Interpretable and Adaptive Data-Driven Root Cause Analysis in Manufacturing through Association Rule Mining

The manufacturing industry faces significant challenges in maintaining quality standards and achieving zero-defect manufacturing (ZDM) to minimize costs and rework. The ongoing trend toward digitization and Industry 4.0 has resulted in an abundance of data throughout the manufacturing process, enabling the automation of tasks such as defect detection and root cause analysis (RCA). However, existing data-driven methods for RCA often lack interpretability, limiting their utility for process experts and failing to adapt to dynamic production environments. This paper presents a novel method for RCA that leverages association rule mining (ARM) by introducing a rule-based classifier that is both adaptive to changes in the production process and directly interpretable by process experts. The method is validated using a real-world automotive final assembly dataset, specifically electronic control unit (ECU) error logs and their associated root causes. This dataset is used to evaluate the performance of the proposed method in comparison to other data-driven models, including random forest (RF), k-nearest neighbors (KNN), and decision trees (DT). Results demonstrate that the proposed method offers adaptability to changes in the manufacturing environment and high interpretability, making it suitable for practical implementation in dynamic and evolving manufacturing environments.