Diagnosis model for bearing faults in rotating machinery by using vibration signals and binary logistic regression

As an important part of rotating machinery, bearing state affects the whole effectiveness and stability of machine components. Recently, many condition monitoring techniques have been developed for bearing fault detection and diagnosis to avoid malfunctioning during operation that might lead to catastrophic failures or even deaths. Vibration monitoring technique is the mostly used as it is cost-effective to detect, locate and estimate bearing faults. Within the technique, the time domain features are favourable to be used for fault machinery faults detection and diagnosis. This is due to its advantages, including it contains all the machine faults information and possibility of using much data for easy and clear fault diagnosis. This study proposes a diagnosis model for bearing faults in rotating machinery based on time domain features and binary logistic regression (BLR) modelling technique of a vibration signals. The steps of the new fault prediction method for bearings are as follows. First, vibration data were collected. Second, the effective time domain parameters extraction from the acquired vibration data sets using multivariate analysis of variance (MANOVA). Third, the data-splitting technique was employed. Here the predictive modelling was performed based on the BLR modelling technique by using the most salient time domain parameters of bearing fault state on the training data set and the selected BLR model was internally validated by using the testing data set. Finally, a comparison was made between the selected BLR model and an artificial neural network model with regards to their accuracy, computational efforts, and effectiveness. The results show the effectiveness and plausibility of the proposed method, which can support timely maintenance decisions in order to facilitate machine performance and fault prediction and to prevent catastrophic failures.