Applying machine learning to predict asset health in hybrid maritime vessels

In recent years there has been a surge of interest in hybrid propulsion technology for commercial and domestic maritime vessels. Factors driving this growth include high maintenance cost of diesel engines and the need for compliance to more stringent environmental regulations. Although electric and hybrid propulsion techniques present an opportunity to address this challenge, today's batteries for large vessels are still expensive and prone to unexpected failure. Therefore, the cost and reliability of batteries still represent risks to the maintenance in hybrid energy systems. In this paper, we first introduce the challenges of asset health management within hybrid energy systems for maritime vessels. We explore the feasibility of data-driven methods to evaluate Remaining Useful Life (RUL) of the system assets to inform a preventative and cost efficient management system. An overview of our proposed solution is presented along with preliminary results of our data-driven model applied to large battery data sets to estimate the battery RUL. The model is based on a state-of-art machine learning technique, Relevance Vector Machine (RVM), which is a powerful tool for resolving uncertainty in large data sets. Initial training of our machine learning algorithm utilizes a public battery life cycle testing dataset, provided from NASA Ames Research Center. Next, we use life cycle analysis of batteries designed for hybrid vessels to evaluate the performance of the algorithm in predicting battery remaining useful life (RUL). The accuracy of the predictions for different batteries are all within 10 cycles (within 8.5% relative error) which encourages us to adopt the same approach in future health management works for other power assets on the hybrid maritime vessels.