Research On Fault Diagnosis Method For Rotating Machinery Based On Manifold Learning And FRFT

With the rapid development of industry,the equipments such as motor trains and,fans and other equipment are becoming more and more complex and precise,and the performance requirements of their components are gradually increasing.However,the complexity and change of the industrial environment,coupled with the lack of adequate shutdown for maintenance and repair,which often make the equipment have a higher risk and failure.Once a part of the equipment fails,it will not only affect the production efficiency,but also cause the consequences of machine crash and casualties.Therefore,as the most common rotating machinery in mechanical equipment,it is particularly important to extract features efficiently and diagnose faults accurately.In this paper,combined fractional Fourier transform and manifold learning algorithm,the fault diagnosis method of single feature and multi-feature under multiorder fractional Fourier transform in fractional domain is studied to solve the problem of fractional order selection.Using the diversity of fractional Fourier transforms,the utilization efficiency of the fractional domain is improved.The fault diagnosis method of manifold learning in time domain,frequency domain and time-frequency domain is studied and compared with the fractional domain.In addition,based on abovementioned algorithm studied,fractional Fourier and manifold learning,and some necessary signal processing methods,this paper has written a set of fault monitoring and diagnosis system for rotating machinery,and the system is compiled by matlab.The detailed research contents are as follows:Firstly,the common fault diagnosis methods based on manifold learning are studied.The basic flow of manifold learning fault diagnosis method is analyzed,And three manifold learning methods in common use are introduced.The characteristics of time domain,frequency domain and time-frequency domain are analyzed,and high-dimensional feature sets of them are constructed.The experimental results of the dimension reduction and features fusion of the manifold learning method are verified by experiments.The results show that the appropriate method and effective feature space are beneficial to improve the accuracy rate of fault diagnosis.Secondly,because of the problem of difficulty in order selection and low efficiency of order using in the fractional Fourier transform method,the multi-order fractional fourier transform method fusion is used instead of searching for the optimal order to realize the automatic selection of the order,instead of the original artificial selection;at the same time,extract the sample entropy feature of multi-order fractional fourier transform and make full use of the diversity of the fractional domain to construct the high-dimensional feature set.The high-dimensional feature set obtained is subjected to manifold learning and the fault diagnosis is finished based on dimension reduction and features fusion,which make the accuracy rate of fault diagnosis.The practical application results show that the multi-order fusion method is superior to the singleorder method in performance of the feature extraction and fault diagnosis,and its performance of anti-noise is also better.Thirdly,for the real-time problem of fault diagnosis and the complexity of sample entropy,the multi-statistic feature,which is calculated simple,is contributed to extract the high-dimensions features set of fractional domain,instead of sample entropy feature.The manifold learning is used to fuse multiple statistical indicators under each order of the fractional domain and fuse multiple order to achieve multi-level fusion of features.The time-consuming of the sample entropy and statistical indicators is carried out by practical application.The results show that the statistical feature method is much more efficient than the sample entropy method.In addition,the experimental result of multistatistics feature of multi-order is compared with experimental results of the submultiple statistic features of single-order and the sub-single statistic features of multiorder,and the effectiveness of the method is further verified.Finally,this paper brings time domain features,frequency domain maps,EEMD,fractional domain features,filtering,manifold learning and other methods together to design a fault detection and diagnosis system.The system is compiled based on the MATLAB,which is simple,convenient and practical.The system can independently perform mechanical signal monitoring and diagnosis,and the user can also perform deeper analysis of signals based on system results.