Unsupervised Fault Detection of Industrial Robot Joints Using Current Signal

Industrial robots have been widely used in various industrial manufacturing companies to improve production efficiency. With the service time gained of an industrial robot, the possibility of failure or fault of an industrial robot joint gains. Due to the motion propagation among the joints, some industrial robot arms show abnormal performance even though there is no fault in their joints. Although some vibration-based detection methods for industrial robot joint faults have been successfully established, it is still difficult to detect industrial robot joint faults by using only the current signal, especially, there is no sufficient label to classify fault or normal current signal. To deal with the above issues, this paper proposes an unsupervised fault detection method based on spectral clustering and the sensitive features of the current signal. To enlarge the samples, the collected current signal in a certain time is divided into several pieces according to the peak finding function. Then widely adopted time-domain features are selected according to the sensitivity. The selected features are fed into the spectral clustering to detect the fault location among the industrial robot joints. The proposed method is validated by a reliability-test industrial robot.