Comparison of Community Detection Algorithms for Reducing Variant Diversity in Production

In production planning and control, discrete-event simulation (DES) is commonly used to address optimization challenges. DES using simgen generally begins with data preprocessing, parameterization, and experiment design. However, due to the complexity of manufacturing environments, DES models require careful parameterization, with empirical experiments designed to ensure efficient execution. This parameterization involves optimizing parameter settings for different materials based on routing, bill-of-materials complexity, and other production process-related features. To achieve optimized parameterization within expected timeframes, reducing variant diversity to eliminate redundant materials is necessary by using data-driven approaches. In this study, to identify representative materials, a network-based approach with five community-detection algorithms is compared for their efficiency in execution time and efficient module detection by constructing bipartite networks of material and routing features for identifying similar material groups and representative materials. The results show that communities and subcommunities identify representative materials by significantly reducing the initial number of materials with a faster approach that can be used for DES parameterization.