Bearing Fault Diagnosis Based On TTVF-EMD Time Scale Denoising Algorithm And DBN Normalized Exponential Model

Rolling bearings is an important part of rotating machinery,so the fault diagnosis of rolling bearings has attracted much attention.This paper conducts research from three aspects of rolling bearing vibration signal preprocessing,effective frequency band signal screening and characteristic matrix construction,and fault pattern recognition,and proposes an effective narrowband screening algorithm based on Time-Varying Filtering Empirical Mode Decomposition(TVF-EMD).And the noise reduction performance and feature matrix classification effect of this algorithm are studied,and combined with the normalized exponential function(also called Softmax function)classification model based on Deep Belief Network(DBN-Softmax),the application of TVF-EMD algorithm and effective narrowband screening algorithm is extended.Firstly,the advantages and limitations of the Empirical Mode Decomposition(EMD)algorithm in time-frequency analysis are analyzed.In view of the limitations of the EMD algorithm,TVF-EMD proposes to use the instantaneous amplitude-frequency information to adaptively design the local cut-off frequency,and then use the non-uniform B-spline approximation as a time-varying filter to improve the frequency division performance and the stability under low sampling rates of the system.Secondely,aiming at signal denoising and eigenmatrix construction,simulation signals and bearing vibration signals were used to analyze the frequency division performance and low sampling rate stability of EMD and TVF-EMD respectively.The EMD algorithm will have mode aliasing when the frequency ratio is higher than 0.65,while TVF-EMD still has good separation performance when the frequency ratio is 0.8～0.9,but at the same time it will decompose a large number of Inherent Modal Functions(IMFs),which increases the workload of data processing.Therefore,this paper judges the signal purity of the narrowband from the three elements: signal energy,noise energy,and waveform complexity,and proposes an effective narrowbands filtering algorithm.This algorithm can filter out several narrowbands with greater signal purity from a set of narrowbands decomposed by TVF-EMD.The Effective Inherent Modal Functions(EIMFs)can be reconstructed to achieve the effect of signal denoising.The feature matrix constructed by EIMFs can effectively improve the accuracy of the pattern recognition of the classifier.In the study of fault classification and recognition methods,machine learning classifier and deep belief network were combined to verify the three dimensions of loss function iteration times and pattern recognition accuracy.The simulation results show that DBNSoftmax model has the best performance in all three aspects.Finally,using Case Western Reserve University(CWRU)bearing failure vibration data and Shanghai Baosteel hot rolling mill drive end bearing data,EIMFs denoising processing and EIMFs volume extraction were performed on different types of faults and different degrees of bearing faults,and the original data set was used as the control group.The DBNSoftmax classifier is used for fault diagnosis.Experiments prove that the effective narrowbands screening algorithm proposed in this paper has good practical application value in bearing fault diagnosis.