Research On Spatio-Temporal Filter Design Methods Of Multiple Sources Aliasing In Wayside Acoustic Fault Diagnosis Of Train Bearing

Mechanical equipment fault diagnosis technology could effectively monitor,early warn,and diagnose the fault of manufacture core equipment,which shows a great economic and social value on avoiding or reducing severe safety accidents.With the development of technology,high-speed rail transportation has become one of the most important transportation ways,whose safety,comfortableness and efficiency are always the focus of attention at home and abroad.Monitoring train operation process through health condition monitoring technology is of great significance for preventing train accidents and ensuring train operation safety.Train wheel set bearing is one of the core components,whose health status directly affects train operation safety.As a common type of failures,the monitoring and diagnosis technology of wheel set bearing defect has been a popular research topic.Train bearing wayside acoustic fault diagnosis system is favored by researchers because of three advantages of non-contact measurement,early failure detection,and relatively low cost.However,complex acoustic environment of multiple moving sound sources aliasing brings a great challenge for the reliability of train bearing wayside acoustic fault diagnosis system.To obtain clear sound spectrum features,this thesis decomposes the multiple moving sound sources aliasing problem into three categories of ipsilateral,coaxial,and composite moving sound source aliasing,and gradually solves the difficulties of moving sound source separation&correction,extraction&enhancement,and number&the relative position estimation through a technical approach of spatio-temporal filter designing to finally form a systematic solution for the train bearing wayside acoustic fault diagnosis.Firstly,according to the measured multiple moving sound sources distribution when train passes,the thesis divided the multiple moving sound sources aliasing problem into three categories,namely,the ipsilateral,coaxial and composite moving sound source aliasing problems.And the characteristics of each category were analyzed,which reveals three technical difficulties in separating multiple moving sound sources,extracting signature signal of defective bearing,and estimating the numbers and relative positions of moving sound sources.With the currently used train bearing,the theory basis of static detection was established to discriminate bearing fault types through fault characteristic frequency.And the wave mechanics was also introduced for bearing fault dynamic acoustics detection.According to the three technical difficulties requireing to be solved,a static and a dynamic experimental platforms were designed for various signal acquisition schemes.By comparing the features of static signal with dynamic signal,the multiple sources aliasing problem was verified in the train bearing wayside acoustic fault diagnosis system.The limitations of traditional methods and array structures in solving the abovementioned difficulties were analyzed to lay a foundation for further research.Secondly,addressing the difficulty of moving sound source separation&correction in ipsilateral moving sound source aliasing problem,the thesis established a geometric model and a signal model based on a uniform linear array(ULA).Regarding this problem as a dynamic blind source separation problem from a novel perspective,this thesis proposed the spatio-temporal filtering rearrangement method by designing parameterized time domain separation matrix and time domain remapping matrix to realize separation and correction of ipsilateral moving sound sources.The simple correlation operation was carried out to remove the noise and the distorted non-target sound source components,so that the estimation of the target sound source could be obtained.The fault diagnosis of the target sound source was realized by envelope spectrum analysis.The sound source separation and correction is realized in time domain.The correlation operation is simple.These two factors lead to the advantages of calculation efficiency and output SNR of the proposed method.Subsequently,addressing the difficulty of fault signal extraction&enhancement in coaxial moving sound sources aliasing problem,the thesis proposed an adaptive sparse filtering method to realize extraction and enhancement of bearing fault signals through establishing a geometric model and signal model based on a cross-ULA,which overcame the essential defect of spatial filtering that could only suppress but not completely eliminate of the interference sound sources and noise in the non-target directions.The thesis combines a traditional beamforming technique with an unsupervised machine learning technique,and completely removes interference sound sources.Signal processing is completed in space and time domains which leads a high computation efficiency.After spatial filtering and sparse filtering,noise has been greatly suppressed so that the output SNR is high.This is an innovative application of combining unsupervised machine learning method with array signal processing method in solving multiple moving sound sources aliasing problem.Finally,addressing the difficulty of moving sound source number&relative position estimation in composite moving sound sources aliasing problem,this thesis made advantage of the Fibonacci sequence to design a Fibonacci array structure with excellent array performance.Based on the Fibonacci array structure,a deconvolution clean spatio-temporal map method was proposed,which realized the number and relative positions estimation of these moving sound sources.Analogous to the"acoustic camera"for stationary sound source localization by two-dimensional.space scanning,a new concept of"focused acoustic camera"was proposed for multiple moving sound source localization by time and space scanning.The thesis designed a Fibonacci array structure with excellent array performance.The microphone array signal was used to estimate the number and relative position of multiple moving sound sources with the clean spatio-temporal map technique.Then the spatial-temporal filtering rearrangement technique is used to realize initial separation and correction of the moving sound sources.Finally the adaptive sparse filtering technique was used to extract and enhance the bearing fault characteristic signal.A clear bearing fault envelope spectrum was obtained to achieve bearing fault diagnosis.A systematic technical solution based on multiple sources aliasing spatio-temporal filter designing,including clean spatio-temporal map method,spatial-temporal filtering rearrangement method,and adaptive sparse filtering method,has been established to provide a certain research basis for finally achieving wayside acoustic fault diagnosis of train bearings.