Machine Learning-Based Efficiency Prediction of Francis Type Hydraulic Turbines Through Comprehensive Performance Testing

Abstract

In this study, the rehabilitation works carried out for the KEPEZ HPP, which has been in operation for over 50 years in Antalya, Turkey, is discussed. Within this scope, the existing turbine components are optimized using the CFD method, and a design that provides higher performance at the required flow rate and head is obtained. Analyses are performed using numerical methods to examine the behavior of the new turbine at different flow rates and heads, and a hill chart is created. In the second stage, model tests are carried out at the TOBB ETU HYDRO Water Turbine Design and Test Center in accordance with IEC60193 standards. Different ML methods are examined for their ability to predict turbine performance, following the development of the hydrid CFD-Experimental methodology. According to the authors knowledge, there is no study in the literature that combines experimental, numerical, and ML methods for turbines, and ML methods have not been applied before for Francis-type turbine performance prediction. The outcomes of the study contribute to the advancement of turbine design and optimization processes, offering valuable insights for the successful implementation of rehabilitation projects in the hydropower sector.