See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/289992133

Multivariate Key Performance Indicator of Baking Process

Article  in  Energy Procedia · December 2015

DOI: 10.1016/j.egypro.2015.11.869

CITATIONS READS

2 272

5 authors, including:

Fabrizio Bonacina Luciano De Propris

Sapienza University of Rome Sapienza University of Rome
6 PUBLICATIONS   16 CITATIONS    14 PUBLICATIONS   332 CITATIONS   

SEE PROFILE SEE PROFILE

Sara Feudo Andrea Marchegiani

Sapienza University of Rome Sapienza University of Rome
8 PUBLICATIONS   13 CITATIONS    14 PUBLICATIONS   131 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Luciano De Propris on 28 February 2016.

The user has requested enhancement of the downloaded file.

 Available online at www.sciencedirect.com

ScienceDirect
Energy Procedia 82 (2015) 554 – 561

ATI 2015 - 70th Conference of the ATI Engineering Association

Multivariate Key Performance Indicator of baking process

Alessandro Corsinia,c; Fabrizio Bonacinaa; Luciano De Proprisa; Sara Feudoa*;
Andrea Marchegiania, c
a
Department of Mechanical and Aeronautics Engineering, Sapienza University of Rome, Via Eudossiana 18, I00184 Roma, Italy.
c
SED srl, Solution for Energy and Diagnostics Srl, Via Colle Baiocco, 03013 Ferentino (FR), Italy

Abstract

Energy efficiency is nowadays a subject deeply discussed in several fields, with a large potential in industry. Here
proper process and energy management routines turns out to be essential to reduce the energy demand while keeping
the control of the product quality. In this respect the bread baking, one of the pillar of food related industry, is an
energy intensive process irrespective to the adopted oven technology or to the primary energy nature.Baking is the
fundamental step of the bread production process and it entails a number of complex chemical and physical
phenomena, critical to the final physical properties of bread, i.e. crust colour, crumb texture and taste.A careful
balance throughout all the steps of the “manufacturing” cycle is vital to ensure the processes synchronization, in order
to produce a consistent and satisfactory loaf of bread. A proper energy management of this process needs to consider
such features to ensure a high quality product. As such process monitoring can not be conveniently described using
customary specific energy metric (correlating the energy demand to the amount of processed material) or full three-
dimensional (3D) physic-based modeling as in computational fluid dynamics (CFD).
In this paper, the energy analysis of the system is carried out with a methodology rooted in the family of non
parametric approaches. The aim of the study is to identify key performance indicators (KPIs) able to assess the
effectiveness of the energy “use” along the baking process. Specifically, the identification of KPIs is carried out using
Principal Component Analysis (PCA) of the available datasets.
© 2015 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
© 2015 The Authors. Published by Elsevier Ltd.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Selection and/or peer-review under responsibility of ATI
Peer-review under responsibility of the Scientific Committee of ATI 2015

Keywords: Energy efficiency, industrial oven; bread baking, KPI, PCA

1. Introduction

European policies are increasingly addressing the use of energy efficiency related technologies
opening new market segments[1]. In this respect, the industry sector represents one of the largest energy
users with a diversity of efficiency challenges. According to recent report on Italian energy structure, the

* Corresponding author: Tel.:+39-0773-476521

E-mail address:sara.feudo@uniroma1.com

1876-6102 © 2015 Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the Scientific Committee of ATI 2015
doi:10.1016/j.egypro.2015.11.869
 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561 555

industry sector accounts for the 42% share of the national energy demand and a significant contribution
(around 10%) is given by the food industry [2].
Bread is a food product widely consumed across the world, with an energy demand increasingly
dependent upon electricity and natural gas, owing to the growth of industrial bread production replacing
the traditional bakery. Bread production includes different fundamental biochemical, chemical and
physical processes, such as: evaporation of water, volume expansion, gelatinization of starch, protein
denaturation, crust formation, carbon dioxide production, formation of a porous structure and browning
reactions [3]. Baking encompasses a range of simultaneous processes of radiative and convective heating,
water and steam transport during the process as the dough is baked into bread. In comparison with other
drying processes, is complicated by the dependence on operational issues, like the temperature gradient
within the oven and the baking time.
Commercial bread ovens are typically of tunnel-type 30 m up to 40 m long. Typically, the oven is split
into three zones in order to control the profile of conditions within the chamber, such as bake time, air
velocity, air temperature and water content. The baking process and the oven heating profiles are
dependent on the type of bakery product [4].Half of the energy used in a bakery is due to the oven
[5],other significant uses include mixers, provers and coolers.
From a methodology viewpoint, large industrial ovens has been extensively studied using
computational fluid-dynamics to simulate the detailed flow and heat conditions. To mention but a few, 3D
CFD has been used as a tool to help baking optimization in different respects as: reduce moisture loss,
optimize temperature [6], improve product quality examining heat transfer coefficient and bread radius
[7], or improve product quality through multi-objective optimization and control vector parameterization
approaches [8].However, few studies in the open literature address the energy analysis of baking
processes, in particular about industrial ovens [9,10].
On the other hand, the methodology widely used for energy efficiency in industry concerns the
construction of energy indicators able to provide information about the energy state of the industrial
process and its dynamics. The International Standard ISO50001[11] suggests the use of dedicated KPIs
called Energy Performance Indicators (EnPIs) to help the performance monitoring in energy management
systems. The typical, and the simplest, indicator measures the performance of the system according to
Service Energy Rate (SER) or Specific Energy Consumption (SEC). Those specific EnPIs correlate the
energy demand to the volume of production and recent studies have indicated their pertinence to any kind
of energy use [12].
Notwithstanding, this metric is not able to fully account for the process information. The energy
consumption of any industrial energy use is dependent upon a large number of variables (even external to
the process itself) [13, 14].Similarly, the baking process is characterized by different physical and
chemical aspects that affect the energy consumption, product quality and production rate. In such
application, as demonstrated by Corsini et al. [15], it is mandatory to attempt the definition of KPIs based
on a proper multivariate data analysis to evaluate the energy efficiency of a production process.
The aim of the present work is to discuss a methodology of identification of KPIs to describe the
energy performance of a bread baking process, and as such the figure of merit of an industrial oven. The
proposed approach makes use of a multi-variate dataset including the principal physical variables
influencing the process, in the attempt to analyze the correlation within them and to identify patterns of
variation [14]. To this end, the Principal Component Analysis (PCA) was first proposed as a methodology
to find mutual correlation in large datasets by reducing the noise from uncorrelated variables. PCA has
been applied in various fields such as taxonomy, biology, pharmacy, finance, agriculture, ecology, health
and architecture[16].
556 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561

The following sections illustrate, first, the data analytics and the data collection methodologies. Then,
the report discusses the results obtained using the non-parametric approach to the definition of energy
KPIs under different oven operating conditions.

2. Methodology

PCA belongs to the Proper Orthogonal Decomposition (POD) family. POD is a method of data
analysis aimed at obtaining low-dimensional approximate description of high-dimensional process. It
provides an efficient way of capture the dominant components of an infinite-dimensional process[17].In
this respect, PCA is a statistical technique first proposed by Pearson and Hotelling [18]. The purpose of
the PCA is to identify the dependence structure behind a multivariate stochastic observation in order to
obtain a compact description of it. It can be interpreted as a variance maximization technique.
The goal is to reduce the dimension of a data set finding a new set of variables, smaller than the
original one, educting the information contained in the data. Here the term information is meant as the
internal variation present in the sample (data), given by the correlations between the original variables
[19]. The extracted variables, called Principal Components (PCs), are linear combinations of the original
variables in which the coefficients of the linear combination depend on the eigenvectors of the covariance
(or correlation) matrix of the original data. PCs are uncorrelated with each other and ordered by
decreasing order of variance[20].
One of the methods often used to graph the results obtained from the PCA is the bi-plot of PCA scores
and loadings. It is the method of representation that displays the data formed by transforming the original
ones into the space of the PCs (scores).The loading provides a measure of the contribution of each
variable to the principal components. In particular, it displays the relationship between the different
variables analyzed and between the latter and the dataset.

3. Test description

As customary in industrial oven temperature profiling, the measurements were acquired directly on the
oven belt conveyor, to describe the entire baking cycle of the bread. The test were carried out using
multiple sensor acquisition system with a sampling frequency of1 Hz (suitable to describe the dynamics
of the process under scrutiny). Namely, the following sensors were used: heat fluxes, air temperature, air
humidity.

250
Oven Teperature (°C)

200

150

100

50

0
Time (s)

Fig. 1. Temperature profile of the typical oven

 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561 557

Notably, the heat flux measurements were able to differentiate radiation and convection contribution,
the latter qc being computed as the difference between the total qtand the radiative qr heat fluxes. The
oven under exam had three sections, with a standard multi-burner direct flame configuration [3]. The
datasets were provided in the ambit of a routine investigation on oven performance [21]. The typical
temperature profile of the oven is illustrated in Figure 1.
Data selected from the database for the present analysis are: the radiative and convective heat fluxes
(W/m2), air temperature (°C) and specific humidity (%). The specific humidity represents the mass of
water vapour in a unit mass of moist air, usually expressed as grams of vapour per kilogram of air.

4. Results

4.1. Rationale and KPI description

The analysis of performance metric was carried out to explore the influence of oven ventilation system
on the heat fluxes. As such two different tests were considered which differ because of the operational
regimes of the re-circulating fans and the extraction fans. Test A, the reference one, featured a setting
with the recirculation fans at 30% of nominal duty point and the extraction fans at 20%.Notably, in this
condition the convective heat flux represents the 54% of the total flux. While, test B featured a setting
with the recirculation fans at 50% and extraction fan at 70% of the nominal duty points with a convective
heat flux component about 61% of the total heat flux.
As a consequence of the rationale of the investigation, the KPI definition was searched to educe
correlation among the convective and radiative heat fluxes (e.g. the energy variables), and air humidity
and temperature (e.g. the process variables). These variables from the production process accounted
globally for the reduction of dough water content as given by the air humidity variation.

4.2. KPI definition

First, the oven energy profiling was based on the specific EnPIs. Namely, the indicator was computed
as the ratio between the natural gas consumption per mass of baked bread, which is the order-zero
parameterization to assess the energy performance of the oven relative to the production volume. Table 1
shows the results obtained for the two tests as well as the by-section share.
It is worth noting that the use of a specific energy index resulted in a similar estimate of the SEC.
When increasing the ventilation settings (either re-circulation and extraction), for instance, the SEC
features an overall increase which is in the order of (10-5) mostly attributable to the oven ends (with a 2%
increase).

Table 1. EnPI for test A and B and percentage in each section of the oven.

Test EnPI (m3methane/n. bread) Section 1 (%) Section 2 (%) Section 3 (%)
A 1.7 x 10-4 37 29.5 33.5
-4
B 1.8 x 10 37.7 28.3 34

The non-parametric analysis is then performed by using the PCA approach. The analysi sentailed four
variables that most influence the process: air temperature (AT), humidity (H), convective (CHF) and
radiative (RHF) thermal heat fluxes. Figure 2, first, illustrates the results of the 4 variable PCA. Notably,
the legend refers to the sections of the oven. Fig. 2 a. shows the test A pattern of the first two PCs, while
Fig. 2 b shows the test B patterns. Air temperature and humidity increase during the baking process as
558 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561

given by the increasing PC1 range. Indeed, in all the tests, the patterns ascribable to the three sections of
the oven are distinct. The variance correlated with PC2, on the other hand, is bounded in the first two
sections (driven by the radiative flux) while it features a pattern discontinuity aligned with the RHF axis
(asymptote) in the final third section where the driving variable is the AT. It is worth of note that, in both
the tested conditions, RHF decreases in the second section of the oven. The farthest points from the
densest section of the graph are the time steps in which the bread undergoes the burner, as given also by
the high radiative heat flux.

Fig. 2. PCA performed on Test A (a) and Test B (b), variables selected: air temperature (AT), humidity (H), convective thermal heat
flux (CHF) and radiative thermal heat flux (RHF).

Fig. 3. PCA performed on Test A (a) and Test B (b). Variables selected: air temperature (AT), humidity (H), convective thermal
heat flux (CHF).
 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561 559

Concerning CHF, it decreases slowly along the first and second sections of the oven and quickly in the
third. The PCA represents the variation of the data into the system, not absolute value, so even if the
points in the second graph appear to be more scattered than in the first, in test B the values of CHF and H
are reduced but the variation of these values increases.
Owing to the found discontinuity in the PCA bi-plot representations, and because of the inefficacy of
the KPI related analysis in view to highlight the A versus B differences, a second PCA was computed
limiting the number of energy variables to CHF only. Fig.3then, presents the results of the 3 variable
PCA, respectively Fig. 3.a shows the test A and Fig. 3.b the test B patterns.

Fig. 4. KPI performed on Test A (a) and Test B (b).

Fig. 5. Comparison of KPIs in linear scale (a) and semi-logarithmic scale (b).
560 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561

The analysis of relationships existing between the PCA variables, brought to the identification of new
latent key variable influencing the process. The key variable detected are the first two components of
dataset investigated, they express the maximum amount of variance in the data. These variables are able
to synthetize the chemical-physical phenomenon of the process enclosed within a wide set of information
very different. The KPI selected is the ratio between the two principal components (PC1/PC2), in order to
obtain a single indicator able to provide an outline of the energy use and the phenomena related to the
baking process along the different sections of the oven. Fig. 4 shows the patterns of the KPI identified,
respectively for the base-line test A (Fig. 4.a) and for the test B (Fig. 4.b). As visible, the KPIs follow the
trend of a tangent function, as it is a relationship between the abscissa and the ordinate of a quadratic
distribution. Interestingly, the middle section of the oven has a constant trend in both tests, while most of
the variation occurs in sections 1 and 3. This supports the view discussed in Fig. 3.
When comparing directly the behavior of the KPI in the two tests (Fig. 5), it is evident how the
proposed metric is able to distinguish between the different baking processes by returning diverse slope of
the KPI evolution diagram within the oven. Observing Fig. 5.a,which shows the functions corresponding
to each test in its entirety, are evident reduced peaks in test A if compared to test B. On the other hand
Fig. 5.b represent the KPIs in semi-logarithmic scale and reveal the different slope between KPIs.
For this reason, it could be inferred about the potentiality of using such a multi-variate KPI approach to
provide information about the efficiency-product quality compromise.

5. Conclusion

In this paper it is intended to identify a KPI able to supply information about the energy use of the
baking process. In particular it is analyzed the energy supplied to the dough during the baking along an
industrial oven. PCA is the methodology applied to detect the KPI. Through this approach it is been
possible to identify the relation between different variables measured along the baking process of the
dough. The analysis of these connections allowed to detect key variables to build a proper KPI.
The indicator was tested changing the ventilation setting of the oven. The KPI is able to provide a
track of the energy use during the baking into different sections of the oven. In this way, identifying the
proper operation mode and product quality it is possible to use this outline as a base line to control the
process.
An interesting development is to analyze the variations of temperature and height of the dough on time, in
order to correlate these informations, concerning the product quality, to the energy use.

Acknowledgements

The authors wish to express their gratitude to Federico Squillante and GrazianoUllucci at SED Soluzioni
for technical contribution.

References

[1] European Climate Foundation (ECF), RoadMap 2050, 2012.

[2] Consumi di energia elettrica in Italia, Database Terna; 2013
[3] Swortfiguer M.J, Dough absorption and moisture retention in bread, Baker Digest, 42 (4) (1968), pp. 42–44.
[4] Therdthai N, Zhou WB, Adamczak T. Optimisation of the temperature profile in bread baking. J Food Eng 2002;55:41–8.
[5] Probert D, Newborough M. Designs, thermal performances and other factors concerning cooking equipment and associated
facilities. Appl Energy 1985;21:81–222.
 Alessandro Corsini et al. / Energy Procedia 82 (2015) 554 – 561 561

[6] Therdthai N, Zhou WB, Adamczak T. Three-dimensional CFD modelling and simulation of the temperature profiles and
airflow patterns during a continuous industrial baking process. J Food Eng 2004;65:599–608.
[7] Purlis E. Bread baking: technological considerations based on process modelling and simulation. J Food Eng 2011;103:92–
102.
[8] Hadiyanto, Boom RM, Van Straten G, Van Boxtel AJB, Esveld DC. Multi-objective optimisation to improve the product
range of baking systems. J Food Process Eng 2009;32: 709–29.
[9] Khatir Z, Paton J, Thompson H, Kapur N, Toropov V. Optimisation of the energy efficiency of bread-baking ovens using a
combined experimental and computational approach. Applaied Energy 2013; p. 918-927.
[10] Khatir Z, Thompson HM, Kapur N, Toropov VV, Paton JB. The application of computational fluid dynamics and design
optimisation in the British breadbaking industry. In: Proceedings of the 8th international conference on CFD in the oil and gas,
metallurgical and process industries, Trondheim Norway; 2011.
[11] ISO 50001:2011(E): Energy Management System.
[12] Dongellini M., Marinosci C., Morini G.L, Energy audit of an industrial site: a case study, Energy Procedia 45, 2014, p. 424
– 433.
[13] Haridy S., Wu Z., Industrial Engineering and Management, Univariate and Multivariate Control Charts for Monitoring
Dynamic-Behavior Processes: a Case Study 2009 – 2(3): p. 464-498.
[14] Haridy S., Wu Z. and Castagliola P., (2011), Univariate and Multivariate Approaches for Evaluating the Capability of
Dynamic-Behavior Processes (Case Study), Statistical Methodology, 8, p. 185-203.
[15] A Corsinia, L. De Propris, S. Feudo, M. Stefanato; Assessment of a diagnostic procedure for the monitoring and control of
industrial processes; 7th International Conference on Applied Energy – ICAE2015.
[16] Principal Component Analysis Multidisciplinary Applications; Parinya Sanguansat; InTech; 2012.
[17] Y C Liang, H P Lee, S P Lim, W Z Lin, C G Wu; Proper Orthogonal Decomposition and its application .- Part I: Theory;
Journal of Sound and Vibration; 2002; pp 527-544.
[18] I T Jolliffe; Principal component Analysi; New York: Springer-Verlag Inc.; 1986.
[19] K.Esbensen, P. Geladi, Principal Component Analysis, Umei University, 1987.
[20] Yeh1 AB, Lin DK.J., McGrath R.N., Multivariate Control Charts for Monitoring Covariance Matrix: A Review,” Quality
Technology and Quantitative Management, 3, 415-436, 2006.
[21] Digitron-Italia Srl private communication.

Biography
Sara Feudo is an Environmental Engineer and PhD fellow in “Industrial and ManagementEngineering”at
Department of Mechanical and Aerospace Engineering, Sapienza Univerity of Rome. She deals with energy
efficiency and mathematical models for multi-variate analyses in industrial systems.

View publication stats