Fault Diagnosis Methods Of The Aero-engine Based On Information Fusion

Aero-engine is the heart of aircraft, flight delays of civil aircraft and grounding incidents of military aircraft often occur due to engine failure. The serious harm and maintenance costs caused by the faults have been widely recognized. The case is especially true in the new machine in development and test. Therefore, this paper puts forward the research topic of engine condition monitoring and fault diagnosis in order to predict the occurrence of failures based on monitored parameters, with the purpose of making correct diagnoses to failures, ensuring the safety of flight and/or ground test, saving maintenance costs and testing costs, and improving the efficiency of maintenance and/or ground test.In combination with the characteristics of engine performance failures diagnosis and the distributed control network for field test, this paper proposes a three-level systematic fusion structure for engine test, by which optimization can be made hierarchically.Through the research of the method of space-time registration for multi-sensor data and the application of technology such as spline interpolation, filtering reconstruction, clock synchronization, and frame acquisition synchronization, this paper proposes a new method of system synchronization integration based on the model of frame synchronization.Using the basic theory of a variety of improved BP network, radial basis function network, probabilistic neural network, self-organizing feature map (SOFM) network and Elman regression neural network, this paper makes a comparative study about the reliability and accuracy of failure diagnoses in different ways of normalization and in different noise conditions.Through the research of diagnosis method of fuzzy clustering, the following results have been obtained: Gaussian membership function is the best choice to make fuzzy diagnosis with membership function. The cosine distance is optimal when distance closeness is used to make fuzzy diagnosis. Through the research of diagnosis method of vector machine support, it has been found that the method has the limitation that it does not correctly identify the distinction between two fault samples with strong linear correlation. This paper proposes the concept of classification rate. The diagnostic decision rule has been improved and a new diagnosis method for data fusion failure based on vector machine support has formed. This method can improve the reliability and accuracy of failure diagnosis.By researching information fusion fault diagnosis method based on D-S evidence theory, it has been found that when making fusion diagnosis of the same failure case, the credibility value of diagnostic output of two evidence samples with two different diagnostic algorithms is different, even conflicting sometimes. This paper, therefore, puts forward the concept of off rate to improve the diagnostic decision rules and algorithms and to further enhance the accuracy of fusion diagnosis.This paper integrates the method to extract the fault characteristics by five layers of wavelet decomposition with the way to make fault diagnosis by probabilistic neural network and therefore opens up the application of wavelet probabilistic neural network in the engine vibration fault diagnosis with satisfactory results.