Research On Intelligent Fault Diagnosis Algorithm Of Rotating Machinery Based On Deep Learning

With the advent of the era of industrial big data,the technology of fault prediction and health management(PHM)has been rapidly developed.As one of the core concepts of PHM,the fault diagnosis technology occupies an important position in the intelligent maintenance of mechanical equipment.Rotating machinery is the most common industrial machinery and equipment.Accurate and reliable monitoring and diagnosis,and then real-time determination of the health status of the whole mechanical system is of great significance to ensure the safety of equipment operation,improve production efficiency and save maintenance costs.In this thesis,rolling bearing and gear in rotating machinery are used as the core research objects,and deep learning technology is used as the core research method.Based on the shortcomings of existing methods,the intelligent fault diagnosis algorithm of rotating machinery based on deep learning was deeply studied from three aspects: convolutional neural network,full convolutional denoising autoencoder,multi-signal source fusion.(1)The background and significance of intelligent fault diagnosis for rotating machinery are introduced in detail,and the domestic and international status quo of three fault diagnosis research branches based on feature engineering,machine learning and deep learning are introduced.It is pointed out that the diagnosis method based on deep learning is the research trend of the current era.Subsequently,the vibration mechanism of two typical core parts of rotating machinery is studied,and the basic structure,fault mode,fault characteristics and reasons of rolling bearing and gear are discussed respectively,and the vibration mechanism and fault characteristic frequency are further analyzed.The basic work is done for the detailed research of the algorithm.(2)Aiming at the shortcomings of traditional signal processing combined with manual analysis methods and feature extraction combined with pattern recognition methods,a fault diagnosis algorithm model based on one-dimensional deep convolutional neural networks(1D-CNN)is constructed.After giving its basic theory,based on the experimentally collected rotating machinery fault data set,the influence of sample length,batch size and optimization algorithm on the network is discussed,and based on algorithm comparison experiments,the diagnostic superiority of the deep learning algorithm is verified.(3)Due to the noise coverage of the fault feature information,the performance of many rotating machinery fault diagnosis methods deteriorates sharply,and the existing deep learning methods cannot take into account strong and weak fault features.An intelligent fault diagnosis algorithm for rotating machinery based on residual dilated pyramid network and fully convolutional denoising autoencoder(RDPN-FCDAE)is proposed.First,a deep two-stage RDPN-FCDAE model is constructed,which is divided into three parts: encoding network,decoding network and classification network.In order to effectively express the data denoising feature of the coding network,the wavelat timefrequency image is first input into the coding and decoding network for unsupervised pretraining.Then the pre-trained coding network and classification network are combined into a residual dilated pyramid full convolutional network,and the parameters are finetuned and tested.This method is applied to the bearing vibration dataset of the test bed under different noise mode.Through comparison with other methods,the results show that the algorithm is superior to other methods in diagnostic accuracy,noise robustness and feature segmentation capabilities.(4)Aiming at the problem of low accuracy and low reliability when diagnosing complex and difficult samples with a single signal source,an end-to-end fault diagnosis network model(IFR-FSAFNet)based on information flow recombination and feature sequence attentive fusion of multiple signal sources is proposed.First,a multi-signal sources information flow reorganization(IFR)mechanism is proposed,which effectively solves the problem of information communication between network branches.Secondly,a feature sequence focused fusion(FSAF)program is designed,which adaptively fuses feature sequence information corresponding to multiple signal sources.Based on two data sets of rotating machinery with different numbers of difficult and simple samples,a large number of experiments have proved that the method is more accurate and reliable in the case of fewer parameters.This method is compared with several other literature methods,showing greater diagnostic advantages,and the model is more competitive in terms of cross-noise adaptability.The method presented in this thesis is not only applicable to the fault diagnosis tasks of rolling bearings and gears,but also to the fault diagnosis tasks of other rotating machinery parts and equipment.