Importance measures in strategic-level supply chain risk management

1
Importance measures in strategic-level supply
chain risk management

• Anssi Käki
• Ahti Salo
• Department of Mathematics and Systems
  AnalysisSchool of Science, Aalto University,
  Finland

2
Introduction

• Diagnosis of risks and evaluation of risk
  mitigation strategies is difficult in large
  supply networks
• Numerous nodes (suppliers, tiers)
• Many uncertainties (demand, quality, lead time)
• High level of dependency (disruptions at
  suppliers suppliers supplier)
• We present how supply network disruptions can be
  evaluated with Probabilistic Risk Analysis (PRA)
  and Bayesian networks
• How to recognize, group, and prioritize risk
  factors?
• How to visualize risks?

3
Executive summary
Risk importance of each supplier illustrated
Material supplier network for Honda Accord
center console1
1 Network adapted from Choi and Hong (2002), Kim
et al. (2011)
4
Why Probabilistic Risk Analysis (PRA) for Supply
Networks?

• Supply chain risks can be captured with
  optimization models
• Stochastic optimization for minimizing expect
  cost under known probability distributions
• Robust optimization for a guaranteed outcome
  without much assumptions of uncertainty
• Tailored for specific decision situations e.g,
  facility location or supplier selection
• Probability based diagnostic analysis serves
  different purposes
• Not focused on particular decisions increases
  visibility and understanding of the whole
• Allows modeling substantially large networks
• Models are not black boxes ? Comprehensible for
  management

Review of optimization models for disruption
management Snyder et al. (2010)
5
PRA importance measures for prioritization A
Fussell-Vesely example
10
S1
10
S3
10

• Probability of disruption at supplier i Pr(Fi)
  10.0 ? Probability for network disruption
  Pr(Fs) 2.2

S2
10
10

• Fussell-Vesely measures the decrease in network
  disruption probability, if a supplier is not
  disrupted

S4
S5
Lower branch Upper branch
FV(S1) FV(S2) FV(S3) FV(S4) FV(S5)
9 9 91 45 45
2.2?2.0
Supplier S3 is the most important, then S4 and
S5, then S1 and S2
?
6
Different importance measures are used to support
different decisions

• There are many importance measures for various
  purposes we consider Fussell-Vesely (FV)
  Risk-Achievement-Worth (RAW)

?
?
The direct effect of supplier i for the network
disruption Fs
Defence in depth - the capability of the
network to resist a disruption at supplier i
FV RAW Potential for improvement Potential for degrading
High High Supplier, network No
High Low Supplier No
Low High Avoid disruptions, network No
Low Low No Supplier, network
Source of table van der Borst Schoonakker
(2001)
7
Bayesian networks can be used to model
probabilistic reliability networks

• Typical PRA methods use logic gates to describe a
  system this can be too rigorous for supply
  chains
• Bayesian network consists of a causality graph
  and conditional probability tables

Logic or-gate Bayesian network
Pr(JFC OK J3 and CVTWood OK) 100 95
Pr(JFC OK J3 or CVTWood OK) 100 50
Pr(JFC OK J3 and CVTWood disrupted) 0 5
Logic diagram Bayesian network
Pr( CVTWood OK) 95 95
8
The Honda Accord center console network

• The Accord net is translated into a Bayesian net
• Assumptions
• A leaf supplier has 5 disruption probability
• Disruption at a parent supplier leaves a 50
  survival probability (due to backup suppliers,
  inventories)
• The disruption probability of suppliers with
  multiple parents is proportional to amount of
  parents disrupted
• Importance measures are calculated for two
  scenarios
• As above
• As above, but with supplier J3 turning risky ?
  Disruption probability is updated from 5 to 50

9
Fussell-Vesely (no disruption at supplier)
First tier suppliers are critical
Scenario J3 becomes risky
For example FV(JFC)32.86FV(Emhart)1.01
Size and color indicate the importance measure
value
10
Risk Achievemet Worth (certain disruption)
Parent supplier CVT is critical
Scenario J3 becomes risky
For example RAW(JFC)3.37RAW(Emhart)1.11
Size and color indicate the importance measure
value
11
Key takeways from different measures

• Fussell-Vesely guides the prioritization of
  improvement actions at individual suppliers
• Improvements at 1st tier suppliers CVTAss and JFC
  increase reliability the most
• When J3 has reliability issues, improvements
  atJFC and J3 become a key priority
• Risk Achievement Worth can be used when improving
  network (design, other suppliers)
• A disruption at CVT (parent of three
  CVT-sub-suppliers) harms reliability the most ?
  Decreasing dependency on CVT is recommended

1.
2.
3.
12
Extensions of the approach

• Estimation of probabilities
• Expert judgment, estimation from statistical
  data, discrete-event simulation
• Dynamic modeling
• Inventory and delays work as supply chain
  buffers they are dynamic in nature
• Once-in-ten-years disruption that lasts 6 months
  vs. Once-a-year disruption that lasts 18 days ?
  Both have (yearly) disruption probability of 5
• Dynamic Bayesian nets and simulation can capture
  such dynamics
• Multi-stage models e.g., Full disruption
  50 capacity Full capacity
• Other importance measures, such as
  joint-importance

13
Conclusions

• Importance measures can be used for various
  purposes
• Fussell-Vesely when planning improvements at
  individual suppliers
• Risk Achievement Worth for changes in network
  design
• and the results can be illustrated in an
  intuitive risk map
• The approach is next validated in real
  applications

14
Thank you!
References

• Choi, T. Y. and Hong, Y. (2002). Unveiling the
  structure of supply networks case studies in
  Honda, Acura, and DaimlerChrysler. Journal of
  Operations Management, 20469493.
• Deleris, L. and Erhun, F. (2011). Quantitative
  risk assessment in supply chains a case study
  based on engineering risk analysis concepts. In
  Planning production and inventories in the
  extended enterprise. Springer ScienceBusiness
  Media.
• Kim, Y., Choi, T. Y., Yan, T., and Dooley, K.
  (2011). Structural investigation of supply
  networks A social network analysis approach.
  Journal of Operations Management, 29194211.
• Schmitt, A. and Singh, M. (2011). A Quantitative
  Analysis of Disruption Risk in a Multi-Echelon
  Supply Chain. Working paper. Center for
  Transportation and Logistics. Massachusetts
  Institute of Technology.
• Snyder, L, Atan, Z., Peng, P., Rong, Y., Schmitt,
  A. and Sinsoyal, B. (2010). OR/MS Models for
  Supply Chain Disruptions A Review. Working
  Paper.
• Van der Borst, M. and Schoonakker, H. (2001). An
  overview of PSA importance measures. Reliability
  Engineering and System Safety, 72 241-245.
• Zio, E. (2011). Risk Importance Measures. In
  Safety and Risk Modeling and Its Applications.
  Springer-Verlag London.

15
Appendix Tabular results