Turbine Design and Reliability
Ocean Current Turbine Health Monitoring
Led by Yufei Tang, Ph.D.
Affiliated Home Campus: Boca Raton
Affiliated Department: Computer and Electrical Engineering and Computer Science
Yahan Chang worked with Dr. Yufei Tang and his team to study ocean current health monitoring systems, with an emphasis on physics-informed machine learning. Maintenance of offshore ocean current turbines face challenges of accessibility, costs, and time constraints due to the harsh ocean environment. To reduce turbine downtime and lower lifecycle costs, prognostics, and health management is used to monitor the current state of the system and predict the remaining useful life (RUL). Furthermore, integrated machine learning can be used with prognostic sensor data to optimize turbine operational performance with the addition of a-priori knowledge from equations and/or probabilistic relations. This approach combines desired qualities from both physics-driven models and data-driven models to decrease overall prediction error and bias. Physics-informed Neural Networks (PINN) specifically take advantage of differential and physics equations such as bearing damage, grease damage, etc. to produce robust health monitoring models that meet the constraints of natural laws. Hybrid models have greater interoperability and generalization due to the theoretical guidance within the training dataset and learning algorithm. This REU project focused on physics-informed Graph Convolutional Networks (GCN) which utilize relational dependencies between individual turbine components and integrated physics models to improve remote health-state prognostics.