Sensor layout optimization by integrating Bayesian approach to diagnose multi-station assembly processes

Abstract

This paper presents a sensor deployment strategy based on a Bayesian network (BN) and information entropy to diagnose multi-station assembly processes. A station-indexed state space model is employed to analyze the influence of fixture faults and part reorientation faults on the process fault. Based on the matrix transformation of the state space model, the inherent rules of process fault propagation in various stations are revealed and the system detectability is quantified by the process fault-detectability index. Subsequently, a BN-based quantified causal graph is developed to model the causal relationship between process faults and sensor measurements, and information entropy is introduced to quantify the uncertainty of process fault diagnosis. Finally, sensor–fault matching algorithms are proposed to minimize information entropy of unit cost and process fault unobservability, under the constraints of detectability, thus achieving optimum sensor placement. An example involving assembly of automobile differential illustrates the methodology.