Fault Detection in Multi-stage Manufacturing to Improve Process Quality

Abstract

Fault detection for a multi-stage manufacturing process is often challenging due to the lack of quality inspection after each individual stage. In most cases, the final product is rated by an end-of-process quality inspection. This leads to a difficult identification of the manufacturing stage in question. This paper presents a novel approach for fault detection of a multi-stage manufacturing process using machine learning. For this approach, an autoregressive model is used, which is enhanced by a neural network to create a residual between process measurements and model predictions. The residual is then evaluated to detect a fault in an individual manufacturing stage and in the experimental study a True Positive Rate of 0.79 is reached for a False Positive Rate of 0.07. The major advantage of the proposed approach is the detection of the fault without an explicit quality inspection for each individual stage.