SCARA mechanical fault identification based on WPM-SE+BPNN method

Abstract

Abstract The Selective Compliance Assembly Robot Arm (SCARA) is becoming increasingly important in industrial manufacturing and its operating condition directly determines the safe operation of production lines. However, the complex structure, the variety of mechanical faults and unstable movements of SCARA make fault identification extremely difficult. Therefore, this study proposed identification of SCARA mechanical faults based on wavelet packet multi-segment entropy (WPM-SE) + back propagation neural network (BPNN). First, the original vibration signal was decomposed into several sub-node signals by wavelet packet transform and its envelope spectrum was obtained by Hilbert transform. Then, the envelope spectrum was divided equally into multiple intervals along the time axis, and the energy of each interval was calculated. Afterwards, the feature information of the envelope spectrum was obtained from the energy of each interval, which is defined as multi-segment entropy (M-SE). Where an envelope spectrum obtains a M-SE, and the number of segmentation intervals determines the value of the M-SE. Finally, a feature vector composed of the values of the M-SE was used as the feature input data of the BPNN for mechanical fault identification in SCARA. The BPNN has been tested to achieve an average recognition accuracy of 99.67% for both single mechanical faults and multiple mechanical faults. The results show that the WPM-SE method can effectively extract the feature information of the vibration signal and achieve fast and accurate identification of mechanical faults in SCARA.