Determining Manufacturing Condition Range Using a Causal Quality Model and Deep Learning

Abstract

This study proposes a method to improve the calculation of probabilities in a Bayesian network. The aim of the Bayesian network in this study is to specify the manufacturing condition range needed to meet a specified level of product quality. In the manufacturing process, workers need to operate within manufacturing condition range to improve product yield. Conventionally, quality standards are established for each manufacturing process, and workers operate according to the tolerances determined by these standards. However, these efforts are not very effective because it is not clear which tolerances actually reduce the failure rate. In addition, product yield is not improved. We develop a method that uses deep learning to estimate the conditional probabilities in Bayesian network modeling. Using the proposed method, the AUC of the model increases from 0.91 to 0.99, indicating that this approach can be used for specifying the tolerances of manufacturing conditions.