Using supervised and one-class automated machine learning for predictive maintenance

Predictive Maintenance (PdM) is a critical area that benefiting from the Industry 4.0 advent. Recently, several attempts have been made to apply Machine Learning (ML) PdM, with majority of research studies assuming an expert-based ML modeling. In contrast these works, this paper explores purely Automated (AutoML) modeling for PdM under two main approaches. Firstly, we adapt and compare ten recent open-source AutoML technologies focused on Supervised Learning. Secondly, propose novel approach One-Class (OC) (AutoOneClass) employs Grammatical Evolution (GE) search best model using three types learners (OC Support Vector Machines, Isolation Forests deep Autoencoders). Using recently collected data Portuguese software company client, performed benchmark comparison study tools proposed AutoOneClass method predict number days until next failure equipment also determine if equipments will fail in fixed amount days. Overall, results were close among compared tools, supervised AutoGluon obtaining all tasks. Moreover, predictive manual approaches (using expert non-ML expert), revealing competitive results.