Multi-joint Industrial Robot Fault Identification using Deep Sparse Auto-Encoder Network with Attitude Data
Ying Hong, Zhenzhong Sun, Xiaohong Zou, Jianyu Long
Abstract
Intelligent fault identification of the mechanical transmission system for multi-joint industrial robots is important to guarantee safe operations. An attitude data-based intelligent fault identification approach is introduced for multi-joint robots in this study. Considering that the attitude change of the last joint can be used to reflect the attitude change of the other connecting rods or joints, an economical data acquisition strategy is proposed by only installing one attitude sensor on the last joint. An intelligent fault identification model is subsequently established by training a deep sparse auto-encoder network (DSAE) on the attitude dataset. To test the performance of the proposed approach, a test-rig was built to collect attitude dataset under different fault conditions. Experimental results show that the proposed fault identification approach has effective performance for multi-joint industrial robots.