“…The well accepted features include frequency based [1], energy based [2] and wavelet coefficients based features [3], most of which usually can be obtained by the Fourier transform (FT) [1], wavelet transform (WT) [3] analysis methods [4]. Besides these common fault features, there are still some other unfamiliar but often effective fault features, such as the fractal dimension [5,6], which often involve the geometrical character of similarity at different scales of the analysed signals. According to the study of Logan and Mathew [6], the fractal dimension can be used as indexes to tell apart the different conditions of working bearings, including the normal, outer race fault and inner race fault.…”