A Hybrid Simulation-Gaussian Process Regression Approach for Performance Prediction and Reconfiguration of Production Layout

Abstract As products become more sophisticated and requirements for superior quality levels become more stringent, the complexity inherent in manufacturing systems increases significantly. In particular, the accuracy of predicting production performance and various operation costs is an increasingly important task under rapidly changing manufacturing conditions. In response, this paper proposes a framework to predict key performance indicators (KPIs) using Gaussian Process Regression (GPR), which can be applied to labor-intensive manufacturing systems in small and medium-sized enterprises with relatively flexible production layout changes. The proposed framework provides several alternatives to the production layout strategies by leveraging Discrete Event Simulation (DES) data from scenario-based Cellular Manufacturing System (CMS) models. Since GPR-based prediction model offers probabilistic results and captures the uncertainty inherent in the manufacturing systems, it can adaptively address the problems related to the impact of production system complexity on KPIs. In this research, three key dependent variables such as cycle time, total buffer capacity, and average manpower efficiency are extracted and utilized from the simulation results to develop a GPR model for production volume prediction under specific manufacturing conditions. This paper describes the creation and comparison of GPR models for three manufacturing layouts and demonstrates the usefulness of the proposed framework in adapting to different production environments.