Mutual Information-Based Sensing Optimization for In-Process Quality Inspection of Multi-Station Assembly Processes

Abstract

Abstract To increase productivity and satisfy diverse customer demands, many assembly processes have been converted from single-model lines to flexible assembly lines. Conventional fixed-location sensing approaches cannot be used to inspect a broad mixture of different quality features for multiple product models effectively. The in-process dimensional quality inspection systems with robot-mounted optical gauging have been becoming increasingly popular for flexible multi-station assembly processes. These new systems enable more powerful quality inspection for various designs and provide more flexibility for dimensional monitoring and diagnosis. Meanwhile, they bring new challenges to sensing efficacy and efficiency because in-process cycle time for quality inspection is limited in dozens of seconds. This paper focuses on sensing optimization problem for multi-station assembly processes. We propose a novel data-driven sensing optimization approach based on in situ quality inspection data, for the purpose of improving variation detection of subsequent assemblies. Based on the dimensional inspection datasets, a mutual information-based model is proposed to measure the correlation relationships among components, intermediate subassemblies and final products. An objective function considering the maximum relevance and the minimum redundancy is developed, and then an optimization algorithm is applied to obtain the optimal sensors. A case study in auto body assembly process is used to show the performance of the proposed method.