This paper proposes a method of decomposing the statistical distribution of manufacturing-condition (e.g., pressure, temperature, and consumable usage time) monitoring data into multiple components using the variational inference method to find the cause of defective products and take measures to control quality. The data come from multiple manufacturing resources (e.g., machines and workers) assumed to be compatible as production capabilities, but some of which are related to the occurrence of defective products. The proposed method determines defective manufacturing resources by calculating their responsibilities for the statistical distribution components with high defective rates. Evaluation results using a dataset modeling actual production lines indicate the effectiveness of the method.
