Using an Explainable Machine Learning Approach to Minimize Opportunistic Maintenance Interventions

Abstract

The industry 4.0 paradigm, with a wide range of sensors, IoT and big data technologies, has facilitated the assessment of faults in complex mechanical systems. In this paper, a fault diagnosis strategy is presented for opportunistic condition-based maintenance decisions of a single failure mode. Focusing on the challenges of the fault identification task, the proposed method was assessed by conducting a case-study using real-world data. To detect symptoms of screen pack degradation in the company’s coextrusion process, the devised strategy was based on an anomaly approach and a technique for explainable artificial intelligence (XAI). Experimental results for two consecutive production runs of an extruder show that the proposed method effectively identifies clustered anomalies as symptoms of a clogged screen pack.