Research On Weak Compound Fault Intelligent Diagnosis Methods Of Rotating Machinery Under Strong Noisy Interference

Rotating machinery is an important part of the mechanical equipment such as gears and bearings.The health condition of rotating machinery is important for guaranteeing the healthy,safety,efficiency and continuous operation of mechanical equipment.Due to the complex structure of mechanical equipment and the poor working environment,the failure of vehicle transmission rotating components,especially for the early fault,is often submerged in the noise environment that may lead to misdiagnosis or missed diagnosis.Therefore,the research on fault diagnosis and health monitoring of vehicle transmission rotating machinery have become popular topics in the current research.With the rapid development of manufacturing technology,mechanical equipment tends to be integrated,intelligent,precise and complex.Therefore,this paper carries out theoretical and experimental research on three issues of vehicle transmission rotating machinery fault diagnosis:weak fault diagnosis,weak compound fault separation,and intelligent fault classification.The investigation focuses on improving the intelligence,robustness and noise adaptability of the diagnosis model.The main content and contributions of this thesis are summarized as follows:（1）The theory of unsupervised sparse feature learning is systematically studied,the sparse representation ability is deduced,and the generalized normalized sparse measure is studied theoretically.The theoretical deficiencies in practical application and the corresponding improved algorithm are proposed.（2）The advantages of unsupervised learning are introduced in weak fault diagnosis under big data environment.The relationship between blind deconvolution and sparse feature learning is studied theoretically.A cross-sparse multi-dimensional blind deconvolution algorithm（C-MBD）is proposed.C-MBD takes the deconvolution process as a sparse feature learning process.First,the sample features are convolutional activated.Then the objective function is constructed and optimized using cross normalization,and the main feature components are determined by the_1/2-norm of the features.The algorithm was verified by simulation and experimental signals.The comparison results show that the proposed method can achieve the feature extraction of impulse and modulation signals,and significantly improve the intelligence,robustness,and noise adaptability of the feature extraction.（3）A compound fault separation method based on Cross Sparse Filtering（Cr-SF）and Cross Kurtosis Pursuit（CKP）is proposed for weak compound fault diagnosis.First,the weight matrix and features are extracted by Cr-SF.And then,CKP is used to determine the main feature components through the spectral Kurtosis of the filters.The separation performance of the proposed method is validated by the simulation and the experimental signals under noisy environment.Compared with Fast Kurtogram,VMD,and CSF algorithms,the results show that the proposed method can learn the different filters from compound fault without any prior experience and perform superior robustness and noisy adaptability to the existing methods.（4）To study the influence of normalized parameters of the sparse feature learning method and solve the leakage of the feature learning process,an intelligent fault diagnosis method based on General Normalized Sparse Filtering（GNSF）and PCA is proposed.First,each training sample is randomly divided into segment matrix.And then PCA is carried out to obtain the training matrix.Finally,the training matrix is trained using GNSF and the fault classification results are given using Softmax.The algorithm was verified by the gear and bearing fault datasets.The effects of normalized parameters on accuracy,calculation efficiency and robustness were studied in detail.The experimental results show that the proposed method can solve the leakage problem in the feature learning process and provides a theoretical basis for the subsequent research.（5）To improve the training efficiency and feature distribution of the intelligent fault diagnosis method.Fast Convolution Activation（FCA）and Pseudo-normalization（PN）methods are proposed.FCA significantly reduces the dimension of the training matrix,achieves direct optimization of sample features.PN of the test features is used to guarantee that all features have similar contributions by dividing the recorded₂-norm of the training features.The experimental results show that the proposed method significantly improves the calculation efficiency and robustness.Compared with the traditional method,the calculation efficiency is improved nearly 10 times with the same accuracy.（5）Based on the theoretical and experimental research in previous sections,an intelligent fault diagnosis method is proposed for rotating machinery under strong noise interference environment.First,the training matrix is constructed by Hankel matrix of training samples.Second,the problem of feature drift is eliminated by weight normalization.Third,_1/2-sparse filtering is used to extract fault features.Fourth,the distribution of features is improved by feature normalization.The algorithm is verified by the fault signals of gear and bearing under noise environment.The classification accuracy and robustness with different SNR values and classifiers are studied.The effect of each improved technology is studied in detail.The experimental results show that the proposed method performs better noise adaptability.