Advancing Turbine Prediction: Harnessing Conformable Artificial Neural Networks for the Fracture Analysis

In this research, an innovative approach was developed using Conformable Artificial Neural Networks (CANN) for detecting and predicting fractures in gas turbine blades made of AISI 410 stainless steel. The methodology integrates conformable transfer functions with advanced modeling and statistical analysis techniques to accurately predict the value of the J-integral, a critical metric in evaluating the resistance of materials to crack propagation. Experimental data collected through fatigue testing were used to train and validate models with various transfer function configurations. The results show superior accuracy in the predictions of the CANN model, reaching a correlation coefficient of 0.99992, an RMSE of 0.00219, and a MAPE of 0.21589. This research represents a significant contribution toward predictive maintenance and efficient management of gas turbines, optimizing safety and operational sustainability in thermal power plants.