TurbineNet: Advancing tidal turbine blade hydrodynamic performance prediction with neural networks

The efficient prediction of system performance is a critical aspect of engineering equipment design, with the traditional methods facing limitations such as high computational demands and precise experimental setups. In response to these limitations, neural network prediction models offer a promising solution due to their lightweight and efficient predictive capabilities. In this context, the evolution of deep learning and computer vision has significantly influenced engineering design applications, particularly in recognizing intricate three-dimensional (3D) structural features. This study addresses the challenges of rapidly and efficiently predicting the performance of horizontal axis tidal turbine (HATT) blades by leveraging artificial intelligence technology. The proposed solution, named TurbineNet, is a neural network specifically designed for predicting the hydrodynamic performance of complex 3D turbine blades. TurbineNet utilizes two descriptors to capture the mesh information and structural features, refining these features through mesh convolution layers. The model establishes a robust connection between the blade features and hydrodynamic performance parameters via two fully connected layers. Through extensive training and validation, TurbineNet demonstrates proficiency in processing and identifying intricate blade surface features, resulting in accurate predictions of HATT hydrodynamic performance. The study showcases the robustness of TurbineNet through extensive testing, revealing its ability to predict hydrodynamic parameters with a relative error of 2%. This exceptional performance positions TurbineNet as a valuable tool for predicting the hydrodynamic performance of complex 3D turbine blade structures, offering a reliable means for assessing engineering equipment performance.