Research On Coupling Faults Of Rotary Machinery Diagnosis Method Based On SVR

Rotating machinery is widely used in various kinds of mechanical equipment, such as steam turbine, turbine, generator, gas turbine, compressor, blower, aviation engine, etc. At present, with the improvement of complexity of rotating machinery, the mechanical equipments are linked more closely, and the faults are becoming more and more complex. The same time the fault may not be a single one, but two or more of the coupling faults. Therefore, coupling faults diagnosis for rotating machinery is an important measure to ensure the reliable and safe operation of the equipments.This paper studies the theory of support vector regression machine, the coupling faults diagnosis method based on SVR is proposed. Firstly, a database is established with various kinds of typical single fault. Then, the database is used for fitting to the diagnosis sample by SVR algorithm, and the weight coefficients are corresponded to the typical fault sample will be achieved. Finally, the proportion of all kinds of fault in the diagnosis sample can be determined according to the weight coefficients, namely this condition of the fault. And the feasibility of the method is validated by the simulation signals.Then, studying the problem of fault feature extraction, the method of feature extraction based on EEMD and permutation entropy is proposed. Seeing the experimental faults data of the rotor test rig as the research objects to extract the feature vectors, the discriminating degree of different classified faults can be judged by the distance of centroid and the sum of average radius between different kinds of fault in the space which is build by feature vectors. And compared with the sample entropy and energy, the result demonstrates the superiority of this method.At last, the three types of coupling faults, rotor mass imbalance-rubbing, rotor misalignment-rubbing and oil whirl-rubbing, are simulated by the Bently-RK4 rotor vibratory test-bed. The feature vectors are extracted by EEMD permutation entropy, than inputted the SVR coupling fault diagnosis model for training and diagnosing. Through the analysis of the achieved weight coefficients can realize the diagnosis of coupling faults. Meanwhile for the single faults diagnosis, also has obtained the good effect.