Optimization methodologies study for the development of prognostic artificial neural network

In this work, we discuss the implementation and optimization of an artificial neural network (ANN) based on the analysis of the back-EMF coefficient capable of making electromechanical actuator (EMA) prognostics. Starting from the pseudorandom generation of failure values related to static rotor eccentricity and partial short circuit of the stator coils, we simulated through a MATLAB-Simulink model the values of currents, voltages, position and angular velocity of the rotor and thanks to these we obtained the back-electromotive force which represents the input layer of the ANN. In this paper, we will turn our attention to optimizing the hyperparameters which influence supervised learning and make it more performing in terms of computational cost and complexity. The results are satisfactory dealing with the number of examples present in the available dataset.