A Data Mining-based Fault-Locating Framework for Automated Production Line

In response to automated production lines with complex process and numerous machines, traditional fault location methods have encountered bottlenecks both in efficiency and accuracy. Therefore, by means of massive data generated from machines' sensors, this paper proposes a fault machine positioning framework based on data mining. In this framework, the main defect type of the product is determined based on multi-dimensional scoring; the association model between dimension-reduced process data and product defect is established by the Xgboost method, then the suspicious machine is backstepping by the feature importance calculated from the model. Meanwhile, the machine alarms are clustered based on the alert type and occurrence time to determine the main alarm of each subset, and the primary problem machine of each subset is located by alert source. Finally, based on the improved Dempster-Shafer evidence theory, the two evidences are combined to obtain the comprehensive conclusion. The proposed framework can be applied to varies automated production lines of complex products, capable of a wide range of practical application.