Machine learning based performance prediction of variable speed hydraulic turbines

Francis type turbines are reaction type hydraulic turbines widely used for energy production in the modern world. These turbines are designed to operate for many years, and thus, having a high hydraulic efficiency is of great importance. Digital twin is a digital form of a real system in a digital environment and the digital form make it possible to see the real system behaviour in digital environment. Tools such as CFD and machine learning are frequently used to create this digital replica. The digital twin structure allows the prediction of a turbine's efficiency and flow parameters under varying conditions in hydroelectric power plants. Within the scope of this thesis, CFD analyses were conducted to examine the variations in hydraulic efficiency, flow rate, and other parameters of an existing Francis-type model test turbine under different operating conditions. To validate the CFD results, a comparison was made with the existing experimental data. Synthetic data was generated through the CFD analyses to train machine learning models. Data obtained from 100 CFD analyses under different angular velocity, head, and flow rate conditions were used to train two distinct machine learning models, thus creating a digital twin structure. Trained ANN and ANFIS models were employed to predict efficiency and flow rate under different angular velocity, head, and flow rate conditions. The discussion also included considerations on optimizing efficiency by adjusting the turbine's angular velocity, particularly in response to varying flow rates and head levels in hydroelectric power plants. Keywords: ANN, ANFIS, CFD, Francis Turbine, Hydraulic Turbine