A methodology is developed for defect type identification in rolling element bearings using the integrated Mahalanobis–Taguchi–Gram–Schmidt (MTGS) method. Vibration data recorded from bearings with seeded defects on outer race, inner race and balls are processed in time, frequency, and time–frequency domains. Eleven damage identification parameters (RMS, Peak, Crest Factor, and Kurtosis in time domain, amplitude of outer race, inner race, and ball defect frequencies in FFT spectrum and HFRT spectrum in frequency domain and peak of HHT spectrum in time–frequency domain) are computed. Using MTGS, these damage identification parameters (DIPs) are fused into a single DIP, Mahalanobis distance (MD), and gain values for the presence of all DIPs are calculated. The gain value is used to identify the usefulness of DIP and the DIPs with positive gain are again fused into MD by using Gram–Schmidt Orthogonalization process (GSP) in order to calculate Gram–Schmidt Vectors (GSVs). Among the remaining DIPs, sign of GSVs of frequency domain DIPs is checked to classify the probable defect. The approach uses MTGS method for combining the damage parameters and in conjunction with the GSV classifies the defect.

A Defect Occurrence Index (DOI) is proposed to rank the probability of existence of a type of bearing damage (ball defect/inner race defect/outer race defect/other anomalies). The methodology is successfully validated on vibration data from a different machine, bearing type and shape/configuration of the defect. The proposed methodology is also applied on the vibration data acquired from the accelerated life test on the bearings, which established the applicability of the method on naturally induced and naturally progressed defect. It is observed that the methodology successfully identifies the correct type of bearing defect. The proposed methodology is also useful in identifying the time of initiation of a defect and has potential for implementation in a real time environment.