Machine Learning Framework for Fault Detection and Diagnosis in Rotating Machinery

Abstract

Rotating machinery are essential elements in industrial systems and strongly present aboard vessels and maritime platforms, whose unexpected failure can lead to significant economic and operational losses, both for the maritime industry and for industry in general. Condition Monitoring (CM), through the analysis of specific parameters, aims to assess equipment health and enable the early detection of deviations from normal operating conditions. Among existing techniques, vibration analysis stands out for its effectiveness. However, when applied to naval environments, it requires human resources and equipment that are not always prepared or available. Aligned with the principles of Industry 4.0, maintenance has been integrating technologies that enhance data collection and analysis, becoming more autonomous and intelligent. The integration of Machine Learning (ML) into CM offers an alternative to conventional approaches, enabling systems to learn real operating behavior and recognize fault patterns with high accuracy and reduced human intervention. Addressing a real industrial challenge, this paper proposes an automatic framework for fault detection and diagnosis using ML models. As a case study, vibration data from rotating machinery were analyzed, encompassing common faults such as unbalance, misalignment, and the combination of both. The obtained results highlight the potential of the proposed framework for CM in maritime environments, modernizing it with new trends and making it more autonomous, efficient, and less dependent on specialized knowledge.