You are on page 1of 33

Digital Twin for Logistics: Machine Learning

and Simulation Approaches


Ivan Kristianto Singgih, Ph. D.

25 September 2021
Institut Supply Chain dan Logistik Indonesia

1
Ivan Kristianto Singgih, Ph.D.
https://www.researchgate.net/profile/Ivan_Singgih
Scopus ID: 57095064900
CV: http://bit.ly/CV_IvanKSinggih
Email: ivanksinggih@gmail.com

Working experience: Projects:


2021-now Postdoctoral Researcher, Quantum Machine Learning 2020-now Smart Manufacturing Education using
Laboratory, Industrial and Management Engineering, Flexible Machine Job Shop System
Korea University, Korea Embedded with Machine Learning
2020-2021 Postdoctoral Researcher, Industrial and Systems Technologies (KAIST and CubicTech)
Engineering, KAIST, Korea 2020 Recommendation System for Optimal Tire
2017-2020 Research Assistant Professor & Postdoctoral Design (KAIST and Hankook Tire)
Researcher, Industrial and Management Engineering, 2019-2020 Torpedo Ladle Car Pairing and Routing
POSTECH, Korea Optimization (POSTECH and POSCO)
2019 Development of Window Frame Cutting
Training experience (as instructor/teaching assistant): Algorithm (POSTECH and LG Electronics)
2020 LG Company with KAIST (“Genetic algorithm in 2015-2017 Technological Development of Low-carbon
manufacturing product design”; “Smart factory using Automated Container Terminals Project
CNN and LEGO Mindstorms”) (Pusan National University and the
Ministry of Oceans and Fisheries, Korea) 2
Research interests:
• Logistics systems (container
terminal, drone routing)
• Operations Research
• Smart systems (smart city, smart port,
smart manufacturing, smart logistics)
• Machine learning (classification
techniques, CNN)
• Serious game development
(blockchain, mathematical
modelling)
Drone 3-dimensional routing Digital twin of semiconductor fab
Skills:
12 particulate matters
• Mathematical modeling, algorithms, (NO, CO,
metaheuristics (genetic algorithm etc) O3, Benzene, …)
• Simulation: Anylogic, Tecnomatix Prediction
Plant Simulation (ML models)
• Programming language: Java
(Eclipse Juno), MATLAB, C++ Air quality
(good, moderate, poor)
(Visual Studio), Python
• Mathematical model solver: LINGO,
Electric vehicle relocation Air quality prediction (machine learning)
3
CPLEX
Collaborations
between Industry and Academia

4
Why we need to collaborate?
• Better results in real life (lower cost, more efficient operations, high
competitiveness, etc):
• Industry: real data and problems, experience dealing with hardware and
physical systems
• Academia: various “sophisticated” methods, flexibility to generate specialized
solutions with various methods

5
How to collaborate in South Korea
• Industry:
• Open proposal, sent to universities (e.g., Samsung Heavy Industries to
POSTECH) -> researchers apply with their proposals
• Visits to universities, knowing that research in universities could save a
significant production cost based on other industry’s experience) (e.g.,
H2Korea to POSTECH)

6
How to collaborate in South Korea
• University:
• Long-term collaboration scheme (e.g., KAIST and Hankook Tire, Synustech, LG)
• Active publication and project list update in websites
• Training about solutions methods to companies (e.g., KAIST to LG)
• Appropriate technology transfer (coding template)

• Research Institute:
• Joint research with universities and companies (e.g., Korea Maritime Institute
with Pusan National University, Donga University, Power MNC manufacturing
company, Total Soft Bank software company)

7
Digital Twin
• Integration between the physical and
virtual worlds

• Using such a virtual model allows:


• monitoring the physical system,
• preventing problems from happening,
• finding new opportunities, and
• planning the future.
Machine Learning in South Korea

Pengubahan data ucapan ke


dalam teks
Sumber: http://www.koreaittimes.com/news/articleView.html?idxno=97629

Pengendalian peralatan
Pembacaan pesan dalam
dengan menggunakan suara
aplikasi chatting kakao &
& rekomendasi pengambilan
penjawaban oleh pengguna
foto menggunakan kamera Sumber: http://www.koreaherald.com/view.php?ud=20200603000761
Sumber: https://www.lg.com/us/discover/thinq, dengan mengatakan “jawab”
(“reply”)
https://www.lg.com/eastafrica/lg-thinq/news/artificial-intelligence-for-
a-better-life 9
Sumber: http://www.koreaherald.com/view.php?ud=20181120000659
Artificial Intelligence Ecosystem in South Korea
• Cooperation from large AI
companies cooperates not
only to foreign companies,
but also to domestic start-
ups to reduce the
technological gap with other
countries with more
advanced AI
• Government-funded
projects to open
collaborations with foreign
AI-tech companies
• Government efforts on
regulations and complicated
Source: https://www.intralinkgroup.com/getmedia/7bca58ca-90d0-4c2d-a25f-af23c057b7b3/Korean-Artificial-Intelligence-Final-Report,-Innovation,-Brochure personal information
protection laws 10
Projects
• Route Recommendation System using Simulation-based Analysis
and Driver’s App Interface
• Simulation-based Electric Truck Investment Analysis
• Semiconductor Production System Evaluation
• Container Terminal’s Truck Congestion Identification and
Resolution using Image Processing
• Container Stowage Planning using Image Processing

11
Route Recommendation System using
Simulation-based Analysis and Driver’s
App Interface Our Research Team:

(Collaborations/Discussions: Jakarta Smart City)

12
System Overview
Simulation-based Car
Movement Analysis
1) Traffic light schedules
- Light durations
- Outflow directions

Route
Recommendation
Through Driver’s
Handphone

Output:
2) Input flow between pair of nodes • Short car travel times
- Source and destination pairs Observed road segments • Less congestions at
- Arrival rates of cars (Blok M, Jakarta city) road network
- Car information (size, speed)
13
Traffic Light Control Method
• Determining duration and directions of traffic lights:
1. Constant cycle/Simple rule
2. Reinforcement learning (RL), one of Machine Learning methods

State:
Current traffic conditions
(e.g.: queue length)

Reward:
• Queue length reduction
• Car’s waiting time
• Number of passing cars

Action:
• Duration of traffic lights
• When to change lights

Source: https://deepdrive.berkeley.edu/project/adaptive-traffic-signal-control-based-deep-reinforcement-learning
14
Simulation Model
Cars waiting for the
traffic light
Traffic light indicator
(red: stop, green: go)

Queue per lane Average car


at each intersection waiting time

Car movement data retrieval


Traffic indicator
from Waze 15
Simulation-based Electric Truck Our Research Team:

Investment Analysis
(Collaborations/Discussions:
Koja & Kualatanjung Container Terminals, Indonesia) Kualatanjung Multipurpose Terminal

16
Container transportation using trucks
Operational decision-making using digital twin
(simulation & IoT-based real condition acquisition)

Simulation

Unexpected
Best
situation
decision change

Real Condition

17
Analysis

Optimization Decisions in
Simulation
https://sumo.dlr.de/docs/index.html
https://medium.com/yonohub/sumo-a-traffic-simulator-over-the-cloud-with-yonohub-f2bbf7f62990

Truck- Container
Truck Container Terminal Simulator
Container Stowage
Routing
Assignment Planning Generator using SUMO Open Source
Truck Battery Truck Battery
Station Location Charge/Swap Energy Consumption
Determination Scheduling Comparison between
General vs. Electric Trucks
18
Our Research Team:
Semiconductor Production
System Evaluation

19
• Complex wafer lot flows: wafer lot re-
entrances into machines

• An advanced technique is required to


capture the system's dynamics.

Source:
https://www.researchgate.net/publication/283119905_Simulation_
Based_Experimental_Investigation_for_Performance_Assessment_
of_Scheduling_Policies_in_Wafer_Fabrication

20
Number of waiting
lots for each wafer
type at each
machine
Production target satisfaction
(84 lots per week)

Number of
produced lots at
each machine Output
• How to predict production target satisfaction
(output) based on production system conditions
Machine working (input) using machine learning?
times
• Two output classes:
(1) Good (more than 84 lots per week)
(2) Bad (less than 84 lots per week)
Input 21
Results (Best prediction model)

22
Results
(Identifying the most important factors
of the production system)

3rd
2nd

1st

23
1st
Container Terminal’s Truck Our Research Team:

Congestion Identification and


Resolution using Image Processing
(Collaborations/Discussions:
Koja & Kualatanjung Container Terminals, Indonesia)
Kualatanjung Multipurpose Terminal

24
Congestion problem
• A container terminal in Jakarta city faces truck congestion at its gate and around
its container storage yard area.
• To reduce the congestion, we propose using image processing techniques to
measure the truck waiting times at critical locations.

25
Image processing for
identifyting trucks’ locations and waiting times

26
Source: https://www.sciencedirect.com/science/article/pii/S0939388918301181
Truck movement management at important points
at the container terminal

https://www.smallworldsocial.com/theory-of-constraints-104-balance-flow-not-capacity/ 27
Our Research Team:

Container Stowage Planning


using Image Processing
(Collaborations/Discussions:
Koja & Kualatanjung Container Terminals, Indonesia)
Kualatanjung Multipurpose Terminal

28
Container Stowage Pattern Analysis

https://ieeexplore.ieee.org/document/8513823

https://ieeexplore.ieee.org/document/9119999

29
Container Stowage Pattern Analysis

Collecting Experts’ Knowledge on


Container Stowage Pattern

Stowage Plan Pattern Analysis using


Deep Learning

Estimating Container Processing


Times Considering Uncertainty
using Optimization Models
Deep Learning Network Optimization
using Metaheuristics
30
Code & Paper Sharing + Training Materials
(Python, Vehicle Routing, Machine Learning, CPLEX)

Link: https://github.com/ivanksinggih 31
Required collaborations:
• [Industry] Potential collaboration discussions, data
sharing Offered collaborations:
[Industry]
• Code sharing (to be implemented in the industry)

• [Researchers] Coders (online discussions)


[Researchers]
• Knowledge sharing (code sharing and tutorials)
• Top journal paper publication
• Joint research supervision
• Paper discussions (Web of Science/top Scopus
(1) Inclusion to projects journal paper study)
Compatibility
(+discussions with • Real problem solving (introductions to related
assessment
companies)
(interests and skills) companies or research institutes)
(2) Code learning scheme

Required Collaborations Offered Collaborations


Thank you!
For collaborations and discussions, please send an email to:
ivanksinggih@gmail.com

33

You might also like