AI-Based Clustering of Numerical Flow Fields for Accelerating the Optimization of an Axial Turbine

Abstract

The growing number of Renewable Energy Sources has increased the demand for innovative and high-performing turbine designs. Due to the increase in computing resources over recent years, numerical optimization using Evolutionary Algorithms (EAs) has become established. Nevertheless, EAs require many expensive Computational Fluid Dynamics (CFD) simulations, and more computational resources are needed with an increasing number of design parameters. In this work, an adapted optimization algorithm is introduced. By employing an Artificial Intelligence (AI)-based design assistant, turbines with a similar flow field are clustered into groups and provide a dataset to train AI models. These AI models can predict the flow field’s clustering before a CFD simulation is performed. The turbine’s efficiency and cavitation volume are predicted by analyzing the turbine’s properties inside the predicted clustering group. Turbines with properties below a certain threshold are not CFD-simulated but estimated by the design assistant. By this procedure, currently, more than 30% of the CFD simulations are avoided, significantly reducing computational costs and accelerating the overall optimization workflow.

Keywords

• optimization
• axial turbine
• clustering
• autoencoder
• artificial intelligence
• evolutionary algorithm