Ensemble Bayesian Network for root cause analysis of product defects via learning from historical production data

Abstract

Root Cause Analysis (RCA) of product defects is crucial to improving manufacturing quality and productivity. However, current efforts to localize root causes are prone to limitations in the aspect of robustness, causality discovery, knowledge representation, stochasticity, and sample size. Therefore, we propose a product-wise Ensemble Bayesian Network (EBN) to provide a robust, intelligent and human-interpretable probabilistic reasoning method for RCA. BN is adopted to enable interpretable probabilistic reasoning under uncertainty. We developed various structure learning algorithms, a parameter learning algorithm, and a Bayesian inference algorithm for BN to learn the root causes of product quality issues from historical product defect records. Our Ensemble Learning (EL) techniques enhance BN base learners with bootstrapped re-sampling and combine the predictions from multiple structure learning algorithms, ensuring a robust performance of BN. The framework structure is modularized by products to reduce the sample size and achieve high efficiency. We proved our method achieved good performance in acquiring causal knowledge, identifying the root cause with probabilities, and predicting quality risks in production, from implementation and extensive experimental testing on real-world data collected from the plastic industry.