Evaluation of principal component analysis and neural network performance for bearing fault diagnosis from vibration signal processed by RS and DF analyses

In this work, signal processing and pattern recognition techniques are combined to diagnose the severity of bearing faults. The signals were pre-processed by detrended-fluctuation analysis (DFA) and rescaled-range analysis (RSA) techniques and investigated by neural networks and principal components analysis in a total of four schemes. Three different levels of bearing fault severities together with a standard no-fault class were studied and compared. Signals were acquired from bearings working under different frequency and load conditions. An evaluation of fault recognition efficiency was performed for each combination of signal processing and pattern recognition techniques. All four schemes of classification yielded reasonably good results and are thus shown to be promising for rolling bearing fault monitoring and diagnosing.