Two-stage Approach to Causality Analysis-Based Quality Problem Solving for Discrete Manufacturing Systems
Haonan Wang,Yuming Xu,Tao Peng,Reuben Seyram Komla Agbozo,Kaizhou Xu,Weipeng Liu,Renzhong Tang
DOI: https://doi.org/10.1080/09544828.2023.2247858
2023-01-01
Journal of Engineering Design
Abstract:In discrete manufacturing systems, the efficiency of quality problem solving always matters, but it can decrease continually due to increasing systematic complexity, uncertainty, and fuzzy internal mechanism. These factors often hinder situation awareness, confuse decision-making, and delay response times in quality problem solving. In such context, Bayesian network performs poorly in precisely locating objects to-be-adjusted in complex causal relationship networks and predicting the impact of interventions to-be-implemented. To address this challenge, a two-stage approach to causality analysis-based quality problem solving is proposed. In the first stage, an improved Bayesian network is proposed to identify the likely root causes that have a direct causality on the quality indicator. In the second stage, causal inference is used to estimate the effect of likely root causes on the quality indicator. The proposed approach compensates for Bayesian network in precisely identifying root causes that should have causation rather than correlation with quality problems, and facilitating a quantitative tuning process to create suitable solutions. To assess the effectiveness of the proposed approach, a quality problem solving case in an aerospace shell parts spinning process was conducted. The results demonstrate that the approach can accurately identify likely root causes and determine the appropriate intervention degree.