Research On The Fault Diagnosis Method Of Rolling Bearing Based On Acoustic Signal

Rolling bearing plays an important role in various industrial fields,but it is also one of the largest sources of mechanical faults.For a long time,researchers have taken vibration signal as the main research object of fault diagnosis,which makes the research of acoustic signal analysis method lag behind vibration signal.However,compared with vibration signal,acoustic signal can achieve non-contact measurement with higher efficiency,simpler sensor installation and easyer signal acquisition.Therefore,the research of fault diagnosis method based on acoustic signal has important practical significance and broad application prospects.The traditional idea of fault diagnosis is to map and transform the original signal,reduce the noise and redundant features,highlight the fault features,and then classify the fault features.However,facing the rolling bearing acoustic signal with low signal-to-noise ratio and weak fault features,the noise reduction processes of feature extraction algorithms based on traditional ideas have become very complex,and they are difficult to meet the real-time requirement.Therefore,combining image feature extraction algorithms and bag-of-words model,omitting the complex noise reduction process,two fault diagnosis methods based on acoustic signal are proposed for single-operating condition and multi-operating conditions respectively.The experimental results show that these methods are effective,can meet the real-time requirements,and have high practical value.Firstly,this thesis introduces the fault mechanism of rolling bearing and the common fault diagnosis methods based on acoustic signal.Based on the analysis of natural frequency and fault characteristic frequency of rolling bearing,the limitation of fault diagnosis by frequency spectrum analysis method is explained.Through the analysis of the relationship between vibration and acoustic signal,the feasibility of fault diagnosis by acoustic signal is proved theoretically.Secondly,based on the research objectives,the overall experimental scheme is developed,acoustic signals are collected,and the acoustic signals are pre-processed.The experiment content is the acoustic signal acquisition of 6 fault types of rolling bearing at 5 rotating speeds.During the signal acquisition experiment,the experimenters walk around the experimental platform and talk with each other to simulate the real scene.The collected acoustic signals are transformed into gray-scale images which are the basic for the following research work.Thirdly,FAST-Unoriented-SIFT feature extraction algorithm is proposed by combining FAST feature point detection algorithm and SIFT feature extraction algorithm.FAST-Unoriented-SIFT algorithm ignores the main direction of feature points and can quickly extract a large number of features.On this basis,a fault diagnosis method based on FAST-Unoriented-SIFT feature extraction and bag-of-words model is proposed.Experiments show the excellent performance of the proposed fault diagnosis method in the accuracy,stability and efficiency of fault recognition,and the advantages of FAST-Unoriented-SIFT algorithm compared with SIFT algorithm in the amount of features and the speed of feature extraction.Then,based on FAST-Unoriented-SIFT feature extraction algorithm,local kurtosis(LK)and local two-dimensional information entropy(L2DIE)features are proposed.To solve the problem of fault diagnosis under multi-operating conditions,an adaptive extended bag-of-words model is proposed,which can adaptively select the number of basic words.LK,L2 DIE and Unoriented-SIFT features are used to construct codebooks of each layer on the three-layer bag-of-words model respectively.Through layer-bylayer diagnosis and screening,the final diagnosis results are obtained.Experiments show that the adaptive extended bag-of-words model has obvious advantages over the traditional bag-of-words model in fault recognition accuracy and operation efficiency.Finally,the work of this research is summarized,and the future development of related research is prospected.This thesis has 37 figures,9 tables and 136 references.