Investigating Data Science Approaches for Predictive Maintenance in Maritime Fleet Operations

Abstract

Operating a maritime fleet impacts global trade and commerce, yet it is often accompanied by issues related to equipment dependability, reliability, and maintenance costs. Traditional reactive or scheduled maintenance methods usually result in considerable resource idleness and unplanned system downtime. This paper focuses on employing data science techniques to facilitate predictive maintenance on maritime vessels. Utilizing the onboard systems' historical and real-time sensor data, we assess the efficacy of Random Forest algorithms, Support Vector Machine algorithms, and Long Short-Term Memory (LSTM) networks in fault detection and failure prediction. The proposed framework comprises data cleansing, feature selection, model training, and evaluation conducted using authentic data obtained from commercial ships. The experimental results validate the assertion that data-driven maintenance models can forecast failure of specific ship components well within the stipulated time, thus aiding planned maintenance and minimizing operational halts. These results underline the importance of predictive analysis on operational maintenance, enhancing fleet dependability, safety, and cost-efficiency in maritime operations.