Different studies have been worked about induction motor bearings fault detection using digital signal processing and pattern recognition techniques. However, performance of these techniques is related with the use of correct features. This paper presents an analysis of the use of filter banks with uniform and nonuniform frequency subbands to features extraction from vibration signals. Classification was developed by an artificial neural network with feedforward connections. Results identifies that the employment of filter banks improve the accuracy in 23percent-flag-change for six considered classes related with faults in bearings.
