Research On The Prediction Method Of Residual Service Life Of Rolling Bearings Based On Deep Learning

As an important part of health management(Prognostic and Health Management,PHM),its purpose is to analyze and observe the performance changes and degradation trend of industrial equipment through historical operation data,and then accurately predict the remaining service life of industrial equipment,so as to provide guidance and support for the maintenance decision of industrial cequipment.With the advent of the era of Internet big data,industrial digitalization and intelligent transformation has become the only way to achieve Made in China 2025.As an important product of industrial digitalization and intelligent transformation,the residual service life prediction method based on data-driven and deep learning has been widely concerned by relevant searchers as soon as it is published.As the name suggests,data-driven is a method to collect massive operation data of related equipment through advanced sensor technology and computer software,integrate and refine the data,and obtain the implicit relationships.Deep learning is a means to dig out the implicit relationships in data-driven data.With the innovation and development of deep learning,it has made remarkable achievements in the field of fault diagnosis and health management highlighting its powerful feature extraction,feature fusion,pattern recognition and life prediction capabilities.As an important component of rotating machinery,rolling bearing is widely used in water conservancy,electric power,machinery,petroleum industry,chemical industry,railway industry,navigation and aerospace industry.The rolling bearing performance plays a crucial role in the safety and stability of the whole rotating mechanical system.Therefore in the context of the era of big data,the deep learning method is applied to predict the remaining service life of rolling bearings,which is of great significance to improve the safety and stability of the whole rotating mechanical system.With rolling bearing as the research object,on the premise of data-driven method,using the deep learning method powerful feature extraction and feature fusion ability,and combined with the health management and residual service life prediction theory,predict the remaining service life of rolling bearing,aims to avoid the accident economic loss to the greatest extent,the later maintenance into prior maintenance,and strive to provide a new theoretical method for industrial equipment health management system.The specific studies of this paper are as follows:(1)In view of the problem that there is a large number of multi-source noise in the original vibration signal,which is difficult to eliminate,a new method of wavelet threshold denoising that can adaptively change the threshold function is proposed.Firstly,the basic knowledge of wavelet theory is introduced to illustrate the development course of wavelet theory.Secondly,an enhanced particle swarm optimization algorithm is proposed,which improves the inertial weight factor in the particle swarm algorithm,improves the search efficiency and reduces the possibility that the algorithm falling into the local optimum.Then,the wavelet threshold function with two shape factors is designed,which can be adapted to change the wavelet threshold by changing the shape factor.Finally,combining the enhanced particle swarm algorithm and wavelet threshold denoising method,the wavelet threshold denoising method based on PSO algorithm.This method can optimize the wavelet threshold function and obtain the best wavelet decomposition and reconstruction effect,so as to realize the effective elimination of multi-source noise in the signal.(2)In view of the difficulty of feature selection and integration,and the interference of too many human factors in the traditional health index construction method,a new method of independent health index construction by computer is put forward.First,the vibration features of the time domain,frequency domain and time frequency domain of the denoising signal are extracted.Second,a deep confidence network structure containing 5 layers of confined Boltzmann machines was constructed to construct the health index HI.Then,weighted fusion evaluation criteria including trend,monotonicity and robustness were designed.Finally,based on the IEEE PHM 2012 data set,comparing the proposed method in this chapter verifies the validity of the proposed method in this chapter.(3)In order to improve the life prediction accuracy of deep learning method and solve the problem that it is difficult to quickly determine the hyperparameters of deep neural network,a residual service life prediction method to improve the long-term and short-term memory network is proposed.First,the direction guidance operator is designed to introduce the mutation probability,which enhances the search efficiency of the ox whiskers search algorithm.Secondly,the traditional long-term and short-term memory network is enhanced,and the learning rate that gradually decreases with the number of iterations is added to improve the convergence rate of the neural network.Then,the enhanced beetle antennae algorithm is used to search the partial hyperparameters of the long short-term memory network.Finally,the optimal network structure allows rolling bearing to predict,which improves the accuracy of the prediction.(4)On the basis of the previous study in this paper,the experimental platform of rolling bearing degradation was built.On the premise of not changing the degradation mechanism of rolling bearing,the degradation experiment of rolling bearing acceleration was carried out,and the data set of the whole life of rolling bearing was obtained.And on the basis of this data set,the method proposed in the preface study,further shows the effectiveness of the proposed method.