Study On Image Recognition Approach For Vibration Fault Diagnosis Of Rotating Machinery

Combined with the project of National Natural Science Foundation of China Fault Diagnosis Method Based on Mathematical Morphology Graphic Recognition of Rotating Machinery and the practical project named Study of Fault Diagnosis Technology of Large-scale Rotating Machinery, the approaches of fault diagnosis based on the mathematical morphology graphic recognition for rotating machinery is investigated in this dissertation.In order to apply three-dimensional spectrum to diagnose the fault of rotating machinery, from the perspective of graphic recognition, based on the detailed analysis of graphic texture characteristic, according to the characteristic of periodic excited of the rotating machinery, the method of applying rotor dynamics theory to construct the basic function of the vibration response so as to deal with graph by means of interpolation reconstruction treatment is studied. On the basis of theoretical research, the application research of pretreatment interpolation of the three-dimensional spectrum graph of turbine is developed, the results indicates that the effect is good when applying the proposed method to treating the graph.On the basis of the analysis of changing law that the amplitude ranges rise and fall with the changes of rotating speed in the three-dimensional vibration graph of rotating machinery. The complex non-linear gray-scale algorithm based on the sine function is studied. Gray scale value of the region which is sensitive to fault is strengthend, improved signal-to-noise ratio. In order to make the treated gray scale histogram tend to be balanced in great dynamic range, the adaptive histogram equalization method is proposed to enhance the graph, improve the overall graphic contrast ratio, expand the dynamic range of pixel grayscale vale and improve graphic grayscale distribution profile. Example analysis on actual graph of rotating machinery is carried on by means of the proposed quantization and enhanced method.On the basis of the analysis of grayscale graphics of rotating machinery and the limitations of extraction method of the texture of the existing graphics edge, the fuzzy soft morphology method of the graph edge texture extraction is proposed. The fuzzy soft morphology filter is constructed to smooth the graph outline, eliminate the graph edge burr and the isolated point, filter the graph background noise and so on. The fuzzy soft structural element that is suited to processing the grayscale graphic for rotating machinery is designed. The edge textural property is extracted, the graph geometrical and topological structure is determined, and the simulation and the instantial algorithmic analysis are carried on.In view of the rotating mechanical parameter graph characteristic, the grayscale-gradient-primitive three dimensional co-occurrence matrix based on the statistical method, the structural method based on texture and the gradient method which characterize the graphic textural direction is studied for describing graphic textural digital characteristic. The rough degree, direction and spatial complex degree and direction of texture are reflected precisely. The graphic grayscale spatial distribution characteristic, spatial statistical dependence and pixel point gradient distributed rule are described accurately. Textural feature information in rotating mechanical state parameter graph is extracted effectively.On the basis of the analysis of the immune negative selection mechanism and the existing negative selection algorithm, variable real threshold immune negative selection algorithm which is suited to rotating machinery vibration parameter graph recognition. The detector's mutation algorithm, are studied. The variable real threshold immue negative selection diagnosis method is applied to analyze the"Iris"data and fault data of turbine. The result shows that the proposed method can detect various faults of turbine accurately.In order to overcome the limitation that the single nature fault characteristic and the single diagnosis method are difficult to be diagnosed accurately in the entire fault state space, the integration diagnosis method based on genetic algorithm is proposed, faults characteristic of different property and different diagnosis methods are fully used to exert each one's advantages, so the accuracy of diagnosis is increased. In this paper, a weighted matrix is established by integrating neural network and artificial immune diagnoses, Wavelet Packet energy and Bispectrum features using genetic algorithm. Experimental results indicate that both diagnosis accuracy and robustness of diagnosis system can be improved by the method.In order to verify the effectiveness of the proposed diagnosis method based on graphic recognition technology, experimental study is carried out on 600MW supercritical modeling turbine rotor bearing system for test-bed, simulating normal state and different faults such as rotor's unbalance, rotors'misalignment, steam-excited vibration, bear's looseness, dynamic static rub and impact ,and so on. Relatively higher accuracy is obtained, when diagnosing faults with diagnosis method based on graphic recognition which is proposed in the article.