Supply chain network design under facility disruptions

Abstract

Key players in the supply chain, including manufacturers, retailers and distributors, have realized the value of comprehensive network planning in which they make detailed plans for constructing new facilities, expanding distribution networks, partnering with new suppliers, and other important logistics activities. During the design phase, many parameters in supply chain design problems are assumed to be fixed. However, the impact of the design decisions spans over a long horizon, during which many parameters such as costs, demands, and capacities will fluctuate. Therefore, it will be dangerous to neglect data uncertainties, since a little change in data input may lead to solutions which are far from optimal in the long run. Possible fluctuations and the reactive strategies have to be taken into account in order to cope with the uncertain environment. Many techniques have been derived to deal with optimization problems with uncertainties, such as sensitivity analysis, stochastic programming methods, robust optimization and so on. Sensitivity analysis procedures are usually tedious to implement, while stochastic programming methods often lead to objective functions that are hard to evaluate. The theoretical framework of robust optimization has been well developed in recent years. It has received more attention in both academy and industry. This thesis studies a variety of problems on designing supply chain networks which are able to achieve well performance in uncertain environment, with methods derived based on recent development in robust optimization techniques. Disruptions represent a form of uncertainty which usually causes capacity loss, transportation blockage, price inflation and other fluctuations to supply chain networks. In this thesis, we first study a strategic supply chain management problem to design reliable networks that perform as well as possible under normal conditions, while achieving relatively well performance when various forms of disruptions strike. We present a mixed-integer programming model whose objective is to minimize the nominal cost (the cost when no disruptions occur) while reducing the disruption risk by applying the p-robustness criterion (which bounds the cost in disruption scenarios). We demonstrate the tradeoff between the nominal cost and system reliability, showing that substantial improvements in reliability are often possible with minimal increases in cost. We also show that our model produces less conservative solutions than those generated by common robustness measures. We propose a hybrid metaheuristic algorithm that is based on genetic algorithms, local improvement, and the shortest augmenting path method to solve the problem. Numerical tests show that the heuristic greatly outperforms CPLEX in terms of solution speed, while still delivering excellent solution quality. The disadvantage of p-robust approach is that it allows less complete description of the scenario space, since the set of scenarios may grow exponentially large as the problem size increases. On one hand, only small problems can be solved due to limited computational power, which makes this model impractical in real world application, where supply chain networks are usually consisted of hundreds of facilities. On the other hand, conserving computational power by considering only a small portion of the total number of scenarios would harm the accuracy of our results. Traditional stochastic programs which are risk neutral in the sense that they consider optimization of expected system-wide cost, are also difficult to solve, since exact evaluation of the expected value is either impossible or prohibitively expensive. To cope with this computational difficulty, we adopt a Monte Carlo simulation based method called sample average approximation (SAA), to solve a stochastic p-robust logistic network design problem in which we minimize the expected total costs, while enforcing p-robust constraints for each scenario. SAA approximates the expected objective function of the stochastic problem by a sample average estimate derived from random samples. It usually results in MIP counterpart problems and can be then solved by deterministic optimization techniques. SAA not only approximates, but also produces confidence intervals on the problem's optimal objective values, which makes this method more attractive. We propose a method based on SAA to solve the stochastic robust model. Statistical lower and upper bounds are derived as well to evaluate the solution quality. Numerical test results show that high quality solutions can be obtained with reasonable computational efforts.

Date of Award	15 Feb 2012
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Leong Chye Andrew LIM (Supervisor)