Advertisement

PCA Method for Debottlenecking of Sustainability Performance in Integrated Biomass Supply Chain

  • Bing Shen How
  • Hon Loong Lam
Original Research Paper

Abstract

This paper presents a novel debottlenecking approach which incorporates principal component analysis (PCA) that can be used to identify sustainability bottlenecks at planning phase (it can be in the form of economic, environmental or social dimensions) and subsequently remove them. Debottlenecking at this preliminary stage (configuration of a system or a plant is yet to be confirmed) aims to reveal root causes that made a given system or design become unpreferable, and subsequently revamping it to improve its overall preferability. It is vital for better insight into a given system or design which enables accurate decision-making at conceptual design stage. The effectiveness of the proposed method is demonstrated by using a palm biomass supply chain case study in Johor, Malaysia. The results show that the proposed debottlenecking method is capable to identify the sustainability bottlenecks of the research problem easily and efficiently. Aside from this, the proposed debottlenecking approach is benchmarked with a P-graph-aided debottlenecking approach which was developed in the previous work. On top of that, the strengths and limitations of these debottlenecking methods are discussed in this paper. This debottlenecking approach is essentially a guide for decision-makers (e.g. researchers, design engineers, project engineers etc.) during planning phase of the biomass industry development.

Keywords

PCA Debottlenecking tool Sustainability bottlenecks Biomass supply chain AHP 

Nomenclature

Abbreviations

ADP

abiotic depletion potential

AHP

analytical hierarchy process

AP

acidification potential

ATP

aquatic toxicity potential

BOD

biological oxygen demand

CAPEX

capital expenditure

CART

classification and regression tree

COD

chemical oxygen demand

DLF

dry long fibre

DORA

Design Operability and Retrofit Analysis

EC

economic

EFB

empty fruit bunches

EN

environmental

FEFP

food-to-energy footprint

GWP

global warming potential

HTPE

human toxicity potential by either inhalation or dermal exposure

HTPI

human toxicity potential by ingestion

ISI

Inherent Safety Index

LF

land footprint

MILP

mixed integer linear programming

NP

nutrification potential

OPEX

operational expenditure

OHSAS

occupational health and safety assessment series

PC

principal component

PCA

principal component analysis

PEIP

palm oil eco-industrial park

POCER

Postgraduate Colloquium for Environmental Research

POCP

photochemical ozone creation potential

SBSCM

sustainable biomass supply chain

SC

social

TTP

terrestrial toxicity potential

WF

water footprint

Indices

a

index for pollutants

b

index for variables

i

index for biomass sources

j

index for processing hubs

k

index for customers

l

index for intermediates

m

index for transportation modes from source i to processing hub j

m

index for transportation modes from processing hub j to customer k

n

index for possible solutions

p

index for products

q

index for environmental categories

r

index for biomass

t

index for technologies to produce intermediates l

t

index for technologies to produce products p

u

index for social impacts

z

index for principal components

Parameters

AHub

total land area that covered by the infrastructure of the processing hub [m2]

C

pairwise comparison matrix in AHP

\( {C}_r^{\mathrm{Biomass}} \)

collection cost of biomass r [RM/t]

CConstruct

construction cost [RM]

\( {C}_t^{\mathrm{CAPEX}\_\mathrm{Tech}} \)

annual capital cost of technology t [RM/t]

\( {C}_{t^{\prime}}^{\mathrm{CAPEX}\_\mathrm{Tech}} \)

annual capital cost of technology t′ [RM/t]

CLand

land cost [RM]

CNP(L), CNP(U)

maximal and minimal net profit [RM/year]

\( {C}_t^{\mathrm{OPEX}\_\mathrm{Tech}} \)

annual operating cost of technology t [RM/t]

\( {C}_{t^{\prime}}^{\mathrm{OPEX}\_\mathrm{Tech}} \)

annual operating cost of technology t′ [RM/t]

\( {C}_p^{\mathrm{Prod}} \)

revenue obtained from product p [RM/t]

\( {\mathrm{Cap}}_{m\left(\mathrm{or}\ {m}^{\prime}\right)} \)

capacity limit (in terms of mass and volume) of transportation mode m or m

CRF

capital recovery factor

DCR

discount rate [%]

\( {\mathrm{EI}}_q^{(L)},{\mathrm{EI}}_q^{(U)} \)

maximum and minimum score for environmental impact q [t-eq/year]

Fr, i

amount of biomass r supplied in source i [t/day]

HTPE(L), HTPE(U)

maximum and minimum score for HTPE [ppm−1]

HTPI(L), HTPI(U)

maximum and minimum score for HTPI [kg/mg]

IChemical

chemical factors for inherent safety of process n

IProcess

process factors for inherent safety of process n

ISI(L), ISI(U)

maximum and minimum score for ISI

JC(L), JC(U)

maximum and minimum job creation [jobs]

LF(L), LF(U)

maximum and minimum score for land footprint [m2]

LD50

lethal dose that caused 50% death of rat specimens by oral ingestion [mg/kg]

LS

life span [years]

OHMax

maximum operating hour [h/day]

OPD

estimated annual working days [days/year]

PFatality(L), PFatality(U)

maximum and minimum value of transportation safety [%]

R

correlation matrix in PCA

S

standardised original data matrix in PCA

TLVTWA

time-weighted averages of threshold limit values of the materials [ppm]

V

average driving speed (or impact speed) [km/h]

\( {\mathrm{Weight}}_{m\left(\ \mathrm{or}\ {m}^{\prime}\right)} \)

vehicle size [t/vehicle]

WF(L), WF(U)

maximum and minimum score for water footprint [m3/year]

wEc, wEn, wSc

relative priority of each objective

wq, wWF, wLF

relative priority of each environmental indicators

wHTPE, wHTPI

relative priority of health aspect

wISI, wTr_S

relative priority of safety aspect

wJC

relative priority of job creation

\( {\overline{x}}_{\alpha } \), \( {\overline{x}}_{\beta } \), \( \overline{x} \)

mean of variables

\( {Y}_t^{\mathrm{Water}} \)

water consumption rate for each technology t [m3/t]

\( {Y}_{t^{\prime}}^{\mathrm{Water}} \)

water consumption rate for each technology t′ [m3/t]

Ψa, q

score of potential environmental impact of pollutant a at category q [t-eq/year]

Ψp, q

score of potential environmental impact of product p [t-eq/year]

\( {\varPsi}_q^{\mathrm{Fossil}} \)

score of potential environmental impact of fossil fuel [t-eq/year]

\( {\sigma}_{x_{\alpha }} \), \( {\sigma}_{x_{\beta }} \), σx

standard deviation of variables

λ

eigenvalue

Variables

Bj

binary variables to denote the selection of processing hub j

CGP

annual gross profit [RM/year]

CInv_Hub

annualised hub investment cost [RM/year]

CLabour

labour cost for transportation system [RM/day]

CMaintc

maintenance cost for transportation system [RM/year]

CMileage

mileage cost [RM/day]

CNP

net profit [RM/year]

CProc

annualised purchasing cost of transportation mode m [RM/year]

CTr

annual transportation cost [RM/year]

Contributionb, z

contribution rate of variable b on zth PC [%]

CVARz

percentage of total variance described by the first z PCs [%]

E

eigenvector in PCA

EIq

environmental impact from impact category q [t-eq/year]

\( {\mathrm{EI}}_q^{\mathrm{Emission}} \)

potential environmental impact due to pollutant emission [t-eq/year]

\( {\mathrm{EI}}_q^{\mathrm{Product}} \)

potential environmental impact caused by product [t-eq/year]

\( {\mathrm{EI}}_q^{\mathrm{Energy}} \)

potential environmental impact caused by energy generation and consumption [t-eq/year]

EnergyGen

generated energy [MJ/year]

EnergyImp

imported energy [MJ/year]

eb, z

eigenvector assigned to variable b on zth PC

FSub

total amount of fossil fuel that that can be substituted by the product [t/year]

Fa

emission of pollutant a [t/year]

\( {F}_{l,{t}^{\prime },j} \)

flowrate of intermediate l in processing hub j to technology t′ [t/day]

\( {F}_{m\left(\mathrm{or}\ {m}^{\prime}\right)} \)

amount of materials delivered (in terms of mass and volume) via transportation mode m or m

Fp

amount of product p produced [t/year]

Fp, j

amount of product p produced in processing hub j [t/day]

Fr, t, j

flowrate of biomass r in processing hub j to technology t [t/day]

HTPE

score for HTPE [ppm−1]

HTPI

score for HTPE [kg/mg]

ISI

inherent safety score

JC

job creation [jobs]

JCDirect

direct job creation [jobs]

JCIndirect

indirect job creation [jobs]

LF

land footprint [m2]

nTrip

number of trips travelled per day [trips/day]

OH

delivery lead time from starting location to the destination [h/trip]

PFatality

risk of pedestrian fatality [%]

w

priority scale matrix

WF

Water footprint [m3/year]

Y

projection matrix in PCA

xα, xβ, x

variables

xstandardised

standardised variables

λEc, λEn, λSc

degree of satisfaction of the biomass supply chain based on economic, environmental and social sustainability

λSCM

overall degree of satisfaction based on the sustainability performance of the biomass supply chain

Introduction

In the last decades, the growing international pressure (Mani et al. 2016) and expanding bioenergy demand (projected to reach 50 EJ by 2035 (IEA 2012)) have been the main driving forces for the recent transition of biomass industry towards sustainable management. In order to cope with this transition, sustainable biomass supply chain management (SBSCM) which concerns on the sustainable operational management of biomass-based material within an interconnected chain (e.g. biomass harvesting site, processing hubs, biomass storage, transportation, product distribution centre, etc.) which converts biomass into value-added products (e.g. biofuels, biochemicals, bioenergy, etc.) is proposed (Hong et al. 2016). In SBSCM, the supply networks are managed in a way that economic viability and social benefits are maximised while the environmental risks are kept at minimal (Wan Alwi et al. 2016).

To date, a number of researches have been conducted to design SBSCM. For instance, Lam et al. (2013) had presented a novel two-stage optimisation model for systematic design of green biomass supply chain. ‘Macro-stage’ focuses on the processing hub determination which aims to minimise both transportation cost and carbon emission, while ‘micro-stage’ concerns on technology selection which aims to maximise the total revenue. This work is further extended by How et al. (2016a) to integrate P-graph framework into the conventional mathematical programming approach. With the aid of this powerful combinatorial graphical optimisation tool, the overall computational efficiency is improved significantly. Lately, Ng et al. (2015a) introduced an algebraic method to solve the palm biomass supply chain problem. The use of quantitative cum graphical analysis in this method enables better performance overview of each possible design of the supply network. Tan et al. (2016) proposed an optimisation-based cooperative game framework for process integration within a palm oil eco-industrial park (PEIP). This PEIP is aimed to maximise the least satisfied objective (either economic or environmental dimension) based on max–min aggregation approach. Teo et al. (2017) then extended this framework to a hybrid model that combines superstructure-based optimisation approach and insight-based automated targeting approach. Apart from economic and environmental aspects, some academicians also integrate social sustainability in the supply chain model. Notably, Čuček et al. (2012) proposed a multi-criteria optimisation of regional biomass-to-energy supply chain through simultaneous maximisation of economic performance and minimisation of both environmental and social footprints. In that work, food-to-energy footprint (FEFP) is used as the social indicator to address the food versus fuel competition issue in the supply network. Ng et al. (2015b) integrated safety risk in designing sustainable biofuel and bioenergy networks within P-graph framework. The risk may be involved in logistics management, pre-treatment facilities and processing facilities. You et al. (2011) addressed the optimal design and planning of bioethanol supply chain with the consideration of sustainability objectives. In this model, the employment opportunity created throughout the lifetime of project is used to measure the social sustainability.

Aside from this, the problem of identifying bottlenecks and subsequently debottlenecking them is another significant topic of research. The term ‘bottleneck’ is defined differently at different phases of supply chain development (see Fig. 1). Most of the previous works focus on debottlenecking at operational-phase which design of a system or a plant is already existing. Debottlenecking at this phase is defined as a strategy of achieving desired performance of a system or plant (e.g. higher yield, purity or productivity), which is currently incapable of with the current design (Schneider 1997). For instance, Voudoris and Ariston Consulting (1996) developed a mixed integer linear programming (MILP) model which helped to detect production bottlenecks of a fine chemical plant and subsequently remove them to attain higher throughput and efficient scheduling. In addition, Ahmad and Polley (1990) presented a work that debottleneck heat exchanger networks in the case of throughput enhancement, with the use of pinch analysis technique. On the other hand, Alshekhli et al. (2010) used a computer-aided process simulation tool to identify possible debottlenecking strategies in a cocoa manufacturing plant for higher profitability and productivity. More recently, Yang et al. (2014) had proposed a classification and regression tree (CART) to explore the impact of process fluctuations on productivity and identify the underlying bottlenecks which limit the throughput in a biomanufacturing plant. Kasivisvanathan et al. (2014) introduced a heuristic framework for identifying and removing process-oriented bottlenecks (bottlenecks which restrict throughput, yield or efficiency) in a palm oil-based integrated biorefinery. Andiappan et al. (2017) then extended the previous work in the novel Design Operability and Retrofit Analysis (DORA) framework to retrofit a given tri-generation design based on benefit-cost ratio. Many of the key developments of debottlenecking approaches from the early 1990s until 2014 are reported in a review paper by Andiappan (2017).
Fig. 1

Debottlenecking at difference phases

Despite the decent contributions of the aforementioned works, none of them has considered the debottlenecking at planning phase which configuration of a system or a plant is yet to be designed. At this phase, debottlenecking refers to the process of revealing root causes that made a given solution become unpreferable, and subsequently revamping it to improve its overall preferability. The debottlenecking at this preliminary stage of design is vital for the better understanding of the potentials embedded in each solution (technology selection, logistics management, operation strategy, etc.), which enables accurate decision-making in selecting appropriate technologies or designs to ensure business sustainability (Foo 2017). On top of that, the term ‘bottleneck’ should not merely limit to economic-related barriers (e.g. throughput (Beer 2015), makespan (Goldratt and Cox 1984), process efficiency (Carlier and Rebai 1996)) but also related to other environmental-related barriers (concern on environmental risks, e.g. extensive land requirement (Oh et al. 2010), gigantic fuel consumption (How et al. 2016b), etc.) and social-related barriers (restriction on social factors, e.g. exposure to various social risks (Yatim et al. 2017), lack of domestic support (Foo 2015), etc.). Thus far, debottlenecking method which capable to identify a diverse form of bottlenecks is still limited. Lam et al. (2017) had developed a novel debottlenecking approach that incorporates P-graph framework to identify underlying bottlenecks that hinder biomass industry from attainment of sustainable paradigm. In this previous work, the sub-optimal solution is debottlenecked based on the overall sustainability performance (i.e. profit for economic dimension; emission or discharge rate of various wastes (gaseous, liquid and solid) for environmental dimension; Inherent Safety Index (ISI) (Hurme and Heikkilä 1998) for social dimension). Figure 2 shows the conceptual illustration of debottlenecking at planning phase. As illustrated, pathway II is less preferable due to the low sustainability performance for the secondary process. However, the optimality of the sub-optimal solution can be improved by removing bottlenecks via implementation of appropriate strategies (e.g. process integration, heat integration, emission abatement planning, regulatory policy amendment, etc.). It is somehow inevitable that most of the relatively new technologies (e.g. green technology) are not as competitive as compared to other conventional and commercialised technologies. Therefore, the proposed debottlenecking approach will serve as a guideline for users (academicians or industry players) in identifying the bottlenecks of these new technologies and subsequently improve them by removing the bottlenecks via some other existing strategies. If there were no existing strategies available to address the respective bottleneck, at the very least, these outcomes can be the guideline for the future research direction (i.e. researchers should put effort in this area). Please note that the term ‘optimal solution’ used in this work refers to the best possible solution (in terms of sustainability performance) which was obtained from an optimisation model (e.g. weighted-sum approach), while the term ‘sub-optimal solutions’ refers to the remaining alternative solutions (e.g. second best option, third best option and so on) which are not suggested by the optimisation model due to its lower favourability compared to the optimal solution.
Fig. 2

Conceptual illustration of debottlenecking at planning phase

In addition to the graph theoretic method (P-graph), principal component analysis (PCA) can be extended to a debottlenecking tool. In general, PCA is a multivariate statistical technique that is often used to reduce the dimensionality of data by converting a series of correlated variables into a set of uncorrelated variables known as principal components (PCs), without losing too much information (Aitchison 1983). As shown in Fig. 3, PCA involves the re-projection of data series on two new axes based on the variation. As a result, the 2D data series can now be redefined based on a single dimension, while ensuring the loss of information is kept at minimal. The earliest PCA application dates back to the early 1960s, when it has been served as an important and powerful tool in colour technology (Tzeng and Berns 2005). Since then, the utility of PCA has been burgeoned into diverse scientific fields (Wold et al. 1987). Some of the extended PCA applications from 2012 to the present are listed in Table 1. Furthermore, PCA technique has gained sufficient establishment, as demonstrated by its appearance in a text book (Flury 1988) and reference books (Jackson 1991; Diamantaras and Kung 1996). While these works are admirable, no study, to our knowledge, has extended PCA to debottlenecking of SBSCM.
Fig. 3

Dimensionality reduction example

Table 1

PCA applications reported in recent publications (from 2012 to present)

PCA application

Authors

Design of chemical supply chain

Pozo et al. (2012)

Energy flow and greenhouse gas emission analysis

Zafiriou et al. (2012)

Pedestrian detection

Nguyen and Kim (2013)

Cyber-attacks detection

Morita et al. (2013)

Modelling of turbulent combustion

Isaac et al. (2014)

Face recognition

Zhou et al. (2014)

Image compression

Dash et al. (2014)

Nuclear structure analysis

Al-Sayed (2015)

Magnetic resonance imaging noise estimation and denoising

Manjón et al. (2015)

Prioritisation of transit projects

Xu and Lin (2016)

Chiller sensor fault detection

Hu et al. (2016)

Dynamic response of commodity markets

Nobi et al. (2017)

3D palmprint identification

Bai et al. (2017)

Optimisation of biomass supply chain

How and Lam (2017a)

This paper represents the extended work of a former work which was recently presented in the 4th Postgraduate Colloquium for Environmental Research (POCER) 2017 (How and Lam 2017b). A novel debottlenecking approach that incorporates a PCA method is proposed as a promising alternative method that aims to identify various forms of bottlenecks in an integrated biomass supply chain during planning phase. In this work, PC scores are served as referencing indicators for the identification of bottlenecks. A case study of palm biomass supply chain in Johor, Malaysia is used to illustrate the effectiveness of the proposed method. Aside from this, since the proposed approach is an alternative method to the P-graph-aided debottlenecking approach (previously developed by Lam et al. (2017)) which was used to identify multiple forms of bottleneck, the debottlenecking results obtained from both methods are therefore presented and compared in this work.

Problem Statement

The problem described in this work aims to identify the bottlenecks during the planning phase for (i) optimal technology selection and (ii) optimal transportation design of an integrated biomass supply chain and subsequently debottleneck them based on the sustainability performance. It is formally stated as follows: given a set of biomass types r supplied from a set of sources i is delivered through a set of transportation modes m to a set of processing hubs j. Then, it is converted into a set of intermediates l and a set of products p via a set of technologies t and t′. Finally, products p will be delivered to a set of demands k through a set of transportation mode m′. Throughout the entire supply chain, a set of pollutants a is released to the environment and causes a set of environmental issues q; at the same time, these activities will lead to a set of social impacts u. The superstructure of the model is illustrated in Fig. 4.
Fig. 4

Generic superstructure of the proposed model (solid connectors: material flow, dotted connectors: non-material relations)

Method

The research flow diagram is shown in Fig. 5. Firstly, the sustainable performance of each possible solution in terms of economic, environmental and social dimensions is determined by using the formulated model based on the collected data. Then, the supply chain model is optimised with the aims of maximising the performance of all sustainability dimensions simultaneously. In this work, the weighted-sum approach which allows to convert a set of objectives into a single objective by assigning a preferred priority scale to each objective is opted to optimise the model. In order to systematically determine these priority scales, the analytical hierarchy process (AHP) which has been abundantly cited as a powerful multi-criteria decision-making tool is proposed in this work. Then, the sub-optimal solutions (i.e. other alternative solutions which are less favourable compared to the optimal solution) are debottlenecked based on a PCA-aided debottlenecking approach, while the debottlenecking results are compared and benchmarked with the P-graph-aided debottlenecking approach which was developed by Lam et al. (2017). Note that a brief description of this P-graph-aided debottlenecking approach is given in Section 5.5.
Fig. 5

Research method

The general flowchart for the proposed PCA-aided debottlenecking approach is shown in Fig. 6. After the sustainability performance of each solution is evaluated, a PCA study is then performed to analyse all these possible solutions. Based on the decision-makers’ interest, one of the sub-optimal solutions is selected for debottlenecking. For instance, it can be merely based on research purpose (i.e. interest in finding root cause which limit one’s favourability); it can also be based on companies’ goal (i.e. seeking of possible strategies to improve their products’ competitiveness). As already mentioned, the debottlenecking at preliminary stage of design is very important in revealing the full potentials embedded in each solution. The principal component scores of the selected sub-optimal solution are compared and benchmarked with the optimum solution obtained from the weighted-sum approach (i.e. with highest satisfaction). The PC that has the largest difference is notified as critical PC, while the variables that contribute a substantial portion to the responding critical PC is notified as critical variables. Fundamentally, the five most contributed variables are selected for the latter analysis, whereas other variables which contributed less portions are neglected since these variables generally pose a lower chance of being the bottlenecks of the supply chain. In fact, the beauty of the proposed method is that the potential bottlenecks are able to be detected systematically and efficiently based on the contribution rate obtained from PCA (in other words, analysis of all variables is not necessary). The critical variable that contributes the most will be the first variable to be improved. The remaining critical variables will be improved accordingly based on their contribution rate (from highest to lowest), until the selected sub-optimal solution is successfully debottlenecked (or all the critical variables are analysed). If the selected sub-optimal solution is not satisfied until this stage, the entire process will be repeated by analysing another critical PC.
Fig. 6

PCA-aided debottlenecking approach

Model Formulation

The description of the formulations for the sustainability evaluations, multi-objective optimisation approach and the proposed PCA-aided debottlenecking approach is given in the subsections below. Note that the definition of sets, variables and parameters of the model is summarised in Nomenclature.

Sustainability Evaluation

The evaluation of each sustainability dimension (i.e. economic, environmental and social sustainability) is presented in this subsection.

Economic Sustainability

Economic sustainability is evaluated based on the annual net profit, CNP [RM/year]. It considers three main economic components, i.e. annual gross profit, CGP [RM/year] (i.e. revenue obtained after deducting the costs associated with production and selling the products), annualised hub investment cost, CInv_Hub [RM/year] and annual transportation cost, CTr [RM/year]. It is defined as follows:
$$ {C}^{\mathrm{NP}}={C}^{\mathrm{GP}}-{C}^{\mathrm{Inv}\_\mathrm{Hub}}-{C}^{\mathrm{Tr}} $$
(1)
CGP is determined by using Eq. (2). From the equation, starting from the left, the first term determines the total revenue obtained from the product where Fp, j [t/day] refers to the daily production rate of product p produced from processing hub j, while \( {C}_p^{\mathrm{Product}} \) [RM/t] refers to the selling price for product p. The second term presents the total biomass procurement cost of biomass, where Fr, i [t/day] refers to the daily processed amount of biomass r in processing hub j, while \( {C}_r^{\mathrm{Biomass}} \) [RM/t] refers to collection cost for biomass r. The third and fourth terms measure the annual operational expenditure, while the last two terms determine the annual capital investment. Note that \( {C}_t^{\mathrm{OPEX}\_\mathrm{Tech}} \) [RM/t] and \( {C}_{t^{\prime}}^{\mathrm{OPEX}\_\mathrm{Tech}} \) [RM/t] represent the operating cost of technologies t and t′; \( {C}_t^{\mathrm{CAPEX}\_\mathrm{Tech}} \) [RM/t] and \( {C}_{t^{\prime}}^{\mathrm{CAPEX}\_\mathrm{Tech}} \) [RM/t] represent the capital cost of technologies t and t′; \( {F}_{l,{t}^{\prime },j} \) [t/day] and Fr, t, j [t/day] refer to the amount of biomass r and intermediate l fed into technology t and t′, respectively, while OPD [days/year] refers to the total annual operating days.
$$ {C}^{\mathrm{GP}}=\left\{{\sum}_p\left({\sum}_j{F}_{p,j}\times {C}_p^{\mathrm{Prod}}\right)-{\sum}_r\left({\sum}_i{F}_{r,i}\times {C}_r^{\mathrm{Biomass}}\right)-{\sum}_t\left({\sum}_r{\sum}_j{F}_{r,t,j}\times {C}_t^{\mathrm{OPEX}\_\mathrm{Tech}}\right)-{\sum}_t\left({\sum}_l{\sum}_j{F}_{l,{t}^{\prime },j}\times {C}_{t^{\prime}}^{\mathrm{OPEX}\_\mathrm{Tech}}\right)-{\sum}_t\left({\sum}_r{\sum}_j{F}_{r,t,j}\times {C}_t^{\mathrm{CAPEX}\_\mathrm{Tech}}\right)-{\sum}_t\left({\sum}_l{\sum}_j{F}_{l,{t}^{\prime },j}\times {C}_{t^{\prime}}^{\mathrm{CAPEX}\_\mathrm{Tech}}\right)\right\}\times \mathrm{OPD} $$
(2)
CInv_Hub is the fixed investment required to build a processing hub, i.e. land cost, CLand [RM] and construction expenses, CConstruct [RM]. It is annualised by using capital recovery factor (CRF) which converts a present value to a stream of equal annual cost over a life span (i.e. LS = 20 years) at a specified discount rate (i.e. DCR = 10%). It is represented in Eqs. (3) and (4):
$$ {C}^{\mathrm{Inv}\_\mathrm{Hub}}={\sum}_j{B}_j\times \left({C}^{\mathrm{Land}}+{C}^{\mathrm{Construct}}\right)\times \mathrm{CRF} $$
(3)
$$ \mathrm{CRF}=\frac{\mathrm{DCR}{\left(1+\mathrm{DCR}\right)}^{\mathrm{LS}}}{{\left(1+\mathrm{DCR}\right)}^{\mathrm{LS}}-1} $$
(4)
CTr is determined by using Eq. (5). It encompasses of four components, i.e. fixed investment cost required for the vehicle procurement, CProc [RM/year], annual maintenance cost (CMaintc [RM/year]), labour wages (CLabour [RM/day]) and fuel consumption cost (CMileage [RM/day]). Please note that the detailed calculation and description of each component have been reported in How et al. (2016a, b).
$$ {C}^{\mathrm{Tr}}=\kern0.5em {C}^{\mathrm{Proc}}+{C}^{\mathrm{Maintc}}+\left({C}^{\mathrm{Labour}}+{C}^{\mathrm{Mileage}}\right)\times \mathrm{OPD} $$
(5)
It is worth noting that Eq. (5) is subjected to two constraints, i.e. (i) delivery time constraint which restricts the total daily delivery time as compliance with regulation (EC) 561/2006 (EC 2014) (see Eq. (6)) and (ii) vehicle capacity constraint which limits the maximal delivered amount of each vehicle at one time (see Eq. (7)).
$$ {n}^{\mathrm{Trip}}\times \mathrm{OH}\le {\mathrm{OH}}^{\mathrm{Max}}\kern1em $$
(6)
$$ {n}^{\mathrm{Trip}}\ge \kern0.5em \left\lceil \frac{F_{m\left(\mathrm{or}\ {m}^{\prime}\right)}}{{\mathrm{Cap}}_{m\left(\mathrm{or}\ {m}^{\prime}\right)}}\right\rceil \kern1.75em \forall m\in M\ \left( or\forall {m}^{\prime}\in {M}^{\prime}\right) $$
(7)
where nTrip [trips/day] refers to the total number of trips travelled per day; OH [h/trip] and OHMax [h] refer to the delivery lead time from starting location to the destination and the maximum allowable daily driving hours; \( {F}_{m\left(\mathrm{or}\ {m}^{\prime}\right)} \) [t/day] represents the amount of materials (in terms of mass and volume) delivered via transportation mode m or m′, while \( {\mathrm{Cap}}_{m\left(\mathrm{or}\ {m}^{\prime}\right)} \) represents the capacity limit (in terms of mass and volume) of transportation mode m or m′. Since stopping in the mid-way is meaningless and undesirable, ceil functions ⌈…⌉ are therefore used to round-up nTrip to a nearest positive integer.

Environmental Sustainability

Environmental sustainability is assessed based on seven environmental impact indicators which are introduced by Heijungs et al. (1992) (i.e. global warming potential (GWP), acidification potential (AP), photochemical ozone creation potential (POCP), nutrification potential (NP), aquatic toxicity potential (ATP), terrestrial toxicity potential (TTP) and abiotic depletion potential (ADP)) and two environmental footprints (i.e. water footprint (WF) and land footprint (LF)). Note that all these indicators are minimised during the multi-objective optimisation.

In general, the environmental impact from impact category q, EI q [t-eq/year] encompasses of three components, i.e. potential environmental impact due to pollutant emissions, \( {\mathrm{EI}}_q^{\mathrm{Emission}} \) [t-eq/year]; potential environmental impact caused due to the characteristics of products, \( {\mathrm{EI}}_q^{\mathrm{Product}} \) [t-eq/year]; potential environmental impact due to energy generation and consumption. It is measured by using Eq. (8):
$$ {\mathrm{EI}}_q={\mathrm{EI}}_q^{\mathrm{Emission}}+{\mathrm{EI}}_q^{\mathrm{Product}}+{\mathrm{EI}}_q^{\mathrm{Energy}}\kern0.5em $$
(8)
\( {\mathrm{EI}}_q^{\mathrm{Emission}} \) measures the environmental impacts caused by the emission of pollutant a, F a [t/year] within the supply chain (emission during pre-treatment process, manufacturing process, transportation, etc.). It is expressed in Eq. (9), where Ψa, q [t-eq/t] refers to the score of potential environmental impact of pollutant a at category q.
$$ {\mathrm{EI}}_q^{\mathrm{Emission}}={\sum}_a\left({F}_a\times {\varPsi}_{a,q}\right)\kern4em \forall q\in Q $$
(9)
\( {\mathrm{EI}}_q^{\mathrm{Product}} \) considers the direct effect (environmental-burdening) and indirect effect (environmental-unburdening) of product p towards the environment. For instance, production of biofuels which can be used to substitute fossil-based fuels will lead to a positive effect on the environment (Čuček et al. 2012). It is determined by using Eq. (10). Note that the first term starting from the left refers to the direct effect, while the second term refers to the indirect effect, where F p [t/year] refers to the annual production rate of product p; \( {F}_p^{\mathrm{Sub}} \) [t/year] refers to the total amount of fossil fuel that that can be substituted by product p, while Ψp, q [t-eq/year] and \( {\varPsi}_q^{\mathrm{Fossil}} \) [t-eq/year] refer to the score of potential environmental impact of product p and fossil fuel respectively.
$$ {\mathrm{EI}}_q^{\mathrm{Product}}={\sum}_p\left({F}_p\times {\varPsi}_{p,q}-{F}_p^{\mathrm{Sub}}\times {\varPsi}_q^{\mathrm{Fossil}}\right)\kern2em \forall q\in Q $$
(10)
\( {\mathrm{EI}}_q^{\mathrm{Energy}} \) evaluates the direct environmental impact which is attributed by imported energy, EnergyImp [MJ/year] (unless the imported energy is renewable energy), and the environmental-unburdening effect of the generated bioenergy, EnergyGen [MJ/year]. It is formulated as follows:
$$ {\mathrm{EI}}_q^{\mathrm{Energy}}=\left({\mathrm{Energy}}^{\mathrm{Imp}}-{\mathrm{Energy}}^{\mathrm{Gen}}\right)\times {\varPsi}_q^{\mathrm{Fossil}}\kern2em \forall q\in Q $$
(11)

Note that all the potential environmental impact scores (i.e. Ψa, q, Ψp, q and \( {\varPsi}_q^{\mathrm{Fossil}} \)) are obtained from WAR GUI, build 1.0.17 which was developed by the US Environmental Protection Agency (WAR GUI 2011).

Aside from the seven impact categories, WF and LF are also used to evaluate the environmental performance. WF [m3/year] measures the total volume of fresh water used in the supply chain. It is determined by using Eq. (12), where \( {Y}_t^{\mathrm{Water}} \) [m3/t] and \( {Y}_{t^{\prime}}^{\mathrm{Water}} \) [m3/t] are the water consumption rate for technology t and t′ respectively.
$$ \mathrm{WF}=\left\{{\sum}_{t^{\prime }}\left({\sum}_l{\sum}_j{F}_{l,{t}^{\prime },j}\times {Y}_{t^{\prime}}^{\mathrm{Water}}\right)+{\sum}_t\left({\sum}_r{\sum}_j{F}_{r,t,j}\times {Y}_t^{\mathrm{Water}}\right)\right\}\kern0.5em \times \mathrm{OPD} $$
(12)
LF [m2] which measures the total land area that is covered by the infrastructure of the processing hub and AHub [m2] are used as indicators to reflect the environmental impact due to land use. It is defined as follows:
$$ \mathrm{LF}=\kern0.5em {\sum}_j{B}_j\kern0.5em \times {A}^{\mathrm{Hub}} $$
(13)

Social Sustainability

Social impacts including health and safety aspects in the processing hubs, transportation safety and job creation are considered in the evaluation of social sustainability. In this work, human toxicity potential (i.e. human toxicity potential by ingestion (HTPI) and human toxicity potential by either inhalation or dermal exposure (HTPE)) is used to evaluate the health aspect. In general, HTPI [kg/mg] was calculated for a chemical if it is existing as a liquid or solid under these conditions, while HTPE [ppm−1] is measured for a chemical if it is existing as a gaseous state at 0 °C and under atmospheric pressure (Young and Cabezas 1999). They are defined as
$$ \mathrm{HTPI}=\frac{1}{{\mathrm{LD}}_{50}} $$
(14)
$$ \mathrm{HTPE}=\frac{1}{TLV^{\mathrm{TWA}}} $$
(15)
where LD50 [mg/kg] refers the lethal dose that caused 50% death of rat specimens by oral ingestion, while TLVTWA [ppm] refers to the time-weighted averages of threshold limit values of the materials.
The safety aspect in the processing hub is evaluated by using the Inherent Safety Index which was introduced by Hurme and Heikkilä (1998). ISI considers two safety factors, i.e. chemical inherent safety factor, IChemical (e.g. heat of reactions, chemical interaction, flammability, explosiveness, toxic exposure, corrosiveness), and process inherent safety factor, IProcess (e.g. inventory, temperature, pressure, equipment safety). It can be measured by using Eq. (16):
$$ \mathrm{ISI}={I}^{\mathrm{Chemical}}+{I}^{\mathrm{Process}} $$
(16)
Other than plant inherent safety, transportation safety is also considered in this evaluation model. In this work, the relationship between driving speed, V [km/h], and the risk of pedestrian fatality, PFatality [%] which is determined by Rosén and Sander (2009), is used to measure the road safety. The relationship is defined as follows:
$$ {P}^{\mathrm{Fatality}}=\frac{1}{1+{e}^{6.9-0.090\ V}}\kern0.5em $$
(17)
Last but by no means least, job creation, JC [jobs] which measures the total employment opportunity created in the entire supply chain, is another significant social indicator. It is determined by using Eq. (18), where \( {\mathrm{JC}}_n^{\mathrm{Direct}} \) [jobs] refers to the employment which is directly related to biomass industry (e.g. operators, engineers, etc.), while \( {\mathrm{JC}}_n^{\mathrm{Indirect}} \) [jobs] refers to the jobs created outside the regional commercial enterprise (e.g. suppliers, logistic companies, retailers, etc.).
$$ \mathrm{JC}={\mathrm{JC}}^{\mathrm{Direct}}+{\mathrm{JC}}^{\mathrm{Indirect}} $$
(18)

Multi-Objective Optimisation Approach

In this work, the weighted-sum approach is opted as the optimisation method. It allows to transform a set of objectives into a single objective by assigning a preferred priority scale to each objective. To achieve this, the analytical hierarchy process is introduced. In short, AHP is a theory of measurement based on pairwise comparisons that relies on experts’ judgements (Saaty 2008). The priority scale of each objective is determined by solving an eigenvalue-eigenvector problem as shown in Eq. (19):
$$ C\ w=\lambda\ w $$
(19)
where C is a pairwise comparison matrix which shows the relative importance of each objective set by the decision-makers, λ is the eigenvalue of Eq. (19), while w represents a column matrix that shows the priority scale of each objective.
The objective function of this work is the overall degree of satisfaction based on the sustainability performance of the biomass supply chain, λSCM. It is defined as follows:
$$ \max\ {\lambda}^{\mathrm{SCM}}={w}^{\mathrm{Ec}}\times {\lambda}^{\mathrm{Ec}}+{w}^{\mathrm{En}}\times {\lambda}^{\mathrm{En}}+{w}^{\mathrm{Sc}}\times {\lambda}^{\mathrm{Sc}} $$
(20)
$$ {w}^{\mathrm{Ec}}+{w}^{\mathrm{En}}+{w}^{\mathrm{Sc}}=1 $$
(21)
where λEc, λEn and λSc refer to the degree of satisfaction of the biomass supply chain based on economic, environmental and social sustainability, respectively, while wEc, wEn and wSc are the priority scales for economic, environmental and social sustainability respectively.
λEc is determined by using Eq. (22). Note that the superscripted ‘U’ indicates the maximal value of a given variable, while the superscripted ‘L’ indicates the minimal value of a given variable. For instance, in Eq. (22), CNP(U) [RM/year] and CNP(L) [RM/year] are the maximal and minimal net profit that can be obtained respectively.
$$ {\lambda}^{\mathrm{Ec}}=\frac{C^{\mathrm{NP}}-{C}^{\mathrm{NP}(L)}}{C^{\mathrm{NP}(U)}-{C}^{\mathrm{NP}(L)}} $$
(22)
λEn is calculated by using Eq. (23), where w q , wWF and wLF are the priority scale assigned to each environmental indicator. Note that these priority scales can be determined by using AHP or either simply assumed that they are equally important. The first term of Eq. (23) determines the satisfaction level based on environmental impact at category q, while the last two terms measure the satisfaction level based on water and land usage respectively.
$$ {\lambda}^{\mathrm{En}}={\sum}_q\left(\frac{{\mathrm{EI}}_q^{(U)}-{\mathrm{EI}}_q}{{\mathrm{EI}}_q^{(U)}-{\mathrm{EI}}_q^{(L)}}\times {w}_q\right)+\frac{{\mathrm{WF}}^{(U)}-\mathrm{WF}}{{\mathrm{WF}}^{(U)}-{\mathrm{WF}}^{(L)}}\times {w}^{\mathrm{WF}}+\frac{{\mathrm{LF}}^{(U)}-\mathrm{LF}}{{\mathrm{LF}}^{(U)}-{\mathrm{LF}}^{(L)}}\times {w}^{\mathrm{LF}} $$
(23)
λSc is measured by using Eq. (24), where wHTPE, wHTPI, wISI, wTr_S and wJC refer to weightage assigned to each social indicator. Similarly, these priority scales can be determined by using AHP or either simply assumed that they are equally important. Note that the first terms of Eq. (24) evaluate the satisfaction level based on health aspect (human toxicity potential), the third and fourth terms measure the satisfaction level based on safety aspects (inherent safety in processing hub and transportation safety), while the last term evaluates the satisfaction level based on societal employment.
$$ {\lambda}^{\mathrm{Sc}}=\frac{{\mathrm{HTPE}}^{(U)}-\mathrm{HTPE}}{{\mathrm{HTPE}}^{(U)}-{\mathrm{HTPE}}^{(L)}}\kern1.75em \times {w}^{\mathrm{HTPE}}+\kern0.75em \frac{{\mathrm{HTPI}}^{(U)}-\mathrm{HTPI}}{{\mathrm{HTPI}}^{(U)}-{\mathrm{HTPI}}^{(L)}}\times {w}^{\mathrm{HTPI}}+\frac{{\mathrm{ISI}}^{(U)}-\mathrm{ISI}}{{\mathrm{ISI}}^{(U)}-{\mathrm{ISI}}^{(L)}}\times {w}^{\mathrm{ISI}}+\kern1.75em \frac{{P^{\mathrm{Fatality}}}^{(U)}-{P}^{\mathrm{Fatality}}}{{P^{\mathrm{Fatality}}}^{(U)}-{P^{\mathrm{Fatality}}}^{(L)}}\times {w}^{\mathrm{Tr}\_\mathrm{S}}\kern0.5em +\kern1.25em \frac{\mathrm{JC}-{\mathrm{JC}}^{(L)}}{{\mathrm{JC}}^{(U)}-{\mathrm{JC}}^{(L)}}\kern1.5em \times {w}^{\mathrm{JC}} $$
(24)

Debottlenecking Approach

A PCA-aided debottlenecking approach is proposed. As already mentioned, the PCA study is performed to analyse all the possible solutions after the sustainability evaluation. As a result, the original series of correlated variables (or indicators) are transformed into a set of uncorrelated variables, namely, principal components (PCs). Usually, the PCs of a data set are determined by solving an eigenvalue-eigenvector problem for the covariance matrix of the data set. However, the properties of PCA have some undesirable features when dealing with variables under different units of measurement (Jolliffe and Cadima 2016). Thus, in order to address this issue, correlation matrix, R which involves standardisation of dataset is used instead of covariance matrix (Al-Sayed 2015). The correlation between variables is defined as Eq. (25), where n refers to the number of possible solutions, x α and x β are the variables, \( {\overline{x}}_{\alpha } \) and \( {\overline{x}}_{\beta } \) are the mean value of these variables, while \( {\sigma}_{x_{\alpha }} \) and \( {\sigma}_{x_{\beta }} \) are the standard deviation of these variables.
$$ \mathrm{corr}\left({x}_{\alpha },{x}_{\beta}\right)=\frac{1}{n-1}{\sum}_1^n\left(\frac{x_{\alpha }-{\overline{x}}_{\alpha }}{\sigma_{x_{\alpha }}}\right)\left(\frac{x_{\beta }-{\overline{x}}_{\beta }}{\sigma_{x_{\beta }}}\right) $$
(25)
Then, an eigenvector, E which shows the magnitudes (or degree of importance) of each variable corresponded to the respective PCs, is computed by solving the eigenvalue-eigenvector problem shown in Eq. (26). Note that the first PC (PC1) is corresponded to the largest λ, indicating that PC1 describes the largest portion of the total variance of the proposed problem, followed by the second PC (PC2), and so on.
$$ R\ E=\lambda\ E $$
(26)
Finally, the sustainability performance of the solutions can now be redefined and represented in the PC space by using the PC scores (also named as factor scores) (Abdi and Williams 2010). In this work, these PC scores are served as the important indicators for debottlenecking (refer to Section 3). It is defined as Eq. (27), where S refers to the standardised original data matrix:
$$ \mathrm{PC}\ \mathrm{Score}=S\ E $$
(27)
Note that the standardised value of data, xstandardised is determined via Equation (28), where \( \overline{x} \) refers to the mean of the original data series, while σ x refers to the standard deviation of the original data series:
$$ {x}^{\mathrm{standardised}}=\frac{x-\overline{x}}{\sigma_x} $$
(28)
On top of that, contribution rate of each variable on each PC, Contributionb, z [%] is used to identify the critical variables of the problem, where eb, z denotes the eigenvector assigned for variable b on zth PC:
$$ {\mathrm{Contribution}}_{b,z}=\frac{{\left({e}_{b,z}\right)}^2}{\sum_b{\left({e}_{b,z}\right)}^2} \times 100\kern1.25em \forall b\in B,\forall z\in Z $$
(29)
Aside from this, a threshold cut of 90% is predefined to ensure the loss of information is kept at an acceptable and significant level. This can be formulated as Eq. (30), where CVAR z [%] refers to the percentage of total variance described by the first z PCs.
$$ {\mathrm{CVAR}}_z\ge 90\% $$
(30)

Case Study Description

The case study used in this work is adapted from the palm biomass supply chain design of the previous works (How and Lam 2017c). In short, this case study considers six empty fruit brunch (EFB) sources and 25 potential processing hubs (see Fig. 7); three available technologies for EFB valorisation (see Fig. 8); and five types of vehicles, i.e. 5-t trailers (m1), 10-t trailer (m2), 20-t trailer (m3), 32-t trailer (m4) and a jumbo tube trailer (m5) (see Table 2) are considered. In the latter discussion, the entire case study is decomposed into two stages: (i) technology selection, which aims to determine the optimal biomass conversion pathway for EFB, and (ii) transportation design which aims to determine the optimal location to set up processing hub and the optimal biomass allocation design for the biomass industry. Essentially, all biomass conversion pathways which are considered in this case study have been designed so that they are in accordance with the local regulatory standards (e.g. biological oxygen demand (BOD), chemical oxygen demand (COD), occupational health and safety assessment series (OHSAS), etc.). Debottlenecking is another level of improvement which aims to further enhance its performance. Other related data and the general description of a P-graph-aided debottlenecking approach which is used as a benchmark approach are stated in the subsections below.
Fig. 7

Geographical location for biomass sources, potential hubs and port (How and Lam 2017b)

Fig. 8

EFB conversion pathway

Table 2

Characteristic of each transportation mode (How et al. 2016b)

Vehicle

Weight limit [t]

Volume limit [m3]

Fuel consumption [L/km]

m1

5

22.77

0.213

m2

10

30.67

0.213

m3

20

43.20

0.235

m4

32

91.20

0.235

m5

4000

0.261

Economic Data

Table 3 shows the material cost for biomass feedstock, product and utility, while Table 4 tabulates the capital expenditure (CAPEX) and operational expenditure (OPEX) for each available EFB conversion technology. Note that other transportation-related economic data is listed in Table 5.
Table 3

Material cost (How et al. 2016a, b)

Material

Cost

EFB

10.8 [RM/t]

DLF

720 [RM/t]

Biochar

1260 [RM/t]

Bio-oil

1.1 [RM/L]

Syngas

0.60 [RM/m3]

Electricity (imported)

0.55 [RM/kWh]

Electricity (exported)

0.43 [RM/kWh]

Table 4

OPEX and CAPEX for each technology (How et al. 2016a, b)

Technology

CAPEX

OPEX

Dry long fibre (DLF) production

32.4 [RM/(t/h)]

66.6 [RM/(t/h)]

Gasification

150 [RM/(t/h)]

180 [RM/(t/h)]

Boiler

9.4 [RM/(t steam/h)]

Turbine

0.18 [RM/kW]

0.18 [RM/kW]

Table 5

Transportation-related data (How et al. 2016b)

Vehicle

Procurement price [RM/vehicle]

Maintenance cost [RM/km]

Fuel price [RM/L]

Wages rate [RM/h]

m1

70,000

0.18

1.90

10.00

m2

90,000

0.22

m3

125,000

0.34

m4

150,000

0.45

m5

170,000

0.45

Environmental Data

The pollutant emission rate for each supply chain activity is presented in Table 6. The environmental impact score for each category can be computed while the energy and water requirement for each activity are listed in Table 7. It is worth noting that the health indicators (i.e. HTPE and HTPI) can also be determined by using the data shown in Table 6.
Table 6

Pollutant emission rates [g/kg biomass] (How and Lam 2017c)

Activity

CO2

CH4

CO

N2O

NOx

SO2

HCs

COD

DLF production

60

Gasification

588.6

0.005

0.080

0.080

0.005

0.005

60

Combustiona

1585

5.820

102

3.11

25.406

0.02

Transportation [g/L fuel]

2600

0.560

276.8

0.028

4.408

0.017

6.851

Imported energy [g/kWh]

967

0.010

0.120

0.015

4.380

7.950

0.213

0.02

aCarbon emission from biomass combustion should be counted as 0 as it will not contribute to the net release of carbon in the carbon cycle

Table 7

Energy and water requirement (How and Lam 2017c)

Activity

Energy requirement [kW/(t/h biomass)]

Water requirement [m3/(t/h biomass)]

DLF production

220

Gasification

280

0.138

Combustion

0.00002

Transportation [g/L fuel]

Imported energy [g/kWh]

0.00007

Social Data

The ISI score and the amount of job created for each technology are listed in Table 8. Aside from this, the relationship between the vehicle size to the fatality risk of fatality is also considered in this case study. As shown in Eq. (31), it is expected that a vehicle with larger size, \( {\mathrm{Weight}}_{m\left(\ \mathrm{or}\ {m}^{\prime}\right)} \) [t] will lead to higher risk of pedestrian fatality since the larger vehicle carries greater kinetic energy compared to the smaller vehicle at the same speed (NHTSA 1997). Figure 9 shows the estimated risk of pedestrian fatality for each transportation mode (i.e. m1, m2, m3, m4, m5). Note that the red dotted line refers to the national speed limit in Malaysia (MOT 2014).
$$ {P}^{\mathrm{Fatality}}\propto 0.5\kern0.5em {\mathrm{Weight}}_{m\left(\ or\ {m}^{\prime}\right)}\ {V}^2\kern3.5em \forall m\in M $$
(31)
Table 8

Social data for each technology

Activity

ISI

Job creation

DLF production

12

0.002 [job/t fibre]

Gasification

34

0.004 [job/m3 bio-oil] (Maia et al. 2011)

Combustion

35

0.5759 [job/MW] (Maia et al. 2011)

Fig. 9

Risk of pedestrian fatality for vehicles in different size

AHP Result

The priority scale assigned to each objective is determined by using AHP technique. The pairwise comparison of each sustainability dimension is given in Table 9. Note that the weightage of each indicator within a given sustainability dimension is assumed equally important (e.g. GWP is equally important to other environmental impacts; HTPI is equally important to other social impacts).
Table 9

Pairwise comparison for each objective

 

EC

EN

SC

Priority scale, w

Rank

EC

1

2

2

0.50

1

EN

½

1

1

0.25

2

SC

½

1

1

0.25

2

CR = 0

Total = 1

 

EC, economic; EN, environmental; SC, social

P-Graph-Aided Debottlenecking Approach for Benchmarking

As mentioned in Section 3, the proposed debottlenecking approach is compared with a P-graph-aided debottlenecking approach. The detailed step-by-step description for this P-graph-aided debottlenecking approach has been reported in Lam et al. (2017). In general, a P-graph model which integrates the formulation of sustainability index is constructed (see Fig. 10). The construction of P-graph model can be divided into three subsequent steps. Firstly, the determined satisfaction level of each indicator is input to the model. Then, O-type vertices (horizontal bar) are used to represent the priority scale assigned to each indicator. To illustrate, since each indicator under the same sustainability dimension is assumed equally important, a conversion ratio of 0.5 is therefore set in the O-type vertices for the two environmental indicators. Subsequently, the priority scale for each sustainability objective (refer to Table 9) which is determined from AHP is inserted into the O-type vertices (in the bottom column).
Fig. 10

P-graph model for debottlenecking (Lam et al. 2017)

In this approach, the bottleneck of the sub-optimal solution is identified by comparing its sustainability performance of each dimension (i.e. economic, environmental and social) with the optimal solution. The dimension that has the largest difference in performance is notified as the potential bottleneck. Then, the least satisfied indicator under this sustainability should be the first targeted indicator to be improved (e.g. if environmental sustainability is noted as the potential bottlenecks, then the satisfaction of each environmental indicator, such as GWP, AP, etc. is analysed). Note that the improvements can be in any form (e.g. amendment on the government policy to promote cleaner production, process integration to improve the economic viability of the technology, etc.). The remaining indicators will be improved accordingly (from lowest score to highest score), until the sub-optimal solution is successfully debottlenecked (increase in ranking) or all the indicators are analysed. If the result is not satisfied until this stage, the entire process will be repeated by analysing another sustainability dimension.

Results and Discussions

As already mentioned, the entire case study is decomposed into two parts: (i) technology selection and (ii) transportation design.

Debottlenecking for Technology Selection

This stage aims to determine the optimal biomass conversion pathway for each biomass. The debottlenecking results obtained from the two debottlenecking approaches (i.e. PCA aided and P-graph aided) are shown below.

PCA-Aided Debottlenecking Approach

The sustainability performance of each solution is determined by using the mathematical model formulated in Section 4.1. Then, this data series is processed through PCA in order to reduce the data redundancy. Figure 11 shows that two PCs are sufficient to describe the data (since CVAR2 > 90%). Therefore, each solution is now represented in terms of PC1 and PC2 (see Table 10). To illustrate the debottlenecking process, EFB combustion is selected to be debottlenecked in this case study, since it is the least satisfied among the three available conversion pathways.
Fig. 11

PCA for technology selection

Table 10

PC score and ranking of each technology before debottlenecking

Technology

PC1

PC2

Rank*

Gasification

3.996

− 0.056

1

DLF production

− 2.166

− 0.968

2

Combustion

− 1.831

1.024

3

*Determined by using Eq. (20)

Figure 12 shows the contribution rate of each indicator on each PC. By comparing the PC scores for combustion to the currently optimal solution (i.e. gasification), it can be clearly seen that PC1 is the critical PC (differs the most), while profit and ISI are the two critical variables for PC1 (high contribution rate). Note that other indicators are not considered due to the insignificance (contribution less than 5%). Since both technologies have similar high ISI scores (i.e. above 30) due to the nature of the process (operate under high-temperature and high-pressure condition), therefore safety aspect is not the bottleneck for this case. In terms of economic dimension, gasification technology poses an ideal position compared to biomass combustion. The economic-unfavourability of biomass combustion is often due to the massive and continuous governmental support for the conventional energy source (Foo 2015). Therefore, regulatory amendments should be carried out in order to advocate the development of biomass industry. The proposed debottlenecking strategies are listed in Table 11 while the ranking of each technology is tabulated in Table 12. It shows that combustion technology is successfully debottlenecked (ranking increased) after the proposed debottlenecking strategies are implemented.
Fig. 12

Contribution rate of each indicator

Table 11

Debottlenecking strategies

Criteria

Initial

Proposed strategy

FiT (feed-in-tariffs)

< 10 MW: RM 0.31/kWh

10-20 MW: RM 0.29/kWh

20–30 MW: RM 0.27/kWh

Increase 50%

Government support for fossil energy

Subsidies, incentives and tax reduction

Eliminated

Remarks

Electricity cost from imported energy is assumed doubled

Table 12

Ranking of each technology after debottlenecking

Technology

Rank*

Gasification

3

DLF production

2

Combustion

1

*Determined by using Eq. (20)

P-Graph-Aided Debottlenecking Approach

The degree of satisfaction of each indicator is determined by using the P-graph model shown in Fig. 13. It is constructed by using P-Graph Studio v5.2.0.7 (P-Graph Studio 2017). The obtained solutions are ranked accordingly based on the sustainability performances (see Table 13).
Fig. 13

P-graph model used for debottlenecking (technology selection)

Table 13

Satisfaction level on each objective and ranking of each technology before debottlenecking

Technology

λ Ec

λ En

λ Sc

Rank*

Gasification

1.000

0.375

0.261

1

DLF production

0.461

0.623

0.750

2

Combustion

0.000

0.682

0.507

3

*Determined by using Eq. (20)

Similarly, the results show that economic sustainability is the key barrier for combustion technology, as λEc of combustion technology is lower compared to gasification (i.e. the best solution). Therefore, after implementing the same debottlenecking strategies suggested in Table 11, the P-graph model is updated with the new degree of satisfaction of each variable (see Table 14). Equivalent results are obtained for both debottlenecking approaches.
Table 14

Satisfaction level on each objective and ranking of each technology after debottlenecking

Technology

λ Ec

λ En

λ Sc

Rank*

Gasification

0.000

0.375

0.261

3

DLF production

0.000

0.623

0.750

2

Combustion

1.000

0.682

0.507

1

*Determined by using Eq. (20)

Debottlenecking for Transportation Design

This stage aims to determine the optimal location to set up a processing hub and the optimal biomass allocation design for the biomass industry. In this case study, the average driving speed, V during the transportation of materials is assumed to be either 50, 60 or 70 km/h. The debottlenecking results obtained from the two debottlenecking approaches (i.e. PCA aided and P-graph aided) are shown below.

PCA-Aided Debottlenecking Approach

Similarly, the sustainability performances of each solution are determined by using the mathematical model formulated in Section 4.1. The PCA results show that two PCs are sufficient to describe more than 90% of the total variance (see Fig. 14). Therefore, each solution is now redefined in terms of PC1 and PC2 (see Table 15). Note that the PC1 and PC2 mentioned in this section is different from the one mentioned in the previous section. The results show that a centralised-mode of operation (biomass from different sources are collected and processed in centre hubs) is less satisfied than the decentralised-mode of operation (biomass from different sources are processed in the local area). To illustrate, debottlenecking is carried out in order to improve the sustainability performances for a two-hub design (see Fig. 15). It is benchmarked with the optimal design, i.e. three-hub design (see Fig. 16).
Fig. 14

PCA for transportation design

Table 15

PC score and ranking of each design before debottlenecking

Design

PC1

PC2

Rank*

Single-hub design

4.731

0.491

5

Two-hub design

0.523

− 1.016

3

Three-hub design

− 1.323

− 0.910

1

Four-hub design

− 1.816

0.131

2

Five-hub design

− 2.116

1.303

4

*Determined by using Eq. (20)

Fig. 15

Two-hub design

Fig. 16

Three-hub design

By comparing the PC scores for these two designs, it can be clearly seen that PC1 is the critical PC (differs the most). As shown in Fig. 17, the critical indicators for PC1 are GWP, AP, POCP, NP, ATP, ADP and HTPE. The high emissions of pollutants through biomass transportation are mainly attributed by the low-density nature of the biomass. In order to mitigate the environmental and social impacts, several strategies can be implemented. For instance, using environmental-benign biodiesel as a substituent transportation fuel has been proven as a promising way to reduce the emission rate. Various studies have estimated that the use of biodiesel as transportation fuel will reduce the greenhouse gas emission by 62% (Ong et al. 2012). However, as a double-edge sword, the increased demand of biodiesel creates another green barrier as the harvesting of crops is driving deforestation and is likely to cause soil erosion due to the massive requirement of fertiliser (Lima et al. 2011). Therefore, in this case study, we proposed to pre-densify the biomass in order to reduce the volume to be transported (e.g. EFB is shredded and compacted) (see Table 16). As a result, the total emissions for a two-hub design are mitigated by 4.5%, while the total transportation cost needed is decreased by 1.5% (equivalent to RM 223,000/year), as the total number of trips required to deliver the biomass to the processing hub is reduced. The debottlenecking result is summarised in Table 17. Despite a three-processing-hub design not being the optimal solution after debottlenecking, its sustainability performances have been improved (ranking increased from third place to second place).
Fig. 17

Contribution rate of each indicator

Table 16

Debottlenecking strategies

Criteria

Initial

Proposed strategy I

Pre-densification

No

Yes

Remarks

Additional cost for pre-densification

RM 20/t EFB (AIM 2013)

Table 17

Ranking of each design after debottlenecking

Design

Rank*

Single-hub design

5

Two-hub design

2

Three-hub design

1

Four-hub design

3

Five-hub design

4

*Determined by using Eq. (20)

P-Graph-Aided Debottlenecking Approach

Again, the degree of satisfaction of each indicator is determined by using the P-graph model shown in Fig. 18. The obtained solutions are ranked accordingly based on the sustainability performances (see Table 18).
Fig. 18

P-graph model used for debottlenecking (transportation design)

Table 18

Satisfaction level on each objective and ranking of each design before debottlenecking

Design

λ Ec

λ En

λ Sc

Rank*

Single-hub design

0.000

0.143

0.000

5

Two-hub design

0.864

0.631

0.608

2

Three-hub design

1.000

0.826

0.879

1

Four-hub design

0.611

0.854

0.953

3

Five-hub design

0.200

0.857

1.000

4

*Determined by using Eq. (20)

Similar to the results obtained from the PCA study, both environmental sustainability and social sustainability are the key bottlenecks for the two-hub design (differs the most). Therefore, after implementing the same debottlenecking strategies suggested in Table 16, the P-graph model is updated with the new degree of satisfaction of each variable (see Table 19). Equivalent results are obtained for both debottlenecking approaches.
Table 19

Satisfaction level on each objective and ranking of each design after debottlenecking

Design

λ Ec

λ En

λ Sc

Rank*

Single-hub design

0.000

0.145

0.000

5

Two-hub design

0.884

0.680

0.656

2

Three-hub design

1.000

0.853

0.905

1

Four-hub design

0.623

0.864

0.963

3

Five-hub design

0.208

0.866

1.000

4

*Determined by using Eq. (20)

Comparison and Limitation

Table 20 summarises the comparison of the two debottlenecking approaches. In general, with the aid of the graph theoretic nature of a P-graph approach, users with minimal mathematical programming background are also able to develop a rigorous model for the research problems easily and determine the optimal and sub-optimal solutions efficiently (Lam et al. 2016). In contrast, debottlenecking via PCA approach required prior algebra knowledge of the users. However, with the aid of the user-friendly closed access Excel add-ins (XLSTAT 2017), users are able to perform PCA easily and efficiently. Aside from this, pre-processing of data is required for both approaches. The data series have to be converted into a correlation matrix (covariance matrix is used if the original variables are expressed in the same unit) in order to perform PCA, whereas a P-graph model has to be constructed in order to calculate the satisfaction level of each solution. Furthermore, the ranking of solutions has to be done manually for a PCA-aided approach, while P-Graph Studio will rank all the solutions automatically for a P-graph-aided approach.
Table 20

Comparison of PCA-aided and P-graph-aided debottlenecking approach

Criteria

PCA aided

P-graph aided

Programming background

Basic knowledge of PCA formulation is required

Minimal knowledge is required

Pre-processing step

Data has to be converted into covariance or correlation matrix

A P-graph model has to be constructed

Ranking of solutions

Rank manually by users

Rank automatically by software

Effectiveness

Capable to identify sustainability bottlenecks efficiently

Limitation

Unable to reveal all underlying bottlenecks

Despite both approaches posing a decent performance as a debottlenecking tool for the debottlenecking of integrated biomass supply chain, there are some underlying bottlenecks in the supply chain that are unable to be identified through the simulation model. For instance, the lack of understanding of risks associated with the biomass industry (includes regulatory risk, low bankability risk, social acceptance risk) is one of the key hurdles that impedes the development of the biomass industry in Malaysia (Yatim et al. 2017). On top of that, the absence of biomass monitoring and tracking system in Malaysia resulted in difficulties for robust assessments. Without these records, academicians can only show the theoretical biomass availability in their work, creating a wrong impression to the stakeholders since the actual availability and accessibility of the biomass is much less than expected (MIGHT 2013). In order to identify all these underlying barriers and debottleneck each of them, collaborative engagement of the stakeholders from diverse fields is necessary. By combining the debottlenecking suggestion obtained from the proposed debottlenecking approaches with the in-depth analysis conducted by the experts from different fields (e.g. policy makers, investors, grassroots citizens, academicians, researchers, etc.), all the stumbling blocks will be removed thoroughly.

Conclusion

This paper has presented an alternative debottlenecking approach which incorporates the PCA method that is able to identify sustainability bottlenecks in an integrated biomass supply chain. The main contributions are stated below:
  1. I.

    A novel mathematical model which considers economic sustainability, environmental sustainability and social sustainability is developed to optimise the sustainability performance of a biomass supply chain.

     
  2. II.

    In the proposed approach, PC scores are served as referencing indicators for the identification of bottlenecks. A case study in Johor, Malaysia is used to demonstrate the effectiveness of this approach.

     
  3. III.

    The proposed PCA-aided debottlenecking approach is benchmarked and compared with the P-graph-aided debottlenecking approach which was initially introduced by Lam et al. (2017).

     
  4. IV.

    The strength and limitation of these two debottlenecking approaches are discussed. It is found that collaborative engagement of experts from different fields is vital to remove the underlying bottlenecks which might be overlooked by the proposed debottlenecking approaches.

     

The demonstrated case study shows that the proposed debottlenecking approach is applicable to debottleneck a research problem efficiently. Decision-makers (e.g. researchers, design engineers, project engineers) can utilise this framework as a tool to efficiently identify the key bottlenecks that hinder the performance of a given process, system or design. However, this approach is still at its infancy. Therefore, it should be extended into a broader framework to test the robustness of the proposed method. This can be achieved by implementing the proposed method in various research problems (e.g. water pinch problem, travelling salesman problem, etc.). Aside from this, the evaluation model should be revised in future to consider several sustainability indicators which are currently being neglected. For instance, agricultural land-use is not considered in the model as all biomass considered in this work are crop residues and process wastes. Furthermore, the agricultural land-use is not originally aimed for biomass harvesting but for food production. However, this indicator should be considered when the biomass industry is commercialised, as additional land is required to harvest biomass in order to cope with the expanding biomass demand.

Notes

Funding Information

The authors would like to acknowledge the financial support from (i) the Ministry of Higher Education, Malaysia, via LRGS Grant (program code LRGS/2013/UKM/PT) and (ii) the University of Nottingham Malaysia Campus (Dean Scholarship).

Compliance with Ethical Standards

Conflict of Interest

The authors declare that they have no conflict of interest.

References

  1. Abdi H, Williams LJ (2010) Principal component analysis. Wiley Interdiscip Rev Comput Stat 2(4):433–459CrossRefGoogle Scholar
  2. Agensi Inovasi Malaysia (AIM) (2013) National Biomass Strategy 2020: new wealth creation for Malaysia’s biomass industry 2013. http://www.nbs2020.gov.my. Accessed 10 Jan 2017
  3. Ahmad S, Polley GT (1990) Debottlenecking of heat exchanger networks. Heat Recov Syst CHP 10(4):369–385CrossRefGoogle Scholar
  4. Aitchison J (1983) Principal component analysis of compositional data. Biometrika 70(1):57–65MathSciNetCrossRefzbMATHGoogle Scholar
  5. Al-Sayed A (2015) Principal component analysis within nuclear structure. Nucl Phys A 933:154–164CrossRefGoogle Scholar
  6. Alshekhli O, Foo DCY, Hii CL, Law CL (2010) Process simulation and debottlenecking for an industrial cocoa manufacturing process. Food Biprod Process 89:528–536CrossRefGoogle Scholar
  7. Andiappan V (2017) State-of-the-art review of mathematical optimisation approaches for synthesis of energy systems. Process Integr Optim Sustain 1(3):165–188CrossRefGoogle Scholar
  8. Andiappan V, Ng DKS, Tan RR (2017) Design Operability and Retrofit Analysis (DORA) framework for energy systems. Energy 134:1038–1052CrossRefGoogle Scholar
  9. Bai X, Gao N, Zhang Z, Zhang D (2017) 3D palmprint identification combining blocked ST and PCA. Pattern Recogn Lett 100:89–95CrossRefGoogle Scholar
  10. Beer JE (2015) Analysis and management of bottlenecks in supply networks: towards a structured approach to stabilization of inbound material flow. Dissertation, University of Stavanger, Stavanger, NorwayGoogle Scholar
  11. Carlier J, Rebai I (1996) Two branch and bound algorithms for the permutation flow. Eur J Oper Res 90:238–251CrossRefzbMATHGoogle Scholar
  12. Čuček L, Varbanov PS, Klemeš JJ, Kravanja Z (2012) Total footprints-based multi-criteria optimisation of regional biomass energy supply chains. Energy 44(1):135–145CrossRefGoogle Scholar
  13. Dash P, Nayak M, Prasad Das G (2014) Principal component analysis using singular value decomposition for image compression. Int J Comput Appl 93(9):21–27Google Scholar
  14. Diamantaras KI, Kung SY (1996) Principal component neural networks: theory and applications. Wiley, New YorkzbMATHGoogle Scholar
  15. European Commission (EC) (2014) Mobility and transport. European Commission. http://ec.europa.eu/transport/modes/road/social_provisions/driving_time/index_en.htm. Accessed 20 May 2015
  16. Flury B (1988) Common principal components and related models. Wiley, New YorkzbMATHGoogle Scholar
  17. Foo KY (2015) A vision on the opportunities, policies and coping strategies for the energy development in Malaysia. Renew Sust Energ Rev 51:1477–1498CrossRefGoogle Scholar
  18. Foo DCY (2017) Extended graphical technique for the evaluation of carbon dioxide emission reduction projects. Process Integr Optim Sustain 1:269–274.  https://doi.org/10.1007/s41660-017-0021-2 CrossRefGoogle Scholar
  19. Goldratt EM, Cox J (1984) The goal. North River Press, Croton-on-HudsonGoogle Scholar
  20. Heijungs R, Guinee JB, Huppes G, Lankreijer RM, Udo de Haes HA, Sleeswijk W (1992) Environmental life cycle assessment of product guide. Centre of Environmental Science, LeidenGoogle Scholar
  21. Hong BH, How BS, Lam HL (2016) Overview of sustainable biomass supply chain: from concept to modelling. Clean Technol Environ Policy 18(7):2173–2194CrossRefGoogle Scholar
  22. How BS, Lam HL (2017a) Novel evaluation approach for biomass supply chain: an extended application of PCA. Chem Eng Trans 61:1591–1596Google Scholar
  23. How BS, Lam HL (2017b) Debottlenecking for sustainable biomass supply chain: PCA approach. 4th Postgraduate Colloquium for Environmental Research 2017 (POCER 2017), 25–26 July 2017, Swiss Garden Hotel, Melaka, MalaysiaGoogle Scholar
  24. How BS, Lam HL (2017c) Integrated palm biomass supply chain toward sustainable management. Chem Prod Process Model 12.  https://doi.org/10.1515/cppm-2017-0024
  25. How BS, Hong BH, Lam HL, Friedler F (2016a) Synthesis of multiple biomass corridor via decomposition approach: a P-graph application. J Clean Prod 130:45–57CrossRefGoogle Scholar
  26. How BS, Tan KY, Lam HL (2016b) Transportation decision tool for optimisation of integrated biomass flow with vehicle capacity constraints. J Clean Prod 136(Part B):197–223CrossRefGoogle Scholar
  27. Hu Y, Li G, Chen H, Li H, Liu J (2016) Sensitivity analysis for PCA-based chiller sensor fault detection. Int J Refrig 63:133–143CrossRefGoogle Scholar
  28. Hurme M, Heikkilä AM (1998) Synthesis of inherently safe chemical processes by using genetic optimisation and case-based reasoning. In: Koikkalainen P, Puuronen S (eds) Human and artificial information processing. Finnish Artificial Intelligence Society, Espoo, pp 134–143Google Scholar
  29. International Energy Agency (IEA) (2012) World energy outlook. International Energy Agency, ParisGoogle Scholar
  30. Isaac BJ, Coussement A, Gicquel O, Smith PJ, Parente A (2014) Reduced-order PCA models for chemical reacting flows. Combust Flame 161(11):2785–2800CrossRefGoogle Scholar
  31. Jackson JE (1991) A user’s guide to principal components. Wiley, New YorkCrossRefzbMATHGoogle Scholar
  32. Jolliffe IT, Cadima J (2016) Principal component analysis: a review and recent developments. Phil Trans R Soc A 374:20150202.  https://doi.org/10.1098/rsta.2015.0202 MathSciNetCrossRefzbMATHGoogle Scholar
  33. Kasivisvanathan H, Tan RR, Ng DKS, Abdul Aziz MK, Foo DCY (2014) Heuristic framework for the debottlenecking of a palm oil-based integrated biorefinery. Chem Eng Res Des 92:2071–2082CrossRefGoogle Scholar
  34. Lam HL, Ng WPQ, Ng RTL, Ng EH, Abdul Aziz MK, Ng DKS (2013) Green strategy for sustainable waste-to-energy supply chain. Energy 57:4–16CrossRefGoogle Scholar
  35. Lam HL, Tan RR, Aviso KB (2016) Implementation of P-graph modules in undergraduate chemical engineering degree programs: experiences in Malaysia and the Philippines. J Clean Prod 136(Part B):254–265CrossRefGoogle Scholar
  36. Lam HL, Chong KH, Tan TK, Ponniah GD, Yeoh TT, How BS (2017) Debottlenecking of the integrated biomass network with sustainability index. Chem Eng Trans 61:1615–1620Google Scholar
  37. Lima M, Skutsch M, de Medeiros Costa G (2011) Deforestation and the social impacts of soy for biodiesel: perspectives of farmers in the South Brazilian Amazon. Ecol Soc 16(4):4CrossRefGoogle Scholar
  38. Maia J, Giordano T, Kelder N, Bardien G, Bodibe M, Du Plooy P, Jafta X, Jarvis D, Kruger-Cloete E, Kuhn G, Lepelle R, Makaulule L, Mosoma K, Neoh S, Netshitomboni N, Ngozo T, Swanepoel J (2011) Green jobs: an estimate of the direct employment potential of a greening South African economy. Industrial Development Corporation, Development Bank of Southern Africa, Trade and Industrial Policy Strategies, Cape TownGoogle Scholar
  39. Malaysian Industry-Government Group for High Technology (MIGHT) (2013) Malaysian biomass industry action plan 2020: driving SMEs towards sustainable future. Malaysian Industry-Government Group for High Technology, SelangorGoogle Scholar
  40. Mani V, Gunasekaran A, Papadopoulos T, Hazen B, Dubey R (2016) Supply chain social sustainability for developing nations: evidence from India. Resour Conserv Recycl 111:42–52CrossRefGoogle Scholar
  41. Manjón JV, Coupé P, Buades A (2015) MRI noise estimation and denoising using non-local PCA. Med Image Anal 22(11):35–47CrossRefGoogle Scholar
  42. Ministry of Transport Malaysia (MOT) (2014) Road safety plan for 2014-2020. Ministry of Transport Malaysia, PutrajayaGoogle Scholar
  43. Morita T, Yogo S, Koike M, Hamaguchi T, Jung S, Koshijima I, Hashimoto Y (2013) Detection of cyber-attacks with zone dividing and PCA. Procedia Comput Sci 22:727–736CrossRefGoogle Scholar
  44. National Highway Traffic Safety Administration (NHTSA) (1997) Relationship of vehicle weight to fatality and injury risk in model year 1985-93 passenger cars and light trucks. U.S. Department of Transportation, Washington, DCGoogle Scholar
  45. Ng WPQ, Promentilla MA, Lam HL (2015a) An algebraic approach for supply network synthesis. J Clean Prod 88:326–335CrossRefGoogle Scholar
  46. Ng RTL, Tan RR, Hassim MH (2015b) P-graph methodology for bi-objective optimisation of bioenergy supply chains: economic and safety perspectives. Chem Eng Trans 45:1357–1362Google Scholar
  47. Nguyen T, Kim H (2013) Novel and efficient pedestrian detection using bidirectional PCA. Pattern Recogn 46(8):2220–2227CrossRefGoogle Scholar
  48. Nobi A, Alam S, Lee JW (2017) Dynamic of consumer groups and response of commodity markets by principal component analysis. Physica A Stat Mech Appl 482:337–344CrossRefGoogle Scholar
  49. Oh TH, Pang SY, Chua SC (2010) Energy policy and alternative energy in Malaysia: issues and challenges for sustainable growth. Renew Sust Energ Rev 14:1241–1252CrossRefGoogle Scholar
  50. Ong HC, Mahlia TMI, Masjuki HH (2012) A review on energy pattern and policy for transportation sector in Malaysia. Renew Sust Energ Rev 16:532–542CrossRefGoogle Scholar
  51. P-Graph Studio (2017) P-Graph Studio. http://pgraph.dcs.uni-pannon.hu/template/. Accessed 3 Mar 2017
  52. Pozo C, Ruiz-Femenia R, Caballero J, Guillen-Gosalbez G, Jimenez L (2012) On the use of principal component analysis for reducing the number of environmental objectives in multi-objective optimization: application to the design of chemical supply chains. Chem Eng Sci 69:146–158CrossRefGoogle Scholar
  53. Rosén E, Sander U (2009) Pedestrian fatality risk as a function of car impact speed. Accid Anal Prev 41:536–542CrossRefGoogle Scholar
  54. Saaty TL (2008) Decision making with the analytic hierarchy process. Int J Serv Sci 1(1):83–98Google Scholar
  55. Schneider EF (1997) Debottlenecking options and optimisation. Stratus Engineering Inc, HoustonGoogle Scholar
  56. Tan RR, Andiappan V, Wan YK, Ng RTL, Ng DKS (2016) An optimization-based cooperative game approach for systematic allocation of costs and benefits in interplant process integration. Chem Eng Res Des 106:43–58CrossRefGoogle Scholar
  57. Teo KY, Ng TY, Wau CM, Liew JTS, Andiappan V, Ng DKS (2017) Hybrid optimisation model for the synthesis of centralised utility system in eco-industrial park. Process Integr Optim Sustain 1(1):33–57CrossRefGoogle Scholar
  58. Tzeng DY, Berns RS (2005) A review of principal component analysis and its applications to color technology. Color Res Appl 30(2):84–98CrossRefGoogle Scholar
  59. Voudouris VT, Ariston Consulting (1996) Mathematical programming techniques to debottleneck the supply chain of fine chemical industries. Comput Chem Eng 20(2):S1269–S1274CrossRefGoogle Scholar
  60. Wan Alwi SR, Klemeš JJ, Varbanov PS (2016) Cleaner energy planning, management and technologies: perspectives of supply-demand side and end-of-pipe management. J Clean Prod 136(Part B):1–13CrossRefGoogle Scholar
  61. WAR GUI (2011) Waste reduction algorithm: chemical process simulation for waste reduction. United States Environmental Protection Agency. http://www.epa.gov/chemical-research/waste-reduction-algorithm-chemical-process-simulation-waste-reduction. Accessed 26 Jul 2016
  62. Wold S, Esbensen K, Geladi P (1987) Principal component analysis. Chemom Intell Lab Syst 2:37–52CrossRefGoogle Scholar
  63. XLSTAT (2017) XLSTAT: your data analysis solution. http://www.xlstat.com/en/company/about-us. Accessed 20 Mar 2017
  64. Xu W, Lin W (2016) Selecting the public transit projects with PCA-DP technique: the example of Xiamen City. Transp Policy 46:56–71CrossRefGoogle Scholar
  65. Yang Y, Farid SS, Thornhill NF (2014) Data mining for rapid prediction of facility fit and debottlenecking of biomanufacturing facilities. J Biotechnol 179:17–25CrossRefGoogle Scholar
  66. Yatim P, Ngan SL, Lam HL, Er AC (2017) Overview of the key risks in the pioneering stage of the Malaysian biomass industry. Clean Techn Environ Policy 19(7):1825–1839CrossRefGoogle Scholar
  67. You F, Tao L, Graziano DJ, Snyder SW (2011) Optimal design of sustainable cellulosic biofuel supply chains: multiobjective optimization coupled with life cycle assessment and input–output analysis. Process Syst Eng 58:1157–1180Google Scholar
  68. Young DM, Cabezas H (1999) Designing sustainable processes with simulation: the waste reduction (WAR) algorithm. Comput Chem Eng 23:1477–1491CrossRefGoogle Scholar
  69. Zafiriou P, Mamolos AP, Menexes GC, Siomos AS, Tsatsarelis CA, Kalburtji KL (2012) Analysis of energy flow and greenhouse gas emissions in organic, integrated and conventional cultivation of white asparagus by PCA and HCA: cases in Greece. J Clean Prod 29-30:20–27CrossRefGoogle Scholar
  70. Zhou C, Wang L, Zhang Q, Wei X (2014) Face recognition based on PCA and logistic regression analysis. Optik Int J Light Electron Opt 125(20):5916–5919CrossRefGoogle Scholar

Copyright information

© Springer Nature Singapore Pte Ltd. 2018

Authors and Affiliations

  1. 1.Department of Chemical and Environmental EngineeringUniversity of Nottingham Malaysia CampusSemenyihMalaysia

Personalised recommendations