
switch d is the Flow Table which contains instructions how a switch
is going to process a matched flow [8]. OpenFlow is the protocol
used in this study.
B. Quality of Service in SDN
Different needs of different types of network flows makes
applying proper QoS mechanisms a challenging task. Prioritiza- tion
and classification are both key factors in enforcing proper QoS
mechanisms for a certain flow [7]. Civanlar et al. [9] introduced a
linear programming-based formula to calculate QoS-rich paths
where video flows are routed into while best-effort traffic is kept on
shortest paths. OpenQoS [10] uses packet header fields in order to
classify incoming flows as either multimedia flows or data flows
sending multimedia flows to QoS-rich paths while data flows are
routed using best- effort routing.
QosFlow [11] manipulates multiple packet schedulers
such as Hierarchical Token Bucket (HTB), Random Early Detection
(RED), and Stochastic Fairness Queuing (SFQ) schedulers. Chato
and Yu [12] [13] also explored the effects Basic Class- based
Queueing (traffic protocol classifications) and Source Class-based
Queueing (source IP address classification). Both mechanisms were
performed under two test cases. The traditional centralized Core-
123
enforced QoS and the distributed SDN approach Leaf-enforced
QoS. CBQ Class Profile CBQ Class Switch QoS
A Previous study by the authors introduced a test Grouping
framework for QoS mechanisms and network topologies in the
context of SDN [1]. Using the introduced test framework, separate Basic at the Leaf Traffic Protocol Client Leaf
studies [2] [3] have been performed in order to explore how layers Switches
in the network topology affect the performance of QoS for both
Leaf-enforced QoS and Core-enforced QoS. This studies involved Basic at the Core Traffic Protocol Core Switch
simulating synthetic HTTP, streaming traffic, and ICMP Ping tests.
Results were then analyzed for IP raw data throughput, HTTP data Source at the Leaf Source IP Address Client Leaf
throughput, streaming bitrate and latency which showed the Switches
advantage of SDN over the traditional centralized QoS enforcement.
Source at the Core Source IP Address Core Switch
Despite having more nodes enforced with QoS in the network,
results have shown that Leaf-enforced QoS generally performed
Destination at the Destination IP Client Leaf
better than Core-enforced QoS. It is important to note that the Core-
Leaf Address Switches
enforced has always been assumed to perform better since it has only
one node enforced with QoS, hence, commonly used. However,
Destination at the Destination IP Core Switch
when more network traffic was generated with the use of iperf3,
Core Address
results showed that all cases of SDN approach QoS were
significantly better than the centralized approach for both raw IP
Source-Destination Source-Destination Client Leaf
data throughput and latency [3].
at the Leaf IP Address Switches

Source-Destination Source-Destination Core Switch

at the Core IP Address
III. FRAMEWORK
This study used the testing framework introduced by the
authors in [1]. Moreover, the same theoretical and conceptual
framework in this study is already described in the previous studies
[2][3]. This study was also performed in an AWS EC2 m5.xlarge
Ubuntu 18.04 instance. Just like previous studies, the network was IV. METHODOLOGY
simulated using a Mininet version 2.3 with Ryu OpenFlow 1.3 as the
remote controller. Figure 1 shows a sample of the network This study is an extension of the previous study [3] where
topologies used in previous studies and also in this study. This has additional traffic was generated with the use of the iperf3. The same
also been already discussed in previous studies. Table I shows QoS simulation flow chart is used as can be seen in Figure 2.
mechanisms and their descriptions, as also discussed in previous
studies [1][2][3], for reference. The configurations of QoS are also
already discussed in the previous studies [1][2][3].

Fig. 2. Methodology Flowchart from [3]

Fig. 1. Sample Virtual Network Topology for 2 layers of client leaf Apache Bench, VLC streaming, ICMP Ping, and iperf3 are still used
switches in this study to generate traffic. Although the focus is only with
TABLE I. CLASS PROFILES regards to HTTP requests, the simulation of VLC streaming, ICMP
Ping packets, and raw data through iperf3 are still important in order
for the network to have:

124
 1) a diversity of traffic types across the network,
2) and in order to have large traffic flows flowing across the
network.
The large traffic flows is necessary as traffic flowing across today’s
networks increases at a rapid pace.

Fig. 4. 3 Layers Results

Fig. 3. 2 Layers Results

V. RESULTS AND DISCUSSION

Results of Completed Requests and Failed Requests per
layer simulation in this study are shown from Figure 3 to Figure 7.
Generally, it is shown that Leaf-enforced QoS mechanisms have
more Completed Requests than Core-enforced QoS mechanisms.
However, there are some instances that show that Core-enforced
QoS perform better than their Leaf-enforced QoS counterpart. In
order to test the statistical significance of such cases, a two-sample
student’s t-test was performed comparing all cases with one another
with significant level set to α=0.05. At the same time, it is also Fig. 5. 4 Layers Results
observed that although Leaf-enforced QoS have more Completed The results for 3 layers of client leaf switches observed in
Requests, these mechanisms also have more Failed Requests. Again, Figure 4 show that for both Completed Requests and Failed
a student’s t-test is also used to test the statistical significance of Requests, Leaf-enforced QoS mechanisms perform better than Core-
these cases. enforced QoS mechanisms. For Completed Requests, Leaf-enforced
For the 2 layers of client leaf switches, all cases of Leaf- QoS mechanisms perform at least 28% better than the Core-enforced
enforced QoS mechanisms performed better than the Core- enforced QoS mechanism counterparts. For the Failed Requests, Leaf-
QoS mechanism counterpart in terms of Completed Requests. For enforced QoS for Source CBQ have 3% less Failed Requests
Basic Class-based Queueing (CBQ), the Leaf- enforced QoS compared to its Core-enforced QoS counterpart while the Source-
performed 26% better than the Core-enforced QoS. For the Source Destination CBQ Leaf-enforced QoS 12% less Failed Requests.
CBQ, the Leaf-enforced QoS performed 38% better than the Core- Leaf-enforced Basic CBQ and Leaf-enforced Destination CBQ
enforced QoS. Destination CBQ at the Leaf performed 47% better showed better results where these cases had at 30% and 28%,
than Destination CBQ at the Core and lastly, Source-Destination respectively, less Failed Requests than their Core-enforced QoS
CBQ at the Leaf performed 11% better than its Core-enforced QoS counterparts.
counterpart. With regards to Failed Requests, the charts show in
Figure 3 that all cases of Core-enforced QoS performed better than
Leaf-enforced QoS. However, the p-values from the two-sample
student’s t-test showed in Table III that no p-value is less than
α=0.05. Hence this disadvantage is statistically insignificant. Leaf-
enforced QoS still perform at least as good as the Core-enforced QoS
counterparts.

125
 Fig. 6. 5 Layers Results

The case for 4 layers of client leaf switches, Figure 6, is TABLE II. 4 LAYERS COMPLETED REQUESTS
the only case where Core-enforced QoS generally performed better P-VALUES.
than Leaf-enforced QoS. Only Basic CBQ Leaf-enforced QoS
case Basic Basic Src Src Dst Dst Src- Src-
resulted in having more Completed Requests than its Core-enforced
at at at at at at Dst at Dst at
QoS counterpart, approximately 18% more Completed Requests.
Leaf Core Leaf Core Leaf Core Leaf Core
The rest showed that the Core-enforced QoS mechanism has more
Completed Requests. However, the p- values for this case in Table I Basic at 1.0 0.350 0.461 0.975 0.531 0.967 0.942 0.967
shows that no p-value is less than α=0.05. Hence, the difference is Leaf 1 8 8 9 3 5
insignificant and the Leaf- enforced QoS still perform as good as
their Core-enforced QoS counterparts. In the same way, where Core- Basic at 1.0 0.834 0.328 0.762 0.357 0.378 0.332
enforced QoS all have less Failed Requests than Leaf-enforced QoS, Core 3 4 7 2 3
the p-values in Table II also show that the differences are sta-
Src at 1.0 0.437 0.921 0.474 0.498 0.439
tistically insignificant. Hence, despite having less Completed
Leaf 3 5 7 1 7
Requests and more Failed Requests, Leaf-enforced QoS still
perform as good as Core-enforced QoS for HTTP requests in the Src at 1.0 0.506 0.942 0.917 0.991
case of 4 layers of client leaf switches. Core 8 2
For 5 layers of client leaf switches, Leaf-enforced QoS
mechanisms once again show to have more Completed Re- quests Dst at 1.0 0.547 0.571 0.506
than the Core-enforced QoS mechanisms counterparts. Although Leaf 1 9
Leaf-enforced Basic CBQ QoS and Leaf-enforced Destination CBQ Dst at 1.0 0.973 0.934
QoS only have 0.5% and 1.7%, respec- tively, more Completed Core 5 9
Requests than their Core-enforced QoS counterparts, Leaf-enforced
Source CBQ QoS and Leaf- enforced Source-Destination CBQ QoS. Src-Dst 1.0 0.909
For Failed Requests, once again Core-enforced QoS have less Failed at Leaf 7
Requests than Leaf-enforced QoS. But the p-value in this case was
0.9735 which is less than α=0.05, hence statistically insignificant. Src-Dst 1.0
For 6 Layers of client leaf switches, despite Core-enforced at Core
Source CBQ and Core-enforced Source-Destination CBQ hav- ing
more Completed Requests than their Leaf-enforced QoS
counterparts, this is shown to be statistically insignificant in Table
V as no p-value is less than α=0.05. Moreover, Leaf- enforced Basic
CBQ QoS and Leaf-enforced Destination CBQ QoS have 16% and
4%, respectively more Completed Requests than their Core-
enforced QoS counterparts. With regards to the Failed Requests, the
Core-enforced Destination CBQ QoS having less Failed Requests
compared to the Leaf-enforced Destination CBQ QoS is shown to
be statistically significant with p-value 0.0309. This is shown in
Table VI.

126
 TABLE III. 2 LAYERS FAILED REQUESTS case Basic Basic Src Src Dst Dst Src- Src-
P-VALUES. at at at at at at Dst at Dst at
case Basic Basic Src Src Dst Dst Src- Src- Leaf Core Leaf Core Leaf Core Leaf Core
at at at at at at Dst at Dst at Basic at 1.0 0.409 0.340 0.355 0.309 0.200 0.541 0.604
Leaf Core Leaf Core Leaf Core Leaf Core Leaf 1 3 9 6 3 4
Basic at 1.0 0.846 0.756 0.531 0.579 0.783 0.585 0.521 Basic at 1.0 0.885 0.922 0.814 0.626 0.871 0.759
Leaf 7 7 7 2 1 7 Core 7 7 4 8 9 6
Basic at 1.0 0.611 0.659 0.449 0.627 0.454 0.640 Src at 1.0 0.960 0.924 0.735 0.770 0.657
Core 1 9 8 6 2 Leaf 8 2 9 9 8
Src at 1.0 0.344 0.805 0.960 0.815 0.345 Src at 1.0 0.885 0.693 0.801 0.686
Leaf 9 5 2 4 1 Core 2 4 9 9
Src at 1.0 0.233 0.341 0.233 0.951 Dst at 1.0 0.819 0.710 0.601
Core 5 6 2 Leaf 8 6
Dst at 1.0 0.759 0.987 0.238 Dst at 1.0 0.546 0.436
Leaf 1 9 Core 4 5
Dst at 1.0 0.768 0.343 Src-Dst 1.0 0.901
Core 7 3 at Leaf 8
Src-Dst 1.0 0.238 Src-Dst 1.0
at Leaf 5 at Core
Src-Dst 1.0
at Core
TABLE VI. 6 LAYERS FAILED REQUESTS
P-VALUES.
Fig. 7. 6 Layers Results
case Basic Basic Src Src Dst Dst Src- Src-
at at at at at at Dst at Dst at
TABLE IV. 4 LAYERS FAILED REQUESTS Leaf Core Leaf Core Leaf Core Leaf Core
P-VALUES.
Basic at 1.0 0.196 0.576 0.447 0.032 0.955 0.011 0.238
case Basic Basic Src Src Dst Dst Src- Src- Leaf 5 2 2 8 3 8 8
at at at at at at Dst at Dst at
Leaf Core Leaf Core Leaf Core Leaf Core Basic at 1.0 0.054 0.591 0.349 0.197 0.182 1.0
Core 6 7 4 1 4
Basic at 1.0 0.866 0.799 0.267 0.209 0.106 0.611 0.190
Leaf 7 4 3 2 1 5 Src at 1.0 0.172 0.006 0.517 0.001 0.083
Leaf 3 1 3 7 2
Basic at 1.0 0.936 0.353 0.279 0.149 0.738 0.253
Core 7 3 4 5 1 Src at 1.0 0.150 0.461 0.067 0.624
Core 7 4 2 2
Src at 1.0 0.375 0.295 0.154 0.790 0.266
Leaf 5 5 9 Dst at 1.0 0.030 0.701 0.386
Leaf 9 3 3
Src at 1.0 0.833 0.548 0.542 0.752
Core 8 2 2 Dst at 1.0 0.010 0.242
Core 6 8
Dst at 1.0 0.710 0.433 0.911
Leaf 5 9 1 Src-Dst 1.0 0.216
at Leaf 3
Dst at 1.0 0.246 0.809
Core 8 3 Src-Dst 1.0
at Core
Src-Dst 1.0 0.390
at Leaf 9
VI. CONCLUSION
Src-Dst 1.0
at Core
This study was performed in order to analyze the effects
of Quality of Service mechanisms in a Software-Defined
TABLE V. 6 LAYERS COMPLETED REQUESTS Networking architecture with regards to HTTP requests.
P-VALUES. Simulations in this study also included VLC streaming, ICMP Ping

127
latency tests, and also iperf3 raw IP data traffic generation. These
tests were also done in multiple layers of client leaf switches where
QoS is distributedly enforced. The purpose was to replicate a real-
world networking scenario where network traffic is large and travel
across multiple hops in order to reach the target destination. Results
in this study have shown that in general, more HTTP requests were
completed with the Leaf-enforced QoS. Despite that, results have
also shown that there are more Failed Requests with Leaf-enforced
QoS for 2 layer, 4 layers and 6 layers of client leaf switches cases.
However, based on a student’s two-sample t-tests were performed in
order to test the statistical significance of the advantage of Core-
enforced, results showed that in general the advantage was
statistically insignificant except for only two Leaf-enforced QoS
mechanisms in 6 layers of client leaf switches case where
Destination CBQ at the Core performed better than Destination CBQ
at the Leaf and Source- Destination CBQ at the Leaf. Furthermore,
it was observed in the results that as layers in the network topology
increased, the Completed Requests for Leaf-enforced QoS relatively
remained the same but the Completed Requests for Core- enforced
QoS steadily increased making the gap from Leaf- enforced QoS
smaller. Despite that, Leaf-enforced QoS still resulted with more
Completed Requests than Core-enforced QoS. This goes to show the
advantage of the distributed nature of Software-Defined Networking
despite having more nodes in the network enforced with QoS.
ACKNOWLEDGEMENT
The completion of this study has been made possible with
the support and guidance of the following people, to whom we
extend our deep gratitude and appreciation: The faculty and staff of
Ateneo De Manila University Department of Information Systems
and Computer Science for support and guidance and our families,
for all their support, understanding and inspiration.
Finally, we thank the Lord God Almighty without whom
none of these would have been possible.
REFERENCES
[1] J. E. T. Regencia and W. E. S. Yu, “Introducing a test
framework for quality of service mechanisms in the context of
software-defined networking,” Submitted to: The 12th
International Multi-Conference on. Complexity, Informatics
and Cybernetics: IMCIC 2021.
[2] J. E. T. Regencia and W. E. S. Yu, “Exploring the effects of
network topology layers on quality of service mechanisms in
the context of software-defined networking,” Accepted:
ICISA2020.
[3] J. E. T. Regencia and W. E. S. Yu,, “Latency and throughput
advantage of leaf-enforced quality of service in software-
defined networking for large traffic flows,” Submitted to: SAI
Computing Conference 2021.
[4] H. Kim and N. Feamster, “Improving network management
with soft- ware defined networking,” IEEE Communications
Magazine, vol. 51, no. 2, pp. 114–119, 2013.
[5] D. Kreutz, F. M. Ramos, P. E. Verissimo, C. E. Rothenberg, S.
Azodol- molky, and S. Uhlig, “Software-defined networking:
A comprehensive survey,” Proceedings of the IEEE, vol. 103,
no. 1, pp. 14–76, 2014.
[6] J. G. V. Pena and W. E. Yu, “Development of a distributed
firewall using software defined networking technology,” in
2014 4th IEEE In- ternational Conference on Information
Science and Technology. IEEE, 2014, pp. 449–452.
[7] M. Karakus and A. Durresi, “Quality of service (qos) in
software defined networking (sdn): A survey,” Journal of
Network and Computer Applications, vol. 80, pp. 200–218,
2017.
[8] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L.
Peterson, J. Rexford, S. Shenker, and J. Turner, “Openflow:
enabling innovation in campus networks,” ACM SIGCOMM
Computer Communication Review, vol. 38, no. 2, pp. 69–74,
2008.
[9] S. Civanlar, M. Parlakisik, A. M. Tekalp, B. Gorkemli, B.
Kaytaz, and E. Onem, “A qos-enabled openflow environment
for scalable video streaming,” in 2010 IEEE Globecom
Workshops. IEEE, 2010, pp. 351– 356.
[10] H. E. Egilmez, S. T. Dane, K. T. Bagci, and A. M. Tekalp,
“Openqos: An openflow controller design for multimedia
delivery with end-to-end quality of service over software-
defined networks,” in Proceedings of the 2012 Asia Pacific
signal and information processing association annual summit
and conference. IEEE, 2012, pp. 1–8.
[11] A.Ishimori,F.Farias,E.Cerqueira,andA.Abele ́m,“Controlofmu
ltiple packet schedulers for improving qos on openflow/sdn
networking,” in
2013SecondEuropeanWorkshoponSoftwareDefinedNetworks.
IEEE, 2013, pp. 81–86.
[12] O. Chato and W. E. S. Yu, “An exploration of various qos
mechanisms in an openflow and sdn environment,” in Accepted
for Presentation in The International Conference on Systems
and Informatics (ICSAI-2016), 2016.
[13] O. Chato and W. E. S. Yu,, “An exploration of various quality
of service mechanisms in an openflow and software defined
networking environment in terms of latency performance,” in
2016 International Conference on Information Science and
Security (ICISS). IEEE, 2016,
128
 Philippine License Plate Localization Using Genetic
Algorithm and Feature Extraction
Patrick Matthew J. Chan, John Anthony Jose, Rhen Anjerome Bedruz, Elmer P. Dadios
Electronics and Communications Enginnering Department
De La Salle University, Manila
patrick_chan@dlsu.edu.ph, rhen.bedruz@dlsu.edu.ph, john.anthony.jose@dlsu.edu.ph, elmer.dadios@dlsu.edu.ph
Abstract— This study aims to focus on the license plate
localization portion of an automated license plate recognition
scheme. In order to achieve this, the study makes use of a genetic
algorithm in order to perform a more systematic search over the
input image; as well as Histogram of Oriented Gradients feature
extraction technique in order to form the fitness function of the
genetic algorithm. When the algorithm was tested, it was able to
successfully localize the license plates within each of the multiple
input images. In fact, even though the study was focused on 2014
series license plates, the program results also show that it was also
able to roughly localize a license plate from another series as well.
Keywords— Genetic Algorithm, Localization, HOG, Feature
Extraction
I. INTRODUCTION
For the Philippines, traffic congestion has been a rampant
problem, and one of the solutions that the government has
implemented to alleviate this, is a contactless apprehension
system that aims to reduce road blockages caused by traffic
violators and poor road discipline. Though numerous cameras
are already in place in order to detect traffic violators from a
control center, the recognition of license plates, which is a very
important aspect of contactless apprehension, is still done by
human eyesight, which proves to be inefficient. Currently, there
are various available studies in automating the detection and
apprehension of traffic violators, such as [1] [2] [3], but in
aiming towards a fully automated contactless apprehension
system, it is important to know the identity of the vehicle itself
and its owner as well. Thus, the first tool needed to track down
and log violators, is to automate the process of reading license
plates from the incoming video feeds.
In fact, aside from smart traffic apprehension systems,
automated reading of license plates, can even be combined with
other algorithms, in order to automated vehicle management
systems as well. An example would be having a fully automated,
and virtually unmonitored parking system management, such as
in [4] [5]. Another possible combination, would be in a fully
automated and tagless Philippine toll gate system, when
combined with vehicle classification such as [6]
Now, in order to be able to achieve this feature, two main
systems are needed as its components, namely: a plate
localization system, and an optical character recognition (OCR)
system [7]. The plate localization system’s purpose is to search
for and isolate potential license plates found within a frame,
whose output will then be fed to an OCR, such as in [8], for a
more robust and accurate character recognition within individual
plate images.
In this paper, the study will be focusing on License Plate
Localization, using Genetic Algorithm (GA), as well as feature
extraction using Histogram of Oriented Gradients (HOG). In this
paper, Section II would give a brief discussion of genetic
algorithm, on which the system algorithm in this study is based
upon. Meanwhile, Section III would further explain the
approach that was used in the study to successfully perform the
plate localization system, Section IV would be discussing the
results of the study, while Section V would provide overall
conclusions and recommendations after performing the study.
II. GENETIC ALGORITHM
Generally, genetic algorithms are typically used to solve
optimization and/or search problems, by simply taking
inspiration on biological events that lead to the evolution of a
species. In the scientific theory of evolution by natural selection,
a generation can be seen to consist of stronger individuals and
weaker individuals, that each compete for the same limited
resources. As a result, stronger individuals are more likely to
survive, compared to the weaker individuals. As enough time
passes by, the surviving individuals start to reproduce, and the
genetic traits that comprise the surviving individuals are
randomly combined and passed on to the children. This makes
the child generation to be mostly comprised of individuals that
mostly have one of the stronger parents’ genes. Then, as the
children form the next generation, generations continue to
generally get stronger and stronger as time passes by. This
concept of each generation getting stronger than the previous
one is the general idea behind the concept of evolution.
129
 In applying these concepts to computer algorithms, a height of 1080 pixels, then the chromosomes can have values
problem is instead solved using a first generation of randomly ranging from (1,1), up until x=1920-91=1829, and y=1080-
generated candidate solutions. Each of this candidate solution, 33=1047; or simply written as (1829, 1047). then starts to
is called a chromosome, which is a data structure that contains perform the genetic algorithm.
the necessary information to be plugged in, like variable values,
to the specified problem. Similarly, in this process of plugging
in each chromosome into the problem, some solutions would
evidently give better results, thus making it a “stronger”
chromosome. Now, this method of knowing how strong a
chromosome is, is called a fitness function, which takes a
chromosome as an input, and takes a numerical value as an
output, indicating the strength of that input chromosome.
Meanwhile, in terms of having each chromosome compete
against one another, this process is called selection, in genetic
algorithm terminology. After this process, similarly, the
chromosomes need to “reproduce”, in order to create the next
generation. At this point, the process of “reproduction” is mainly
designed by the programmer, on how two (or more) parents
produce a new “child” chromosome. In genetic algorithm
terminology, this process of “reproduction” is called crossover.
Lastly, it is important to note that some of the strong traits in
currently evolved creatures in biology, resulted in a mutation, or
a mistake in the copying of genes, leading to new behavior.
Similarly, in genetic algorithm, mutation can also be defined by
programmer, usually by manually editing the values inside a
chromosome, along with a mutation probability that defines how
often this kind of chromosome alteration is likely to happen.
III. METHODOLOGY Fig. 1. System Algorithm Flowchart showing Genetic Algorithm
For the plate localization system, a Genetic Algorithm was
utilized to perform the localization process. With this, the idea
is that various chromosomes, each representing different B. Fitness Function
cropped parts of the input image, will be competing against one
Another important element in the design of a Genetic
another on how much each looks like a license plate; and, as the
Algorithm, is its fitness function. For this study, in relation to
winners “reproduce” and make the next set of chromosomes, the
license plates, although there is no single exact image depiction
whole population will keep evolving, until it eventually
of a license plate, there is a certain set of features that can instead
produces one, that points to the actual position of the plate
be used to describe one. Typically, one can say that a license
number within the picture.
plate is a rectangular-shaped and usually plainly colored flat
In order to further describe how the program algorithm object, containing huge numbers and/or letters in the middle area
works, a high-level flowchart of the general processes involved in easily readable font. Thus, in order to take advantage of plate
are shown in Figure 1. In the first step, the system basically numbers having a common set of features, this study used HOG
reads the image input, and loads it into memory for processing. feature descriptors, as a way to check whether parts of the input
For simplification, the plate size in the image was set to a image resembles a license plate, or not, based on the features of
constant value of 91-by-33. Now, after the image is loaded, the a known license plate. With this in mind, the general idea behind
machine then starts to perform the genetic algorithm. the fitness function used in this study is the difference in HOG
features between a basis image, which is a sample image
A. Chromosome containing only a plate number by itself, and a specific part of
For the chromosomes used in the system, each of them the input image, as specified by a chromosome.
simply contained two decimal values, which denoted x and y
coordinates. These coordinates supposedly point to where the Now, in order to be able to obtain a numerical value for the
top-left corner pixel of the plate number is, in the picture, fitness function, the process begins with first extracting the
according to that chromosome. For this coordinate system, the HOG features of both images, each of which result in a huge
top leftmost pixel is described as (1,1), with the x-coordinate array of normalized numbers, representing the HOG description
increasing in value as one goes rightward, and with the y- of that image. Then, in order to obtain how far or how close the
coordinate increasing in value as one goes downward instead. features of the two images are from one another, a simple
Thus, coordinate values can range from one, to the lower computation using L2 distance, or also known as Euclidean
rightmost pixel within an image, wherein one can still draw a distance, is used, taking the square root of the sum of squared
full-sized license plate rectangle from, thus covering every differences, between corresponding pairs of numbers, between
possible spot within the image, to be a license plate. For the two sets of extracted features. This effectively condenses the
example, if the input image size has a width of 1920, and a two huge arrays of numbers representing two pictures, to a

130
single value, representing the error between thee features of the
two images; thus, successfully giving us a numerical value for
our fitness function. Hence, the fitness function for this genetic
algorithm is described as:
𝑁
2 it more likely to be removed from the current generation. This
𝐹𝑖𝑡𝑛𝑒𝑠𝑠 = √∑ (𝑓𝑏𝑖 − 𝑓𝑐𝑖 ) ()
𝑖=1
where N is the total number of elements of the HOG array,
𝑓𝑏𝑖 is the value of HOG feature number i of the basis image, and
𝑓𝑐𝑖 is the value of HOG feature number i of the image described
by the chromosome.
Note that both the basis image and the image described by
the chromosome should be of the same size, in order to produce
the same number of elements in the HOG feature number array,
in order for this fitness function to successfully give an
appropriate value. In addition, it should also be noted that in the
system, a lower fitness value indicates a fitter chromosome,
since the fitness function measures the amount of error between
the basis image and the image pointed at by the chromosome.
Now, going back into the flowchart, after the image is
loaded, the initial population is generated, giving us a number of
randomly generated ordered pairs of numbers, depending on the
initial population size. After this, each chromosome of the
current generation, which in this case is the initial randomly
generated one, is evaluated for its corresponding fitness value,
mainly based on the portion of the input image each of them
represent. After fitness evaluation, the program then checks for
the stopping condition. In this study, the stopping condition used
was when either the fittest chromosome gets a fitness value of
zero, or when a set amount of child generations have been
generated.
C. Selection
Whenever the stopping condition is not met, the selection
process is done over the current generation, in order to prepare
it for making the next generation. In this Plate Localization
Genetic Algorithm, the selection process is done through fitness
proportionate selection, which is also known as roulette wheel
selection, wherein the probability of a chromosome to be
removed, or “killed”, from a generation is directly proportional
to how unfit the chromosome is, when compared to all other
chromosomes in the same generation. However, a slight
modification was made in implementing this genetic operator in
this study, such that the fittest chromosome is guaranteed to be
retained, by excluding it in the selection process. This is done,
in order to eliminate the possibility of the algorithm to “forget”
the best solution as the generations pass by. Thus, in terms of the
fitness values explained earlier in this study, the probability of
each non-best-fit chromosome to be removed from the current
population is expressed as:
𝐹𝑖𝑡𝑛𝑒𝑠𝑠𝑗
𝑃𝑗 = (𝑃𝑜𝑝𝑆𝑧−1) ()
∑𝑘=1 𝐹𝑖𝑡𝑛𝑒𝑠𝑠𝑘
where 𝑃𝑗 is the probability of chromosome j to be removed,
𝐹𝑖𝑡𝑛𝑒𝑠𝑠𝑗 is the Fitness value of chromosome j, and 𝑃𝑜𝑝𝑆𝑧 is the
population size of the generation.
This easily makes sense, since, as mentioned before, a higher
fitness value represents a more unfit chromosome; thus, making
part of the algorithm continues to remove chromosomes from
the population, until only around half of the original population
has remained from the generation.
D. Crossover and Mutation
After selection has finished, the remaining population,
including the fittest chromosome, is left to “reproduce”, in order
to create the next generation. For this system, the two parents of
each child chromosome were randomly chosen, although a rule
was placed that prohibited a single chromosome to become both
parents of a child. Now, since each of the chromosomes contain
x and y coordinate information, the crossover method used in
this study involved computing a weighted center point between
the two parents, based on their individual fitness values,
resulting with the child chromosome’s x and y coordinates being
slightly closer to the fitter of the two parents in the image.
After this crossover process, the mutation possibility of each
child immediately follows. For this study, the researcher wanted
an exponential formula that greatly increased the chances of
mutation as the parent chromosomes’ coordinates get very close
to one another in the image. This was done, in order to avoid the
genetic algorithm from converging to a single value as
generations pass by, leaving room for enough diversity of
chromosomes in the next generation as the remnants of the
parent generation becomes too close to one another in the image.
The mutation process is done as follows: first, a “chance” value
is computed based on the distance between the two parents; then,
that “chance” value is applied twice for the child chromosome
on whether it is to be mutated or not, with one involved with
changing the x-coordinate to a random value, and the other
involved with changing the y-coordinate to a random value.
Thus, for example, if the “chance” value is computed to be 20%,
then the actual probability of the child to have a single mutation,
is [ 0.2*(1-0.2)+(1-0.2)*0.2]=0.32, or 32%. On the other hand,
the actual probability of the child to have a double mutation,
(which means that both x and y coordinate values are fully
changed,) is 0.2*0.2=0.04, or 4%. Moreover, the formula used
to obtain the “chance” value is written as:
2 −(𝑦 −𝑦 )2 )
chance = (1.5)(10−√(𝑥2 −𝑥1 ) 2 1 % ()
where(𝑥1 , 𝑦1 ) and (𝑥2 , 𝑦2 )are the x and y chromosomes of
the two parents.
This formula was chosen, so that the mutation chance would
only be low, at values less then 1.5%, up until the two parents
become closer than 10 pixels apart, then quickly rises in value.
It should be noted that a maximum chance value of 25% was
implemented in the program for this study as well.
Furthermore, another feature that is placed during the
crossover and mutation process, was that the best performer of
the parent generation is retained into the child generation. This
means that a slot in the child population is actually reserved for
131
the clone of the previous best performer. This is, once again,
done in order to guarantee that the best solution over the whole
run of the algorithm, gets passed on until the end, wherein the
stopping condition is met, and the best performer is only taken
from the latest generation as the GA plate localization program
output..
IV. DATA AND RESULTS
For this study, the License Plate Localization System using
Genetic Algorithm was implemented using MATLAB. In
testing the program algorithm with different image inputs, for
the sake of this study, the population size per generation was set
to a fixed value of 250 chromosomes for all runs, while the
Fig. 2. Program Output
number of child generations was set to 1000 generations. These
numbers are used, since it was experimentally noticed that it is
slightly more efficient to allocate a bit more computation time
to producing more generations, rather than when compared to
focusing on the population size over number of generations.
Having a constant setting was done, in order to focus more on
observing examples on how close the program can get to
Fig. 3. Cropped Plate Number from Original Picture
localizing the plate number within the picture, even though the
basis plate image stays the same.
The result of the first run is shown in Figure 2. In here, it can
be confirmed by manual inspection that the system has
successfully localized the plate number within the image. Of
course, in terms of program data, this output is simply the fittest
chromosome of the latest generation, which is simply
represented as the coordinates (1011,474) of a 1920 by 1080
image in this case and has a computed fitness value of around
0.0228. Thus, the program output visualization in Figure 2
simply drew a bounding rectangle over the original input image,
by simply knowing the top left pixel of the output chromosome,
as well as the plate size for the rectangle dimensions. Similarly,
when instructed to only extract the plate number from the picture
from the chromosome, MATLAB also displayed Figure 3 as an
output.
In Figure 4, it shows that the most fit chromosome
practically stayed as is at around generation 81 of the whole
1000-generation run. It can also be observed here that even
though the best fit of each generation seems to be stable, either Fig. 4. Best One’s Fitness vs. Generation (Full, and Zoomed)
steadily being the same value, or going down (or getting fitter);
while on the other hand, we get the second most fit of each In the visualizations on Figure 5 and Figure 6, one can see
generation to somewhat be a fluctuating value as the generations observe as the search process is done over the whole image.
pass by. This property is actually a direct result of the Obviously, unless by lucky chance, the randomly generated
modifications done in the selection, crossover and mutation initial generation would not immediately be able to find the plate
phases as discussed earlier that guarantees the preservation of number immediately within the image. This is only achieved as
the best chromosome overall so far, over future generations. fitter chromosomes are found as generations pass by. So,
Without the application of this rule, one can basically see its following the new best chromosomes as displayed in Figure 5,
effect on the second best of each generation, on which the value one can observe the corresponding “drops” in Figure 4, in terms
can greatly differ over various generations, depending on the of how it gradually moves to spots in the image that resemble a
results of the generation of each of the next generation with the license plate more, all thanks to the HOG feature descriptor as a
crossover and mutation processes. metric indicating this closeness. On the other hand, the
fluctuating values of the second best in Figure 4 is also reflected
in Figure 6, where various parts of the image became second
fittest of its generation, though with most of the ones at the end
being around the plate number as well, as the algorithm just
cannot find a spot within the image that is fitter than the ones
closer to the actual license plate, which is as intended of the
system program.

132
 leading to around 80 to 110 of the whole 250-chromosome
population to actually have single mutations, and around 10 to
20 of the whole 250 to actually have double mutations;
therefore, proving the diversity of the later generations. As what
was said earlier, the added diversity by thee mutated children
just happened to perform too badly, that they are practically cut
off from the recombination process by their selection
probability.

Fig. 5. Best Fit Ones per Generation

Fig. 7. Average Fitness vs. Generation (Full, and Zoomed)

Fig. 6. 2nd Best Fit Ones per Generation

Now, in Figure 7, one can see the effects of the selection

process. Looking at the graph, one can see that the average
chromosome fitness over the generations are typically lowered
after the whole selection process is finished, which, as discussed
before, removes around most of the bad performing half of each
generation population. One can also observe that around
generation 85, even though the crossover and mutation process
produces the next generations to have various fitness values in
order to garner an average fitness of around 10 to 15, the added
diverse chromosomes just end up to be performing too badly
relative to the ones close to the current value, that they are
practically removed by the selection process itself based on their
selection probability alone; thus, leading to an average fitness
value after selection to practically stay around the zero area for
the rest of the succeeding generations.
This observation can also be further confirmed by looking at
Figure 8, which confirms that by around generation 85, the
resulting chromosomes after selection end up being too close to
one another, thus driving the mutation chance to its maximum
values, as the black “chance” value can be seen in the graph to
practically stay at 0.25, or 25%, which corresponds to what was
discussed earlier in the mutation rules used. The thing is, it’s not Fig. 8. Mutation Probability vs. Generation (Full, and Zoomed)
like each of the later generations all only contain values which
converge around the fittest chromosome of the 81st generation.
As can be seen in Figure 9, corresponding to the rise in mutation
chance at around generation 80 to 85 in Figure 8, the number of
mutated children has also risen over those generations as well,

133
 program algorithm eventually preferred the plate number on the
right over the one on the left by the end of the whole run, due to
the one on the right being of the same series, and thus bearing
more similar characteristics and features.
For this program run described by the program output
displayed in Figure 13, one can see from the program graphs in
Figure 15 that the program did not get to its final chromosome
location as fast as it did during the first run shown in this study
in Figure 4; with this run being only able to obtain a final fitness
value of 8.4503 at the 503rd generation. Even so, the average
fitness graph on the lower left shows that the populations past
this point still ended up being very diverse, even after the
selection process has been done. So, as a result, the “chance”
values ended up only within the bounds of around 0.01, or 1%,
and 0.12, or 12%, by the end of the run; thus, resulting in
mutated children to be only limited to around 0 to 45 of the
whole 250-chromosome population, over the whole program
run.

Fig. 9. Number of Mutated Children vs. Generation (Full, and Zoomed)

In addition, in this study, the same input image was once

again placed into the algorithm for two more full runs, and each
gave similar results: with the second run staying with a fittest
value of 0.0228 by generation 140, and the third run staying with
a fittest value of 0.0228 as well, but by generation 205.
Visualizations of the best fit ones over generations are also
shown in Figure 10, and Figure 11, for the second, and the third
run, respectively. Note that the slight variations over different
runs of the input image on how the search goes closer to the plate
number within the picture, as well as how many generations it
took before the fittest chromosome seemed to settle to a single Fig. 10. Best Fit Ones per Generation (2nd run)
point, is mainly a result of the random aspect of generic
algorithms in general, along the initialization, selection,
crossover, and mutation phases.
In Figure 12, it shows the results when the Plate Localization
is run using different input images which both have plate
numbers that belong to the same Philippine plate number series,
which in this case is the 2014 plate number series. In these
figures, it is shown that the program was also able to
successfully locate the plate number within the image as well.
Now, looking at Figure 13, which displays the visualization of
the program output from another input image, the program was
able to detect was able to detect the plate number in the middle,
which also belongs to the 2014 Philippine plate number series
as well. But then, it is worth noting that it wasn’t able to also
grab the location of the send plate number to the left. This is
mainly due the current program design only being limited to a Fig. 11. Best Fit Ones per Generation (3rd run)
single plate number output, thus limiting it to only one plate
number detected at a time. As can be seen in Figure 14, one of Lastly, by looking at Figure 16, one can see in this program
the second-best fit chromosomes was actually able to mark the run that even though the plate number series included in the
second plate number on the left on one of the earlier generations input image is once again from a different series, which is the
of this run. So, one possible reason to explain this, is that the 2003 series in this case, the program was still able to roughly
plate number on the left is from a different Philippine plate detect the plate from the input image, mainly due to the fact that
number series, which is the 1981 series in this case, thus having unlike in Figure 13, it is the only plate number that can bee found
a slightly different set of features. Since the basis image plate in the input image in this case. In this specific run, the best
number was a 2014 series plate number, one can say that the

134
chromosome settled at the 23rd generation, and the final fitness V. CONCLUSION
value was 7.7916. This paper shows that proposed system algorithm was able
to successfully localize the license plate within different input
pictures, even though the basis image stayed the same
throughout the study, which was a 2014 series Philippine license
plate. In fact, it was even able to roughly localize some license
plates from different Philippine license plate series as well,
despite the series limitation in the basis image.

Fig. 15. Sample Graphs when Convergence is Later

Fig. 12. Other Runs of the Same Plate Series

Fig. 13. Program Output (Two different plates) Fig. 16. Detection of Plate from a Different Series

As opposed to a constant mutation chance, having a varying

mutation chance that gets higher as the parent chromosomes
become very similar to one another, actually helps in keeping
the population diverse until the last generation of the overall
program run, rather than converging to a all-similar set of
chromosomes that can possibly be only local minima, and not
close to the global minima to the range of possible solutions.
Although it is worth noting that an upper limit must be placed
on the mutation chance, in order to still be able to get the benefits
of inherited strong genes from the previous generations.
As was also pointed out over the data and results, the
algorithm used in the study was only limited to being able to
localize a single license plate within the input frame. Thus,
improvements can be further made in order to look into various
Fig. 14. Second-Best Fit Ones per Generation

135
local minima as well, and be able to determine whether another
license plate exists within the picture as well.
Furthermore, the robustness of the algorithm may be further
improved by using a basis image which consist of the average of
multiple HOG descriptor feature from the same license plate
series, for a more exclusive purpose, or for all license plate
series, to act as more of a one-size-fits-all solution.
ACKNOWLEDGMENT
The authors would like to recognize the Engineering for
Research and Development for Technology (ERDT) Program of
Department of Science and Technology (DOST) for funding this
study.
REFERENCES
[1] R. A. Bedruz, A. C. P. Uy, A. R. Quiros, R. K. Billones, E. Sybingco, A.
Bandala, and E. P. Dadios, “A Robotic Model Approach of an Automated
Traffic Violation Detection System with Apprehension,” 2018 IEEE 10th
International Conference on Humanoid, Nanotechnology, Information
Technology,Communication and Control, Environment and Management
(HNICEM), 2018.
[2] E. A. Roxas, J. N. Acilo, R. R. P. Vicerra, E. P. Dadios, and A. A.
Bandala, “Vision-based traffic sign compliance evaluation using
convolutional neural network,” 2018 IEEE International Conference on
Applied System Invention (ICASI), 2018.
[3] R. K. C. Billones, A. A. Bandala, E. Sybingco, L. A. G. Lim, and E. P.
Dadios, “Intelligent system architecture for a vision-based contactless
apprehension of traffic violations,” 2016 IEEE Region 10 Conference
(TENCON), 2016. (null) G. Eason, B. Noble, and I. N. Sneddon, “On
certain integrals of Lipschitz-Hankel type involving products of Bessel
functions,” Phil. Trans. Roy. Soc. London, vol. A247, pp. 529–551, April
1955. (references)
[4] J. Lin, S.-Y. Chen, C.-Y. Chang, and G. Chen, “SPA: Smart Parking
Algorithm Based on Driver Behavior and Parking Traffic Predictions,”
IEEE Access, vol. 7, pp. 34275–34288, 2019.
[5] Y. He and Y. Jiang, “The Design and Implementation of Residential
Parking Spaces Management and Information Issuing System,” 2013
Fourth International Conference on Digital Manufacturing &
Automation, 2013.
[6] R. N. Celso, Z. B. Ting, D. J. R. D. Carmen, and R. D. Cajote, “Two-Step
Vehicle Classification System for Traffic Monitoring in the Philippines,”
TENCON 2018 - 2018 IEEE Region 10 Conference, 2018.
[7] P. Szilveszter and G. Csongor, "Parking surveillance and number plate
recognition application," in Intelligent Systems and Informatics (SISY),
2010
[8] A. R. F. Quiros, R. A. Bedruz, A. C. Uy, A. Abad, A. Bandala, E. P.
Dadios, A. Fernando, and D. L. Salle, “A kNN-based approach for the
machine vision of character recognition of license plate numbers,”
TENCON 2017 - 2017 IEEE Region 10 Conference, 2017.
136
 Evolution of Social Networks in a Social Learning
Management System
Orven E. Llantos1,2,a *, Maria Regina Justina E. Estuar1,b
1
Department of Information Systems and Computer Science, Ateneo de Manila University, Quezon City
2
Department of Computer Science, Mindanao State University-Iligan Institute of Technology, Iligan City
a
orven.llantos@g.msuiite.edu.ph, brestuar@ateneo.ph

Abstract—Interaction is a result of interplay between two or

more entities where interplay could be behavior, or some other
variable being measured. This paper tries to discover the evolution
of the social network in a perceived social LMS environment with
principal, teacher, parent, and student as agents. Social network
analysis was applied on the interaction data sourced from the
deployment of my.eskwela system between October 15, 2019 and
January 31, 2020.
Results showed that the principal’s interaction influenced
the density of the social network as more teachers, students, and
parents interact with the system. The network density is much lesser
when different roles interact with the system because of the
implemented communication hierarchy following the real-world
organizational structure.
Keywords—learning interactions, social learning
management system, social network analysis, pattern interaction
analysis
I. INTRODUCTION
Learning Management System (LMS) is grounded on
constructivism theory which emphasizes on the self-directed
learning approach to education. This method allows each learner
to achieve optimal self-paced learning [1]. This approach was
further extended by adding social component thus making it social
constructivism theory. The theory believes that collaboration
among actors in the learning environment is much more powerful
than learning alone and this forms the foundation for social LMS
(sLMS) [2]. Studies have shown that utilizing social network in
learning environments have more potential to spread information
and is associated with high learner’s satisfaction observed through
active participation and higher level of cognitive thinking process
[2]–[5]. sLMS facilitate learning interactions which is broadly
defined as actions performed by an actor to another actor(s),
encouraging collaboration to guide the students’ learning
outcomes.
my.eskwela is a social Learning Management System
(sLMS) that brings together administrators, teachers, students,
and parents forming a community that supports collaborative
learning through learning interactions and active monitoring
driving student learning success. Recent versions of
my.eskwela comply with the communication structure as
specified by a Philippine Law under Republic Act 9155,
otherwise known as “Governance of Basic Education Act of
2001” [6]. Users view my.eskwela positively, especially in the
provision of “feeling of socialness” and the reduction of effort in
performing a task [7].
This study answers the question: How can learning
interactions of different roles manifest influence in the network
over time?
II. METHODOLOGY
The study uses activity logs in my.eskwela version 3.0
software. The study has been permitted by the Department of
Education Iligan City Division Office to proceed with any
member schools willing to be part of the study. Two schools
indicated the participation by signing the Informed Consent
Form (ICF) granting access to interaction data for the study. Out
of two schools, only one school has interaction data from the
parents. Consequently, this study focused on the school with
parent’s interactions.
A. Data Source
All of the interaction activities are recorded in the
timeline table. Another table was created to record user’s view of
different information propagated by the system. This table was
updated every time the user accesses the timeline, load more
items, compute level of proficiency, and other related tasks
performed on the system. comments and reactions tables are
included as sources of interactions.
1) Participants: The school principal and majority of
Teachers signed the ICF and their system identification were
encoded in a whitelist table. Through the principal, the parents
were also requested to sign the ICF and was informed that their
signature also means that they allow the interaction data of their
children to be used in this study. The number of participants
included in the whitelist resulted to 119.
2) DepEd School Calendar: The Department of Education
Central Office issues an Order specifying the calendar of
activities every school year. For the school year 2019-2020,
DepEd published order no 007 s. 2019. In line with the approved
Ethics Clearance, coverage of discussion is from October 15,
2019, to February 28, 2020.
3) Ethical Considerations: Ethics clearance was approved by
the Ateneo de Manila University Research Ethics Council.
Participants were given an Informed Consent Form (ICF)
regarding the study. A signature in the form indicates approval on
the utilization of interactions data in my.eskwela v3.0. for this
study.
B. Data Transformation
The data from the timeline, view, comments, and reactions
table were converted to agentxagent and agentxtask matrices to
produce the social network.
C. SNA Tool

137
 The agentxagent and agentxtask matrices are converted as
entries to CASOS-ORA XML definition file. Each entry for
initiator, receiver, and task is encoded as nodes for the network
and the edges as part of the agentxagent and agentxtask matrices.
D. Pattern Interaction Analysis
Pattern Interaction Analysis (PIA) is a category of SNA
that studies the interaction patterns derived from the sLMS
networks. There are two main approaches to study PIA that are
inherent in SNA: mathematical analysis and visualization. The
mathematical analysis includes computations for centrality
measures, density, cliques, and clusters. Mathematical analysis is
useful in discovering the structural properties of networks and
communities.
The SNA tools provide all the information needed to
conduct PIA both in mathematical analysis and visualization. PIA
allows for the discovery of preferences for communication
channels, participation levels, collaboration patterns, and
members contributions.
III. RESULTS AND DISCUSSION
To look at how the network evolves, Table I details in
each social network, the density, links, and node count.
TABLE I: Network Evolution
Measure Oct Nov Dec Jan 1st Jan 2nd
half half
Density 0.04 0.04 0.25 0.24 0.03
All Links 197 157 14 11 754
Node Count 78 72 12 11 180
Network density refers to the measure of how the network
nodes connect to the other nodes. The measure of network density
is one when each node connects to every other node in the network.
The highest network density value is from December 2019 to the
first half of January 2020. During these times, fewer faculty
members and students have interacted with the system. Thus, the
hierarchical structure is relaxed, providing more connectivity to
the nodes. However, when hierarchical structure emerges as in the
interaction of the principal in the second half of January, network
density is at its lowest. The low network density value means that
more nodes do not connect to other nodes like the principal cannot
be directly related to the students and parents, and not all students
relate to all the teachers in the school.
IV. CONCLUSION
The evolution of the sLMS social network emphasizes
engagement as the primary driver for network connectedness.
Engagement also affects to the centrality score of the roles. Thus,
for a social network with imposed hierarchies, critical roles need to
engage in the network to encourage roles below the hierarchy to do
the same.
The principal’s learning interactions promote faculty
member’s compliance with the principal vision of integrating ICT in
teaching. The principal’s involvement is transformative because of
the actual hands-on on the system is an act of lead by example, and
the Transformational Leadership Theory backs such a phenomenon.
The teachers also signified that the involvement of the principal
affects their decision to use the system [8].
V. FUTURE STUDIES
The promotion of the integration of online learning in
pedagogy to beneficiary schools in the DepEd is timely,
especially during these challenging times of the coronavirus
disease 2019 (COVID-19) pandemic. It would be interesting how
the network evolves during these challenging times.
ACKNOWLEDGEMENT
The authors would like to acknowledge DOST-ERDT
for the financial support provided for this study, DepEd Iligan,
and Tambo Central School for participating in this study.
REFERENCES
[1] A. Amory, W. Gachie, P. Clarke, and S. Yates, “Needs analysis,
development and hci evaluation of a new learning management
system,” in Proceedings of EdMedia + Innovate Learning 2004.
Association for the Advancement of Computing in Education
(AACE), 2004, pp. 373–380. [Online]. Available:
https://www.learntechlib.org/p/12960
[2] M. Raspopovic, S. Cvetanovic, I. Medan, and D. Ljubojevic, “The
effects of integrating social learning environment with online
learning,” International Review of Research in Open and
Distributed Learning, vol. 18, no. 1, pp. 141–160, 02 2017.
[3] H. M. Mora, M. T. S. Pont, G. D. M. Casado, and V. G. Iglesias,
“Management of social networks in the educational process,”
Computers in Human Behavior, vol. 51, pp. 890 – 895, 2015.
[Online]. Available:
http://www.sciencedirect.com/science/article/pii/S07475632140058
95
[4] K. S. Scott, K. H. Sorokti, and J. D. Merrell, “Learning “beyond the
classroom” within an enterprise social network system,” The
Internet and Higher Education, vol. 29, pp. 75 – 90, 2016. [Online].
Available:
http://www.sciencedirect.com/science/article/pii/S1096751615300
117
[5] H. El-Ghalayini, M. Abu-Arqoub, G. Issa, and A. F. Shubita,
“Graduation-project management system: A social network analysis
perspective,” Journal of Theoretical and Applied Information
Technology, vol. 95, no. 4, pp. 810 – 819, 2017. [Online]. Available:
http://www.jatit.org/volumes/Vol95No4/9Vol95No4.pdf
[6] O. E. Llantos and M. R. J. E. Estuar, “Characterizing instructional
leader interactions in a social learning management system using
social network analysis,” Procedia Computer Science, vol. 160, pp.
149 – 156, 2019.
[7] O. E. Llantos and M. R. J. E. Estuar, “my.eskwela: Designing an
enterprise learning management system to increase social network
and reduce cognitive load,” Procedia Computer Science, vol. 138, pp.
595 – 602, 2018.
[8] O. E. Llantos, “Cloudification of my.eskwela for e-governance in
philippine education,” Procedia Computer Science, vol. 109, pp. 680
– 685, 2017.
138
Manufacturing and
Machinery
Session Chair : Dr. Arthur L. Fajardo
Moderator : Dr. Napoleon Enteria
Panelists : Dr. Sherwin Guirnaldo
Dr. Carl Salaan
Dr. Omar F. Zubia
 Development of an Oyster Mushroom (Pleurotus
Spp.) Substrate Compactor-Bagger
Jordan L. Abad1,a *, Helen F. Gavino2,b, Romeo B. Gavino2,c and Melba M. Denson2
1
Department of Science and Technology Regional Office 1, San Fernando City, La Union
2
Department of Agricultural and Biosystems Engineering, College of Engineering, Central Luzon State University,
Science City of Munoz, Nueva Ecija
a
jlabad@region1.dost.gov.ph
b
hfgavino0313@gmail.com, caeigsstudent@gmail.com
*
corresponding author

Abstract— The low supply of mushroom in the country
evident to 90% importation is ironic to the abundant biomass
resource the country could offer. The low biomass utilization rate
affected by the lack of appropriate technologies also applies to the
low productivity in bagging of substrate in the mushroom industry
using manual labor with a capacity of 25 bags/hr. These problems
enthuse the study to design, fabricate and evaluate a mushroom
substrate compactor-bagger adoptable to local practices and
mushroom production technologies in the country.
The 1432mm x 741mm x 2040mm (l x w x h) 140kg
machine was designed using 152 x 305mm, 0.02 mil polypropylene
plastic bags with compaction height of 203mm and 95mm diameter.
A weight of 750 grams of substrate per bag was determined as
setting using the CLSU formulation (70-30 RS-S) and BSU (80-15-
5 S-RB-L) formulation.
Performance evaluation of the machine using the two
formulations was tested with different loading weights of 10kg,
15kg and 20kg per batch. It was evaluated using simple CRD and
means were compared using LSD test. Results of the study revealed
that CLSU formulation at 15kg loading gained 71.67% bagging
efficiency, 70.78% filling efficiency, 204 bags/hr bagging capacity
and with an average of 740.90 grams/bag. Using BSU formulation,
the machine loaded with 20kg produced the highest bagging and
compacting efficiencies of 88% and 83.48%. It had a capacity of 307
bags/hr with an average of 759 grams/bag. The average weight per
bag was within the acceptable weight of 750 ±50 grams.
The cost analysis of the machine for a year of operation at
six hours per day, three times a week with a custom rate of
PhP1.5/bag could generate an income of PhP96,736.00-
PhP191,749.00 per year and a payback period of 0.30-0.60 years for
the BSU and CLSU formulations, respectively.
With the developed machine, there would be increased
biomass waste utilization, more production of oyster mushroom at
lower cost and hence, more benefits to the farmers and the
mushroom industry.
Keywords— biomass utilization, compactor-bagger,
oyster mushroom, substrate, formulation
I. INTRODUCTION
According to economic analysts, mushrooms are seen as
another significant crop of the country due to its growing local
demand [1]. Being an agricultural country with well-endowed
livestock and forestry products and cultivates major crops like rice,
coconut, and sugarcane, the state can generate substantial amount of
residues favorable for the production of mushrooms. However, the
Philippine mushroom industry set its lowest production volume of
355 metric tons (MT) in 2009 and worsened since 1995 [2] as almost
90% of the mushroom consumed was imported from different
countries of South East Asia [3].
The lack of appropriate research for mushroom
mechanization facing the country's small-scale and or village-level
mushroom industry could be observed in the bagging of substrate
materials where manual labor is employed and described to be labor-
intensive [4]. As [5] explained it, more people are needed to do the
filling of bags than in other operations before bagging.
Mechanization technology suited for the scale of
production in the bagging or filling of substrate bags and other
related operations could encourage the use of available biomass
resource and could further boost the mushroom industry towards
self-sufficiency of mushroom supply in the country.
With these problems of low mushroom supply and the
potentials of untapped biomass resource in the country, a need for
efficient and cost-effective biomass utilization technologies and
equipment compared to existing production technologies could be
seen to encourage the cultivation of mushroom. Thus, the study
aimed to develop an oyster mushroom (Pleurotus spp.) substrate
compactor-bagger using locally available materials and fabrication
technologies. The specific objectives of the study include the
establishment of the optimum weight of substrate in bags; to design
a substrate bagger adaptable to existing technologies of mushroom
production in the Philippines; to fabricate the machine using local
manufacturing technologies; to evaluate the operating performance
of the machine in terms of capacity, efficiency, and uniformity in
weight of substrate bags; and, to perform cost analysis in using the
machine.
With the study, bagging of the mushroom substrate would
be made easier resulting in higher output in terms of mushroom
substrate in bags.
The creation of the machine could also lead to increasing
biomass resource utilization which was not given too much
attention. It would also provide opportunity to the local machinery
industry as this machine could be added to the list of machines that
could be fabricated locally, thereby may also increase the income of
local machinery shops.
Thus, the study would help improve the production of the
mushroom substrate and the mushroom itself creating a more

140
profitable farmer but also in the utilization of biomass resource for
food production and profit creation.
II. METHODOLOGY
The study was set to design, fabricate, and evaluate a
machine mechanizing the compacting and bagging of mushroom
fruiting bags using two of the most commonly used formulation, the
(a) 70 parts rice straw and 30 parts sawdust or CLSU Technology
and (b) 80-15-5 parts of sawdust-rice bran-lime or BSU
Technology.
The study started with the gathering of benchmark data to
support the claim of mechanizing the bagging of the mushroom
substrate. The data gathered in the form of a survey questionnaire
was conducted related to the existing technologies in the country on
bag cultivation. Data gathered like the dimension of polypropylene
plastic (PP) bag, the capacity of laborer and weight of substrate
served as basis or design considerations for the fabrication of the
machine.
A. Establishment of Optimum Weight of Substrate in Bags
To standardize the weight of the substrate bagged by the
machine, the establishment of the optimum weight of substrate in
bags was conducted. It was used in the conceptualization of some
settings for the fabrication of the machine.
The planting spawn weighs 40 grams per bag placed in
102mm x 152mm PP bags plugged with cotton through a PVC neck.
It was sterilized in an autoclave for one and a half hour at 120
degrees Celsius, 21psi.
Formulation of the substrate follows CLSU technology
and BSU technology. The substrate was manually bagged using
152mm x 305mm PP bags with 0.02 mil thickness. The bags were
approximately uniform in volume since height was maintained at
203mm while diameter at 95mm. It was inoculated with the prepared
spawn and was incubated in a dark room. It was then transferred to
the fruiting area after full ramification.
Data gathered include the number of days for full mycelial
ramification, yield per bag, and flushing interval. Meanwhile, the
statistical analysis used was (2x3) factorial by Completely
Randomized Design. Factor A of the study included the types of
substrate mixture: (F1) 70% Rice Straw and 30% sawdust; and (F2)
80% sawdust, 15% rice bran, 5% lime while Factor B was the
different weight per substrate bag: (A) 500 grams, (B) 750 grams,
and (C) 1000 grams compacted in the same volume (95mm diameter
x 203mm height).
B. Design of the Mushroom Substrate Compactor-Bagger
The machine had the following components: the flywheel
and compactor assembly, the conveyor-hopper assembly, the
rotating disc assembly, the mechanism to rotate the rotating disc
assembly, the frame assembly and the power transmission system.
The hopper was designed to accommodate 32-35kg of
substrate per batch. The volume of the hopper was determined using
Equation 1 using density of uncompacted substrate for BSU
formulation of 347kg/m3 and CLSU formulation with 320kg/m3 .
W1 (1)
Vh =
ρ C=
where: 𝑉ℎ = 𝑣𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 ℎ𝑜𝑝𝑝𝑒𝑟, 𝑚3
𝑊1 = 𝑤𝑒𝑖𝑔ℎ𝑡 𝑜𝑓 𝑠𝑢𝑏𝑠𝑡𝑟𝑎𝑡𝑒 𝑝𝑒𝑟 𝑏𝑎𝑡𝑐ℎ, 𝑘𝑔
𝜌 = 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 𝑜𝑓 𝑠𝑢𝑏𝑠𝑡𝑟𝑎𝑡𝑒, 𝑘𝑔/𝑚3
The design of the spiral conveyor was based from [6] for
screw conveyors. The diameter of the shaft was calculated using
Equation 2 where the density of sawdust was used, and the diameter
of the spiral conveyor was 4”.
CT = (D2-d2)/36.6) x P x RPM (2)
where: CT = theoretical capacity (x 60%), ft3/hr
D = spiral diameter, in
d = shaft diameter, in
P = pitch diameter, in (usually equals to D)
RPM = speed of shaft, rpm
The size of the PP bags dictated that the diameter of the
compression chamber was 95mm and the height was 295mm lower
than the computed value of 305mm to facilitate easier mounting of
PP bag wherein the remaining gap would be augmented by using
the adjustable stopper depending on the type of substrate.
After the dimensions and sizes of components were
calculated, a final design plan was drawn using SolidWorks
software.
C. Fabrication and Assembly of the Machine
Local technologies and skills of manufacturing industry
in the country was utilized in the fabrication using mainly the
instruments found in a local fabricating shop. Some of the
equipment used were welding machine, grinding machine, drilling
machine, oxy-acetylene gas, and cutting disc. The assembled
components were set to their proper timing before conducting the
preliminary test.
D. Machine Testing and Evaluation
After the machine was fabricated, initial testing was
conducted to validate the timing and functions of all components.
The study was laid out in Completely Randomized
Design, ANOVA was used to evaluate the effects of different
weights per batch in the machine using two formulations. Since the
two formulations used differed in physical property especially the
bulk density of uncompacted materials, a separate evaluation of the
machine using the two substrates was conducted. There were three
varying weight per batch per kind of formulation. With this, there
were three replications per treatment for each of the two
formulations.
The methods of preparation for the substrate materials in
the testing for the machine were similar with the preparation of a
substrate in the establishment of optimum weight per substrate bag.
Data Gathered during the evaluation include productive
time, unproductive time, total bagging time, number of bags within
the acceptable limit, total number of bags produced, and total weight
of loaded inputs or substrate.
The bagging capacity of the machine was determined
using the Equation 3:
Bin
(3)
T
141
where: C= Bagging Capacity (bags/hr) To conduct the cost analysis of using the machine,
𝐵𝑖𝑛 = Number of Bags Produced within the limit (bags) assumptions were laid to facilitate computation. Six hours were
T = Total bagging time (hr) allotted for the total operating hours per day of the machine,
considering 8-hr operation for the laborer. The labor cost per laborer
The bagging and filling efficiencies of the machine were is PhP300.00/day at 6hr/day.
calculated using Equations 4 and 5. Assuming that in a community with several mushroom
and fruiting bag producers, 3 days/week was allotted for custom
Bin (4) hiring at PhP1.50/ bag. The custom rate is 75% of the manual
EffB = x 100
BT bagging cost of PhP2.00 considering the cost of manual bagging for
25 bags/hour. If the machine is used for a year, it would operate for
(5)
Win 156 days or 939 hours.
EffF = x 100 Considering that the machine uses 0.368kW of motor at
WT
7.13 pesos per kilowatt hour rate, the cost per hour of operation is
where: 𝐸𝑓𝑓𝐵 = 𝐵𝑎𝑔𝑔𝑖𝑛𝑔 𝐸𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑐𝑦 PhP2.64.
𝐵𝑖𝑛 = Number of Bags Produced within the limit (bags)
𝐵𝑇 = 𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑏𝑎𝑔𝑠 𝑝𝑟𝑜𝑑𝑢𝑐𝑒𝑑
𝑊𝑖𝑛 = 𝑇𝑜𝑡𝑎𝑙 𝑤𝑒𝑖𝑔ℎ𝑡 𝑜𝑓 𝐵𝑎𝑔𝑠 𝑊𝑖𝑡ℎ𝑖𝑛 𝑡ℎ𝑒 𝑙𝑖𝑚𝑖𝑡 (𝑔𝑟𝑎𝑚𝑠) III. RESULTS AND DISCUSSION
𝑊𝑇 = 𝑇𝑜𝑡𝑎𝑙 𝑊𝑒𝑖𝑔ℎ𝑡 𝑜𝑓 𝑙𝑜𝑎𝑑 𝑖𝑛𝑝𝑢𝑡 (𝑔𝑟𝑎𝑚𝑠)
A. Optimum Weight of Substrates
E. Cost Analysis For the average mycelial ramification, the analysis of
To evaluate the cost of using the machine, the following variance revealed significant difference between the formulation,
equations by [7] as presented in Table 1 were used. weight of substrate, and their interactions. The comparison of means
revealed that the 500g weight per bag with 26 days of full
Table 1. Formula used for the simple cost analysis of the study. ramification vary significantly to that of the 750g and 1000g bags
PARTICULARS FORMULA with 31 days and 40.6 days. The mean ramification of 27.33 days
Salvage value, (10% IC) SV = 10% IC for CLSU formulation varied significantly to the 37.73 days of BSU
Depreciation, PhP/yr D= (IC-SV)/n formulation. For the interaction of the factors, Table 2 shows that
Interest on Investment, PhP/yr I= ((IC+SV)/2)i the least ramification days was recorded in CLSU formulation at
Fixed Cost, PhP/yr Fc = D+I+TIS 500g/bag with 20 days only while highest at 41.20 days in
Variable Cost, PhP/yr Vc = Cp+Cl+RM+Club 1000g/bag BSU formulation. The simple effects vary significantly
Total bagging cost, PhP/yr Tc = Fc+Vc with each other except to that of BSU at 750g and CLSU at 1000g
Bagging Cost, PhP/bag Bc = Vc/C*T with 40 days ramification period.
Breakeven point, bag/yr BEP = Fc/Cr-Bc
Net Income per year, PhP/yr NI = C*T*Cr-Tc Table 2. Average mycelial ramification (days) per formulation as
Payback Period, yr PBP = IC/IN affected by different weight of substrate.
WEIGHT, g MEAN
where: Fc = Annual fixed cost, PhP/yr FORMULATION
A-500g B-750g C-1000g
D = depreciation, PhP/yr
I = interest on investment, PhP/yr CLSU 20 z 22 y 40.00w 27.33b
TIS = tax, insurance and shelter, PhP/yr BSU 32 x 40 w 41.20v 37.73a
Vc = variable cost, PhP/yr MEAN 26l 31k 40.60j 32.53
Lu = lubricant cost, PhP/yr Means with the same letter are not significantly different
Cp = fuel cost, PhP/yr
RM = repair and maintenance cost, PhP/yr The results revealed that the higher the weight of substrate
Cl = labor cost, PhP/yr in bags the longer it would take for the mycelial run for both
Tc = annual cost, PhP/yr substrate formulations. This result is supported by a similar study on
C = Bagging capacity, bags/hr the mycelial run for Pleurotus florida by [8] which resulted in lesser
Vc = variable cost, PhP/yr time for colonization in rice straw-based substrate than the
Bc = Bagging Cost, PhP/bag traditional substrate of sawdust.
Cu = cost of using the machine, PhP/bag The results of the analysis of variance of the average yield
T = Annual operating Time, hrs/yr per bag as affected by substrate formulation and weight per bag
Cr = Custom Rate, PhP/bag show significant difference between the weight per bag and
Break-even point = the number of bags that the machine interaction while no significant difference in formulation. The
would properly compact and bag to recover the capital comparison among means as affected by weight per bag presented
investment in Table 3 would show that the 750g produced the average yield per
Payback period = the time it takes to recover the investment bag of 130.25g and was significantly different to the 113.65g/bag of
cost the 500g substrate bag and the 76.75g yield of the 1000g. For the
interactions, yield for both formulations at 500g and 750g did not

142
differ significantly with each other but differ significantly to that of use of the hanging method if hanged one after the other. Meanwhile,
the 1000g/bag. the heaviest weight recorded the highest span of ramification in
days, lowest yields and the highest flushing intervals. The
Table 3. Average yield (g) per formulation as affected by different recommended setting of 750 grams for both formulations has been
weight of substrate. seen to prove the not too tight not too loose recommendation of
WEIGHT, g MEAN various authors and books on mushroom cultivation.
FORMULATION
A-500g B-750g C-1000g
B. Description and Operation of the Machine
CLSU 133.50v 73.40w 105.37 The major components of the machine were the following: (1.)
109.21v
BSU 118.10v 127.00v 80.10w 108.40 spiral-conveyor and hopper assembly, (2.) the four-cylinder rotating
MEAN 113.65k 130.25j 76.75l 106.89 disc assembly, (3.) the power transmission system, (4.) the piston
compressor assembly, (5.) the frame assembly, (6.) push-to-rotate
Means with the same letter are not significantly different lever assembly and (7.) the molder or compaction chamber
assembly as shown in Figure 1. These components were fabricated
The result in yield was similar to the results of [9] on oyster and assembled to form an operational compactor-bagger machine.
mushroom yield using tray method of cultivation whereby from 2kg
to 10kg per tray, 6kg was seen to be favorable for growing. The
results emphasized that the decrease in yield below the optimum
weight was attributable to earlier depletion of nutrients due to
limited substrate quantity while increasing the substrate quality did
not increase in mushroom production but only a waste of resources.
The comparison among means in Table 4 shows that the
flushing interval means of BSU formulation with 11.40 days
significantly differed to CLSU with 18.07 days flushing interval. For
the weight of bags, 750g per bag resulted to lowest flushing interval
of 8.75 significantly different to the 14 days of 500g and 21.45 days
of the 1000g. The interaction effects of the two factors were seen to
be significantly different with each other where CLSU at 750g with
8 days flushing interval differ significantly to all other treatment
combinations except with BSU at 750g with 9.50 days. The BSU
formulation at 750g also did not vary significantly to BSU at 500g
with 11.10 days but differed to other treatment combinations.
Meanwhile, BSU formulation at 500g and 1000g did not differ
significantly with each other but differed from the rest of the
treatment combinations.
Figure 1. The mushroom bagger-compactor
Table 4. Average flushing interval (days) per formulation group
The machine was made of locally available materials and
as affected by different levels of weight.
was fabricated using local manufacturing technologies. It was
WEIGHT MEAN
FORMULATION powered using a 0.367kW electric motor and was equipped with
A-500g B-750g C-1000g wheels for mobility in the mushroom house.
CLSU 16.90 w 8.00 z 29.30 v 18.07a The machine was operated by three persons with a
BSU 11.10 xy 9.50 yz 13.60 x 11.40b specified task. One who supervised the level of the substrate in the
MEAN 14.00k 8.75l 21.45j 14.71 feeding hopper, the other in putting the PP bag in the molder, and
Means with the same letter are not significantly different the last person in charge of the exit of the substrate bags or the
collector of the compacted and bagged substrate.
Based on the results of the statistical analysis conducted,
both formulations were affected by the different weight levels of C. Performance Characteristics of the Machine Using CLSU
substrate in bags. In selecting what type of formulation would be Formulation
more favorable than the other, CLSU formulation revealed better The machine was evaluated in terms of bagging efficiency,
results in terms of mycelial ramification (days) and flushing interval filling efficiency and bagging capacity.
(days) than BSU mixture but had no difference in yield for two The average bagging efficiency of the machine as affected
flushing periods. by different weight per batch using the CLSU formulation is
For the weight of substrate used in both formulations, presented in Table 5. Analysis of variance for the bagging efficiency
although the 500g substrate recorded the least ramification span for revealed that the means of bagging efficiency as affected by the
both formulations, 750g had favorable results in terms of yield and weight per batch differ significantly with each other at 5% level of
flushing intervals. During the study, it was also observed that the significance.
500g substrate formulation was difficult to manage as they were Comparison among means using Least Significant
easily disintegrated and substrate materials would shrink with the Difference (LSD) at 5% level of significance shown in Table 5

143
revealed that weight per batch of 15kg with highest bagging The highest bagging efficiency was computed in the 20kg weight
efficiency vary significantly with the 10kg and 20kg weight per per batch with 88% decreasing as the weight per batch decrease with
batch. 15kg in 84.21% and 10kg at 83.33%. Analysis of variance shows
that the weight per batch had no significant effect in the bagging
Table 5. Average bagging efficiency (%) as affected by different efficiency of the machine when using the BSU formulation at 5%
weight per batch. level of significance.
WEIGHT PER REPLICATE MEAN
BATCH, kg I II III Table 8. Average bagging efficiency (%) at different weight per
10 46.15 38.46 53.85 46.15 b batch.
15 75.00 75.00 65.00 71.67 a WEIGHT PER REPLICATE MEAN
20 2.63 17.65 21.88 14.05 c BATCH, kg I II III
Means with the same letter are not significantly different 10kg 83.33 83.33 93.33 83.33
15kg 84.21 84.21 84.21 84.21
It is also noted that using the weight per batch type, the 20kg 88 88 88 88.00
machine produced acceptable weight per bag ranging from 708.13
to 740.90g/bag and with an average of 725.42g/bag when using the Also, the bagged substrate using the BSU formulation
CLSU formulation. resulted to uniform weight per bag of substrate within the acceptable
On the filling efficiency of the machine as affected by limit with an average of 755.25g/bag close to the 750g/bag.
weight per batch using CLSU formulation, analysis of variance Meanwhile, results of the analysis of variance for filling
showed that there was a significant difference in the mean effects efficiency revealed that the treatment means varied significantly
of weight per batch in the filling efficiency of the machine at 5% with each other. Comparison among means using LSD as presented
level of significance. in Table 9 shows that the three weight per batch differed
Comparison among means presented in Table 6 revealed significantly with each other with 20kg as the highest filling
that weight per batch of 15kg with 70.78% vary significantly with efficiency of 83.48%.
the 10kg and 20kg weight per batch. The lowest recorded filling
efficiency is at 20kg weight per batch with only 16.75%. Table 9. Average filling efficiency (%) as affected by different
weight per batch.
Table 6. Average filling efficiency (%) as affected by different Weight per batch, kg REPLICATE MEAN
weight per batch. I II III
WEIGHT PER REPLICATE MEAN 10 75.90 75.90 75.80 75.87c
BATCH, kg I II III 15 79.60 79.60 80.20 79.80b
10 42.40 35.6 49.40 42.47 b 20 84.00 83.35 83.10 83.48a
15 73.60 74.13 64.60 70.78 a Means with the same letter are not significantly different
20 3.75 21.60 24.91 16.75 c
Means with the same letter are not significantly different Based on Table 10, the bagging capacity as affected by
weight per batch in the BSU formulation recorded 306.59~ 307
The summary of the machine’s bagging capacity as bags/hr when using the 20kg weight per batch. It was followed by
affected by different weight per batch is presented in Table 7. The the 15kg weight per batch of 302.11 bags/hr and 10kg with 292
analysis of variance of the bagging capacity affected by the weight bags/hr. Results of the analysis of variance revealed significant
per batch at 5% level of significance revealed no significant difference in the mean bagging capacity as affected by weight per
difference between means at 15kg and 10kg weight per batch but batch.
varied significantly with 20kg weight per batch. Comparison of means presented in Table 10 shows that
the bagging capacity in 20kg and 15kg weight per batch did not vary
Table 7. Average bagging capacity (bags/hr) as affected by significantly but differed significantly to the 10kg weight per batch.
different weights per batch.
WEIGHT PER REPLICATE MEAN Table 10. Average bagging capacity (bags/hr) as affected by
BATCH, kg different weights per batch
I II III
10 171.43 130.43 201.6 167.82 a WEIGHTS PER REPLICATE MEAN
a BATCH, kg I II III
15 214.29 215.14 180 203.14
b 10 300.00 288.00 288.00 292.00b
20 8.18 53.33 64.62 42.04 15 303.16 300.00 303.16 302.11a
Means with the same letter are not significantly different 20 306.98 304.62 308.17 306.59a
Means with the same letter are not significantly different

D. Performance Characteristics of the Machine Using BSU Other observations during the evaluation of the machine
Formulation include the elimination of PP bag breakage which was observed in
The average bagging efficiency of the machine as affected the manual bagging of substrates. This could be attributed to the
by weight per batch and BSU formulation is presented in Table 8.

144
controlled force of compaction as compared to manual labor with The cost curve of using the two formulations in the
varying forces exerted during bagging. machine is shown in Figure 2. The figure shows that the area
covered by the net income of the BSU formulation in using the
machine is greater than that of the CLSU formulation and that the
E. Cost of Using the Machine for the two types of Formulation custom rate is higher than the cost of producing per bag for the two
To determine whether the machine was cost-effective in formulations.
bagging substrates, a simple cost analysis was conducted.
The analysis used the highest capacity recorded for both
formulations as a result of evaluating the performance of the
machine using the treatments. For the CLSU formulation, 204
bags/hour were used as the bagging capacity as it showed significant
results based on the findings of the study while 307 bags/hour when
using the BSU formulation.
The machine was assumed to be operational thrice a week
with a total of 156 days/year or 939 hours per year total operating
time for every six-hour per day operation.
There were three operators for the machine with
PhP300.00 per laborer as labor cost. The custom rate of PhP1.5/bag
was based on the 75% cost per bag for manual bagging with
PhP2.00/bag pegged at PhP300.00 per day at 6-hour per day. It was
used since there was no existing bagger in the locality. While the
calculated values are presented in Table 11.
Table 11 shows that the total bagging cost per year in
using the machine is PhP190,731.92 with labor as the main
contributor with PhP140,785.71 for the three operators for a year.
Figure 2. The Cost Curve of the Machine using the two
The bagging cost for the CLSU formulation resulted to
formulations
0.995~1.00 peso per bag while 0.66 pesos per bag for BSU
formulation. The difference is due to the higher capacity of the later
contributing to a higher net income per year with a total of
PhP191,749.00. Meanwhile, the net income for CLSU formulation IV. CONCLUSION
is PhP96,736.00.
Taking into consideration the objectives of the study and
The computed breakeven point for the CLSU formulation
the conducted evaluation of the machine based on the parameters
is 32,997 bags per year with a projected payback period of 0.60 years
set, the following are therefore concluded:
of machine operation for the investment to be recovered.
When using the BSU formulation, 0.304 years is the
1. Compaction of substrate bags for CLSU formulation and BSU
payback period of the machine considering the assumptions. The
formulation using 152mm x 305mm PP bag is at optimum weight of
breakeven point for bags per year is calculated at 22,119 bags/year
750 grams/bag to attain optimum days of mycelial run, flushing
using the custom rate of PhP1.50/bag.
intervals and yield in oyster mushroom production;

Table 11. Comparison of cost analysis results for the two substrate
2. The designed compactor-bagger for oyster mushroom considers
formulations.
the mushroom technologies of the country to compact 750 grams of
PARTICULARS CLSU BSU substrate material in a 152mm x 305mm PP bag utilizing piston-type
Specifics of compression. It is designed to bag at least 100% greater than the
Salvage value, (10% IC) SV 5,825.7 5,825.7 manual bagging capacity of 25 bags/hr;
Depreciation, PhP/yr D 10,486.26 10,486.26
Interest on Investment, PhP/yr I 6,728.68 6,728.68 3. The fabricated machine could be fabricated using local
Taxes, Insurance and Shelter, TIS 2,912.85 2,912.85 manufacturing technologies and is made of locally available
PhP/yr (5% IC) materials. It is recommended for CLSU and BSU formulation;
Fixed Cost, PhP/yr Fc 20,127.79 20,127.79
Variable Cost, PhP/hr Vc 181.77 181.77 4. The capacities of the machine divided by three operators is still
Total bagging cost, PhP/yr Tc 190,731.92 190,731.92 more competitive than manual bagging capacity of 25 bags/hr with
Bagging Cost, PhP/bag Bc 0.995 0.66 68 bags/hour-102.33 bags/hour. The machine is also seen to
eliminate the breakage of plastic bags which was observed in manual
Breakeven point, bag/yr BEP 32,997.00 22,119.00
bagging;
Net Income per year, PhP/yr NI 96,736.00 191,749.00
Payback Period, yr PBP 0.60 0.304
5. The use of the machine could produce bags within the acceptable
limit of ±50 grams of 750 grams ranging from 740.90-758.94 grams
depending on the type of formulation used. The machine decreases

145
the gap of ununiformed weight per bags in manual bagging that
could result to inconsistent days of ramification, yield, and flushing
intervals; and,
6. The cost of production per bag using manual labor is decreased
from PhP2.00 to PhP0.66-PhP0.995/bag. When the machine is used
six-hour a day, thrice a week, at custom rate of PhP1.50/bag income
to be generated may range from PhP96,736.00-191,749.00/year and
a payback period of 0.304-0.60 years when using either CLSU and
BSU formulation.
ACKNOWLEDGEMENT
The support of DOST, the Department of Agricultural and
Biosystems Engineering of the College of Engineering through the
project leader of the CLSU-DOST-ERDT Program Dr. Ireneo C.
Agulto, and the Central Luzon State University- Center for Tropical
Mushroom Research and Development (CLSU-CTMRD) with its
project leader- Dr. Sofronio Kalaw, and the Benguet State
University -Mushroom Project is highly recognized in this paper.
REFERENCES
[1] Philippine News Agency. (2010). Raise mushroom; DOST
urges farmers. Retrieved from www.balita.ph:
http://balita.ph/2010/07/06/raise-mushroom-dost-urges-
farmers/
[2] Chang, H. Y., Jeon, S. W., Cosadio, A. L., Icalina, C. L.,
Panganiban, R., Quirino, R., & Song, Y. (2014). Status and
Prospect of Mushroom industry in the Philippines. Journal of
Philippine Association of Institutions for Research.
[3] Business Diary Ph (2015, January 31). DA puts up P8 million
Mushroom Technology Center in Tarlac to boost production,
substitute mushroom import from Taiwan. Retrieved from
Make Money – Business Diary Ph.:
http://businessdiary.com.ph
[4] De la Torre, E. (2010). Strengthening the Capabilities of
climate change risk management and Disaster Preparedness in
Selected Provinces of the Philippines (Bicol Region). FAO,
DA.
[5] Oei, P. (2005). Small-scale mushroom. Digigrafi, Wageningen,
The Netherlands: Agromisa Foundation and CTA.
[6] Henderson, S. Milton., Perry, R. L. (1955). Agricultural
process engineering. New York: Wiley.
[7] Hunt, D. O. (2001). Farm Power and Machinery Management.
Iowa State University Pres. Ames, Iowa. 10th Edition. Reissued
2008 by Waveland Press Inc.
[8] Mondal, S., Rehana, J., Norman, M., and Adhikary, S. (2010).
Comparative study on growth and yield performance of oyster
mushroom on differerent substrates. journal of Bangladesh
Agriculture University.
[9] Zireva, D.T., M. Fanadzo and Mashingaidze, A.B. (2007).
Effect of Substrate Quantity and Shelf Position on Yield of
Oyster Mushroom (Pleurotus sajor caju). Pakistan Journal of
Biological Sciences, 10: 3458-3461.
146
 Development of a Compost Tea Brewer Machine
Elman C. Torres1, a*, Emmanuel V. Sicat2, Marvin M. Cinense3 and Carolyn Grace G. Somera4
1,2,3 & 4
Central Luzon State University, Science City of Muñoz, Nueva Ecija
a
engr.ectorres@gmail.com
Abstract—This study was carried out to develop and
evaluate the performance of the compost tea brewer machine. The
specific goals were to solve the problem of the previous design of
the compost tea brewer machine with regards to the inability to
maintain the aerobic condition of compost tea during brewing
process, inability to extract available nutrients and microorganisms
available in compost, and design problems which result to difficulty
in properly cleaning the machine. The developed machine was
evaluated in terms of the quality of compost tea produced. Three
settings of the rotational speed of the mixing vessel (1rpm, 3rpm,
and 5rpm) were evaluated on brewing compost and its influence on
the quality of compost tea produced. Data gathered were analyzed
using Random Complete Block Design and Least Significant
Difference was used for comparison of treatment means. Blocking
was done on each batch of produced vermicompost used in the
study. Results showed that 5rpm setting of vessel mixing
vermicompost gave a good quality of produced compost tea
(dissolved oxygen of 6.43mg/L, electrical conductivity of 1.530mS-
cm-1, pH level of 8.27, total bacterial count of 7.02log10 CFU/mL,
total NPK content of 900mg/L and seed germination index of
133.97%) which is above the m values set in this study. Based on
the compost tea brewer machine’s cost of Php 43,033.00 with a
capacity of 100 Liters of compost tea per brewing process, cost
analysis indicated a breakeven of 45 brewing cycles per year and a
payback period of 2.4 years.
Keywords—Aerobic Condition, Aeration System,
Compost Tea Brewer Machine.
I. INTRODUCTION
Composting, a process of decomposition of organic
materials, plays a key role and is the major process in making
organic fertilizer. It is the natural process of rotting or decomposing
organic matter by microorganisms under controlled conditions. Raw
organic materials such as crop residues, animal waste, food garbage,
some municipal waste and suitable industrial wastes, enhance their
suitability for application to the soil as a fertilizing source, after
having undergone composting [1].
Since compost is in solid form, it is difficult for plants to
absorb the nutrients out of it. To ensure faster absorption of nutrients
by plants, it is necessary to make compost tea. Compost tea is
produced by mixing compost with water and culturing to a defined
period, either actively aerating (aerated compost tea, ACT) or not
(non-aerated compost tea, NCT) and with or without additives that
are intended to increase microbial population densities during
production [2].
Both extraction methods involve the steeping of compost
in water for a defined period at room temperature. ACT requires
aeration throughout the extraction period [3]. Weltzein, as a pioneer
in this area, focused primarily on the non-aerated method of compost
tea production in the late 1980s and early 1990s. However, in recent
years, interest has shifted to the ACT method [4]. From a grower’s
perspective, ACT has a distinct advantage because it can be prepared
in 1 to 2 days and results in fewer odor problems, whereas NCT
requires 1 to 2 weeks steeping time [5]. However, NCT does not
require any special technology beyond a steeping vessel and is
associated with low cost or low energy input, whereas ACT requires
constant stirring and aerating of large volumes of liquid.
Aerated compost tea a more recent approach adopted by
compost tea brewers involves an active process that relies on the use
of an aerator to oxygenate the mixture of compost and water during
the brewing process, thereby producing aerated compost tea in a
shorter brewing time ranging from 12 hours to 72 hours.
However, there are documented problems with the recent
design of compost tea brewer machines available in the market. For
instance, the inability of several machines to achieve aerobic
conditions (6mg/L of dissolved oxygen level and above), inability
to extract available nutrients and microorganisms from the compost
and design of the machine that leads to the inability to properly clean
the machine.
Locally available design of aerated compost tea brewer
machine includes an air diffuser for increasing oxygen level and
having a vessel to hold the compost during the brewing time. Aside
from the attached air diffuser and several openings for the
interaction of water and compost on the vessel, there are no other
mechanisms that will cause the agitation of compost inside the
vessel which can lead to the inability of the machine to extract
available nutrients and microorganisms and it can lead to the
anaerobic condition of the compost. To overcome the constraints
discussed above, it is necessary to develop a new compost tea brewer
machine that will address the above problem using locally available
materials.
A. Objectives of the Study
Generally, this study aimed to develop a compost tea
brewer machine. Specifically, it aimed to design and fabricate a
compost tea brewer machine, evaluate the quality of produced
compost tea of the designed machine in terms of dissolved oxygen
level, electrical conductivity, pH level, total bacterial count and seed
germination index, and NPK content and analyze the cost of using
compost tea brewer machine.
II. METHODOLOGY
A. Design and Fabrication of the Machine
The machine was developed to address the problem in
current designs of available compost tea brewer machines. The
design and dimensions of each component were based on a desired
capacity and proportions of brewing materials (water and compost
in terms of volume). The proportion of compost to the water of the
machine used was 1:10 [6]. The duration of the brewing process
using the designed machine is 24 hours [5]. The designed compost
tea brewer was able to make 100 liters of compost tea per 24 hours
brewing process.
The primary components of the compost tea brewer
machine are the brewing tank, compost vessel, aeration system,
transmission system and frame. The sizes of the machine elements
of each component were designed based on estimated loads and
147
allowable stresses of the selected materials. The design drawing was
drafted using the SolidWorks software. Further analyses of the
major elements were also done using such features of SolidWorks.
Based on the approved design plan, the machine was
fabricated by an accredited local machinery fabricator that had
simple and local manufacturing tools and equipment. Stainless steel
materials and HDPE plastics were used in the components that are
in contact with the compost tea. Simple manufacturing processes,
namely: measuring, cutting, joining of different parts by welding,
bolts and nuts, assembly of the different components, finishing and
painting were employed by the local manufacturer.
B. Preliminary Testing of the Machine
Preliminary tests were conducted to determine the
working conditions of different components of the machine.
Necessary adjustments and/or minor modifications were done until
the machine satisfied the required rpm and DO level to be
maintained. Visual inspection was made on parts of the machine.
This also allowed the familiarization of the researcher with the
machine. A 24-hour test run was conducted to ensure that the
machine will be able to attain the required rotational speed and
minimum level of dissolved oxygen during the performance
evaluation.
C. Compost Preparation
The type of compost used in the study was vermicompost.
Vermicompost is the product of composting process using various
species of worms to create a heterogenous mixture of decomposing
organic materials [7]. The method used in the production of
vermicompost was recommended on Ramon Magsaysay Center for
Agricultural Resources and Environment Studies (RM-CARES) at
Central Luzon State University (CLSU). The materials were
collected at the vicinity of CLSU. Table 1 shows the materials used
in vermicomposting.
Table 1. Test materials used in vermicomposting.
Item Ratio Quantity in kg
Carabao manure 6 60
Dried mango leaves 2 20
Rice straw 2 20
Chopped banana trunks 1 10
Fresh madre cacao 1 10
D. Compost Tea Production
One week old vermicompost (after harvest) was used to
make compost tea using the designed compost tea brewing machine.
The dilution ratio (in terms of volume) of vermicompost and water
was 1:10 per brewing process [6]. The brewing materials used
during the evaluation of the machine are listed in Table 2.
Table 2. Brewing materials used during the performance evaluation
of machine.
ITEM QUANTITY
Vermicompost 6.83kg* (0.01 m3)
Tap Water 100 liters (0.10 m3)
*weight of vermicompost is based on highest significant value of
bulk density of vermicompost reported [9].
E. Compost Tea Brewing Procedure and Data Gathering
Compost tea brewer machine with all necessary
instrument, apparatus, and brewing materials were placed in a
covered area with electrical power source. Brewing tank was filled
with 100 liters of water. Power switch of the air pump was turned
on. Water inside the brewing tank was aerated for 2 hours to raise
the DO level from 4mg/L to 8 to 12mg/L and to dissipate the
chlorine from water. The vermicompost (6.83kg) was placed on
compost vessel and placed in the machine. The machine was allowed
to brew compost tea for 24 hours. After 24 hours of brewing process,
the machine was switched off and the gathering of samples of
compost tea was done. The same procedure was done for each
setting of rotational speed.
During brewing process, DO level was recorded every six
(6) hours using pen type dissolved oxygen meter from the start up to
the end of brewing process at the surface liquid in brewing tank.
Data gathered was used to determine the trend of DO level over time
of brewing process. Electrical conductivity and pH level were
recorded using electrical conductivity meter and pH meter at the end
of brewing process at the surface liquid in brewing tank. One (1)
sample of 350 mL compost tea was collected at the discharge after
brewing process and was placed on unused distilled water container.
Each sample taken was subjected to laboratory testing to determine
the total percentage of NPK in compost tea.
For the total bacterial count, one (1) sample of 350 mL of
compost tea was collected every six (6) hours from the start up to
the end of brewing process at the surface liquid inside the brewing
tank and was placed on unused distilled water container. One
milliliter of well stirred sample was taken, dilution series prepared,
and microorganisms were cultured using pour plate technique with
three replicates. Nutrient agar was used as the culture media. Then
cultured sample were kept in an incubator at 37oC for 24 hr. All
microbial colonies were counted separately, and total bacterial count
(CFU/mL) was calculated.
For the seed germination index, one (1) sample of 350 mL
compost tea was collected after brewing process and was placed on
unused distilled water container. Each sample taken was subjected
to seed germination bioassay to determine the seed germination
index of tomato (Lycopersicon esculentum cultivar Diamante Max
F1) applied with compost tea. Five (5) milliliter of compost tea was
pipetted into a sterile petri dish lined with filter paper. Ten seeds of
tomato were evenly distributed on the filter paper and placed in dark
room. . For control, five (5) milliliter of distilled water was pipetted
into a sterile petri dish lined with filter paper. Each sample were run
in triplicate. Sterile petri dishes were left in dark room and after 5
days, all the tomato seeds successfully germinated and total length
of each tomato root was recorded.
The power consumption by the machine was recorded
using digital power meter. Power consumption was calculated by
multiplying the power recorded to the time consumed by the
machine to brew the compost tea. For the NPK of compost as test
material, a 500 g of sample was collected from compost before
subjecting to brewing process. Samples was placed in individual
plastic container. Each sample taken was subjected to laboratory
testing to determine the total percentage of NPK in compost.
F. Performance Evaluation of the Machine
The machine was evaluated based on the extraction
efficiency and quality of produced compost tea as affected by
rotational speed of compost vessel. Data were gathered and recorded
148
on a prepared data sheet. Performance parameter that was used as B. The Principle of Operation
reference in the evaluation is shown in Table 3.
The main function of compost tea brewer machine is to
Table 3. Minimum compost tea quality requirements. extract microorganisms, nutrients and organic compounds available
on compost into a water solution. Compost vessel held compost and
Item Unit Values the lower portion of compost vessel was suspended into the tank
Dissolved Oxygen Level, [5] mg/L 6 or higher , with water throughout brewing process. A motor caused the compost
pH Level, [8] - 6.5 – 8.4 vessel to rotate through chain and sprocket connection to mix the
Electrical Conductivity, [6] & [9] mS-cm–1 1.1 or higher compost inside the vessel and to allow extraction of microorganism,
257mg/L or nutrients and organic compounds present in the compost. Air pump
NPK Content, [10] mg/L
higher directed the air through PVC pipes into the tank which stirred its
Log10 contents and kept the consistency of the compost tea relatively
Population of Bacteria, [11] 5.00 – 9.00 uniform. It provided oxygen to support the growth of microorganism
CFU/ml
Germination Index, [12] % 60 or higher in compost tea. Brewing tank held the compost tea throughout the
brewing process.

G. Experimental Design Table 4. Specifications of the compost tea brewer machine.

Item Specification
The statistical analysis used was a single factor
experiment and it was laid out in randomized complete block design A. Capacity, Liters/brewing process 100
(RCBD). Analysis of variance (ANOVA) was used to determine B. Main Structure
whether the differences among the treatment means are significantly Dimensions and Weight
different. Blocking was done to eliminate the variance in the time of Length, mm 1,590
production of each batch of vermicompost. Least significant Width, mm 690
difference (LSD) was used for the comparison of means of the Height, mm 1,080
different treatments. There were three (3) settings of continuous Weight, Kg 105
rotational speed (1, 3, and 5 rpm) of mixing compost in compost C. Brewing Tank
vessel that served as the treatments in this study. Materials HDPE Plastic
Length, mm 970
Diameter, mm 550
III. RESULTS AND DISCUSSION D. Compost Vessel
Materials HDPE Plastic
Length, mm 800
A. Description of the Machine
Diameter, mm 260
The compost tea brewer machine was designed using E. Prime mover
locally available materials and it was fabricated adopting local Manufacturer SPG Co., Ltd.
manufacturing technology. The major components of the device Type Induction Motor
were brewing tank, transmission system, aeration system, compost Rated power, W 25
vessel and frame. The fabricated compost tea brewer machine, as Rated speed, rpm 1,700
shown in Figure 1 in a batch type manner, where the compost was
Gear Head Ratio 1:36
loaded in compost vessel. One operator was required during loading
Weight, kg 4.2
of materials, brewing process and cleaning of machine after
brewing. Table 4 shows the specification of the compost tea brewer F. Air Pump
machine. Manufacturer Resun
Air Flow Rate, L/min 75 x 2
Power Rating,W 58
Type Electro-magnetic
Air Pump
Weight, kg 2.5 kg

C. The Characteristic of Compost as Test Material

Table 5 shows the chemical characteristics of compost
used to produce compost tea using the designed machine. The total
nitrogen, total phosphorus and total potassium of compost in all
batch were within the ideal range of good compost with a values of
3,000 – 15,000 mg/Kg, 1,000 – 10,000 mg/Kg and 1,500 – 5,600
mg/Kg respectively [1].

Table 5. Chemical characteristics of compost used in brewing

process (mg/Kg).
Batch of Mean Values (mg/Kg) Total NPK
Compost Nitrogen Phosphorus Potassium (mg/Kg)
Batch 1 5,700 4,900 2,200 12,700
Fig. 1. Compost tea brewer and its part
Batch 2 5,500 4,800 2,500 12,900
Batch 3 5,200 4,200 2,700 12,100

149
D. Results of Preliminary Testing compost and water with ratio of 1:10 were used to test and evaluate
the performance of compost tea brewer machine.
After fabrication of the machine, preliminary testing was
done to check if the machine worked according to its purpose.
During the first preliminary testing, rotational speed of compost
vessel was established through series of different number of teeth of
the sprocket and using motor controller that regulated the rotational
speed of motor. A power switch was also provided at the right side
of the device for safety purposes and ease of operation.

Second preliminary testing was done to test the capability

of aeration system of the machine to maintain the aerobic conditions
of compost tea and the transmission system of the machine to
continuously mix and agitate the compost inside the vessel. It was
tested using 5rpm (highest rpm setting of the machine) rotational
speed of compost vessel. Test materials used was vermicompost
produced by RM-CARES, water, molasses and vegetable oil. Fig. 3. Trend line of dissolved oxygen level of compost tea during
Dissolved oxygen level was recorded every three (3) hours using DO 3rd preliminary testing of the machine
meter to observe its trend over time. After 24 hours of brewing
process, a downward trend Figure 2 was observed in the dissolved
oxygen over time during the first 12 hours of operation (from E. Performance Characteristics of the Machine
7.4mg/L to 1.2mg/L) and an upward trend during the last 12 hours
of operation (from 1.2mg/L to 4.2mg/L). The machine was not able The overall performance of the machine was determined
to maintain the aerobic condition of the compost tea so research was by measuring the different parameters considered in the study. Data
done to determine the cause of this event. on the quality of compost tea in terms of dissolved oxygen level,
NPK Content, electrical conductivity, pH level, population of
microorganisms and seed germination index of compost tea brewer
machine were gathered and analyzed.

Dissolved Oxygen Level

The dissolved oxygen level of the compost tea as affected

by varying rotational speed of compost vessel is shown in Table 6.
The highest mean of DO level was at 5rpm setting with 6.43mg/L
while the lowest mean was at 1rpm setting with 6.20mg/L. The
values recorded at three (3) rotational speed were above the
minimum value of 6.00mg/L in which the aerobic bacteria required
[6].

Table 6. Dissolved oxygen level (mg/L) of compost tea as

Fig. 2. Trend line of dissolved oxygen level of compost tea during affected by rotational speed of compost vessel.
2nd preliminary testing of the machine Rotational Block
speed, Mean
Reference [13] and [5] stated that using molasses in I II III
RPM
brewing process will cause the microorganisms present in the 1 6.3 6.2 6.1 6.20
compost to grow faster and consume more oxygen and as a result 3 6.4 6.5 6.3 6.40
the microorganism present used up oxygen at a rate faster than the 5 6.3 6.6 6.4 6.43
machine adding oxygen into the compost. As a result, dissolved
oxygen level in the water dropped. It was decided to remove the
Analysis of variance revealed an insignificant effect of
molasses and vegetable oil in the composition of test materials to be
rotational speed of compost vessel to the dissolved oxygen level of
used in the brewing process. Reference [9] and [10] used only
the compost tea during the brewing process. The dissolved oxygen
compost and water to observe the microbiological and chemical
of compost tea produced were 6.20mg/L, 6.40mg/L, and 6.43mg/L
properties of compost tea.
for 1rpm, 3rpm and 5rpm respectively. Therefore, the DO level of
compost tea was the same whether it was produced using 1rpm,
Third preliminary testing was done in order to test the
3rpm or 5rpm setting of the rotational speed of compost vessel. DO
machine in terms of maintaining 6mg/L of dissolved oxygen level.
level of compost tea produced was mainly affected by the aeration
Using the same setting in 2nd preliminary testing of the machine and
system of the machine which was set in fixed air flow rate of 150
test material of vermicompost and water only with a ratio of 1:10,
liters/minute.
the machine was able to maintain the dissolved oxygen level of
above 6mg/L Figure 3. A downward trend was also observed in DO
Electrical Conductivity
level over 24 hours of brewing process. It was a normal condition
due to the growth of microorganism population present in the
The effect of varying rotational speed of compost vessel
compost tea.
to the electrical conductivity of compost tea produced is shown in
Table 7. The highest mean of electrical conductivity was at 5rpm
Considering the problem of decreasing DO level with
setting with 1.530mS-cm–1 while the lowest mean of electrical
regards to the test materials to be used in brewing process, only
conductivity was at 1rpm setting with 1.470mS-cm–1. The values

150
recorded at three (3) rotational speed were above the minimum value aerated compost tea did not vary greatly (5.07 – 5.16) as affected by
of 1.100mS-cm-1 in which had a greater plant growth response [9]. duration of brewing process.
Analysis of variance revealed a statistically significant indication at
α=0.01 to show that there was a highly significant difference in Total Bacterial Count
electrical conductivity among the three (3) rotational speeds of
compost vessel to mix the compost. Comparison of among means of Total bacterial count (log10 CFU/mL) of compost tea as
the electrical conductivity as influenced by rotational speed of influenced by varying rotational speed of compost vessel is shown
compost vessel revealed that electrical conductivity of compost tea in Table 9. Highest mean of total bacterial count of compost tea with
produced at 1rpm setting was significantly lower compared to the 7.13log10 CFU/mL was recorded at 3rpm setting while lowest mean
electrical conductivity of compost tea produced at 3rpm and 5rpm total bacterial count was recorded at 1rpm with 6.99log10 CFU/mL.
settings. The electrical conductivity of compost tea produced at The values of total bacterial count of compost tea produced at three
3rpm and 5rpm did not differ significantly. It could be concluded different settings were within the ideal range of total bacterial count
that 3rpm and 5rpm setting produced compost tea with higher of compost tea (5.00 to 9.00log10 CFU/mL). The values of total
electrical conductivity compared to 1rpm setting. bacterial count were also in agreement with the range of results (6.43
to 8.62log10 CFU/mL) of other study of compost tea [10].
Table 7. Electrical conductivity (mS-cm–1) of compost tea as
affected by rotational speed of compost vessel. Table 9. Total bacterial count (log 10 CFU/mL) of compost tea
Rotational Block as affected by rotational speed of compost vessel.
speed, Mean Rotational Block
I II III speed, Mean
RPM
I II III
1 1.490 1.467 1.448 1.470b RPM
3 1.544 1.500 1.500 1.510a 1 7.091 7.15 6.73 6.99
5 1.550 1.540 1.510 1.530a 3 7.180 6.92 7.29 7.13
Means not sharing same letter in a column differ significantly at 1% 5 7.120 6.96 6.98 7.02
by Least Significant Difference (LSD).
Analysis of variance revealed no significant difference on
The higher rotational speed of vessel mixing compost the total bacterial count of compost tea produced among the three
helped the extraction of available soluble salts in the compost. rotational speed of compost vessel. The mean total bacterial count
Source of soluble salts in composts such as carabao manure consists of compost tea were 6.99log10 CFU/mL, 7.13log10 CFU/mL and
of ammonium, nitrate and phosphate and other major soluble salts 7.02log10 CFU/mL for 1rpm, 3rpm and 5rpm respectively.
such as sodium, magnesium, calcium and sulfate. Reference [14] Therefore, the total bacterial count of compost tea produced at 1rpm,
suggested that the electrical conductivity of compost tea should not 3rpm and 5rpm were the same.
exceed to 2.00mS-cm–1 or if excessive soluble salts exist, plant
injury, such as reduced germination rates and leaf burning, or death This could be attributed also to the compost being used in
might occur. the study and level of DO level which the air flow rate given aeration
system was constant. Reference [5] suggested that if mixing of
pH Level compost was not adequate, the compost would compact and
eventually would go to anaerobic condition because it did not have
The pH levels of compost tea as influenced by varying aerated water moving through the compost. The brewing process in
rotational speed of compost vessel is shown in Table 8. Highest all experimental units were above the minimum aerobic range of DO
recorded mean of pH level of compost tea was produced at 1 RPM level of about 6mg/L and above.
setting with 8.4 while lowest recorded mean of pH level was
produced at 3 RPM setting. The values of pH level of compost tea Total NPK Content
produced at three rotational speed setting were within the
recommended range of pH level (6.50 – 8.40). The total NPK content (%) of compost tea as influenced
by varying rotational speed of compost vessel is shown in Table 10.
Table 8. pH level of compost tea as affected by rotational speed The highest mean of total NPK content of compost tea was produced
of compost vessel. at 5rpm setting while the lowest mean of compost tea was produced
Rotational Block at 1rpm setting. The values in the experimental unit were above the
speed, Mean baseline set in the minimum total NPK (257mg/L) content based on
I II III other published study of compost tea [6] & [10].
RPM
1 8.4 8.4 8.4 8.40
Table 10. Total NPK content (mg/L) of compost tea as affected
3 8.3 8.1 8.1 8.13
by rotational speed of compost vessel.
5 8.2 8.2 8.4 8.27
Rotational Block
speed, Mean
Analysis of variance showed that there was no significant I II III
RPM
effect of varying rotational speed of compost vessel in the pH level
of compost tea produced. The pH level of compost tea were 8.40, 1 800 640 400 613.33
8.13 and 8.27 for 1rpm, 3rpm and 5rpm respectively. Therefore, it 3 800 900 900 866.67
can be concluded that pH level of the produced compost tea is the 5 700 800 1200 900.00
same whether it was produced at 1rpm, 3rpm and 5rpm setting.
Analysis of variance revealed no significant difference on
The pH level across the experimental unit did not vary the total NPK content of compost tea produced at three (3) different
greatly due to the same compost materials being used in each rotational speed setting. The mean values of compost tea were
experimental unit of the study. In reference [10], the pH level of 613.33mg/L, 866.67mg/L and 900mg/L for 1rpm, 3rpm and 5rpm

151
setting respectively. Therefore, it could be concluded that total NPK compost tea in all three (3) rpm setting. Downward trend of DO level
content of the produced compost tea was the same whether it was and upward trend of TBC could be attributed to the growth on
produced at 1rpm, 3rpm or 5rpm setting. population of bacteria in the compost tea. As the population of
bacteria increased the demand for oxygen increased and as a result,
Despite the insignificant result of analysis of variance, it the DO level decreased in the compost tea. The critical part of
could be observed that the higher rpm would have higher NPK brewing process is when the DO level of compost tea reached below
content compared to lower rpm, but several authors suggested that 5mg/L, anaerobic condition ensued and materials that were toxic to
solubility of nutrients could also be attributed to the quality of plant growth were produced in the compost tea.
compost and different levels of concentration of nutrients present in
compost to water [15]. In addition, it was observed that the highest
and lowest total NPK content and electrical conductivity of
produced compost tea were at 5rpm and 1rpm respectively. It only
indicated that the higher the rotational speed of compost vessel the
more nutrients would be present in compost tea due to agitation and
mixing of compost.

Seed Germination Index

The seed germination index of tomato (sn. Solanum

Lycopersicum cv. Diamante Max F1) applied with compost tea
produced at three (3) different settings of the rotational speed of
compost vessel was shown in Table 11. The highest mean of seed
germination index was at 5 RPM setting with 133.97% while the
lowest mean of seed germination index was at 1 RPM setting with
92.18%. All data of seed germination index recorded in each
experimental unit at three (3) rotational speed were above the
minimum threshold of 60.00% as suggested for a safe soil Fig. 4. Average dissolved oxygen level (mg/L) and total bacterial
application [12]. count (Log10 CFU/mL) of compost tea as affected by varying speed
of compost vessel during 24 hours of brewing process
Analysis of variance reveals a statistically significant
indication at α=0.05 to show that there is a difference in the Power Consumption
germination index of tomato applied with compost tea produced at
three (3) different rotational speed setting. Comparison of among The power consumption of the machine was determined
means of the seed germination index as an influenced by the using digital power meter. The power consumption of compost tea
rotational speed of compost vessel revealed that seed germination brewer machine as affected by rotational speed of compost vessel
index of compost tea produced at 1 RPM was significantly lower was shown in Figure 5. It could be seen that the power consumption
compared to seed germination index of compost tea produced at 3 ranged from 2.522 to 3.072kW. Highest power consumption was
RPM and 5 RPM setting. The seed germination index of compost recorded at 1rpm while the lowest power consumption was recorded
tea produced at 3 RPM and 5 RPM setting does not differ at 5rpm. Power consumption was lower at highest rotational speed
significantly. Therefore, 3 RPM and 5 RPM settings produced setting of compost vessel compared to lowest rotational speed. It was
compost tea with a higher seed germination index compare to 1 RPM due to the use of AC motor speed controller that regulated the
setting. current flow to the motor and as a result, it was able to control the
desired speed at given current flow output by AC motor speed
Table 11. Germination index (%) of tomato (sn. Solanum controller. AC motor speed controller was a close loop speed
lycopersicum cv. Diamante Max F1) applied with controller used for controlling motor speed in response to varying
compost tea produced at three (3) different rotational current supply. Electric motor decreased its efficiency as the current
speed of compost vessel. flow increased caused by AC motor speed controller. Using a lower
setting speed setting in AC motor speed controller the current flow
Rotational Block
in electric motor increased and as a result, the power consumption
speed, Mean
I II III increased.
RPM
1 107.26 99.26 70.03 92.18b
3 128.73 116.33 131.58 125.55a
5 133.26 128.60 140.06 133.97a
Means not sharing same letter in a column differ significantly at 1%
by Least Significant Difference (LSD).

Interaction between Dissolved Oxygen and Total Bacterial Count of

Compost Tea over Time of Brewing Process.

The desirable level of dissolved oxygen (DO) for brewing

process of compost tea is 6 mg/L (Ingham, 2005). The machine was
able to maintain the desired level of dissolved oxygen level (6.20 to
7.50mg/L) all throughout the brewing process, however, downward
trend line (as shown in Figure 4) was observed on dissolved oxygen Fig. 5. Power consumption of machine as affected by varying speed
level of compost tea in all three (3) rpm settings of the machine. It of compost vessel during 24 hours of brewing process
observed an upward trend line on total bacterial count (TBC) of

152
Cost Analysis of the Machine
The cost of materials and fabrication of compost tea
brewer machine was Php 45,033.00 with expected life span of 5
years. Considering all the assumptions made including the annual
used of machine per year of 98 brewing cycle per year and price per
liter of 8.00 Php/Liter, cost analysis indicated a breakeven of 45
brewing cycle per year and a payback period of 2.4 years. Figure 6
shows the breakeven graph analysis of using compost tea brewing
machine. It is revealed that the compost tea brewer machine could
earn a net income when the machine operates more than 45 brewing
cycle per year.
Fig. 6. Breakeven graph analysis of using compost tea brewer
machine
IV. CONCLUSION AND
RECOMMENDATION
Based on the results of the study, the following conclusion
were drawn: A compost tea brewer machine with five (5) major
components namely: compost vessel, brewing tank, aeration system,
transmission system, and main frame was successfully fabricated
following the design calculations and considerations. It operated on
the principle of extraction of beneficial organism and nutrients
through rotational motion of vessel containing compost. The
machine with an overall dimension of 1,365mm x 665mm x
1,370mm was completely fabricated using available materials from
local manufacturing shop.
The machine performed satisfactorily based on its
performance parameters in terms of quality of the produced compost
tea (dissolved oxygen level, electrical conductivity, pH level, total
bacterial count and total NPK content). Generally, the compost tea
brewer machine best operated at operating speed of 5rpm with
dissolved oxygen level of 6.43mg/L, electrical conductivity of
1.530mS-cm–1, pH level of 8.27, total bacterial count of 7.02 log10
CFU/mL, total NPK content of 900mg/L and seed germination index
of 133.97%. The cost analysis on the use of the machine revealed
that it is financially viable if it is used for selling produced compost
tea to the farmer. The cost of fabricating the device was Php
45,033.00. The average annual net income was 18,739.78Php/year,
with a computed breakeven brewing cycle per year and payback
period of 45 brewing cycle/year and 2.4 years respectively.
Based on the results of the study, the following were
hereby recommended: The compost tea brewer machine should be
used at operating speed of 5rpm and brewing time of 24 hours. The
machine could be evaluated using other types of compost. The
machine could be evaluated using shorter duration of brewing
process. articles.
ACKNOWLEDGEMENT
The Engineering Research and Development for
Technology (ERDT) Scholarship, Department of Agricultural and
Biosystems Engineering, College of Engineering and Central Luzon
State University are hereby acknowledged in this research.
REFERENCES
[1] brackets [1]. The TNAU Agritech Portal. (2009). Organic
Farming: Compost, URL:
http://www.agritech.tnau.ac.in/org_farm/orgfarm_composti
ng.html
[2] Scheuerell, S.J. (2004). Compost Tea Production Practices,
Microbial Properties, and Plant Disease Suppression. Soil
and Compost Eco Biology, International Conference of Soil
and Compost Eco-Biology, Leon, Spain, pp. 1.
[3] Weltzien, H. C. (1991). Biocontrol of foliar fungal disease
with compost extracts, Microbial Ecology of Leaves.
Springer-Verlag, New York, p. 430-450.
[4] Scheurell, S.J., & Mahafee, W.F. (2002). Compost Tea
Principles and Prospects for Plant Disease Control. Compost
Science and Utilization, Volume 10, 2002, pp. 313-338
[5] Ingham E.R. (2005). The Compost Tea Brewing Manual.
Soil Food Web Incorporated. Corvellis Oregon 97333. Fifth
Edition, pp. 1-5.
[6] Pant, A., Radovich, T.J.K., Hue, N.V. & Arancon, N.Q.,
(2011). Effects of vermicompost tea (aqueous extract) on
pak-choi yield, quality, and on Soil biological properties.
Compost Sci. Util. 19 (4), pp. 279–292.
[7] Ndegwa PM, SA Thompson and KC Das, 1998. Effects of
Stocking Density and Feeding Rate on Vermicomposting of
Biosolids. Bioresource Technology, 71: 5-12.
[8] Ayers, R.S., & Westcot, D.W. (1976) Water Quality for
Agriculture. Food and Agriculture Organization of United
Nations. Rome, Italy 1976, pp 7
[9] El-Haddad, M.E., Zayed, M.S., El-Sayed, G.A.M.,
Hassanein, M.K. & El-Satar A.M.A. (2014). Evaluation of
Compost, Vermicompost and their Teas Produced from Rice
Straw as Affected by Addition of Different Supplements.
Annals of Agricultural Science (2014) 59(2), pp. 243-251
[10] Martin, C.C.G. (2014). Potential of Compost Tea for
Suppressing Plant Diseases. CAB Reviews. 9(32), pp. 1-38.
[11] Ngampimol, H., & V. Kunathigan, (2008). The study of
shelf life for liquid biofertilizers from vegetable waste.
Assumption Univ. J. Technol., 11, pp. 204–210
[12] Zucconi, F. and De Bertoldi, M. (1987) Compost
Specifications for the Production and Characterization of
Compost from Municipal Solid Waste. In: De Bertoldi, M.,
Ferranti, M.P., L′Hermite, M.P. and Zucconi, F., Eds.,
Compost: Production, Quality and Use, Elsevier, London,
pp. 276-295.
[13] Western Sustainable Agriculture Research and Education,
(2017). Aerated Compost Tea: A Field Guide to Production
Methods, Formulas and Application Protocols,
Albuquerque, New Mexico 87197, pp. 3-5
[14] Gartley, K.L. (2011). Chapter 10: Recommended Methods
for Measuring Soluble Salts in Soils. Recommended Soil
Testing Procedures for the Northeastern United States.
USA, pp. 88-93
[15] Bernal, M.P., Paredes C., Sanchez-Monedero, M.A.,
Cegarra, J ., Maturity and stability parameters of composts
prepared with a wide range of organic wastes, Bioresource
Technology, 63 (1998), pp. 91-99
153
 Development of an Impact-type Cacao
(Theobroma cacao L.) Bean Huller
Herald Kam G. Hernando1,a*, Carolyn Grace G. Somera2,b, Jeffrey A. Lavarias2,c, and Emmanuel V. Sicat2,d.
Department of Agricultural and Biosystems Engineering, College of Engineering, Central Luzon State
University, Science City of Munoz, Nueva Ecija
heraldkam@gmail.com1,a*, cggsomera@clsu.edu.ph2,b, jeffreylavarias@clsu.edu.ph2,c,
emmanuelsicat@yahoo.com2,d
Abstract - The study mainly aimed to develop an impact type
cacao bean hulling machine and evaluate its performance in
producing cacao nibs in terms of input capacity, hulling
capacity, hulling efficiency, nib recovery, large nib recovery,
purity, noise level and power consumption. The cacao huller
was subjected to three varying tangential velocities in
cracking mechanism, specifically at 40m/s, 47m/s, and 54m/s,
and with three different volumes of air in winnowing
mechanism with 41m3/hr, 45m3/hr, and 49m3/hr. The data
gathered were arranged in a CRD Two Factorial and
comparison among means using Statistical Tool for
Agricultural Research (STAR) software. The analysis showed
that tangential speed of the machine significantly affected all
parameters except purity and noise level; moreover, the
volume of air significantly affected the hulling efficiency, nib
recovery, large nib recovery and purity. Impact-type cacao
bean huller with 47m/s and 41m3/hr gave the highest input
capacity and hulling capacity of 195.60kg/hr and 156.53kg/hr,
respectively. Furthermore, the highest hulling efficiency, nib
recovery, and large nib recovery of 96.84%, 96.98%, and
86.75%, respectively, were obtained from tangential velocity
of 40m/s and volume of air of 41m3/hr. The lowest power
consumption and noise level were obtained from the
combination of 40m/s and 45m3/hr with 3.99kwh and
88.20dB, respectively. When considering mass production of
cacao nibs, the most desirable set up of the machine was the
combination of 40m/s tangential velocity and 41m3/hr volume
of air.
Key Words - cacao beans, cacao nibs, hulling, nib
recovery, winnowing.
I. INTRODUCTION
Mechanization across the different areas of the
value chain is necessary in gaining competitive and relative
advantage in cacao industry. Improvement of postharvest
facilities is the primary challenge in the Philippine cacao
industry given its growing demand particularly for the high
value added products like cacao nibs, tablea, cacao powder
and cacao butter. On cacao production and management,
majority of local farmers and cacao traders do not go beyond
selling beans in the market due to high investments incurred
in the hulling operation of cocoa beans. Cacao farmers and
traders are still using the conventional way of cracking and
winnowing roasted cacao beans to produce nibs. Cracking the
cacao beans is carried out to produce nibs using a roller pin
that is mostly used in bakeries. Meanwhile, winnowing
cracked cacao beans is accomplished by tossing the beans and
shells up into the air on a windy day using a winnowing
basket also known as “bilao” which takes an hour to finish
three to five kilograms of roasted cacao beans. As a result,
most local farmers and traders are not able to produce enough
nibs to supply the local chocolate manufacturers. Existing
machines in the Philippine cacao industry employ cracking
operation only wherein winnowing is done separately using
conventional way. Moreover, existing design of cacao hulling
machines uses multiple contacts to the beans including roller,
screw/auger, serrated cone, and hammer types that cause more
breakage thus developing more dust and smaller particles in
the operation which decreases the recovery of large nibs with
above 3mm particle size. Hence, the availability of a food
grade compliant, efficient impact-type cacao hulling machine
may address the aforementioned problem.
II. METHODOLOGY
A. Preparation of Samples
Locally available dried cacao beans were used as
samples for the testing and evaluation of hulling machine.
Three kilograms of cacao beans in 27 trials was used in final
testing the machine. A food grade material was used to roast
the cacao beans until the beans started to crack (a cracking
sound is usually heard) and the outer shell became darker in
color. Moisture content of the roasted cacao bean was
recorded and not longer an hour to be fed into the impact-type
cacao bean huller due to cooling that may also lead to
absorbing moisture.
B. Fabrication of the Machine
The proposed mechanical cacao bean huller was
fabricated by a certified agricultural machinery manufacturer
in accordance to the approved drawings and design
specifications. Locally available materials were used in the
fabrication of the machine. The machine underwent
preliminary testing to determine if the machine worked
according to the design purpose and if it satisfied the set
criteria and to assess the working conditions and possible
operation faults of each component. Visual observations were
made to assess the safety, proper alignment, and actual
clearances that may affect the operation of the machine. The
final testing was conducted after the machine was desirable
for its intended function and ready for performance
evaluation. Running-in test was also conducted on an area
with stable and level ground and adequate working space.
C. Data Collection
During the test and evaluation, the following data
were gathered for analysis and determination of the
performance parameters of the machine namely; moisture
content, initial and final weight, weight of large nib, duration
of test, noise level, speed of rotating components, air velocity,
and power consumption.
D. Statistical Analysis
The machine was evaluated based on the three
tangential velocities of the rotating cracking mechanism and
three different volumes of air in the winnowing system. It was
replicated three times. Each experimental unit used 3 kg of
roasted cacao beans. The speed of the rotating mechanism was
based on the resulting optimum speed during the preliminary
testing of the machine which was 2400rpm. The Analysis of
Variance was also computed using STAR software and
Comparison among Means was done to parameter means with
significant effects.
154
 III. RESULTS AND DISCUSSION 93.33%. Additionally, no significant effect was recorded in
noise level of the machine; however, tangential velocity of the
A. Description of the Machine cracking system significantly affected the power consumption
The impact-type cacao bean hulling machine is of the machine at 5% level of significance with 4.42kwh as
composed of hopper; impeller type cracking component; the highest record in the combination of 47m/s (tangential
winnowing component with aspirator and a cyclone that velocity) and 49m3 /hr(volume of air).
separates the cacao nibs to hulls and other impurities;
transmission system that consists of v-belts and pulleys which IV. CONCLUSIONS
was coupled to the main shaft powered by an electric motor;
nib and hull discharge; and the support frame. The fabricated Based on the result and observations of the study, the
impact-type cacao bean hulling machine as shown in Figure 1, following conclusions were drawn:
functions in a continuous manner wherein the material is a. The impact-type cacao bean hulling machine with
constantly fed into the hopper. combination of cracking and winnowing systems was
The hopper could carry a successfully designed based on different considerations;
capacity of 15 kgs of roasted b. The impact-type cacao bean hulling machine was
cacao beans when full. An
fabricated using locally available materials and local
impeller type cracking
chamber receives the input manufacturing technologies and functions in a continuous
material and is then passed to manner wherein the material is constantly fed into the hopper
the aspirator with a cyclone that could carry a capacity of 15 kgs of roasted cacao beans
type vacuum which separates when full;
the cracked beans to its hulls. c. The machine performed satisfactorily with highest
The cacao hulls are collected in input capacity and hulling capacity using the combination of
the cyclone while the cacao
Figure1 .Cacao Bean Huller 40m/s and 41m3/hr. Moreover, the combination of 40m/s and
nibs are then collected with a
pail supported by inclined nib outlet. The machine can be 49m3/hr obtained the highest hulling efficiency, nib recovery,
operated continuously by one person. All materials used in the and large nib recovery. Additionally, lowest power
design and fabrication were of food grade quality except the consumption and noise level of were recorded from the
support frame and cover, most of which was stainless steel combination of 40m/s and 45m3/hr. highest purity of roasted
especially components that are in direct contact of the roasted cacao nibs was recorded from the combination of 47m/s and
cacao beans. 49m3/hr. The machine delivered the most desirable
performance when set at 40 m/s and 41m3/hr.
B. Performance Evaluation
HIGHEST LOWEST
PARAMETERS ACKNOWLEDGEMENTS
VALUE VALUE
The Department of Science and Technology through
Input Capacity 195.60 kg/hr 166.39 kg/hr
Engineering Research and Development for Technology
Output Capacity 169.69 kg/hr 136.43 kg/hr
(DOST-ERDT) is highly recognized in this paper.
Hulling capacity 156.53 kg/hr 124.99 kg/hr
Hulling Efficiency 96.84%, 76.86% REFERENCES
Hulling Recovery 96.98% 45.29% [1] Adejuyigbe, S. B. & Bolaji, B. O. (2005). Design
Large Nib Recovery 86.75% 69.36% fabrication and performance evaluation of beans
Purity of Nibs 99.00% 93.33% dehuller. Journal of Science and Technology. Volume
Noise Level 91.17 dB 88.20% 25, No.1. http//:doi.org//10.4314/just.v25i1.32938
Power Consumption 4.42 kwh 3.99 kwh [2] Akinnuli, B. O., S. P. Ayodeji & Omieza, A. J. (2014).
Computer aided design for cocoa beans processing yield
Statistical analysis showed that tangential velocity of the prediction. International Journal of Applied Science and
cracking mechanism significantly affected input capacity and Technology, Vol. 4, No. 5.
hulling capacity of the machine at 5% level of significance. [3] Akinnuli, B. O., O. S. Bekunmi, & Osueke, C. O. (2015).
Moreover, the highest input capacity (195.60kg/hr) and Design concepts towards cocoa winnowing
hulling capacity (156.53kg/hr) of the machine were generated mechanization for nibs production in manufacturing
from the tangential velocities and volumes of air of 47 industries. British Journal of Applied Science and
m/sand41m3 /hr, respectively. Also, tangential velocities and Technology 8(1):35-45, 2015, Article no.
volumes of air of40m/sand 49m3 /hr garnered the lowest input BJAST.2015.184.
capacity and hulling capacities of 166.39kg/hr and http//:doi.org//10.9734/BJAST/2015/16161
124.99kg/hr, respectively. [4] Bureau of Plant Industry – Department of Agriculture
For the winnowing mechanism, statistical analysis (2017). Philippine cacao industry road map 2017-2022.
revealed that volume of air significantly affected the hulling [5] Lavarias, J. A. (2015). Optimization of mechanical cacao
efficiency, nib recovery, and large nib recovery at 5% level of shelling parameters using unroasted cocoa beans.
significance. The combination of 40m/s (tangential velocity) International Journal of Agriculture and Biosystems
and 41m3 /hr (volume of air) of the machine generated the Engineering. Vol:9, No:6, 2015.
highest hulling efficiency, nib recovery, and large nib waste.org/abstracts/29726
recoveries of 96.84%, 96.98%, and 86.75%, respectively. [6] Philippine National Standard (PNS) – Bureau of
The interaction of the two treatments significantly Agriculture and Fisheries Standards (BAFS) for
affected large nib recovery and purity of the roasted cacao Agricultural Machinery (2018). Cacao Huller
nibs. The combination of 47m/s (tangential velocity) and Specifications and Methods of Test.
49m3 /hr (volume of air) gained the highest purity of 99.00%,
while 54m/s and 41m3 /hr garnered the lowest nib purity of

155
 Development of Dry-Scale Cleaner cum Grader for
Stored Onion (Allium cepa L.)
Jamel M. Salo1, a *, Emmanuel V. Sicat2,b, Helen F. Gavino3,c and Theody B. Sayco4,d
1
Carmen, Bohol
2
Science City of Muñoz, Nueva Ecija
3
Science City of Muñoz, Nueva Ecija
4
Science City of Muñoz, Nueva Ecija
a
jamsalo565@gmail.com, bemmanuelsicat@yahoo.com, chfgavino0313@gmail.com, dtbsayco@clsu.edu.ph
*graduate student Central Luzon State University
Abstract— The study was conducted to develop a dry-
scale cleaner cum grader for stored onion bulbs. The machine was
tested by using three levels of speed in the cleaning and grading
assembly (e.g. 53rpm & 13rpm, 61rpm & 15rpm and 71rpm
&17rpm) as factor. Based on the results, the most desirable machine
set up was established at cleaning speed of 71rpm and grading speed
of 17rpm. Evaluated performance parameters was cleaning capacity
(535.28kg/hr), grading capacity (556.53kg/hr), overall capacity
(493.29kg/hr), cleaning efficiency (88.50%), grading efficiency
(82.20%), overall efficiency (72.71%), mechanical damage (2.09%),
power consumption (14.41Wh), and noise level (82.51dB). The total
machine cost was ₱125,000.00 with an operating cost of ₱0.40/kg.
It will break-even at 33,262.16kg at a custom rate of ₱1/kg in a year.
Based on the results, the machine can lessen the manual labour
utilization from 0.03 man-day/kg to 0.0008 man-machine-day/kg.
Keywords— dry-scale cleaner cum grader, stored onion
bulbs, desirable machine set up, performance parameters
I. INTRODUCTION
Onion (Red Creole and Yellow Granex) is widely
cultivated in the Philippines with an earnings of P5.5B in which
Central Luzon is the major producer [1]. However, farmers have a
difficult time competing to imported onions in the market [2]. As a
conventional practice, onion bulbs are stored to preserve its quality
and acquire higher price during off seasons [3]. However, it requires
cleaning of excess dry-scale and grading base on its size to increase
its marketability [4]. At present, manual cleaning and grading are
still traditionally practiced. This processes requires an average labor
utilization of 0.03man-day/kg [5]. Resulting to low output capacity,
inconsistency and non-availability of labors during peak seasons [6].
The study can help simplify and hasten the onion cleaning and
grading. It can also help to improve the uniformity, adds value,
lessen the man-day requirement and cost of operation. Hence, there
is a need to develop a machine that can clean by removing the excess
dry scales after storage and efficiently grade them according to size.
II. METHODOLOGY
A. Design and Fabrication of the Machine
The machine was designed to perform cleaning by
removing excess dry scales and grade red onion bulbs into small (15
to 30 mm), medium (31 to 50 mm) and large (51 to 80 mm) [7]. It
was composed of five major components namely: hopper, cleaning
assembly, grading assembly, frame, and transmission assembly.
Materials was selected based on its cost and availability in the local
market. Each sizes was designed based on the allowable stress and
considering the factor of safety of the operation to ensure no failure
in each elements. A detailed drawing was used as a guide in the
fabrication of the machine at an accredited local fabricator in San
Rafael Village, Navotas City. Modification on the major parts was
done until its functionality and design considerations was achieved.
B. Performance Evaluation
Samples used were un-cleaned and ungraded onion bulbs
with visible loose dry scales due to curing and storing. Each sample
(10kg each) was randomly selected considering that no rotten bulbs
were included. Machine performance was evaluated based on 3
levels of speed combinations in the cleaning and grading assembly
(e.g. 53rpm & 13rpm; 61rpm & 15rpm; and 71 rpm & 17rpm) and
was replicated 3 times. Raw data such as initial weight, operating
time, impurities left and removed, weight in each outlet chutes,
weight of correct and correctly graded, weight of mechanically
damage, power consumption and noise emitted was collected to
determine and analyze the performance parameters such as capacity,
efficiency, mechanically damage, power consumption and noise
level. The economic performance such as unit cost of operation,
break-even cost, and payback period were also determined.
C. Performance Analysis
Performance evaluation were set up according to
Complete Randomized Design (CRD). Analysis of Variance
(ANOVA) was used to indicate statistical significance among
treatments and Post- hoc test using Fisher’s Least Significant
Difference (LSD) was conducted to test which speed variations
shows significant results. The establishment of the desirable
machine set up was based on simple rating system. The machine’s
performance in three speed variations were rated in a scale of 1 to 3,
1 being the highest and 3 being the lowest based on the results of the
statistical analysis. The ratings were averaged and the result that is
closer to 1 was the desirable set up of the machine[8].
III. RESULTS AND DISCUSSION
A. Machine Description and Principle of Operation
The machine was composed of five major components
namely: hopper, cleaning unit, grading unit, frame, and transmission
assembly powered by a 0.5hp electric motor. The loading of the
uncleaned and ungraded onion bulbs takes place in the hopper. The
cleaning unit, composed of six rollers, was responsible in removing
excess dry-scale and remained one to two dry-scales. Each rollers
rotates in opposite direction using spur gears. After cleaning, onions
are conveyed to the roller-type grading assembly for size grading.
156
 The total cost of fabricating the machine was ₱125,
000.00. The unit cost of operating the machine was ₱196.22/hr or
₱0.40/kg. As shown in Fig. 2, the machine needs to clean and grade
33,262.16kg/yr to break-even all the costs in 1 year at a custom rate
of ₱1.0/kg in order to consider ownership otherwise it will be
expensive of use. Furthermore, the use of the machine requires 3
persons as operators which lessen the utilization of manual labor
from 0.03 man-day/kg to 0.0008 man-machine-day/kg.

Fig. 1. Dry-scale cleaner cum grader for stored onions
B. Performance of the Machine
Statistically, the machine performance was affected
significantly by speed in terms of cleaning capacity, grading
capacity, overall capacity, and power consumption at 1% level of
confidence. Results shows that in terms of cleaning capacity,
grading capacity, and overall capacity, speed in T3 was significantly
higher among T1 and T2. While T1 and T2 were not significantly
different. On the other hand, power consumption reveals that T3 was
significantly lower than T1 and T2. The mean power consumption
when using speed combination of T1 and T2 were not significantly
different. Similarly, the means of the percent mechanical damage as
influence by different speed variations showed significant difference
at 5% level of confidence. Results shows that speed combination of
T1 was significantly lower compared to T3. On the other hand,
means between T1 and T2 as well as between T2 and T3 were not
significantly different. Furthermore, in terms of cleaning efficiency,
grading efficiency, overall efficiency, and noise level, no significant
difference was observed in the speed variations on the performance
of the machine. Results revealed that satisfactory machine
performance was expected when the speed combination were 71rpm
and 17rpm.
IV. CONCLUSION
Based on the results, it can be concluded that:
1. a dry-scale cleaner cum grader for stored red onion bulbs
was successfully developed based on the considerations
and was fabricated using locally available materials;
2. the machine showed satisfactory performance in cleaning
capacity, grading capacity, overall capacity, cleaning
efficiency, grading efficiency, overall efficiency, percent
mechanical damage, power consumption, and noise level
when set up to 71rpm in cleaning and 17rpm in grading;
and,
3. ownership of the machine should be by group of farmers
and an individual ownership can also be an option,
however, it should be subjected for custom service.
ACKNOWLEDGEMENT
The researcher wants to express his sincere gratitude to his
advisory committee Emmanuel V. Sicat Ph.D., Helen F. Gavino
Ph.D., and Theody B. Sayco Ph.D. for their support and guidance
that leads to the completion of this study. Also, to the Department of
Science and Technology-Engineering Research and Development
for Technology for the scholarship granted to the researcher.

TABLE I. MACHINE PERFORMANCE AS INFLUENCE REFERENCES

BY DIFFERENT SPEED VARIATIONS
[1] PSA “Selected Statistics on Agriculture 2018,” Philipp.
T1,rpm T2, rpm T3,rpm Stat. Auth., 2018.
Machine Parameters [2] BPI “Onion Production Guide,” Bur. Plant Ind., pp.1–6,
53 & 13 61 & 15 71 &17
2013.
Cleaning capacity, kg/hr 427.52 b 458.83 b 535.28 a
[3] D. Del Carmen, A.M. Espigol, P. Nuevo and G.
Grading capacity, kg/hr 458.86 b 489.16 b 556.53 a Masilungan “Morphological and Physico-Chemical
Overall capacity, kg/hr 401.72 b 430.08 b 493.30 a Characteristics of “Red Creole” Allium cepa L. in Three
Cleaning efficiency, % 80.84 87.16 88.50 Production Areas in the Philippines,” Crop Prot. Newsl.,
Grading efficiency, % 87.46 85.79 82.20 vol.41, pp.13–9, 2016.
[4] G. Calica and Z.L.Cabanayan, “Assessment of the
Overall efficiency, % 70.70 74.77 72.75
postharvest systems and losses of bulb onions in Nueva
Mechanical damage, & 1.23 b 1.60 ab 2.09 a Ecilja, Phillipines,” Asian J. Postharvest Mech., vol.1, pp.
Noise level, dB 78.87 80.96 82.51 47–55, 2018.
Power consumption, Wh 17.17 a 16.28 a 14.41 b [5] P. Castillo, C. Domingo, R. Gutierrez and R. Paz, “Pilot
Testing of BPRE Quick Sorter,” Department of
Means with the same letter are not significantly different Agriculture:Bureau of Postharvest Research and
. Extension, 2009.
[6] D. S. Narvankar, S. K. Jha and A. Singh, “Development of
Rotating Screen Grader for Selected Orchard Crops,” J.
Agric. Eng. vol.42, pp. 60–4, 2005.
[7] PNS 14:2004, “Fresh Vegetables- Common onion (Allium
cepa L. cv. Group Common Onion) and shallot (Allium
cepa Aggregatum),” Specifications Philipp. Natl. Stand.,
2015.
[8] R. K. B. Gallegos, “Design , fabrication and performance
evaluation of a small capacity mungbean sheller,”
University of the Philippines, 2010.

Fig. 2. Cost curve of using the machine

157
 Semiconductor,
Materials and
Electronics
Session Chair : Dr. Jonathan W.L. Salvacion
Moderator : Engr. Ericson D. Dimaunahan
Panelists : Dr. Michael C. Pacis
Dr. Jennifer C. Dela Cruz
Engr. Febus Reidj G. Cruz
 i-LAB-u: Internet-connected Laboratory Units with
Over-The-Air (OTA) Programming Capability
Alexander C. Abad
ECE Department, John Gokongwei Jr. College of Engineering,
De La Salle University, Manila
alexander.abad@dlsu.edu.ph
Abstract—This 2-page abstract provides a step-by-step
procedures on how to set-up a teleoperated electronics laboratory
with Over-The-Air (OTA) programming capability using
Raspberry Pi Zero with camera, and Particle devices such as
Photon and Argon. This study might help in the online learning
scheme as we battle against the covid19 pandemic by providing
students a hands-on experience on programming remote
microcontrollers to control different devices such as servos and
LEDs. This will motivate students in learning the basics of OTA
programming and Internet-of-Things (IoT). This study can be
applied to telepresence, remote sensing and monitoring as well as
in telerobotics applications.
Keywords— Over-The-Air (OTA) programming, Particle
devices, Raspberry Pi, Internet-of-Things (IoT)
I. INTRODUCTION
Due to covid19 pandemic, most schools were forced to
go online learning or distance learning. Unfortunately, some
laboratory experiments need actual demonstration or hands-on
experience. This study aims to help mitigate the hardships in online
teaching and online learning by providing step-by-step process on
how to set up a teleoperated electronics laboratory with over-the-
air programming capability. The setup in this paper is known as i-
LAB-u for internet-connected LABoratory units. This setup can be
used not only for online classes during covid19 pandemic but can
also be as a teaching and learning platform for Over-The-Air
(OTA) programming and Internet-of-Things (IoT) in the new
normal. The setup can be teleoperated anywhere in the world.
II. METHODOLOGY
A. The Setup
The top view and front view of the teleoperated
electronics lab is shown in Fig. 1. It uses Particle devices Photon
[1] and Argon [2] with an online camera feedback from Raspberry
Pi Zero [3] with Raspberry Zero Pi Mini Camera [4].
Fig. 1. i-LAB-u setup.
More detatils about the setup will be discussed in the following
sections.
B. Hardware
The hardware component is based on Particle devices. I
used Photon and Argon because they have WiFi capability and can
be programmed anywhere through the use of Web Integrated
Development Environment (IDE) [5]. To demonstrate some output,
2-degrees of freedom (dof) mini-robot made from 2 mini servos
(SG90) is connected directly to Photon1. A tutorial with codes on
how to control a servo using Particle Photon can be found in these
links [6,7]. Moreover, to demonstrate that we can control more
than 2 servos, I used PCA9685 (16-Channel 12-bit PWM/Servo
Driver - I2C interface) [8] connected to Photon2 to drive a 4-dof
mini robotic arm made from 4 mini servos and popsicle sticks.
There are many video tutorials on how to create a 4-dof mini
robotic arm such as this one [9]. Discussion on how to integrate
Photon device with PCA9685 can be found here [10].
To demonstrate visual output, I used 8x8 SPI LED
MATRIX with MAX7219 module to Particle Argon device and
uploaded the code from this link [11]. The text messages scrolling
can be changed through OTA programming using Web IDE [5].
Aside from servos and LEDs, sensors such as temperature and
humidity sensor can be connected to Particle devices and their
readings can be read remotely like in this link [12]. In order to have
a real-time video feedback to see what is happening on the remote
setup, I used Raspberry Pi Zero with camera. The total cost of my
setup is around P7,000.00. All the devices can be purchased locally
or online. A rough estimate cost of my setup in PHP:
Particle Photon = P 1,200 per unit
Particle Argon = P 2,000 per unit
Raspberry Pi Zero = P 850 per unit
Raspberry Zero Pi Mini Camera = P 700 per unit
Matrix LED 8x8 Display module MAX7219 = P 150
Tower Pro SG90 9g Gear Steering Servo (6PCS) = P 350
PCA9685 = P 200
Breadboard and jumper wires = P 100
C. Software
My setup uses UNIX for the Raspberry Pi, C++, HTML,
and CSS for the Particle Devices. This paper provides links to
many tutorials to help the reader replicate the setup.
There are so many step-by-step tutorials on how to set up
a Raspberry Pi Zero such as the video tutorial in this link [13] and
this link [14]. After you set up the Raspberry Pi Zero with a
camera, you can make the video feed appear in a browser by
following the instructions in this link [15]. You need to port
forward your local IP network so that you can access or share to
others your camera feed anywhere. Aside from port forwarding,
you can use tunneling through ngrok [16] by signing up a free
account. I used ngrok to stream the video of my remote setup.
159
 The Particle devices can be programmed like Arduino teaching and learning platform for Over-The-Air (OTA)
using C++ language. Through the use of Web IDE, we can programming and Internet-of-Things (IoT) for the new normal.
program Particle Photon and Argon anywhere. Using HTML and This study can be applied to telepresence, remote sensing and
CSS files like in this link [17], we can create remote control button monitoring as well as in telerobotics applications.
in integrate in a webpage.
Step-by-step procedures on how to setup Particle Photon ACKNOWLEDGEMENT
and Argon can be found here [18] and here [19] respectively.
I would like to thank DOST-ERDT for helping me in my studies.
III. RESULTS AND DISCUSSION
REFERENCES
[1] PARTICLE PHOTON: WI-FI, https://docs.particle.io/photon/
If the tutorial in [14] has been successfully executed, the
(accessed on October 22, 2020)
live video feed from the Raspberry Pi Zero cam can be viewed on
[2] PARTICLE ARGON: WI-FI + BLUETOOTH,
your internet browser connected to your local network as shown in
https://docs.particle.io/argon/ (accessed on October 22, 2020)
Fig. 2. Using ngrok tunneling, you can get public IP and the live
[3] Raspberry Pi Zero,
video feed can be access beyond your local network.
https://www.raspberrypi.org/products/raspberry-pi-
zero/?resellerType=home (accessed on October 22, 2020)
[4] Raspberry Zero Pi Mini Camera,
https://www.lazada.com.ph/products/raspberry-zero-pi-mini-
camera-i207843638-
s263818843.html?spm=a2o4l.searchlist.list.4.2c642c1djDYR
gB&search=1 (accessed on October 22, 2020)
[5] FLASH APPS WITH THE PARTICLE WEB IDE,
https://docs.particle.io/tutorials/developer-tools/build/
(accessed on October 22, 2020)
[6] MAKER KIT TUTORIALS,
Fig. 2. Live video feed from Raspberry Pi Zero Camera. https://docs.particle.io/tutorials/hardware-projects/maker-kit/
(accessed on October 22, 2020)
If the tunneling using ngrok is successful, you will get a link [7] Tutorial: Spark Variable and Function on One Web Page,
similar to this one: http://0.tcp.ngrok.io:13175 that you can use or https://community.particle.io/t/tutorial-spark-variable-and-
share to others for them to see your setup. The tunnel address function-on-one-web-page/4181 (accessed on October 22,
changes whenever you close the Raspberry Pi terminal or when 2020)
you restart it. The live video feed can be embedded inside an [8] Adafruit PCA9685 16-Channel Servo Driver,
HTML by putting these lines inside the body of your HTML code: https://learn.adafruit.com/16-channel-pwm-servo-driver
<center><iframe width="550" height="365" (accessed on October 22, 2020)
src="http://0.tcp.ngrok.io:13175” [9] How to make Micro Servo Robotic arm arduino based simple
</iframe> DIY, https://www.youtube.com/watch?v=MCYJAyIXBdA
(accessed on October 22, 2020)
The live video feed can be combined to the control buttons as [10] Photon + PCA9685 Library,
shown in Fig. 3. https://community.particle.io/t/photon-pca9685-library/22794
(accessed on October 22, 2020)
[11] LedControl-MAX7219-
MAX7221/firmware/examples/matrixpaneldemo.ino,
https://github.com/chrisloris/LedControl-MAX7219-
MAX7221/blob/master/firmware/examples/matrixpaneldemo.
ino (accessed on October 22, 2020)
[12] IOT Particle Temperature Sensor,
https://www.hackster.io/93944/iot-particle-temperature-
sensor-9820b2 (accessed on October 22, 2020)
[13] How to Setup Raspberry Pi Zero W for Headless
https://www.youtube.com/watch?v=LlCr09B2HZI (accessed
on October 22, 2020)
[14] RASPBERRY PI ZERO MODEL W TUTORIAL 2:
MAKING A WIRELESS PORTABLE IP CAMERA
https://toptechboy.com/raspberry-pi-zero-model-w-tutorial-2-
making-a-wireless-portable-ip-camera/ (accessed on October
Fig. 3. Live video feed embedded in a webpage together with 22, 2020)
control buttons. [15] RPi-Cam-Web-Interface https://elinux.org/RPi-Cam-Web-
Interface (accessed on October 22, 2020)
[16] ngrok. https://ngrok.com/ (accessed on October 22, 2020)
III. CONCLUSION [17] Light a IOT LED from web with Particle Photon
https://knsakib.com/2015/12/28/light-a-led-over-the-net-with-
This paper presented step-by-step procedures including
particle/ (accessed on October 22, 2020)
clickable link references on how to setup a teleoperated electronics
[18] Quick start: Photon, https://docs.particle.io/quickstart/photon/
laboratory using Raspberry Pi and Particle devices with over-the-
(accessed on October 22, 2020)
air programming capability. This setup can be used not only for
[19] Quick start: Argon, https://docs.particle.io/quickstart/argon/
online classes during covid19 pandemic but can also be as a
(accessed on October 22, 2020)

160
 Design and Development of a Neural Network-based
Coconut Maturity Detector using Sound Signatures
Nemilyn A. Fadchar1, a *, Jennifer C. Dela Cruz2,b
1
Mapua University
2
Mapua University
a
nemilyn_ece@yahoo.com, bjcdelacruz@mapua.edu.ph
Abstract— Fruit maturity detection is one of the factors
that influence the quality of post-harvest products. One of the known
practices in fruit maturity detection particularly in coconut is by
manual tapping or knocking of the surface of the coconut fruit and
listening to its sound. Hence, this practice tends to be subjective and
requires skills that is earned through years of experience.
Modernizing the postharvest process using low-cost non-destructive
approaches such as acoustics is a promising frontier to uplift the
conventional farming practices. Moreover, implementation of
machine learning tools could further increase the accuracy of such
process. This study attempts to develop a prototype that uses the
sound signatures to classify the maturity level of the young coconut
by using neural network. Specifically, it aims to: (1) extract the
acoustic features from the sound signal; (2) train the data set and
develop the classification model using neural network; (3) design
and develop the prototype; and (4) test and evaluate the prototype.
The major parts of the prototype were composed of the vibration
motor, vibration sensor, rpi 3b+ microcontroller, battery and LCD.
Result of the neural network model obtained an over-all
classification accuracy of 91.3 %. This implied that the neural
network model has a high prediction rate to accurately determine the
maturity level of the coconut using the sound signatures. Lastly,
final evaluation results showed that the prototype has a higher
percentage of prediction accuracy as compared to the manual
process.
Keywords— coconut maturity, sound signatures, machine
learning, neural network, raspberry pi
I. INTRODUCTION
Coconut postharvest industry in the Philippines
tremendously improved as new technologies were introduced.
Coconut which is scientifically known as cocos nucifera L., is one
of the most valuable crops in the Philippines because different
products could be derived from this tree particularly in its fruit. The
quality of these coconut-based products was maintained through
proper selection of coconut fruit according to its maturity stage. One
of the essential factors in postharvest industry is maturity detection
and classification. Maturity stage of a fruit or vegetable is said to be
directly related to the product quality.. A coconut fruit must be
harvested on its proper maturity stage in order to serve its intended
use as well as maintain its quality. According to [1], an interview
with some of the coconut farmers from the Philippines there were
actually three known maturity stage of young coconut namely:
Malauhog, Malakanin and Malakatad and these classifications were
according to meat thickness. The mucus-like thickness which is thin
and has relatively transparent color is called Malauhog while the
second maturity stage was the cooked rice-like, has thicker and solid
meat which is called Malakanin. The third maturity stage which is
named to be Malakatad has the thickest coconut meat, leather-like
and have stiff to chew quality. Local farmers identify the maturity
stage of coconut by tapping the fruits of “buko” and listening to its
sound. Hence, this practice introduced bias judgement and requires
several years of experience or practice. Since, quality is directly
associated with maturity it is therefore important to establish
systematic procedure to determine the maturity stage. Therefore,
predicting the maturity stage of coconut fruit using sound signatures
could impact the post-harvest standards in the agricultural sector.
Moreover, implementation of machine learning techniques could
establish models that could be used to develop platforms that will
improve the traditional manual procedures of the coconut farmers.
This could also facilitate the industrial sorting system as well as
maintain the quality of the end products.
According to [2] acoustic properties has been widely used
to assess the quality of agricultural products. It plays an important
role in determining different attributes related to maturity in a non-
destructive approach. Moreover, the adaptation of this technique in
the post-harvest industry uplift the quality of products and sustain
consumer demands with high quality which is beneficial to the
farmers. Several attempts to describe the association between
properties and maturity of some fruits has also been explored in
different studies. Acoustic signals have been used to classify
maturity of different fruits such as coconut [1], [3], watermelon [4],
mango [5], cocoa [6], pomelo [7], melon [8] and tomato [9]. Several
studies also utilized the acoustic response to non-destructively
determine the internal quality and firmness of different fruits such
as apple [10], pear [11], pineapple [12], [13] and kiwi [14]. The
initial study on coconut maturity detection in the Philippines was
pioneered by the study of [1] to establish a scientific method of
determining the maturity level of a coconut fruit. In this study, the
young coconut fruit or “buko” were tapped manually by the farmer
and the sound were collected using a microphone.
The study of [1] use the knocking trick at the coconut shell
which is the common method of determining the ripeness. The study
utilized an improvised sound box as shown in Figure 1 to collect
sounds from tapping. A microphone was installed inside the sound
box and is directly connected to a tape recorder. The analog sound
collected was converted using and analog-to- digital converter
(ADC) and was later converted into its frequency domain using the
Fast Fourier Transform (FFT). Results revealed that the pitch of
Malauhog was different from Malakanin and Malakatad which
indicates that using the pitch, the Malauhog can be identified. In
order to further classify the young coconut, the average power was
also used. Results of the experiments done by [1] indicated that
Malakatad has the highest average power and a finer extent of main
spectral component while Malauhog has obtained the lowest average
power which is supported by its low pitch. This characteristic was
used to differentiate the Malakatad and Malauhog from Malakanin.
Further enhancement were done by the succeeding studies such as
determination of physical, acoustical, mechanical and chemical
properties of fresh and young coconut for maturity separation [3],
the automated grading of coconut using audioception [15], coconut
maturity classification using fuzzy logic [16] and the
microcontroller-based coconut maturity prediction [17]. In the
study of [15], the proposed methodology uses the concept of
audioception to term the usability of a coconut, using the sound
produced. A coconut can be segregated into the good or bad
category. The procedure starts by dehusking the coconuts of
standard size and shape, then dropping it in a granite slab from a
height of one foot within a noise free environment that has been set-
161
up. The sound produced on impact were recorded. The Fast Fourier
Transform (FFT) acts as the base for all signal related feature
extraction. FFT was used to convert a signal in wave form into
spectral information from which most distinguished features can be
extracted using Fourier domain. Features were extracted and
classified using the following supervised learning methods namely:
SMO, Dagging and Naïve Bayes according to their pre-defined
class. Results shows that sound signals were able to establish a
model for coconut grading which segregates the good and bad
coconut. Likewise, in the experiment conducted by [16], image and
sound were used to classify the coconut into three (3) stages namely:
Malauhog, Malakanin and Malakatad.
The two input parameters in the fuzzy interference system
were obtained using the image analysis and sound spectral analysis
as shown in the conceptual framework in Figure 2. The classification
system used image analysis and sound spectral analysis prior having
these parameters connected into a fuzzy inference system. The fuzzy
logic classifier will then process the information as per its
knowledge base and come up with a decision which corresponds to
a maturity stage. The output of the system is any of the coconut
maturity stage: malauhog, malakanin, and malakatad. Based on the
simulation results, the fuzzy logic classifier was able to determine
the maturity stage based of the coconut fruit based on color and
sound. Meanwhile, the effects of maturity, fruit surface and hitting
point on natural frequency were examined by [3] using the acoustic
property tester. Each sample was applied with an impact force to the
sample’s equator and the sound signal was recorded through a
microphone mounted a few millimeters above the sample surface
180° around the perimeter from the point of impact. Results revealed
that the greatest natural frequency difference between the immature
and the mature stages was 76 Hz and that between the mature and
the over-mature stages was observed to be 126 Hz. The natural
frequencies observed for impact at a certain point for the immature,
mature, and the over-mature stages were 870, 946, and 1072 Hz,
respectively. The recent study of [17], developed a microcontroller-
based device for coconut maturity prediction using the Naïve Bayes
method. The device is comprised of a noise sensor which is
connected to a stethoscope diaphragm for collecting the sound signal
before it is processed in the microcontroller. A digital filter was used
to reduce the noise and determine the average frequency spread of
sound using the Fast Fourier Transform (FFT) created by knocking
the coconut using the improvised knocking tool. The device was also
equipped with an SD card to record the knocking sound frequency.
In classifying the coconut maturity, the inputs were three (3) average
frequency values of knocking sound and the size were fed to the
Naïve Bayes algorithm in order to perform the prediction. Results
revealed that the device was capable of classifying the coconut fruit
into three (3) stages namely: young, fairly mature and mature. The
system success rate of 80% was obtained from a total of twenty (20)
test sample and twenty-four (24) training data. Previous literatures
shows that classifying ripeness [6], maturity [7] and [4], firmness
[11] and internal quality [12] uses supervised learning tools in order
to develop a classification model or establish the regression model
based on the acquired data set. The study conducted by [6], uses the
Support Vector Machine (SVM) and K-nearest Neighbor (KNN)
classifier in training the 392 data samples. Eleven (11) feature
vectors with known class labels were used to classify ripe and unripe
cocoa pods. Results revealed that SVM has a higher accuracy
percentage in predicting the ripeness classification as compared to
KNN. Likewise, the research conducted by [4] shows that the
collected acoustic response from watermelons samples were pre-
processed to extract the acoustic features that were used to develop
the a non-destructive approach to develop classification system for
determining watermelon maturity. Results revealed that the neural
network model has a higher accuracy in classifying the watermelon
ripeness as compared with the expert’s prediction.
Traditional practice of determining the maturity of several
fruits including coconut is usually done manually. Hence, this
results in human errors, subjective judgement, productivity issues
and other several factors that limits the production outputs. As result
of, a growing demand for automated platforms that replace the
traditional method of determining the maturity of several post-
harvest commodity specially coconut is a challenge to the post-
harvest industry. Modernizing the post-harvest process using low-
cost non-destructive approaches such as acoustics is a promising
frontier to elevate the conventional farming practices as well
incorporating machine learning tools to further enhance the accuracy
of the system.
The general objective of this study was to design and
develop a neural network-based coconut maturity detector using
sound signature. Specifically, it aims to: (1) extract the acoustic
features from the sound signal; (2) train the data set and develop the
classification model using neural network; (3) design and develop
the prototype; and (4) test and evaluate the protype.
Threats in the global food security is one of the
consequences of low standards in maintaining the quality of post-
harvest commodities. Since maturity is likely to dictate the quality
of the harvest, it is important that a device that surmount the
limitations of the traditional manual practice be developed to
contribute in uplifting the farming standards. Likewise, this attempt
paves the way in applying non-destructive methods such as
acoustics in determining the maturity level of other post-harvest
fruits. Moreover, this study could reveal the potential of using
machine leaning tools increasing the accuracy of automated
systems. Lastly, other protypes for specific could also be developed
as well as to deploy the predictive models in automated platforms to
enhance the accuracy of existing systems.
This study covers the design and development of a
prototype for predicting the maturity stage of young coconut. It was
focused on extracting the acoustic vibration, pre-processing the
sound signal to obtain the acoustic frequency using the FFT, and
training the dataset using neural network to create a classification
model in order to determine the maturity stage of young coconuts.
The maturity stage of young coconut fruit considered were namely
Malauhog, Malakanin, and Malakatad. The basic components that
were used to develop the prototype were the following: (1) vibration
motor and vibration sensor for the sound signal acquisition; (2) the
raspberry pi 3b+ microcontroller was used in order to perform
classification using the neural network model; (3) an LCD
component to display the reading of the maturity phases read by the
system that is “Malauhog”, “Malakanin”, and “Malakatad”; (4) a
rechargeable battery for design portability. Python programming
and tensor flow was used to develop a program for the
microcontroller. However, there were some limitation in the study
such other parameters or factors affecting the age of coconuts; the
detection of maturity level of a fully-grown coconut fruit, also
termed as Niyog; the identification between usable and rotten
coconuts; and the effect of storage duration after harvesting the
coconut fruit.
II. METHODOLOGY
This section explains the systematic procedure employed in the
design and development of the coconut maturity detector protype.
Figure 4 shows the major procedures involve in the conduct of the
study namely: sound signature acquisition, neural network
modelling, program development, hardware fabrication and lastly
the implementation which includes the testing and evaluation of the
prototype.
162
 is done layer-by-layer. Each layer was generated by a neuron
and pass on to the next and so on.

Figure 1. Block diagram of the study.

A. Sound Signatures Acquisition
Sound signatures were obtained from a total of 300
young coconut fruits shown in which is composed of 100
samples from each maturity stage. According to [1], there
were three (3) known classification of buko in the Philippines
referred to as Malauhog, Malakanin and Malakatad. The
Face 2
sound signatures were gathered by introducing a vibration
using a vibration motor in the three (3) faces of the coconut
fruit twice which makes six (6) acoustic features in each
coconut sample.
A vibration sensor as was interfaced with an
Arduino and laptop to collect and record the sound signal
from the coconut fruit as shown in Figure 2.
Figure 2. Sound signature acquisition set-up.
Figure 3. Neural network architecture.
According to[18], the rule for learning algorithm is based
on gradient-descent which means that the error is computed
after a forward computation is reduced after each iteration.
The Matlab R2019b was used to train the neural network
using the pattern recognition app as the classifier. There were,
four (4) major steps in building the ANN model [19]. First is
the selection of possible inputs and outputs. After the data
were prepared, it is pre-process in order to eliminate the
outliers. Next, it is divided into training, testing and
validation data sets. After this, the network architecture is
selected and training process starts. Lastly, the performance
evaluation of the trained network is done in order to
determine if there is a need for retraining or additional data
may be required to increase the model accuracy. The network
model generated in the neural network pattern recognition
and classification app was used in developing the final
program for the prototype.
C. Program Development
Phyton programming language was used in developing the
program codes because it is an open source language and has
simple syntax. Figure 4 illustrates the program flowchart that
was used in developing the software for the prototype.

The collected signal was pre-processed by transforming the

time-domain into frequency domain using the Fast Fourier
Transform (FFT) to determine the average spread of the
frequency spectrum of each sample. Data set were tabulated
according to their predetermined class labels for training.
B. Neural Network Modelling
Neural network is one the prevalent supervised learning
technique and a dominant tool in predictive modelling. This
is because this algorithm has the capability to perform data
training similarly to the human brains. Like the human brain,
a neural network is basically composed of the so-called
neurons which represents the mathematical function that
facilitates learning process. Figure 3 shows the two-layer
feed forward network with 10 neurons using the
backpropagation algorithm that was used to develop the
classification model in this study. The process of data training Figure 4. Program flowchart

163
 Upon turning the device on and pressing the scan button, III. RESULTS AND DISCUSSION
the program starts to read the starts to capture the vibration
signal in the tapping, then it will proceed to pre-processing by A. Sound Signatures
obtaining the frequency information though FFT. Hence, if Based on the results the following acoustic frequency
were obtained upon applying FFT on the sound signal. Figure 11
the signal is unclear it will display “TRY AGAIN” in the shows the difference between the spectral magnitudes of Malauhog,
LCD and the user has to repeat the initial procedure. The Malakanin and Malakatad after several series of transformations. It
frequency information is entered to the network model which reveals that Malauhog has the lowest average power and then
performs the classification. But if the frequency information followed by Malakanin, while Malakatad has the highest average
is not obtained it will display “TRY AGAIN” in the LCD and power. This supports the results of [1], that explains the transition
the user has to repeat the tapping process again Finally, the from solid to hollow sound as the coconut matures from Malauhog
program will output the maturity stage and displayed the to Malakatad.
result in the LCD.
The MATLAB neural network model was translated into a
tensor flow model because there is no straightforward method
that allows a MATLAB script to be incorporated in a phyton
script. Then the python script implements the tensor flow
model and reads the data from the input and perform the
classification to produce an output.
D. Hardware Fabrication
The coconut maturity detector was composed of five (5) (a)
major components namely: vibration motor, vibration sensor,
raspberry pi 3b+ microcontroller, rechargeable battery and
the LCD. The vibration motor was used to serve as the
constant source that initiates vibration to the surface of the
coconut fruit while the vibration sensor was responsible for
detecting the sound signal. The Raspberry Pi 3 model B+ was
used because it is capable to perform the machine learning
function that will use the classification model in order to
determine the maturity stage of the coconut fruit and makes
the device stand-alone. A rechargeable LiPo battery was used (b)
to make the device independent from using the AC power
source during its operation for portability purposes. Finally,
an LCD was used to display the maturity stage of the coconut
fruit.
E. Device Calibration
The coconut maturity detector was calibrated to ensure the
accuracy and reliability of the vibration motor and vibration
sensor used in the prototype. The DT2234C non-contact
high/low speed digital tachometer for small motors was used
to determine the motor speed accuracy as well as the
measurement reliability of the vibration sensor. The effect of (c)
distance between the vibration motor and vibration sensor Figure 5. Spectral magnitude of (a) Malauhog; (b) Malakanin;
(c)Malakatad
was also tested prior to the device construction to determine
the appropriate distance between the motor and the sensor. B. Neural Network Model
F. Implementation A total of 300 samples having six (6) acoustic features
each was trained using the neural network. The datasets were then
A preliminary testing was conducted prior to the final modelled through neural network algorithms on MATLAB R2019b.
evaluation of the prototype. In the preliminary testing, all There were 300x6 dataset used as input data for training and 300x3
parts and function were tested if it is working properly. dataset used as target or output data. Data was randomly divided in
Calibration and troubleshooting were also conducted to ten the following allocation: 70% training dataset, 15% validation
(10) coconut fruits that were randomly selected. Lastly, the dataset and 15% testing dataset. Output class 1 stands for Malauhog,
final evaluation was conducted from fifty (50) coconut fruit class 2 is for Malakanin and class 3 is for Malakatad. During the
training stage the following results were obtained 70 samples were
with different maturity stage. Each coconut was labelled
classified as Malauhog, 69 samples were classified as Malakanin
based on the result obtained from the prototype. To further and 53 samples were classified as Malakatad which were accounted
validate the accuracy of the device each coconut was cracked for the accuracy of percentage of 91.4% as shown in Figure 12.
open to determine if it was correctly classified by the device.

164
 Figure 7 shows the actual prototype for the coconut
maturity detector. The dimensions of the device were 15 cm x 9 cm
x 3.5 cm. The enclosure was made of a flexible plastic to reduce the
weight but maintain the durability of the device

Figure 6. Confusion Matrices

Meanwhile, the testing and validation stage obtained a
classification accuracy of 86.7% and 95.6% respectively. For the
over-all accuracy, there were 98 out 100 coconuts correctly
classified as Malauhog, 90 out of 100 coconuts correctly classified
as Malakanin and 86 out of 100 coconuts were correctly classified
as Malakatad. An over-all the classification accuracy of of 91.3%
was obtained for the neural network model. Results indicate that the
model has a high percentage of predicting the maturity stage of the
young coconut fruit accurately using the given datasets. Moreover,
this implies that the model could be implemented to further evaluate Figure 7. Coconut maturity detector prototype.
the accuracy on actual conditions. D. Implementation
A pilot testing was conducted to ten (10) randomly
C. Coconut Maturity Detector Prototype selected newly harvested young coconut fruits. Results of the
Figure 6 illustrates the schematic diagram of the circuit. initial testing were tabulated in Table I which showed that device
It shows the interfacing of the major components such as the has predicted 9 out 10 correct maturity stage. This indicate that all
vibration motor, vibration sensor, rechargeable LiPo battery, the components of the prototype were functioning well and sensor has
powerboost 1000C charger for the charging port, the on/off switch, been calibrated accordingly.
the push button for scanning, the LCD to display the result and the TABLE I
Raspberry Pi 3 model 3b+ that will perform the machine learning Summary of Predictions Done Manual vs. Prototype During
algorithm in order to classify the maturity stage of the coconut. Pilot Testing
Sample Manual Prototype Actual
No. Prediction Predictions Maturity Stage
1 Malauhog Malakanin Malakanin
2 Malakanin Malauhog Malakanin
3 Malakanin Malakanin Malakanin
4 Malakatad Malakatad Malakatad
5 Malauhog Malauhog Malauhog
6 Malauhog Malauhog Malauhog
7 Malakatad Malakatad Malakatad
8 Malakatad Malakanin Malakanin
9 Malakanin Malakanin Malakanin
10 Malauhog Malauhog Malauhog

After the pilot testing the prototype were subjected to the

final evaluation. A total of fifty (50) randomly selected coconut
fruits of unknown maturity stage were harvested. It was labelled and
then predictions were done by side by side using the manual
approach and employing the coconut maturity detector prototype.
Some of the predictions obtained were shown in Table II. Results
revealed that the predictions done manually obtained a prediction
accuracy of 85% for Malauhog, 90 % for Malakanin and 94% for
Figure 6. Schematic diagram of the coconut maturity detector circuit.
Malakatad. Meanwhile, the prototype acquired the following

165
prediction accuracy: 98% for Malauhog, 95% for Malakanin and Phyton program by using the Tensor Flow. Evaluation results
96% for Malakatad. showed that the prototype has a higher accuracy rating of 96.33% as
TABLE II compared to the manual practice with a rating of 89.67.
Some of the Predictions Done Manual vs. Prototype During This study proves that acoustic feature could be extracted
Final Evaluation from sound signatures to determine the maturity stage of young
Sample Manual Prototype Actual coconut fruit. Furthermore, it also reveals that neural network is a
No. Prediction Predictions Maturity Stage significant tool in developing classification models that could be
1 Malauhog Malakanin Malakanin implemented into automated platforms to improve the efficiency of
5 Malakanin Malauhog Malakatad the system.
10 Malakatad Malakatad Malakatad
15 Malakanin Malakanin Malakanin ACKNOWLEDGEMENT
20 Malakanin Malakanin Malakanin
25 Malakatad Malakanin Malakanin The authors would like to acknowledge the Cavite coconut
30 Malauhog Malakanin Malauhog farmers for providing the samples for the study.
35 Malauhog Malauhog Malauhog
40 Malakatad Malakatad Malakatad REFERENCES
45 Malauhog Malakanin Malakanin [1] M. M. Gatchalian, S. Y. De Leon, and T. Yano,
50 Malakanin Malakanin Malakanin “Measurement of young coconut (Cocos nucifera, L.)
maturity by sound waves,” J. Food Eng., vol. 23, no. 3, pp.
253–276, 1994.
100 [2] M. Aboonajmi, M. Jahangiri, and S. R. Hassan-Beygi, “A
Review on Application of Acoustic Analysis in Quality
Evaluation of Agro-food Products,” J. Food Process.
95 Preserv., vol. 39, no. 6, pp. 3175–3188, Dec. 2015.
[3] M. R. Mitchell et al., “Determination of Physical,
Accuracy Percentage

Acoustical, Mechanical, and Chemical Properties of Fresh

90
Manual
85
Coconut Maturity Detector
80
75
Malauhog Malakanin Malakatad
Coconut maturity stages
Figure 8. Comparison of prediction accuracy
Result in Figure 8 shows the comparison of the accuracy
of prediction conducted manually by a coconut farmer and the
device. It reveals that the device has a higher over-all accuracy rate
of 96.33 % as compared to the prediction done manually which has
an over-all rating of 89.67%.
IV. CONCLUSION
This study has successfully developed a prototype that is
capable of predicting the coconut maturity levels based on sound
signatures. The accuracy of prediction is further enhanced through
the aid of supervised machine learning tool specifically the neural
network. The three (3) commonly known maturity levels considered
in this paper were Malauhog, Malakanin and Malakatad.
Based on the results, sounds signatures that were acquired
has been used to extract the frequency information that was used as
the feature vectors in the dataset for in developing the classification
model. The neural network was able train the dataset and developed
the classification model with an over-all classification accuracy of
91.3% which indicates a high prediction accuracy rate. The
developed prototype uses the vibration motor and vibration sensor
interfaced with the Rasberry Pi model 3b+ microcontroller to
facilitate the use of neural network model in maturity classification.
The developed prototype for coconut maturity detection using sound
signatures was able to implement the classification model in the
Young Coconut Fruit for Maturity Separation,” J. Test.
Eval., 2010.
[4] Y. X. Zhang, J. L. Han, and W. Yao, “Non-destructive
watermelon maturity detection by acoustic response,” in
2nd International Conference on Information Engineering
and Computer Science - Proceedings, ICIECS 2010, 2010.
[5] A. Zakaria et al., “Improved maturity and ripeness
classifications of Magnifera Indica cv. harumanis mangoes
through sensor fusion of an electronic nose and acoustic
sensor,” Sensors (Switzerland), vol. 12, no. 5, pp. 6023–
6048, 2012.
[6] D. Z. H. Arenga and J. C. Dela Cruz, “Ripeness
classification of cocoa through acoustic sensing and
machine learning,” in HNICEM 2017 - 9th International
Conference on Humanoid, Nanotechnology, Information
Technology, Communication and Control, Environment
and Management, 2018.
[7] J. Hongwiangjan, A. Terdwongworakul, and K.
Krisanapook, “Evaluation of pomelo maturity based on
acoustic response and peel properties,” Int. J. Food Sci.
Technol., 2015.
[8] F. Khoshnam, M. Namjoo, and H. Golbakhshi, “Acoustic
testing for melon fruit ripeness evaluation during different
stages of ripening,” Agric. Conspec. Sci., vol. 80, no. 4, pp.
197–204, Dec. 2015.
[9] A. Baltazar, J. I. Aranda, and G. González-Aguilar,
“Bayesian classification of ripening stages of tomato fruit
using acoustic impact and colorimeter sensor data,”
Comput. Electron. Agric., vol. 60, no. 2, pp. 113–121,
2008.
[10] S. Landahl, P. T. Jancsók, and J. De Baerdemaeker,
“Firmness of developing apples on the tree and shortly after
harvest measured with the acoustic impulse response
technique,” in Acta Horticulturae, 2003, vol. 600, pp. 63–
70.
[11] N. De Belie, S. Schotte, J. Lammertyn, B. Nicolai, and J.
De Baerdemaeker, “Firmness changes of pear fruit before
and after harvest with the acoustic impulse response
technique,” J. Agric. Eng. Res., 2000.
[12] C. Ay, “Evaluating the internal quality of pineapple by

166
 acoustics,” in 2000 ASAE Annual Intenational Meeting,
Technical Papers: Engineering Solutions for a New
Century, 2000, vol. 2, pp. 1229–1244.
[13] S. Boonmung, B. Chomtee, and K. Kanlayasiri,
“Evaluation of artificial neural networks for pineapple
grading,” J. Texture Stud., 2006.
[14] T. Y. Yen and D. J. Yao, “Freshness detection of kiwifruit
by gas sensing array based on surface acoustic wave
technique,” in NEMS 2018 - 13th Annual IEEE
International Conference on Nano/Micro Engineered and
Molecular Systems, 2018, pp. 98–101.
[15] P. Thomas and H. B. Anita, “A novel automated method
for coconut grading based on audioception,” J. Theor. Appl.
Inf. Technol., vol. 95, no. 10, pp. 2265–2273, 2017.
[16] I. M. Javel, A. A. Bandala, R. C. Salvador, R. A. R. Bedruz,
E. P. Dadios, and R. R. P. Vicerra, “Coconut fruit maturity
classification using fuzzy logic,” in 2018 IEEE 10th
International Conference on Humanoid, Nanotechnology,
Information Technology, Communication and Control,
Environment and Management, HNICEM 2018, 2019.
[17] D. Rahmawati, H. Haryanto, and F. Sakariya, “THE
DESIGN OF COCONUT MATURITY PREDICTION
DEVICE WITH ACOUSTIC FREQUENCY
DETECTION USING NAIVE BAYES METHOD
BASED MICROCONTROLLER,” JEEMECS (Journal
Electr. Eng. Mechatron. Comput. Sci., 2019.
[18] J. W. Orillo, J. Dela Cruz, L. Agapito, P. J. Satimbre, and
I. Valenzuela, “Identification of diseases in rice plant
(oryza sativa) using back propagation Artificial Neural
Network,” in 2014 International Conference on Humanoid,
Nanotechnology, Information Technology, Communication
and Control, Environment and Management, HNICEM
2014 - 7th HNICEM 2014 Joint with 6th International
Symposium on Computational Intelligence and Intelligent
In, 2014.
[19] T. M. Amado and J. C. Dela Cruz, “Development of
Machine Learning-based Predictive Models for Air Quality
Monitoring and Characterization,” in IEEE Region 10
Annual International Conference, Proceedings/TENCON,
2019, vol. 2018-October, pp. 668–672.

167
 Transfer Learning with Convolutional Neural
Network Towards Fungi Classification in Microscopy
Images
Josh Daniel Lao Ong1, a *, Luis Gabriel Cajucom1,b, Erinn Giannice Abigan1,c, Patricia Angela R. Abu1,d and Ma.
Regina Justina E. Estuar1,e
1
Ateneo de Manila University
a
josh.ong@obf.ateneo.edu, bluis.cajucom@obf.ateneo.edu, cerinn.abigan@obf.ateneo.edu, dpabu@ateneo.edu,
e
restuar@ateneo.edu

Abstract— The Panama disease, also known as the
Fusarium wilt, is a deadly disease known to affect banana plants all
over the world. Caused by a fungal pathogen known as Fusarium
oxysporum f. sp. cubense (Foc), the disease has been a constant
threat to banana producers considering that it cannot be eradicated
once it has infected the soil. A new strain that emerged in 1989
called Tropical Race 4 (TR4) is now threatening the Cavendish
cultivar, the most popular banana variety being grown today.
Furthermore, symptoms of the disease are not visible until late stages
of the infection. While there are methods that accurately determine
the presence of Foc in a soil sample, these are costly and inaccessible
to most banana producers. Thus, we propose the use of
convolutional neural networks in the automatic detection of Foc
TR4 in soil samples with the use of microscopy. This study utilized
a dataset containing microscopy images of various fungal species
captured using three distinct microscopy techniques: brightfield,
darkfield, and fluorescent. Transfer learning has shown to make
significant improvements to the performance of the models in
classifying microscopic fungi. The best performing model was
trained exclusively on brightfield images and has achieved an
accuracy score of 93.11%.
Keywords—convolutional neural networks, transfer
learning, image classification, fusarium oxysporum, fungi,
microscopy
I. INTRODUCTION
The Fusarium wilt is a disease that has long plagued
banana plantations all over the world. It is caused by a fungal
pathogen known as Fusarium oxysporum f. sp. cubense (Foc) [1].
Since its discovery in the 20th century, numerous efforts have been
conducted to combat the spread of the disease. While previous
variants of the wilt have successfully been mitigated, a new strain
that emerged in 1989 has been a sizable threat until today [2]. This
strain, known as tropical race 4 (TR4), is known to affect the
Cavendish cultivar of banana which is the most common variety of
banana being traded and consumed in the world as of writing.
The main challenge in addressing the problems brought
about by the disease is that there is no known method of completely
eradicating it once it has infected the soil [3]. Furthermore,
symptoms of the disease that would be visible to the naked eye are
typically only observed in the later stages of the infection [4]. As
such, there is great importance in methods that would be able to
detect the presence of the pathogen in an earlier manner.
There are methods that can be used in the accurate
detection of the disease. One such method is to use polymerase chain
reaction. However, these methods require expensive equipment and
materials that most small-scale banana producers would not have
access to [5].
Even before infecting a plant host, Foc can be detected in
infected soil through the use of microscopy. While Foc itself does
not consist of any unique fungal structures that would be a sure
indicator, the presence of all known structures could suggest that it
may be present in a given soil sample. This may then be used as a
low-cost preliminary method in detecting the disease before
proceeding with a more comprehensive detection method.
While microscopy would seem to be a feasible method, it
does require some training in recognizing fungal structures. The
viability of manually detecting Foc through a microscope also
heavily depends on the scale of the banana production. Because of
this, automating this process with the use of image processing and
machine learning may be an apt solution. In this study, we propose
the use of a convolutional neural network (CNN) in detecting the
presence of Foc TR4 in microscopy images.
II. METHODOLOGY
A. Dataset Overview and Description
The dataset used in this study is composed of images taken
with a lab microscope of lab-cultivated soil samples. These samples
contain various species of fungi and are labeled accordingly. This
study will focus on the detection of Foc TR4. As such, the images
were divided into four distinct labels: Clean, Foc TR4, Foc Non-
TR4, and Other Fungi. The number of images under each of these
labels is shown in Table 1. While the number of images under each
label is not balanced, the use of transfer learning makes it so that
balancing the number of images would not be necessary. Clean
images refer to image captures which do not contain any fungal
structure at all while the other labels refer to images that do contain
fungal structures and are captured from microscope slides that are
known to contain their respective fungal species.
TABLE I. DATASET OVERVIEW
Label Microscopy Technique Total
Brightfield Darkfield Fluorescent
Clean 771 771 771 2313
Foc TR4 771 771 771 2313

168
 details in fungal structures as the fluorescence itself, if sufficiently
Foc Non- 876 876 876 2628
intense, may introduce some noise in the image capture. On the other
TR4
hand, brightfield images seem to present the structures in the clearest
detail.
Other 598 598 598 1794
TABLE II. MODEL ACCURACY
Fungi

ResNet50 Model Test Accuracy

Furthermore, each image capture used three distinct
microscopy techniques: brightfield, darkfield, and fluorescent.
These techniques were able to produce images that were
nonidentical despite being of the same capture area. Examples of Brightfield Only Model 93.11%
these can be seen in Fig. 1.

Darkfield Only Model 91.45%

Fluorescent Only Model 86.87%

Fig. 1. Example of an image captured using three microscopy All Techniques Model 90.89%
techniques (from left to right): Brightfield, Darkfield, Fluorescent.

Table III shows the accuracy scores achieved by the

B. Image Sampling models when initialized with random weights and when using
The images were organized according to the microscopy ImageNet weights. As outlined in the Table, there is a significant
technique used. For this study, the dataset was split into a 80-20 improvement in model accuracy. This is most likely due to the small
train-test split. Furthermore, the images were downscaled to 224 x size of the dataset that would not allow for the model to sufficiently
224 pixels primarily to reduce computational costs. learn the features needed for the classification task. Since the
ImageNet dataset is focused on object detection and classification,
all the CNNs need are images that will help fine tune the model to
C. Modelling work with microscopic fungi images. It is also worth noting that the
All Techniques Model was able to benefit greatly from transfer
This study will utilize the ResNet50 architecture for the
learning relative to the other models. This may indicate that the
CNN. Since the dataset is divided into three different lighting
model with ImageNet weights was able to identify objects in the
techniques, three separate instances of the model will be developed
images despite the color variances in looking at all microscopy
to work on those respectively. In addition, a fourth one will be
techniques.
developed to classify the images irrespective of the microscopy
technique used. The models will then be evaluated using
TABLE III. TRANSFER LEARNING COMPARISON
performance metrics such as accuracy, precision, recall, and F1-
score.
ResNet50 Model Random Initial ImageNet
Weights Weights
D. Transfer Learning
Due to the small size of the dataset, it is expected that a Brightfield Only Model 78.23% 93.11%
CNN trained from scratch would not perform optimally [5]. In order
to mitigate this drawback, transfer learning will be applied to the
models. The ResNet50 models used in this study were trained on Darkfield Only Model 75.67% 91.45%
ImageNet images and will be fine-tuned using the microscopy
images. Transfer learning also eliminates the need to balance the
dataset. Fluorescent Only Model 69.37% 86.87%
To show the improvements that transfer learning can bring
about in this classification problem, separate instances of the models
will be trained using randomly initialized weights. The performance All Techniques Model 69.93% 90.89%
of these models will then be compared to the ones achieved by the
models that used the ImageNet weights.

B. Precision, Recall, and F1-scores

III. RESULTS AND DISCUSSION
A. Model Accuracy
The CNNs were able to achieve satisfactory accuracy
scores as shown in Table II. The brightfield model was able to
achieve the highest accuracy of 93.11%. The fluorescent model was
the worst performing model with an accuracy of 86.87%. The
fluorescence may make it challenging for the model to see finer
The precision, recall, and F1-scores achieved by the
brightfield model are shown in Table IV. From the values in the
table, it can be seen that the model was able to classify Foc images
well. The recall score for Other Fungi images was significantly
lower than the rest of the labels. This may be caused by the number
of different species under the Other Fungi label making it difficult
for the models to accurately determine what features exactly belong
to that class.

169
 TABLE IV. BRIGHTFIELD MODEL METRICS

TABLE VI. FLUORESCENT MODEL METRICS

Label Precision Recall F1-score

Label Precision Recall F1-score

Clean 0.87 0.97 0.92

Clean 0.81 0.95 0.87

Foc TR4 0.98 0.92 0.95

Foc TR4 0.90 0.83 0.86

Foc Non-TR4 0.94 0.99 0.97

Foc Non-TR4 0.88 0.89 0.88

Other Fungi 0.97 0.80 0.88

Other Fungi 0.90 0.73 0.81

The precision, recall, and F1-scores achieved by the
darkfield model are shown in Table V. The darkfield model
performed similarly to the brightfield model. There are differences
in that the darkfield model was able to get slightly better metrics on
the Other Fungi class while getting lower metrics on both the Clean
and Foc TR4 labels.
TABLE V. DARKFIELD MODEL METRICS
The precision, recall, and F1-scores achieved by the model
trained on all techniques are shown in Table VII. The trends that can
be observed in this table are similar to those seen in the one for the
brightfield model albeit the scores are slightly lower across the board
except for when labeling Foc TR4 images where it performs slightly
better than the brightfield model.

TABLE VII. ALL TECHNIQUES MODEL METRICS

Label Precision Recall F1-score

Label Precision Recall F1-score

Clean 0.84 0.93 0.88

Clean 0.83 0.97 0.89

Foc TR4 0.94 0.87 0.91

Foc TR4 0.94 0.92 0.93

Foc Non-TR4 0.92 0.97 0.95

Foc Non-TR4 0.95 0.91 0.93

Other Fungi 0.96 0.84 0.90

Other Fungi 0.95 0.81 0.87

The precision, recall, and F1-scores achieved by the
fluorescent model are shown in Table VI. The most noteworthy
metric in this table is the recall score on Clean images. The model
was able to correctly predict about 95% of the Clean images but has
a tendency to mislabel other images as Clean as well. This may be
explained by the fact there is some loss in detail and information
when looking at the fluorescent images in this dataset. An example
of this can be seen in Fig. 2.
As observed in the performance metrics of the models, the
models consistently attain lower scores in labeling Other Fungi
labels. An example of images under the label is shown in Fig. 3. The
inclusion of multiple different fungi species in the Other Fungi label
produced images that contain a wide variety of fungal shapes and
structures that are not consistent in appearance, thus complicating
the features that belong to the label.

Fig. 3. Example of images under the Other Fungi label showing

significant variance in fungal shapes and features

Fig. 2. Example of a sample taken using the Brightfield and

Fluorescent microscopy techniques.

170
 IV. CONCLUSION ACKNOWLEDGEMENT

The early detection of Fusarium oxysporum f. sp. cubense The authors would like to thank the Department of
Tropical Race 4 is crucial towards the mitigation of damages caused Science and Technology - Science Education Institute (DOST-SEI),
by the disease to banana plantations all over the world. Microscopy the Cloud-based Intelligent Total Analysis System (CITAS) Project,
is one method that can be used for the detection of the pathogen, and and the Ateneo Center for Computing Competency and Research
the use of convolutional neural networks is proposed in the (ACCCRe) for their support in this study.
automation of this process. The microscopy images used in this
study utilized three different microscopy techniques in capturing REFERENCES
images of the samples: brightfield, darkfield, and fluorescent
[1] G. Fourie, E. T. Steenkamp, R. C. Ploetz, T. R. Gordon, and A.
microscopy. The effectiveness of these techniques in detecting the
Viljoen,“Current status of the taxonomic position of fusarium
fungi was assessed through the development of four models trained
oxysporum formae specialis cubense within the fusarium
on different subsets of the dataset. Based on the results obtained in
oxysporum complex,” Infection, Genetics and Evolution, no.
this study, brightfield images are more helpful in classifying fungi
11, p. 533–542, 2011.
in microscopy images among the three techniques used in this
[2] M. Reynolds, “A fungus could wipe out the banana forever,”
dataset. It is worth noting that this is true only in the context of the
https://www.wired.com/story/fungus-could-wipe-out-banana-
images used since these do not contain any dirt or other irrelevant
forever/, August 2019.
artifacts. Transfer learning has shown to make significant
[3] R. C. Ploetz, “Panama disease: A classic and destructive
improvements to the performance of the models in classifying
disease of banana,” Plant Health Progress, 2000.
microscopic fungi. The model trained exclusively on brightfield
[4] M. Dita, M. Barquero, D. Heck, E. S. Mizubuti, and C. P.
images was able to attain an accuracy of 93.11% and was able to
Staver, “Fusarium wilt of banana: Current knowledge on
classify three out of the four labels well. Though the model can be
epidemiology and research needs towards sustainable
further improved, it can still function as a low-cost solution in the
management,” Frontiers in Plant Science, no. 9, 2018.
early detection of Foc given that the model was able to distinguish
[5] M. Fakruddin, K. S. Mannan, A. Chowdhury, R. M.
Foc from other species of Fungi relatively well. The model can also
Mazumdar, M. N. Hossain, S. Islam, and M. A. Chowdhury,
be used as a preliminary test before using comprehensive tests that
“Nucleic acid amplification: Alternative methods of
require costly equipment and materials. Future work could include
polymerase chain reaction,”Journal of Pharmacy Bioallied
more experimentation on CNN architectures used, pre-processing
Sciences, no. 5, p. 245–252, 2013.
techniques, and testing on real-world samples gathered from the
[6] C. Tan, F. Sun, T. Kong, W. Zhang, C. Yang, and C. Liu, “A
field.
survey on deep transfer learning,” ICANN 2018, August 2018.

171
 Vision-based approach for checkpoint status analysis
Czeritonnie Gail V. Ya-on1, a *, Jonathan Paul C. Cempron1,b, and Joel P. Ilao1,c
1
De La Salle University, Manila
a
czeritonnie_ya-on@dlsu.edu.ph, bjonathan_cempron@dlsu.edu.ph, c joel.ilao@dlsu.edu.ph
Abstract—This paper presents a vision-based technique to
calculate the service rate, arrival rate, and saturation level of
checkpoints surrounding Metro Manila. The method is applied to
different inspection locations: EDSA Starmall, Megamall, and ADB
Guadix. Results shows that the approach is able to identify the lane
membership of trajectories of vehicle trajectories to estimate the rate
and saturation levels of queues on clear camera view. In conlusion,
the technique works and it could also be used to automate the
estimation of traffic congestion levels at sections covered by
surveillance cameras.
Keywords— vision-based checkpoints analysis, computer
vision, service rate, arrival area, queue saturation
I. INTRODUCTION
Checkpoints or quarantine control points (QCP) are areas placed in
the boundaries of cities and municipalities to regulate and limit the
mobility of people to essential travel, to arrest the spread of the
novel coronavirus. The existence of a checkpoint, like any other
queues, introduces delays and bottlenecks. The goal is to balance the
trade off between maximizing inspection personnel and minimizing
the amount of time it takes to screen vehicles through security
checkpoints. A fundamental step is to identify the present condition
of checkpoints in selected regions.
In this paper, we provide useful insights in assessing the
performance of QCP inspection areas in terms of the number of
customers that request service at a given time step, the arrival
process for the customer, the number of servers which control how
many customers leave the queue, etc. Our work aims to estimate the
fundamental parameters that are related to Queueing approach
practiced in Traffic Engineering [1].
Similar works we have found that employ queue analysis
particularly on data provided by Metro Manila Development
Authority (MMDA) are: [2] used a traffic light simulation to
determine the hourly traffic density. They used the number of cars
in each lane, length of the junction, and the traffic light time in
España Boulevard. [3] focused on analyzing the boarding and
unloading of passengers in bus stops areas along EDSA-Shaw. Our
approach, on the other hand, used a vision-based technique to
calculate the service rate, arrival rate, and saturation level of QCP
checkpoints traffic video feeds surrounding Metro Manila. This
study aims to discuss the potential benefits of this approach to
estimate traffic congestion levels at areas covered by surveillance
cameras.
Fig. 1. Checkpoint on EDSA ADB Guadix from MMDA
footage
II. METHODOLOGY
A. Object Tracking
We developed a detection based tracker from our previous work [4]
where the inputs are video files and YOLO [5] was used as a detector
for each input video frame. The output of the detection consists of
[x coordinate, y coordinate, width , height] position and
corresponding vehicular classification.
Tracking was implemented using a predict-match-update
tracking framework. The prediction is based on a linear motion
model. Matching is performed using motion constraint and
template-based matching. Updating is accomplished by way of
simplified implementation of multiple hypotheses tracking. The
output of the tracking process is a file containing a list of tracklets,
henceforth tracklists. The tracklets contain information position,
size, and classification of the tracked object; it serves as a basis for
constructing the object's trajectory during the queue analysis.
A tracklet is a list of trackpoints. Trackpoint is a triple of
frame numbers, classifications and bounding box, video frame
number, and classifications. For the queue analysis, we only need
the bounding box position and classification. The centers of the
bounding boxes are calculated and used for lane identification. We
refer this information as tracklet points.
B. Queue Analysis
Queueing is commonly defined as a process wherein
people, goods or information need to wait at a certain time to receive
a service. [6]. In this study, the checkpoint activity is considered to
be the queueing system. Processing areas in the input video are
either a series or a parallel queue when multiple customers are
served at a time. The queue system is composed of a server and a
customer. In the context of vehicular traffic queues, inspection teams
operating in checkpoints are the servers while the vehicles passing
these checkpoints are the customers.
The number of vehicles joining the queue over a period of
time is called the rate of arrival which can be represented as variable
q. The number of vehicles processed at the queue system for each
unit of time is called the service rate denoted as variable Q.
Figure 2 illustrates the series queue wherein multiple
vehicles are served in a single lane per unit of time. Figure 3 shows
172
an example of a parallel queue serving multiple vehicles in multiple
lanes over a period of time.
Fig. 2. Example of a Series Queue System
Fig. 3. Example of a Parallel Queue System
In parallel queues, each lane Li has inspection areas. The
variable i in Li stands for the number of lanes. The inspection
personnel assigned to inspection areas are represented as P(i,t)
which are the number of personnels assigned to Li for time t.
Service rate [5] is the rate of processing in Li for time unit
t, and denoted as Q(i,t) as shown in (1):
(1)
where tj is a window of time.
We generate the trajectories by analyzing the output file
containing the list of tracklets. Lanes are defined as regions in the
scene passed by vehicle objects that correspond to a cluster of
trajectories. Inspection areas are points at the lanes that signify a
processing region. We compute the service rate by counting the
number of vehicles that pass by the inspection area for a given time
t. The Q(i,t) is the mean of all Q(i,tj) and time tj can be arbitrarily
set, i.e. per hour, half-day, per half-hour, etc.
Arrival rate [5] is the rate of vehicles joining in Li for time
unit t and denoted as q(i,t), as shown in (2):
(2)
where tj is a window of time .
The end of the queue is not typically visible from the
camera especially when the vehicles are too far from the camera
view as shown in Figure 4. We approximate the tail position of the
line as the first detection of the object being tracked as it arrives in
the queue.
Fig. 4. A queue with an undetectable tail
A queue is considered as under-saturated when the
number of vehicles processed in the system Q is greater than the
number of vehicles joining the queue q:
𝑄>𝑞
A queue, on the other hand, is over-saturated when the
number of vehicles processed in the system Q is less than the number
of vehicles joining the queue q:
𝑄<𝑞
We use these categories to identify the characteristics of
lanes for each checkpoint.
C. Lane Identification
Data preparation involves partitioning of the tracklet data
points. Partition was implemented because the trajectory starts to get
inaccurate when a vehicle is not completely visible particularly
when it starts to exit the scene. The corresponding vehicle’s center
becomes incorrect when the vehicle is only partially visible. In this
kind of scenario, the cluster results are affected. So for more robust
analysis, tracklet data points were partitioned according to three
regions: upper, middle, and lower. Thresholds for separating the
aforementioned regions were based on manual observation of the
tracklet points and the scene. Each partition was clustered using K-
Means [7]. We choose the cluster partition that produces the best
outcome which result distinctly represents the lanes. The selected
cluster is used as a basis for imposing membership on the lane.
Figure 5 shows a clustered portion of the tracklet points. Colorbar
represents the cluster grouping or the lane in the scene. We based
the lane membership on majority polling. If most tracklet points in
a partition belong to a certain cluster, say cluster 1, then the rest of
the tracklet points will be imposed on this cluster.
173
 D. Queue Analysis in Quarantine Checkpoints
We gauged the queue status based on the service rate and
arrival rate values of each lanes. Different checkpoint sites were
used in the test to validate how the proposed method works on
different lanes. In Figure 7, we can see in a) that vehicles moves
towards the camera while in c) the vehicles moves away from the
camera. The camera view in b) covers both movements.
In this case, we will use the definition of upstream and downstream
to describe the roads throughout the analysis. Downstream is the
flow of vehicles moving towards the camera. Upstream, on the other
hand, is a flow when vehicles move away from the camera.

a)

(a)

b)

Fig. 5. Partition clustering of a track a) upper and b) middle. The

colorbar represents the cluster grouping or the lane in the scene.

Service rate is calculated by collecting the number of

trajectories that intersect with inspection areas. Inspection areas, in
queue analysis context, are specific locations in the lanes where
processing is assumed to take place. It can be any form of processing
like checkpoints.
Arrival rate is based on the first detection of a trajectory (b)
in a lane. In Figure 6, the inspection areas are marked in black circles
and the start of the queue is represented by the first data point in the
trajectory. In this figure, these are the lowest points along the
centroid x-axis. It must be noted that not all trajectories are exiting
the queue. There are instances that vehicles do not leave the scene
or there are vehicles that do not enter the queue at the same time. We
have to calculate the service and arrival rate of each lane, for this
reason, to track these kinds of scenarios.

(c)

Fig. 7. a) downstream, b) both, c) upstream

Given that vehicle movements are changing depending on the

location, it is assumed that checkpoint areas follows a parallel queue
Fig. 6. A scene with three lanes, processing areas, and start of
for all sites. Checkpoints were manually placed “in front of queue”
queue. The colorbar represents the cluster grouping or the lane in
for downstream scenarios and “at the back of queue” for upstream
the scene
scenarios.

174
 EDSA Starmall
Lane
Saturation
Q q
Level

Cluster 0 288 360 Under-saturated

Cluster 1 252 252 Optimal

Cluster 2 372 372 Optimal

a.
Video recorded on April 23, 2020 between 9AM-11AM
Fig. 8. Marikina Infanta Highway
TABLE II. QUEUE ANALYSIS RESULTS FOR
One limitation of the experiment is we did not include the CHECKPOINT 2
scenes where the inspection areas are not visible to the camera,. In
Fig 8, we observed that inspection area is undetectable because the
vehicles are too small in the scene. Megamall
Lane
Saturation
Q q
Level

Cluster 0 432 432 Optimal

Cluster 1 132 132 Optimal

Cluster 2 228 228 Optimal

a.
Video recorded on May 18, 2020 between 9AM-11AM

Fig. 9. EDSA GT Tower

Similarly, scenes like in Fig 9 where the camera is too high are not TABLE III. QUEUE ANALYSIS RESULTS FOR
included in the analysis because the vehicles are undetectable at this CHECKPOINT 3
state.

Guadix
Lane
III. RESULTS AND DISCUSSION Saturation
Q q
Level

The method introduced in this study is investigated using videos Cluster 0 192 216 Under-saturated
showing a checkpoint operations held at EDSA Starmall, Megamall,
and Guadix. The videos were recorded on April 23 (ECQ) , May
18 (MECQ), and June 01 (GCQ), respectively. We used 5-minute Cluster 1 1800 1800 Optimal
chunks of the input videos between 9 AM to 11 AM. The experiment
covers the calculation of service rate, arrival rate, and queue Cluster 2 492 492 Optimal
saturation for each locations; and analysis of trajectories generated
by the algorithm. The colorbars in the figures represent the cluster a.
Video recorded on June 01, 2020 between 9AM-11AM
grouping or the lane in the scene.
In Table 1, the clusters 1 and 2 are optimal; and cluster 0 is over
saturated. Table 2 shows that all clusters are optimal, while Table 3
A. Queue Analysis shows that cluster 0 is under saturated while clusters 1 and 2 are both
optimal.
Tables 1, 2, and 3 show the per-hour results of estimating the service
rate Q and arrival rate q for the selected case site. We evaluate the
queue saturation of each lane by comparing the values of Q and q, B. Trajectory Analysis
according to our queue saturation levels defined in Section B.
Figure 10 shows the generated trajectories for a EDSA Starmall. All
TABLE I. QUEUE ANALYSIS RESULTS FOR clusters are downstream in this case. The lanes in cluster 1 and 0 are
CHECKPOINT 1 merging because there was a roadblock before the inspection area in
cluster 1. It is expected to cause cluster 0 to process additional

175
trajectories. Looking at Table 1, we expected that Q in cluster 0
should significantly increase due to merging. However, it is
observed that some points are uncounted in cluster 0, especially in
the regions where the data points overlap.

Fig. 10. EDSA Starmall inspection area

Figure 11 shows the generated trajectories for a

Megamall. Cluster 0 is a downstream; clusters 1 and 2 are upstream.
All the clusters yielded an optimal saturation level in Table 2. It is
noticeable that the size of the queue is shorter compare to other
locations. The point where vehicles entered the queue are not clear
for all clusters. A fence blocks the arrival scene for cluster 0 while
the camera view for clusters 1 and 2 are shorten.
Fig. 11. Megamall inspection area
Figure 12 shows the generated trajectories for a Guadix.
All clusters are upstream. Comparing the results to Table 3, we can
see that cluster 0 is under saturated and; clusters 1 and 2 are optimal.
The arrival of vehicles in cluster 0 is visible to the camera but it
becomes undetectable as it exists the queue. The switching of lanes
between cluster 0 and cluster 2 is also not accounted for the same
reason.

176
 REFERENCES

[1] F. Mannering and S. Washburn. 2013. Principles of Highway

Engineering and Traffic Analysis, 5th Edition. John Wiley &
Sons, USA.
[2] P. D. L. Delica, M. R. U. Landicho, J. A. A. Tabliga, L. R.
Virtus and R. Anacan, "Development of an intelligent traffic
control system using NI LabVIEW," 2017 7th IEEE
International Conference on Control System, Computing and
Engineering (ICCSCE), Penang, 2017, pp. 17-22, doi:
10.1109/ICCSCE.2017.8284372.
[3] R. K. C. Billones et al., "Vision-Based Passenger Activity
Analysis System in Public Transport and Bus Stop Areas,"
2018 IEEE 10th International Conference on Humanoid,
Nanotechnology, Information Technology,Communication
and Control, Environment and Management (HNICEM),
Baguio City, Philippines, 2018, pp. 1-6, doi:
10.1109/HNICEM.2018.8666357.
[4] J. Cempron, and J. Ilao. 2019. VEMON: Vision-based Vehicle
Counter for Traffic Monitoring, De La Salle University
Research Congress
[5] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi. 2016. You
Only Look Once: Unified, Real-Time Object Detection.
arXiv.org [Online]. Available:
https://arxiv.org/abs/1506.02640. [Accessed: 14-Nov-2020].
[6] K. Teknomo. 2012. Queuing Rule of Thumb based on M/M/s
Queuing Theory with Applications in Construction
Management, Civil Engineering Dimension, Vol. 14, No. 3.
[7] A. Al-Masri, “How Does k-Means Clustering in Machine
Learning Work?,” Medium, 15-May-2019. [Online].
Fig. 12. Guadix inspection area Available: https://towardsdatascience.com/how-does-k-
means-clustering-in-machine-learning-work-fdaaaf5acfa0.
IV. CONCLUSION [Accessed: 14-Nov-2020]

This study presents an algorithm to calculate service rate,

arrival rate and saturation level using vision-based techniques.
Experiment make use of the MMDA video feeds of several different
checkpoints during quarantine. The algorithm works to identify the
lane membership of trajectories under different locations; and it is
able to classify checkpoints based on vehicular volume in the queue.
The study is also able to determine the inspection service efficiency,
lane capacity and behaviour of the observed locations.
Enhancement are to be made on handling of overlapping
trajectories on the lanes, setting of thresholds for the queue system,
and recognizing the service areas in the actual input video. The
measurement of the algorithm’s effectiveness is accomplished in
another study.
Our proposed method could be used to automate the
estimation of traffic congestion levels at sections covered by
surveillance cameras. It can also be used to measure the level of
service in managing vehicular traffic.

ACKNOWLEDGEMENT

This project is funded by the Department of Science and Technology

– Philippine Council for Industry, Energy, and Emerging
Technology Research (project number 08784) and the De La Salle
University Research Coordination Office (project number IR
1TAY18-1TAY19). The authors are also grateful to the
Metropolitan Manila Development Authority (MMDA) for
providing the samples of traffic videos captured by their CCTV
systems.; and DLSU Engineering R&D for Technology , for the
graduate scholarship of some authors.

177
The ERDT National Conference 2020 was conducted via the video conferencing platform,
Zoom.

178