Unbalance Rotor Parameters Detection Based on Artificial Neural Network

Abstract

Unbalance is an important fault that can damage or shut down vital rotary systems such as the gas turbine, compressors, and others, so to avoid this trouble, the balancing process is very crucial, even though it is time-consuming and costly. Thus, having a technique which can predict the unbalance location and its parameters will be valuable and practical. The current study represents a model that can identify the unbalance’s mass, radius, and location of the eccentric mass based on the artificial neural network (ANN) model. The inputs of the proposed ANN, which is based on a feed forward with back propagation model, is the bearing acceleration signal in the frequency domain. It has 10 hidden layers with 10 neurons through each layer. The accuracy in prediction was acquired at 96%, 96%, and 94% for the disc number (plane), the eccentric radius, and eccentric mass values, respectively.