Time series forecasting for predictive maintenance of refrigeration systems

Predictive maintenance is a fundamental task in the context of Industry 4.0 to achieve high quality standards by optimizing interventions, before the actual occurrence of faults. Over the year, several machine learning techniques have been exploited to obtain models providing high fault detection accuracy, and, in general, proposed solutions consider it either as a classification or a regression task. Generally speaking, regression approaches requires more data but can obtain more refined results when it comes to the prediction of when a fault will happen. In this paper, we provide a contribution in this context by focusing on a scenario composed of industrial refrigeration systems, typically located in supermarkets, and studying the possibility of applying a Time Series Prediction approach to build an unsupervised predictive maintenance solution. In our investigation, we considered the SARIMAX model and verified, through an experimental campaign on real data, its adequateness for Automatic Fault Detection and Diagnostic (AFDD).