Data-driven fault identification method of RV reducer used in industrial robot

The RV reducers are complex and sealed mechanical systems that are difficult to conduct fault diagnosis in advance. The previous research worked on the fault identification of RV reducer were mainly carried out on the test platforms instead of real complex working conditions. Most of faults were intentionally created in laboratory instead of real malfunction caused by factory daily operation. In the present paper, the actual failure mode of RV reducer for the industrial robots in factory is taken as the goal of fault diagnosis. The constant speed segment data extraction method is designed to overcome the difficulty of frequency domain analysis caused by non-uniform rotation in working conditions and ensure the quality and effectiveness of features extraction. Several machine learning classification models are selected regarding their inherent features. The proper DNN binary classification model shows the best performance that can meet the requirements of fault identification in industrial environment.