Research On Fault Diagnosis Technology Of Belt Conveyor Bearing Based On Data Driven

As the main device of bulk material transportation,belt conveyor is widely used in mining,metallurgy,port and other industrial fields.It has the advantages of strong conveying capacity,simple structure,low cost and good versatility.With the continuous advancement of industrialization and the rapid development of technology,large belt conveyor with long distance,large capacity and high belt speed has become the main direction of development.However,due to the bad working conditions of belt conveyor,heavy load,fatigue,corrosion,high temperature and other factors,the core parts of belt conveyor will inevitably have different degrees of failure.Bearing is the main supporting part of the roller,and its health seriously affects the running state and efficiency of the equipment.Compared with other parts,the closed structure is often difficult to judge its health status.Once the fault occurs,it will affect the whole production operation,even lead to production paralysis,resulting in irreparable economic losses.Therefore,it is of great significance to study bearing fault diagnosis based on vibration signal.This paper takes the bearing of the main supporting part of belt conveyor as the research object.Through studying the characteristics of bearing vibration signal,noise interference and speed change in operating condition,the bearing fault diagnosis technology based on vibration signal is formed,which provides theoretical support and technical solution for the safe operation of bearing and belt conveyor.The main content includes:(1)According to the specific mechanical mechanism and working mode of the bearing,the vibration signal characteristics of the bearing fault and the influence of the actual operating conditions on the vibration signal are explored.Based on the vibration signal,the problems that need to be solved and the difficulties of the problem are analyzed in the fault diagnosis of belt conveyor bearing.At the same time,the source of sample data,collection methods and self-made experimental platform are introduced in detail.(2)Aiming at the problem of noise interference in bearing vibration signal,this paper proposes a bearing feature frequency extraction algorithm based on frequency band feature from the perspective of frequency spectrum.Firstly,by constructing Hankel matrix to decompose the time-domain signal,the diversity noise in the signal is extracted.Then spectrum fusion method is used to reconstruct the spectrum,which effectively eliminates the random characteristics of background noise and reduces the interference of noise to the resonance band in the spectrum.At the same time,the PSO algorithm is introduced into the spectrum reconstruction.The standard deviation of spectrum amplitude is set as the fitness function,and the delay parameters and embedding dimensions in Hankel are optimized.The optimal time domain signal splitting matrix is obtained to improve the utilization of noise random characteristics in the sample.Finally,MOMEDA algorithm is used to design the optimal filter and construct kurtosis spectrum for the specified frequency interval,which can effectively extract the periodic fault components contained in the sideband and realize the accurate extraction of bearing fault characteristic frequency.(3)Aiming at the problem that single fault frequency characteristic index can not distinguish fault degree,a bearing fault degree diagnosis algorithm based on step-bystep VMD and multi feature is proposed.Firstly,an improved step-by-step VMD algorithm is proposed to solve the problem that the decomposition parameters of different signal components cannot be unified.In this method,single component and under decomposition are used to extract the resonance frequency band of the signal step by step,which effectively avoids the problem of poor decomposition effect of different signals under specific decomposition parameters.Then,an initialization method of center frequency based on segment standard deviation is proposed.The stability of the center frequency position search results is improved by sliding to intercept segments and calculate the standard deviation.Finally,by extracting the basic features from the signal components and constructing the fault recognition neural network,the mapping relationship between the multi features and the bearing class labels is effectively established,and the accurate diagnosis of different degrees of bearing faults is effectively realized.(4)In order to solve the problems of parameter selection of VMD decomposition and complex feature extraction process,a method to construct a complete center frequency feature matrix is proposed based on the center frequency position feature of VMD component.Firstly,the influence of different number of components and sideband constraints on signal decomposition is studied to set appropriate combination of complete signal decomposition parameters.Then,based on the combination of decomposition parameters,the original signal is decomposed by VMD in turn,and the center frequency eigenvectors of each signal component are extracted.Each group of center frequency position features are recombined to build a complete center frequency feature vector.This method can effectively avoid the influence of the difference of VMD decomposition parameters on the signal decomposition effect,and further improve the completeness of the center frequency position distribution feature.Finally,the mapping relationship between the complete central frequency characteristics and category tags is constructed by neural network,and the effective diagnosis of different bearing failure degree is realized.(5)In order to solve the problem that the same fault features of different bearing speeds are quite different,and the bearing fault diagnosis under multi speed can not be realized,a bearing fault diagnosis algorithm based on image features and CNN is proposed.Firstly,through the analysis of the signal spectrum under different speeds,an image feature extraction method based on VMD is proposed.The method extracts the most prominent resonance band component in the signal by VMD single component search,and intercepts the feature image from the original signal based on the center frequency of the component.This method not only effectively extracts the fault features of the signal,but also avoids the loss of side frequency details in VMD component reconstruction.Finally,the extracted image features at different speeds are labeled with the same category label.Through the construction of deep convolution neural network and sample training,the optimization of network parameters,the further extraction of internal features and the accurate construction of the mapping relationship between signal and label are realized,and finally the bearing fault diagnosis under different speeds is realized.At the end of this paper,the works of this paper are summarized,and the related research technology is prospected.This paper has 87 figures,10 tables and 170 references.