Fault diagnosis within multistage machining processes using linear discriminant analysis: a case study in automotive industry

Abstract

Statistical process control provides useful tools to improve the quality of multistage machining processes, specifically in continuous manufacturing lines, where product characteristics are measured at the final station. In order to reduce process errors, variation source identification has been widely applied in machining processes. Although statistical estimation and pattern matching-based methods have been utilized to monitor and diagnose machining processes, most of these methods focus on stage-by-stage inspection using complex models and patterns. However, because of the existence of high rate alarms and the complexity of the machining processes, a surrogate modelling is needed to solve quality control problems. Here, a novel approach based on variation propagation modelling and discriminant analysis of set-up errors is proposed to diagnose faults in multistage machining processes. In this approach, the future deviation is also allocated to the classification rule of process errors and finally the source of deviation is identified within machining process. The applicability and the performance of the proposed within stage fault diagnosis is investigated using an illustrative case study. The proposed approach can be used in vast multistage machining processes such as aerospace and automotive industries.