Evaluating Fast and Efficient Modeling Methods for Simulation-Based Optimization

Sammanfattning: As the industry in general, and the automotive industry in particular, is transforming -- due to new technologies and changes in market demands through electrification, digitalization, and globalization -- maintaining a competitive edge will require better predictions. Better predictions of production performance allows companies to capitalize on opportunities, avoid costly mistakes, and be proactive about change.A commonly used tool in manufacturing for the prediction of production performance is discrete-event simulation. In combination with artificial intelligence methods such as multi-objective optimization, in literature often referred to as simulation-based optimization, and knowledge extraction, bottlenecks in the production process can be identified and recipes for optimal improvement order can be obtained. These recipes support the decision-maker in both understanding the production system and improving it optimally in terms of resource efficiency and investment cost. Even though the use of simulation-based optimization is widespread on the production line level, use on the factory level is more scarce. Improvements on the production line level, without a holistic view of factory performance, can be suboptimal and may only lead to increased storage levels instead of increased output to the customer.The main obstacle for applying simulation-based optimization to the factory level is the complexity of its constituent parts, i.e., detailed production line models. Connecting several detailed production line models to create a factory model results in an overly complicated, albeit, accurate model. A single factory model running for one minute would equate to almost 140 days required for an optimization project, too long to provide decision-support relevant to manufacturing decision-making. This can be mitigated by parallel computing, but a more effective approach is to simplify the production line models to decrease the runtime while trying to maintain accuracy. Model simplification methods are approaches to reduce model complexity in new and existing simulation models. Previous research has provided an accurate and runtime efficient simplification method by use of a generic model structure built by common modeling components. Although the method seems promising in a few publications, it was lacking external and internal validity.This project presents simulation-based optimization on the factory level enabled by a model simplification method. By following the design science research methodology, several case-studies mainly in the automotive industry identify issues with the current implementation, propose additions to the method, and validates them.