A real-time semi-supervised anomaly detection framework for fault diagnosis of marine machinery systems

Maritime companies are currently working to ensure a digital revolution within the maritime industry. Smart maintenance is pivotal in leading this transition, the aim of which is to employ internet of ships to perform real-time data collection through the utilisation of smart sensors, reliable communications, and seamless integration in order to apply predictive maintenance with the application of artificial intelligence and provision of relevant information. Therefore, regular diagnosis and prognosis can be performed to assess the current and future health of machinery to assist in decision-making processes. To enhance the current practices in this area, an innovative anomaly detection framework implementing LSTM-based VAE is proposed to address the challenges identified within this sector. A case study of a diesel generator of a tanker ship is introduced to assess the proposed methodology. Results demonstrated the capability of identifying anomalous instances under various simulated scenarios, thus achieving the maximum precision and recall when the context considers significant anomaly dimensions.