A Transfer Learning approach for Anomaly Detection within a Collaborative Prognostic Framework for advanced maintenance services

Abstract

Original Equipment Manufacturers of industrial machineries are shifting toward a servitization strategy aimed at proposing various maintenance solutions delivered as a Service leveraging on advanced digitalised technologies. Given this trend, the present research aims at studying the feasibility of adopting a transfer learning approach, within a collaborative prognostic framework, to support the maintenance servitization strategy of an Original Equipment Manufacturer. The maintenance task that the research focuses on is anomaly detection. The research is carried out considering the need for a cost-effective maintenance management solution delivered to support a fleet of assets. The application of the approach to an industrial case is correspondingly developed, allowing to validate the transfer learning approach in the context of an Original Equipment Manufacturer that provides an advanced maintenance service offering, and can leverage the proposed solution to create business-grade anomaly detection models, with limited effort in terms of resources and time. The validation allows to derive different managerial implications.