Fault Diagnosis of Multistage Manufacturing Processes by Using State Space Approach

Abstract

This paper presents a methodology for diagnostics of fixture failures in multistage manufacturing processes (MMP). The diagnostic methodology is based on the state-space model of the MMP process, which includes part fixturing layout geometry and sensor location. The state space model of the MMP characterizes the propagation of fixture fault variation along the production stream, and is used to generate a set of predetermined fault variation patterns. Fixture faults are then isolated by using mapping procedure that combines the Principal Component Analysis (PCA) with pattern recognition approach. The fault diagnosability conditions for three levels: (a) within single station, (b) between stations, and (c) for the overall process, are developed. The presented analysis integrates the state space model of the process and matrix perturbation theory to estimate the upper bound for isolationability of fault pattern vectors caused by correlated and uncorrelated noises. A case study illustrates the proposed method.