Comprehensive Predictive Maintenance of Marine Equipment using LSTM Neural Networks

Critical and auxiliary equipment aboard ships is maintained using a combination of preventive and corrective maintenance. These policies lead to over maintenance and unanticipated failures which can be very costly since they com-promise operational readiness of ships. Predictive maintenance based on monitoring the current health of the component can reduce unanticipated failures and reduce costs associated with unnecessary maintenance. In this paper, An AI and IoT based comprehensive Predictive Maintenance Framework is proposed for equipment aboard ships, in particular for a water pump. Components which have a high frequency of failure and result in significant downtime are chosen for Predictive Maintenance. Vibration, acoustics, temperature, current, pressure and flow are continuously monitored in real time by installing sensors at multiple pick points on the equipment. A novel single layer LSTM model with an attention module is used to predict the degradation level of the components and the most common failures of the pump. The optimal value of the degradation level at which maintenance should be performed for the subsystem is calculated using cost analysis of maintenance. However the failing component is likely to be interdependent on several other components. A cost based grouping model clusters components taking into account the predicted health of the components, the economic dependencies among the components and the location of the components within the system. In the case of a maintenance alert the cluster of components would undergo joint maintenance to optimize maintenance costs