CONFIDENTIAL. Limited circulation. For review only.

Multi tier supplier selection for a sustainable global supply chain

N. Viswanadham and A. Samvedi

Abstract: Using the green supply chain supply In this paper, our approach is first to present in
ecosystem consisting of the supply chain, resources section 2, the ecosystem map with four elements
involved, the government and social factors and the that completely describe all the features of a green
delivery mechanism, we formulate the risk and supply chain. We present the performance
performance criteria as qualitative as well as
measures and the risks that are relevant for green
quantitative measures. Then we solve the multi tier
sustainable supplier selection problem using grey supply chain networks using the ecosystem
relational analysis approach. framework. In section 3, we present the grey
relational analysis approach and then show the
I. INTRODUCTION applicability of the approach to such problems by
solving a suitable example problem in section 4.
The most often quoted definition of sustainability is Finally we conclude in section 5 giving few future
“development that meets the needs of the present research directions.
without compromising the ability of future
generations to meet their needs.” Organizations are II. THE ECOSYSTEM
implementing sustainable supply chain policies due
to either reactive regulatory reasons or rising prices Figure 1 shows the ecosystem of a green supply
of energy and raw materials, to proactive strategic chain comprising of a) forward-backward supply
and competitive advantage reasons. Green supply chains, b) resources, c) institutions, and d) delivery
chain emphasizes minimum consumption of technologies and mechanism. The green supply
resource and energy, and the minimum green house chain has both forward and backward supply
gas emissions. Green supply chain adds a role of chains. The reverse features add additional
recycler function, the reuse of products or parts, complexity to the supply chain design. The
and the recycling of material and energy, which delivery mechanisms include inbound, outbound
forms the closed-loop of material flow. The and reverse logistics components. The reverse
performance of the green supply chain is judged in logistics is a new feature in green supply chains.
terms of energy consumption, percentage of The transport mode selection, outsourcing or
product that is recoverable and its cost and time, owning the fleet, JIT or Inventory management are
toxic and hazardous materials used or generated, all decisions that affect the contribution of the
Green House Gases (GHG) generated etc [1]. green house gases. Institutions are playing a major
role in enforcing and enabling the compliance of
There is lot of literature on green supply chain environmental regulations. There are three
networks including excellent survey papers by [1] mechanisms to achieve the reduction in emissions:
and [9]. There are MIP formulations of selection of carbon pricing, clean development mechanism
suppliers in green supply chain networks [5, 7 and (CDM), and joint implementation (JI) [11]. Pricing
8]. There are also AHP and grey net work carbon has become widely acknowledged as a
formulations [4]. Despite the clear motivation significant catalyst in international efforts to reduce
regarding green supply chain management, there is greenhouse gas emissions. There are two
no established clear cut approach on how to do it mechanisms for delivering carbon prices: carbon
[10]. This unstructured approach then brings in tax and carbon trading. While both are institutional
different kinds of new risk to the supply chain enforcements, carbon trading is a market based
which are in addition to the risks the chain already approach that makes carbon a tradable utility and
faces. Thus deploying a green initiative is not risk- hence a vital resource in the ecosystem [2]. In
free and hence a good risk management is also addition to GHG emissions, industries significantly
needed [12]. influence the environment in the use of raw
materials, energy, water, and land. The carbon
N. Viswanadham is INAE Distinguished Professor at trading markets have made carbon also a resource
Department of Computer Science and Automation, Indian though it is emission rather than consumption that
Institute of Science Bangalore, Pin – 560012, India, email:
[email protected]
is priced. In a global supply chain with many
facilities, one can judiciously choose the facilities
A. Samvedi is a researcher at Department of Computer Science in different regions such that the emissions can be
and Automation, Indian Institute of Science Bangalore, Pin – traded among subsidiaries by balancing carbon
560012, India, email: [email protected]
reductions with economic justifications.

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.
 CONFIDENTIAL. Limited circulation. For review only.

Figure 1: Green Supply Chain Ecosystem [11]

PERFORMANCE AND RISK shortcomings in the upstream tier of the chain will
propagate quickly and will have negative
The commonly used metric of greenness is carbon consequences. Thus it becomes important to test
footprint. A carbon footprint is the total set of the upstream subchains of these alternatives than
greenhouse gas emissions caused directly and just these alternatives. Figure 2 shows in detail the
indirectly by an individual, organization, event or supplier selection process.
product, expressed as carbon dioxide equivalent. In
general, the carbon footprint should be measured C1) Cost: The cost of products that the company
over the lifecycle of manufacturing, transportation, will have to bear if it chooses the particular
usage, and recycling or disposal [6]. In addition, supplier. This includes the purchase cost plus
green supply chains may include new processes other life cycle costs such as maintenance
related to repair, re-use, reverse logistics, re- costs, consultant expenses and other
manufacturing, and recycling. A green supply chain infrastructure costs.
ensures greenness in all the three business
processes: procurement, manufacturing, and C2) Lead time: How fast the particular supplier can
distribution. Green supply chain is essentially deliver the order.
taking into account factors like carbon footprint
along with conventional drivers like cost, quality, C3) Reverse logistics ready: This criterion tests the
and lead time in all the three business processes. A readiness of the supplier and its subchain for
green supply chain also bring several risks due to reverse logistics. That is do they use the parts
market uncertainty, unclear government regulations of used products for further manufacturing.
such as green transportation or promoting electric
cars and community campaigns as on nuclear C4) Carbon Foot Print: The full footprint of an
energy [11]. Table 1 list out these risks. organization encompasses a wide range of
emissions sources, from direct use of fuels to
SELECTION CRITERIA indirect impacts such as employee travel or
emissions from other organizations within the
The following criteria are chosen to be ones on supply chain. The carbon footprint as the
which the alternatives will be evaluated. As can be performance metric thus should be measured
seen, the last two criteria are specifically for the over the entire supply chain, from raw
upstream subchains of the alternatives. This is materials to final packing and delivery.
because of the fact that the affect of any

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.
 CONFIDENTIAL. Limited circulation. For review only.

Cost Lead Time

Carbon
footprint
Risk mitigation
Supplier readiness
Subchain green Green
Selection Risk
performance Performance
Criteria
Subchain risk
Reverse readiness
logistics ready

Supplier 1 Supplier 2 Supplier 3 ⋯ Supplier n

Figure 2: Combined performance risk based supplier selection hierarchy

Table 1: Risk classification by ecosystem approach [11]

Risk classification Risk subclassification
• Pollutants during the production
• Waste disposal, hazardous waste liability, Recycling
Planning and product • Unstable governance structure: opportunistic behavior by partners
related risks • Community influence on buying patterns & legislative process
• Perceived non-commitment by top management
• Product recalls and after effects
• Criminal / Insurance liability for violations and accidents
• Inability to identify and remedy non-compliance or risk problems
Resources related risk • Accidents due to a lack of training or awareness
• Public pressures to Ban or restrict raw materials due to non-compliance
penalties
• Political/social pressures for regulations
Institutional risk • Policy Changes
• Changes during elections
Due to delivery • Reverse logistics and Waste disposal infrastructure
infrastructure • Operational readiness for accidents

C5) Risk mitigation readiness: How best this III. PROPOSED METHOD
supplier can handle the situations when the
normal operations of a supply chain has been The theory and applications of grey systems has
disrupted and also how well it can help to been gaining increasing attention these days. This
mitigate such risks is in line with the growing recognition that the real
life decisions involve uncertainty. The method of
C6) Subchain green performance: How green are grey analysis is an approach to handle this
the operations of the upstream subchain of that uncertainty. The theory is based on the degree of
alternative. information known. The advantage of grey theory
over fuzzy theory is that grey theory takes into
C7) Subchain risk readiness: How risk prepared are account the condition of the fuzziness; that is, grey
the operations of the upstream subchain of that theory can deal flexibly with the fuzziness situation
alternative. [4]. It is used for the mathematical analysis of
systems with discrete data and uncertain
Table 2 explains how does a supply chain perform information. A MCDM methodology, based on
under green criteria when influenced by enablers of grey relational analysis has been proposed here.
a sustainable supply chain from all parts of The stepwise discussion on the proposed
ecosystem. methodology is given below:

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.
 CONFIDENTIAL. Limited circulation. For review only.

Table 2: Enablers and supply chain performance

Product & Value Trade Policies Delivery Resource
chain Infrastructure Management
Enablers Remanufacturing, Green Regulations, Reverse Logistics, Water, Green Power,
Closed loop Supply Carbon Trading, Gas Carbon Efficient Clusters, Carbon Trade
Chains, GHG Emission Limits, ISO Transportation, Smart
Certification & Green Warehousing
Cost High High High initial High initial investment,
investment, lower lower costs thereafter
costs thereafter
Lead time High High High High
GHG emissions Low if product is Low Low Low
refurbished or else
high
Green Less pollution per High SC motivation Green transportation Less polluting
Performance product for greening like electric vehicles resources used like
nuclear power

Step 1: Construct a committee of decision makers obtained reflect the maximum weighted value
and determine the IS alternatives and selection obtained in the data set for that attribute.
criteria to be considered. The latter has been
defined in Section 3.1 for this study. Step 7: The next step is to find the relational
coefficient between each alternative and the
Step 2: Obtain the decision matrix by identifying reference vector. Suppose there are two sequences
the criteria values as triangular fuzzy numbers or denoted by ! (") &  ("). Then
linguistic terms. ∆%& + '∆%*
#!, (") =
∆ (") + '∆%*
Step 3: Normalize the decision matrix as shown in where
Karsak (2002). The normalized values for fuzzy ∆ (") = |! (") −  (")|
data denoted by triangular fuzzy numbers as (aij, bij,
cij), for benefit-related criteria (B) and cost-related is the absolute difference between two comparing
criteria (C) are given as: sequences.
∆%* =  - ∆ (")
 =  ,  ,   ∆%& =  - ∆ (")
 −   −   − 
⎧ ∗ , , , ∈ are respectively the maximum and minimum values
⎪  −  ∗ −  ∗ −  of the absolute differences of the comparing
= ∗ ∗
⎨  −   −  ∗ −  sequences, and τ ε [0,1] is a distinguishing
⎪  ∗ , ∗ , ,  ∈ coefficient, the purpose of which is to weaken the
⎩  −   −  ∗ − 
effect of Δmax when it gets too big, and thus
enlarges the difference significance of the relational
where ∗ =   and  =   coefficient, RC0,i(k) reflect the degree of closeness
between the two comparing sequences at k. At Δmin,
Step 4: Convert the fuzzy triangular numbers into RC0,i = 1, that is, the relational coefficient attains
interval numbers using the α-cut. its largest value. While at Δmax, RC0,i, attains the
smallest value. Hence 0 < RC0,i <1, for all i.
Step 5: The weighted interval data is calculated
next. The priority weights of the criterion are Step 8: Grey relational analysis compares relations
multiplied to the performance values of alternatives of sequences in their appropriate metric spaces. If
under that criterion. The values are calculated by two sequences agree at each point then the grey
multiplying the minimum of the weight interval to relational coefficient is 1 everywhere and so their
the minimum of the performance interval to get the grey relational grade should be 1. Thus we can say
start value of the interval. Similarly the end value that the relational grade between two sequences is
of the interval is calculated by multiplying the two the mean of relational coefficient values at different
maximum. points. It is given as
2
1
Step 6: The reference number vector is found by #.!, = 0 #!, (")
using the optimal values from the weighted interval /
-34
values for every alternative. This can be explained
as the maximum value of all the starting values of The RG being the relational grade between two
intervals under a given criterion and also the sequences and ‘p’ being the length of the two
maximum of all the end values. The values sequences.

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.
 CONFIDENTIAL. Limited circulation. For review only.

Step 9: The alternative with the highest relational linguistic terms have been used. The fuzzy numbers
grade is chosen as the best alternative. gives us a range in which the value for the said
criterion should fall, as well as the most likely
The next section illustrates the proposed value. The linguistic terms (poor, fair, good, very
methodology numerically by applying it to a good) are represented such, as they make decision
problem set suggested in a previous study. makers more comfortable in providing their
assessments to these tangible criteria. They can be
IV. NUMERICAL ILLUSTRATION converted to fuzzy triangular numbers for the
calculation purposes. The membership value graph
After determining all the selection criteria and the shown in Figure 3 is used for this conversion. All
methodology to be applied, the supplier alternatives the criteria used here do not carry equal importance
which will be considered for evaluation are to the selection and thus weights must be assigned
finalized. A total of seven alternatives are used here to these criteria. This is done by asking the experts
for the illustration of the proposed methodology. about what they feel the weights should be and they
The criterion values that fill up the decision matrix are asked to register their answers in linguistic
have been picked up from [3]. The decision matrix forms (medium, high, very high). The membership
is shown in the Table 3. The assessments for value graphs for these have been shown in Figure
criterion cost and lead time have been given in 4.
fuzzy triangular numbers whereas for other criteria

Table 3: Data used to evaluate

Alternatives C1 ($ millions) C2 (days) C3 C4 C5 C6 C7
S1 (3.3,3.8,4.3) (40,42,45) G F G F P
S2 (2.9,3.1,3.5) (31,37,39) G G F VG F
S3 (4.5,5.0,5.5) (29,32,34) VG G G G G
S4 (6.1,6.4,6.9) (52,54,56) G F F G VG
S5 (2.9,3.6,3.9) (39,44,46) P G G P G
S6 (6.1,6.7,7.0) (57,59,64) VG VG F F F
S7 (3.0,3.4,3.8) (34,37,40) F P P G VG

normalization process makes the criterion values

unit free and comparable. The values in the
normalized decision matrix are still in form of
fuzzy triangular numbers. This should be converted
to intervals so that we can apply GRA to it. This is
done by applying α-cut method. The value of α =
0.5 is used here.

Figure 3: Membership functions for criteria values Table 4: Importance weights and aggregate weights
[3]. of the criteria [3].
Criteria DM1 DM2 DM3 Aggregate weights
Cost H VH H (0.600,0.800,1.000)
Lead time M M H (0.300,0.567,0.867)
Risk
mitigation H VH M (0.500,0.733,0.933)
readiness
Reverse
H VH M (0.500,0.733,0.933)
logistics ready
Carbon
M M VH (0.400,0.667,0.867)
footprint
Subchain
Figure 4: Membership functions for importance green H H VH (0.600,0.800,1.000)
performance
weights [3]. Subchain risk
H VH H (0.600,0.800,1.000)
readiness
The values given by three experts from the decision
making committee and the aggregated weights have The interval numbers are then multiplied with the
been tabulated in Table 4. The problem set defined corresponding criterion weights to get the weighted
above has been now solved using the proposed normalized numbers. These numbers are then used
methodology in the previous section. The values in to get the reference number sequence given in
the decision matrix are first normalized, before Table 5. Distances here are defined as the
applying the grey relational analysis to it. The

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.
 CONFIDENTIAL. Limited circulation. For review only.

maximum between each interval value and the values are tabulated in the Table 6. It lists out the
extremes generated. The maximum distance for average weighted distances for the alternatives. The
each alternative to the ideal is identified as the rankings based on these values are given as S3 > S2
largest distance calculation. The reference point is > S7 > S5 > S1 > S4 > S6. The values also tell that
the minimum of all minima and maximum of all there is not much of a difference in S3 and S2.
maxima distance for each alternative. The reference Thus, if so required due to any reason, S2 can be
point is [0, 0.7805]. The maximum distance chosen over S3 without much deterioration in
between the reference point and each of the performance. The rankings are also matched up
weighted matrix values is given by the formula with those given in the paper from where the data
which gives us the relational coefficients. Here the has been picked up. The rankings are almost
value of resolving coefficient is 0.3. The average of similar and this validates the proposed
these weighted values gives us the total score of the methodology.
alternative and is used to rank the alternatives. The
Table 5: Reference number vector
C1 C2 C3 C4 C5 C6 C7
Max(Min) 0.3659 0.0000 0.2000 0.2000 0.0794 0.3000 0.3000
Max(Max) 1.0000 0.8100 0.9043 0.9043 0.8100 1.0000 1.0000

Table 6: Weighted distances to reference point

C1 C2 C3 C4 C5 C6 C7 Average
S1 0.7059 0.4791 0.7214 0.4633 1.0000 0.4384 0.3189 0.5896
S2 1.0000 0.8349 0.7214 0.7214 0.5654 1.0000 0.4384 0.7545
S3 0.3750 1.0000 1.0000 0.7214 1.0000 0.7007 0.7007 0.7854
S4 0.2308 0.3055 0.7214 0.4633 0.5654 0.7007 1.0000 0.5696
S5 0.8485 0.5029 0.3412 0.7214 1.0000 0.3189 0.7007 0.6334
S6 0.2308 0.2654 1.0000 1.0000 0.5654 0.4384 0.4384 0.5626
S7 0.8819 0.6693 0.4633 0.3412 0.3941 0.7007 1.0000 0.6358

V. CONCLUSION [4] Li, G.D., Yamaguchi, D. and Nagai, M. (2007). A grey-

based decision making approach to the supplier selection
problem. Mathematical and Computer Modelling, 46,
In this paper we use the ecosystem map for a green 573–581.
supply chain, to identify the risk and performance [5] Nagurney, A., and Nagurney, L.S. (2010). Sustainable
parameters in both qualitative and quantitative supply chain network design: A multicriteria
perspective. International Journal of Sustainable
forms and use these for supplier section. The Engineering, 3(3), 189-197.
method can be extended to include more risk and [6] Neuhoff, K. (2008). Tackling Carbon: How to Price
performance measures. Our contribution is the Carbon for Climate Policy. Report, University of
methodology. Several optimization and machine Cambridge.
[7] Paksoy, T., Özceylan, E. and Weber, G.W. (2011). A multi
learning techniques can be applied towards objective model for optimization of a green supply chain
selecting the suppliers. This method is immensely network. Transaction on Evolutionary algorithm and
useful in the current day scenario of supply chains continuous optimization ISSN: 2229-8711 Online
being subjected to various pressures from Publication, 84-96; www.pcoglobal.com/gjto.htm
[8] Ramudhin, A., Chaabane, A., and Paquet, M. (2010).
governments, resource shortages and logistics Carbon market sensitive sustainable supply chain network
issues. design. International Journal of Management Science and
Engineering Management, 5(1), 30-38.
Acknowledgement: The first author would like to thank INAE [9] Srivastava, S.K. (2007). Green Supply Chain
for all its support. Second author thanks Prof. Y. Narahari Management: A State of the Art Literature Review.
(Chairman CSA Dept., IISc) and DST for their support and International Journal of Management Reviews, 9(1), 53-
funds, which helped in successful completion of this study. 80.
[10] Vachon, S. and Klassen, R.D. (2006). Extending green
REFERENCES practices across the supply chain: the impact of upstream
and downstream integration. International Journal of
[1] Beamon, B.M. (1999). Designing the green supply chain. Operations & Production Management, 26 (7), 795–821.
Logistics Information Management, 12 (4), pp. 332 – 342. [11] Viswanadham, N. and Kameshwaran, S., (2013).
[2] Capoor, K. and Ambrosi, P. (2008). States and Trends of Ecosystem Aware Global Supply Chain Management.
the carbon market. The World Bank. World Scientific Publishing.
[3] Karsak, E.E. (2002). Distance-based fuzzy MCDM [12] Wang, X., Chan, H.K., Yee, R.W., and Diaz-Rainey, I.
approach for evaluating flexible manufacturing system (2012). A two-stage fuzzy-AHP model for risk assessment
alternatives. International Journal of Production of implementing green initiatives in the fashion supply
Research, 40(13), 3167-3181. chain. International Journal of Production
Economics, 135(2), 595-606.

Preprint submitted to 9th IEEE International Conference on Automation

Science and Engineering. Received March 9, 2013.