Effect of Different Training Algorithms in Artificial Neural Networks on the Rotor Unbalance Diagnosis

A neural network simulator is designed to predict unbalance in a turbine-bearing system. Artificial intelligence (AI) techniques for machine diagnostics and prognostics are gaining traction and are predicted to prove fruitful shortly. In this research, multi-layer feedforward neural networks (FNNs) with different training algorithms have been employed for rotor unbalance diagnosis. The rotor is derived from physical theories and bearing parameters are identified via the particle swarm optimization (PSO) method using the operational data acquired from bearing sensors. Following that, the physical model is used to generate numerous sets of simulated data sets for FNN. The results show that Levenberg-Marquardt (LM) named trainlm algorithm that single layer with 50 neurons show the least RMSE and superiority testing performance over the other algorithms.