Norm-based data labelling in supervised learning for fault detection and diagnostics of rotating elements towards maintenance servitisation

Abstract

In the current industrial context, the wide availability of data from the shopfloor is enabling companies to develop condition-based maintenance (CBM) and predictive maintenance (PdM) solutions towards production performance improvements. However, the path is not straightforward and several technological and managerial challenges have to be faced. Specifically, the current challenge is to mix the high-performance, yet difficult-to-interpret results, of AI (Artificial Intelligence) algorithms with the vast available domain knowledge provided by scientific literature and norms. It is the goal of this work to propose a norm-based data labelling to implement a supervised model which leverages on time-domain features to guarantee the interpretability of results for maintenance operators and technicians for FDD (Fault Detection and Diagnostics). The proposed approach is tested and a complete CBM solution is deployed in a case of an OEM (Original Equipment Manufacturer) of rotating elements. Through it, the company could move towards a fully-fledged maintenance service offering, already integrating norm-related knowledge.