Research On Time-varying Array Analysis Of Fault Spectrum Identification In Trackside Acoustic Diagnosis Of Train Bearing

The mechanical fault diagnosis technology provides a support to the safety assurance of the emerging manufacturing and avoids the occurrence of major accidents,which shows a huge economic and social value.High-speed rail as a synonym for China's emerging equipment manufacturing industry,its safety,comfort,efficiency has been widespread concern at home and abroad.Therefore,monitoring and diagnosing the running status of the train and preventing the occurrence of train accidents are of great significance.Among them,train bearing failure as a common fault,its related monitoring and diagnostic research such as trackside acoustic diagnosis has been a hot spot.Because the trackside acoustic diagnosis system uses a non-contact measurement means and can monitor the incipient failure,it has a great potential economic and application value.However,due to its unique way of acquiring the signal,it is inevitable that there are some measurement problems.This dissertation used the time-varying array analysis as the main means to obtain a clearly identifiable bearing failure sound spectrum in the trackside acoustic diagnosis system.It focused on the acquisition of sound signals in the presence of sound distortion,sound aliasing and weak sound to obtain reliable and accurate train bearing diagnostic results.Firstly,we analyzed the geometrical model of the acoustic diagnosis system and discussed the causes of the three measurement problems,i.e.,sound aliasing,sound distortion and sound weakness.By analyzing the structure of the train bearing,the foundation of identifying the bearing fault type based on bearing fault frequency was established.And the static acoustic acquisition scheme of train bearing and the dynamic acoustic acquisition scheme based on a microphone array of train bearing were designed respectively.By analyzing the static and dynamic experimental signals characteristics,the spectrum aliasing and distortion issue was verified.In this dissertation,the time-frequency filtering and time-frequency amplitude matching method were taken as an example to analyze the limitations of single-microphone separation and correction methods.It was pointed out that the single microphone signal lacking spatial position information is the main reason of failure.Secondly,according to the far-field array model,this dissertation proposed a time-varying spatial filtering rearrangement for separating and correcting the aliasing distortion signal.This method used the zero-angle spatial filter to obtain the time-center of different sound sources,and finally implemented aliasing source separation and correction through the time-varying spatial filter rearrangement.Since the time-varying spatial filter was almost independent of the signal energy,the array scheme had an advantage over the single-microphone method in the weak source separation and correction.In addition,experiments showed that the array scheme was also well suited for band-like and energy-discrete aliasing signal separation.Then,this dissertation further studied the array-based distortion signal correction scheme,and proposed a time-varying multiple signal classification and angle interpolation resampling method to realize non-parametric correction.The method obtained the source real-time position by time-varying multiple signal classification and established the resampling time series through the one-to-one mapping relationship between the source emission time and the receiving time to correct the distortion signal.Compared with the traditional method,the proposed method had many advantages such as no prerequisite knowledge,low computational complexity,strong robustness and be suitable for variable speed,which showed a potential in practical system application.Finally,a spectral feature enhancement method for periodic transient signals based on time-varying singular value decomposition is proposed,which constructs a pseudo-array signal by the Hankel matrix.This dissertation focuses on studying the basic properties of the time-varying singular value decomposition,and establishes a new method for fault frequency enhancement and recognition.Related research indicates that the technique not only can eliminate noise and improve the spectral characteristics,but also can retain the harmonic components of the periodic fault signal.Compared with the traditional method,it has a more significant advantage.In addition,the analysis of train bearing signal results show that the scheme has a significant effect in improving the sound spectrum characteristics of the fault signal.This dissertation took the single-and multi-source signals obtained by the microphone array as the processing objects.Based on the system acquisition model and the signal characteristic,a complete technical route of sound aliasing separation,sound distortion correction and sound spectrum enhancement with the time-varying array analysis as the main body is established,which lays a foundation for the final identification of the fault sound spectrum in trackside acoustic diagnosis system.