Optimization of Production Layout in Multi-Floor Industrial Buildings

In a world with limited real estate and a volatile economic landscape, flexible production chains are transitioning from a competitive advantage to an outright necessity for a variety of industries. One major step in the direction of high adaptability in production processes is the reliance on flexible production facilities. These facilities promise to maximize resilience against changing demands and consistently optimize, often conflicting, common key performance indicators such as material handling costs and total land required. In the last decades, the scientific community has increasingly considered so-called Facility Layout Problems (FLP) in a multi-floor context (MFFLP). As promising as digitally optimized multi-story production facilities seem, they also pose entirely new challenges. This thesis presents a novel port-based two-phase evolutionary algorithm to the MFFLP with Elevators (MFFLPE), considering five distinct objectives, dynamically placed elevators, and fixed-size departments. The algorithm operates on a discrete search space of possible cube locations and connections, and was written in the C# programming language. The procedure is evaluated both quantitatively and qualitatively. First, it is analyzed with respect to the extensive data recorded throughout simulation experiments conducted with two distinct input sets. This includes the comparison of two well-known fitness evaluation measures: Evaluation based on Pareto-dominance and evaluation based on the (unweighted) scalarization of objectives. Secondly, a supplementary questionnaire answered by a civil engineering expert helps to further investigate the strengths, weaknesses, and overall performance of the proposed algorithm. Our results indicate that the port-based approach is suitable to generate coherent multi-floor production layouts with satisfactory objective values, independent of the evaluation method utilized. While we observed differences between the two strategies in terms of objective completion, we also identified issues in the utilized normalization mechanism for the scalarization-based implementation. Thus, our data does not suggest any evaluation strategy to be more appropriate for the task at hand. As both optimization strategies pose their own advantages, we conclude that a hybrid approach seems the most promising for MFFLPEs, given proper objective normalization. Even though the results indicate that the scalarization-based implementation is faster for smaller problem instances, both evaluation measures generally required similar runtimes. Feedback provided by the expert revealed that certain objectives have yet to be investigated in future work. Lastly, future work needs to explore a wider variety of different input sets to make well-founded statements about the robustness of MFFLPE optimization procedures alike.