AudioForesight: A Process Model for Audio Predictive Maintenance in Industrial Environments

Nowadays, the digitalization and automation of industrial processes is already a standard in many companies. Especially maintenance is an essential process with huge impact in business trying to minimize downtimes of industrial equipment. Maintenance can be separated in several different categories including predictive maintenance which we will focus on. In this paper we present AudioForesight, an audio-based approach for industrial equipment predictive maintenance. The approach is based on a combination of two machine learning techniques: Anomaly detection and classification. The anomaly detection is used to predict abnormal behavior in comparison to the normal behavior observed after setting up the system. After detecting an anomaly, a classifier tries to identify the system error based on predefined failure classes. In the case study, we present an example implementation of our approach for an industrial metrology machine provided by our partner from industry. The case study shows the feasibility of the approach and allowed the prediction of machine failures before they occurred using an autoregression model based on mean-squard-errors from the anomaly detection and classification. We present the results, challenges we uncovered and an outlook of future work.