A fault diagnosis approach to rolling element bearings based on improved hilbert-huang transformation and linear neural network

It maybe fails to detect rolling element bearing defects using Hilbert-huang transformation due to its deficiency that its empirical mode decomposition (EMD) may generate undesirable intrinsic mode functions (IMFs) at low-frequency region. Here, the unwanted IMFs are removed based on the ratio of energy of each IMF to the total energy of the original signal (energy radio). Then, calculate the marginal spectrum of the IMF whose energy ratio is biggest and construct a fault feature vector with a ratio of the frequency where the marginal spectrum has a peak value to the shaft rotation frequency. Ultimately, the fault feature vector is served as the input parameter of a linear neural network to identify fault patterns of rolling element bearings. Experiment result shows that the proposed approach is effective.