Failure Warning of Harmonic Reducer Based on Power Prediction

—Harmonic reducer is the core component of industrial robots. During its operation, the power signal is a key parameter that embodies the performance of the harmonic reducer. Therefore, accurate power prediction of the harmonic reducer has instructive significance for its failure warning and performance prediction. In this paper, a hybrid deep neural network (DCBNN) based on CNN and BiLSTM was proposed to process the condition monitoring data of the harmonic reducer and improve the prediction accuracy of power signal. First, the operating parameters were pre-processed and the data sets were divided. Then, the pre-processed data were input into DCBNN, and the spatial characteristics and bidirectional timing dependencies of the condition monitoring data are captured by CNN and BiLSTM. On this basis, the absolute value of the residual of the actual power and the predicted value is obtained according to the prediction result, and the residual curve is fitted by the distribution fitting method to obtain the alarm threshold of the harmonic reducer failure warning. Finally, 8 different data sets constructed using the experimental data of the harmonic reducer are used to verify the effectiveness and superiority of the proposed method. The test results on the complete data set show that the DCBNN model can complete effectively failure warning of the harmonic reducer.