Fault Diagnosis Of RV Reducer Based On Time-frequency Decomposition Noise Reduction Combined With Deep Learning Model

As industrial intelligence advances swiftly,the capacity of industrial cobots has soared,driving up the demand for core components.Since,RV reducers occupy the center of core component sales due to their better reduction ratio,higher reliability and stiffness,and more rigid structure.Therefore,the use of condition monitoring technology to observe its operating status can identify fault problems in time so that effective solutions can be taken,which is of great importance to the stability and reliability of industrial robots.In this study,the health condition of RV reducer is investigated using a combination of time-frequency decomposition noise reduction and improved convolutional neural networks,Fault diagnosis based on parameter optimization VMD combined with MOMEDA,fault diagnosis based on time-frequency decomposition noise reduction and improved deep learning model,and fault diagnosis based on Laplace wavelet convolution and attention mechanism.The main studies are as follows:(1)Because of the running conditions are complicated,industrial noise disturbance,another device operation and other reasons make the RV reducer rolling bearing fault diagnosis is full of great challenges,and it is hard to obtain results based on traditional,single signal processing methods.This chapter proposes a parameter-optimized timefrequency decomposition algorithm combined with deconvolution for fault feature extraction,which uses the autocorrelation function impulse harmonic noise ratio index to optimize the decomposition parameters of VMD,selects the optimal component modal reconstruction for noise reduction,and achieves signal impact feature enhancement by MOMEDA deconvolution.The effectiveness of the method in this chapter is verified by simulated bearing data and QPZZ-II type fault simulation test bench data.(2)Due to the complicated construction of the RV reducer and the time-varying operating state,its signal frequency element is complicated and highly time-variable.This article puts forward a fault diagnosis approach using a wavelet packet power noise mitigation and an extended convolutional neural network to classify the RV gearboxes in various fault conditions.The diagnostic accuracy and ubiquity of the approach was demonstrated by two sets of testbed figures.(3)In order to respond with the adaptive diagnostic capability of model features under variable load and to deal with the design from the perspective of a priori knowledge of signal processing,this article suggests a fault diagnosis approach combined with Laplace wavelet convolution and attention mechanism,and verifies the fault diagnosis capability of RV reducer data under variable load by experimental data from the model itself combined with a priori knowledge of signal.