Over the past decades, companies have become more and more concerned with maximizing company profits as well as maximizing their customer service level by efficiently managing supply chains. This thesis focuses on three major supply chain decisions that are highly related to supply chain profit, namely facility location, safety stock, and lot sizing. The first decision is strategic in nature, as a part of the supply chain network design, while the other two can be classified as operational decisions. The current study attempts to deploy concepts borrowed from the Theory of Constraints (TOC) to maximize the overall throughput of the supply chain rather than maximizing the throughput of every process separately. By utilizing this concept, supply chain network design decisions should be made to minimize the amount of inventory at each region in order to reach the global optimization.
This thesis is intended to formulate, solve, and implement a mathematical model that incorporates the concepts of TOC into multi-echelon supply chain network design under demand uncertainty to determine the optimal location of facilities and the optimal lot size. The objective is to minimize the total cost across the supply chain network as well as to minimize the total inventory capacity at each facility. For that purpose, the researcher follows an incremental methodology in which a series of mathematical models are developed and solution algorithms are proposed and tested leading to a final comprehensive model. The first model focused on locating a new distribution center under demand uncertainty. It is, then, extended in the second model to consider the location of a new distribution center with varying inventory capacity. Another model is then applied to tackle the joint economic lot sizing problem (JELS) under demand uncertainty. The fourth model further integrates the facility location problem with the joint economic lot sizing problem under demand uncertainty. Finally, the fifth model utilizes the TOC in integrated facility location and joint economic lot sizing problems.
In order to solve these models, several computational intelligence optimization techniques are proposed, implemented, and compared: particle swarm optimization (PSO), gravitational system search (GSA), cuckoo search (CS), charged system search (CSS), non-dominated sorting charged system search (NSCSS), non-dominated sorting particle swarm (NSPSO) algorithm, and non-dominated genetic-II (NSGA-II).
Given the constrained nature of the problem, the thesis proposes and compares four methods (i.e. naive, penalty, penalty with feasibility rules, and heuristic) for handling problem constraints; moreover, it proposes and testes a mixed integer procedure (MI) for converting continuous stochastic algorithms into mixed integer stochastic algorithms.
The numerical analysis of results is accordingly presented, including comparisons with models established in the literature. In general, results show that applying the TOC reduces the total cost of the whole supply chain by about 50%. Results also demonstrate that solving the multi-objective problems by using multi-objective stochastic algorithms is better than solving them by using single objective stochastic algorithms, thereby reducing the total cost more than 80% and reducing the total inventory more than 18%; the best algorithm is, thus, the NSPSO with 99.71% of its final solutions ranked on the first front. Additionally, using varying inventory reduces the total inventory amount by about 20%. The suggested penalty with feasibility rules constraint handling method is considered better than the penalty or feasibility rules alone. The findings also assert that the CSS algorithm is the best algorithm for solving JELS problems.
]