A MULTI-LAYER PERCEPTRON APPROACH FOR INDIRECT MEASUREMENT OF EROSIVE CAVITATION IN HYDRAULIC TURBINES

Erosive cavitation generates wear in hydraulic turbines, leading to costly maintenance and extended downtime in hydroelectric power plants. This study introduces an indirect measurement approach to predict the intensity of erosive cavitation and its degradation trajectory in hydraulic turbines, using a Multi-Layer Perceptron (MLP) model that predicts the instantaneous rate of cavitation based on operational parameters while also quantifying predictive uncertainty. Although cavitation may appear difficult to predict and complex, this study demonstrates that it can be effectively modeled and predicted using a relatively simple model like MLP, achieving good performances. By accumulating the predictions, we model the cumulative degradation caused by cavitation over time, enabling more informed maintenance scheduling. Additionally, this study explores transfer learning to extend the predictive capability of a model trained on one turbine to predict the erosive cavitation behavior of another turbine.