Fault Recognition for Mechanical Arm by Using Relative Margin SVM

Monitoring and detecting faults during the operation of the manipulator is the prerequisite for fault recognition and safe operation. Accurate classification of mechanical arm faults can support to effectively eliminate mechanical arm faults. In this paper, the authors utilize a relative margin support vector machine (RMSVM) to classify and monitor the faults for mechanical arm. First, the status of mechanical arm are represented a high dimensional vector which consists of the mean, variance, correlation coefficient of the residual momentum signal in time domain, and the wavelet packet energy spectrum in frequency domain. The collected feature vectors for mechanical arm status are used to train RMSVM. A virtual prototype of mechanical arm is used to analyze the changes in the features of the residual momentum caused by fault and evaluate the RMSVM model for future mechanical arm status. The simulation results show that RMSVM can effectively detect the faults during the operation of manipulator.