Bearing Fault Detection and Classification Based on Vibration Signal Analysis and ANFIS Classifier

Automatic Bearing defects are able to lead to deterioration of the operating conditions of the rotating machine, how to extract the most informative characteristics of the fault from the vibration signals and classify the bearing fault have become a critical problem and addressing this problem is an imperative for ensuring the safe operation of the rotating machines. This paper proposes a hybrid method that uses the Empirical Mode Decomposition (EMD) technique for the extraction of the most informative characteristics of the bearing faults using calculated energy and entropy and the ANFIS algorithm as an intelligent classifier for rolling bearings fault classification. Firstly, the non-stationary features of the vibration signal are extracted by applying the EMD that is applied for decomposing the measured signal into a fixed amount of stationary intrinsic mode functions (IMFs), and then the energy and entropy of the IMFs are considered to form the parameters vector used in the classification stage of the proposed procedure. In fact, the parameters vector is first used as an input for the ANFIS classifier, but after choosing from it the best extracted features adapted to bearing fault diagnosis through a wrapper algorithm. The proposed method is tested on experiment using real bearing vibration signals for different health conditions (bearing with inner-race, out-race and ball faults) by considering 12 fault classes that are determined according to fault type and severity. The results approve that the proposed technique reached a good classification accuracy.