Vibration Fault Dinagnostics And Prognostics Of Multi-State Information Fusion Based On Wavelet Time-Division Scale Level Moment

This research was supported by the NSFC (National Natural Science Foundation of China), project No. 50775083. This paper pointed out the shortcoming of the state-based fault diagnosis methods and proposed a novel fault diagnosis and prognosis method based on multi-stat information fusion, and developed a rotating machinery fault diagnosis system, on which some research were applied.At first, the fault simulation rotor testbed of turbine rotor shaft system was designed, and several typical faulty signals during speed rising were collected from this rotor test rig. For each fault , several tests were done to verify the repeatability. This established the analysis foundation in this paper.At the second, the new feature extraction method were studied. in this paper, based on an in-depth study of wavelet gray moment, both time-division scale level moment and Sub-scale-invariant momen were definition. On this basis, the paper discussed the factors which affect the fault diagnosis ability of time-division scale level moment. The analysis results to the caculation of six typical fault signals show that the time-division scale level moment can be used to display the detailed information of wavelet gray level image, extract the signal's characteristics effectively, and distinguish the vibration fault. Compared to the method of wave gray moment vector, the method mentioned in this paper can provide higher calculation speed and higher capacity of fault identification, hence it is more suitable for online fault diagnosis for rotating machinery.At the third, in order to achieve the online intelligent fault diagnosis, gray relational analysis and probabilistic neural networks were used to diagnose the fault. The concept of correction factor was proposed to improve the traditional gray relational algorithm, and had a good result. In addition, the probabilistic neural network for fault diagnosis was constructed, and its iput were time-division scale level moment, the result showed that probabilistic neural network was a good method for fault diagnosis. This paper also pointed out that in the lack of adequate samples, the gray relational analysis method was used to get a cumulative fault sample. And when enough samples have accumulated, and then using probabilistic neural network could be used for fault diagnosis in order to improve the accuracy of diagnosis.The fourth, the distribution of time-division scale level moment on start-up process in a typical fault condition was studied, and the problem of the start-up process fault diagnosis could convert into the close degree and similarity of two curves, and fault gray correlation degree algorithm was proposed, and verified. On this basis, In order to avoid the shortcomings of the start-up process-based fault diagnosis method in the application, a novel fault diagnosis method based on muiti-state information fusion was proposed. The specific algorithm has been implemented and provided in this paper with the wavelet time-division scale level moment as the fault features, wich can be got with wavelet transform . A quantitative discriminant fault indicator, Fault Comparability Degree, is also proposed. The experiment results show that the Failure Comparability Degree is able to better distinguish between different fault categories, and thus improved diagnostic accuracy can be achieved.The fifth, in order to be able to consider the dynamics of failure, failure prediction method had been a preliminary study. and this paper putforward a modified gray prediction mode and its iput was the time-division scale level moment. The prediction result of typical failure experimental data proved validity of the GM Gray Prediction Mode based on the time-division scale level moment.Finally, the configured online monitoring and diagnosis system was to achieve in accordance with the development trend. And application of the wavelet transform scalogram and the time-division scale level moment were researched and used in an actual diagnosis system. and this system can be realized for different rotating machinery.