On spectral component estimation using neural networks for rolling bearing fault diagnosis

In general, the functioning state of rotating internal parts of a machine, which is inaccessible without dismantling the machine, can be obtained by indirect methods. Hence, fault diagnosis of rotating components, like rolling bearings, can be done by analyzing the external relevant information obtained by measurements, in order to evaluate their internal states. The most widely used method for detection and fault diagnosis of rolling bearings is based on vibration measurements. In signal processing, there are many applications of neural networks (NN) combined with spectral analysis, using Fast Fourier Transform (FFT) algorithm. The function approximation using neural networks is a complex task for problems with high dimension of the input space, like those based on signal spectral analysis. In this paper, some neural estimation aspects of spectral components of vibration signals for rolling bearings fault diagnosis are discussed. Inner race defect of bearings is considered in simulations. The goal is to find a feed-forward neural network (FFNN) model for estimating spectral components, with computational complexity comparable with FFT algorithm, but easier to implement in hardware. Different FFNN architectures, with different data sets and training conditions, are analyzed.