An Application Research In Fault Feature Extraction Of Gear Systems With Sparse Representation Encoder Driven By Mechanism

As important parts of modern industry,rolling bearing and fixed shaft gear are easy to cause different kinds of faults under complex and bad working conditions for long time service,endangering the safety of users’ lives and property.The fault diagnosis of rolling bearing and fixed shaft gear has important application value.The premise of reliable fault diagnosis is to extract the fault characteristic parameters with definite physical meaning accurately from the fault signal.Based on the mechanism model of fault vibration response signal,combined with Auto-Encoder and sparse representation method,the fault feature extraction of rolling bearing and fixed shaft gear is carried out respectively.Two kinds of sparse representation Encoder are designed to extract the characteristic parameters of steady and impact faults.The fusion mechanism of Auto-Encoder and sparse representation method is analyzed,a sparse representation Encoder structure which can extract fault characteristic parameters is designed.Based on the mechanism of steady and impact faults,the atomic formulas and fault characteristic parameters of steady and impact fault response signals are determined respectively.According to the characteristic parameters,the optimization algorithms of steady and impact faults are designed respectively.Then the sparse representation Encoder of steady modulation signal and impact response signal are obtained respectively.The interpretable reconstruction and feature learning process of steady and impact fault signals and extract the fault characteristic parameters with physical significance and adaptive optimization ability are realized.Combined with sparse representation Encoder,a neural network for rolling bearing fault feature extraction is designed.Based on the analysis of rolling bearing fault response signal characteristics,the fault feature extraction layer is designed by combining impact response signal sparse representation Encoder.The feature optimization layer is designed to improve the accuracy of fault signal reconstruction and damping ratio.Then the neural network for rolling bearing fault analysis is obtained.It is verified by simulation and experiment that the signal reconstruction and feature learning process of sparse representation Encoder is interpretable at the level of impact fault response mechanism,and high-precision impact fault characteristic parameters are extracted from time-domain signals.Compared with the comparison method,the proposed method has higher signal reconstruction and characteristic parameter accuracy.Combined with sparse representation Encoder,two neural networks are designed for steady and impact faults feature extraction of fixed shaft gear.Based on the analysis of steady fault response signal characteristics of fixed shaft gear,the feature extraction layer is designed by combining steady modulation signal sparse representation Encoder.The feature optimization layer is designed to improve the accuracy of fault signal reconstruction and atomic parameters.Then the steady fault analysis neural network of fixed shaft gear is obtained.Impact fault response signal characteristics of fixed shaft gear is analyzed.Based on the rolling bearing fault analysis neural network,the damping ratio screening process is added to design the fixed shaft gear impact fault analysis neural network.It is verified by simulation and experiment that the signal reconstruction and feature learning process of sparse representation Encoder is interpretable at the level of steady fault response mechanism,and high-precision steady and impact fault characteristic parameters are extracted from time-domain signals.Compared with the comparison method,the proposed method has higher signal reconstruction and characteristic parameter accuracy.