Research On Mechanical Fault Feature Extraction Method Based On Time-frequency Analysis And Image Processing

Rotating machinery is the most common in machinery and its status in the industry is also very significant.Because it is usually operated under harsh working conditions and for a long time,it is easy to damage and the probability of failure is extremely high.Economic production has brought serious threats and losses.Therefore,it is extremely important to diagnose and detect mechanical faults.Taking bearings in rotating machinery as an example,the vibration signals are detected and studied.In this field,the extraction of fault features has always been a hot topic of research.In this paper,the research of this problem is carried out from time-frequency analysis and image processing,the purpose is to achieve double detection,to minimize the hidden dangers of safety.The specific research content is as follows:(1)Because EMD has an important disadvantage when decomposing signals,that is,the modal aliasing problem,to improve this problem,it is proposed to use HVD to improve it.This method is better than the typical EEMD and CEEMDAN in signal decomposition.There are advantages in analysis,high resolution and completely avoiding modal aliasing.In the wavelet threshold remove noise analysis,the improved threshold remove noise is based on the wavelet threshold and combined with the "three ? criteria".The remove noise effect is better than the traditional wavelet threshold.Remove noise contrast has been verified.(2)In the feature extraction research based on time-frequency analysis,a new remove noise algorithm based on HVD and improved threshold is proposed to analyze the fault signal and provide guarantee for subsequent fault feature extraction.This algorithm is applied in the United States Kai Prior to the analysis of bearing data of SICU,the analog signal was removed noise and compared with the methods of EMD,EEMD,CEEMDAN and improved threshold remove noise respectively.The subjective visual and objective evaluations consistently show the effectiveness of the method in this paper.it is good.Finally,in order to obtain the fault feature,the envelope spectrum analysis of the inner and outer ring signal after removed noise is supported by HHT technology.From the frequency comparison obtained by the above four methods,it is concluded that this method is the best in terms of the amplitude and resolution of the extracted spectrum.(3)In the study of feature parameter extraction based on image processing,firstly,for the two different noises of Gauss,pepper and salt,the remove noise method based on wavelet and anGaussian filtering and the remove noise method based on wavelet and AMF filtering are respectively proposed.Noise points and window fixed are improved by adaptive median filtering is AMF,and then the multi-resolution analysis of wavelet is used to improve the problem of edge blurring after remove noise.(4)When extracting feature parameters from the removed noise fault image,a feature parameter extraction method based on the canny operator is proposed.The extracted fault edges are calculated using the functions bwperim and bwarea,which are the two basic parameters.Finally,the degree of failure is judged by the roundness and deformity.