Data-driven Blind Source Separation Method For Multiple Bearing Fault Diagnosis Of Rolling Bearings

In the actual operation of the petrochemical equipment industry,the sensors that monitor its operation often collect a large amount of important information that contains status information and fault characteristics,which can be used for equipment condition monitoring and fault diagnosis.Therefore,the premise and key factor in condition monitoring and fault diagnosis is how to extract the state characteristics from the strong interference of the mechanical signals,and it is absolutely useful to objectively evaluating and diagnosing the object.However,in the actual industrial field,there is a lot of background interference such as a variety of mechanical structure and the unknown noise source signal coupling,resulting in the number of sensors less than the number of source signals often exists,also coupled with the transmission process of unknown factors like observation signal collected by the sensor often contains any possible factors and fault,to estimate the fault source signal,we can directly obtain from the observed signal valuable information.Therefore,in order to extract the fault source signal accurately and efficiently,we must suppress the background noise and other interference signals as much as possible.In view of the above problems,the paper faces the problems in the fault diagnosis of the equipment operation in the petrochemical industry.Taking the intelligent fault diagnosis system of the petrochemical equipment bearing as the starting point,the demand analysis of the fault diagnosis system of the petrochemical equipment is firstly carried out,and the overall system is determined by vary analysis.Through the requirement analysis of the petrochemical equipment rolling bearing intelligent fault diagnosis system framework,we hackle the logical framework analysis that the accuracy of the data,the effectiveness is the basis of intelligent fault diagnosis.Combining the current analysis and realization of petrochemical equipment's rolling bearing data acquisition technology in intelligent fault diagnosis process,we summarize the advantage and disadvantage of data driven below:the logical framework of intelligent fault diagnosis system data preprocessing,data driven fault diagnosis based on artificial intelligence and data driven based on blind source separation statistical estimation method.The main work and innovations of this paper are:1.Data-driven denoising rolling bearing fault signals by the improved wavelet analysis method.Based on the principle of traditional wavelet threshold denoising,a new method of wavelet threshold denoising is proposed.we design the standard deviation estimation index of the wavelet coefficients at each scale,and based on this,we construct a new adaptive threshold filtering function to realized filtering and denoising pretreatment of the collected signal.Finally by the simulation and example simulation,we verify the effectiveness of noise reduction.2.Data-driven decoupling rolling bearing multiple faults signals.In view of the coupling interaction characteristics of multiple fault data,we proposed linear discriminant analysis theory to decouple data mapping,based on this,we put data-driven artificial intelligence BP diagnosis method for decoupling mode identification.Considering the high coupling degree of multiple fault data,we propose the kernel function linear discriminant analysis method.Under this function,the multiple faults are decoupled,and furtherly the extreme learning machine is used for pattern recognition.At last,we compared these two models to verify the improved validity.3.Data-Driven separating rolling bearings multiple faults signals by blind source separation method.For the de-noised multiple fault data,the blind source separation method is designed to separate,the actual system error and observation error are introduced,the total least-squares method is further introduced to optimize the separated signal,and the experimental simulation is designed to verify the ability of the algorithm to separate the identification signal.