Real Time Machinery Health Detection Via Thermal Imaging

Marine vessels are equipped with advanced and integrated complex machines that operate in challenging conditions, where the failure rate is high. One of the most vital elements of these machines is fault diagnosis and maintaining 100 percent operational states. Any machinery breakdown at sea during maritime operations can have severe consequences leading to operational impacts and reducing combat efficiency. The existing maintenance routines - PPM (Planned Preventive Maintenance) and CBPM (Condition Based Predictive Maintenance) are reactive approaches that may miss early signs of material degradation, leading to unplanned machinery breakdown at sea. An efficient method to counter this issue in real time is thermal imaging. The dataset required to develop a thermal prognostic model for real-time machinery health detection is not available. This paper presents a systematic approach to develop a prognostic model for real time machinery health detection via thermal imaging. We have developed a custom thermal image dataset representing various machinery conditions-normal, defective and degraded. We have developed our real-time prognostic model ‘Thermal Prognostic Model’. This model targets fuel systems to prevent major fires and accidents at sea by predicting potential failures and taking corrective actions with mAP50 0.956, precision 0.945 and recall 0.957.