Performance measures based optimization of supply chain network resilience: A NSGA-II + Co-Kriging approach

•Maximization of supply chain network resilience through performance measures: unfilled demand and transportation cost.•Multi-Objective Stochastic Mixed-Integer Programming (MOS-MIP) is formulated.•Computationally cheaper NSGA-II + Co-Kriging approach handles MOS-MIP formulation effectively.•Enriched Pareto frontier for better decision making is obtained through Co-Kriging interpolation.•Variance plot provides information about degree of uncertainty associated with a decision point.

Increased vulnerability of supply chain networks, due to globalization of trade, has solicited attention of researchers and practitioners towards enhanced risk and disaster management. This has resulted in evolution of extant literature and new practices to construct resilient networks. Resilience is the ability of a network to regain its original state post-disaster. In this work, it is measured by the expected value of the fraction of demand that gets satisfied post-disaster. Most studies in literature capture resilience through qualitative dimensions. Even quantitative based researches, compute resilience through structural dimensions which characterize network density, complexity or excess resource availability. This has resulted in inadequate emphasis on two important performance measures of a supply chain network: the percentage of unfulfilled demand and the total transportation cost post-disaster.This work addresses above gap through a Multi-Objective Stochastic Mixed-Integer Programming (MOS-MIP) model with above two performance measures as objective functions. To address high computational complexity of MOS-MIP model, a two stage approach of NSGA-II + Co-Kriging is adopted. NSGA-II generates initial points of Pareto frontier which form input for surrogate modelling through Co-Kriging. As compared to conventional simulation, the proposed approach is computationally cheaper and can handle multi-objective formulation effectively. Co-Kriging quickly performs interpolation to provide enriched Pareto frontier. Additionally, it provides variance plot to define degree of uncertainty or confidence associated with accuracy of prediction of each point of Pareto frontier. Subsequently, managers can make informed choices by evaluating tradeoff between objective functions through enriched Pareto frontier with associated degree of confidence of prediction accuracy.