Research On Key Techniques Of Fault Self Diagnosis For Novel Atmospheric Data Sensing System

The new Air Data Sensing System can not only simultaneously measure the height, the speed, the Angle of attack, the sideslip Angle and other flight parameters, but also complete self-validating by recognizing from their own online state. The system fully inherits the advantages of the flush airdata sensing technology and the self-validating sensing technology, meets the high stealth, high maneuverability and high reliability requirements for modern aircraft. The project centers to study the fault detection, fault location and fault diagnosis and other status self-validating method, to solve some key technical problems for the new Air Data Sensing System. The main research content is as follows:(1) to the fault propagation problems for the new Air Data Sensing System, a kind of fault propagation analysis method based on fuzzy probability Petri net has been explored. With powerful modeling and logical reasoning performance of Petri net, the maximum probability fault propagation path in the system is analyzed, the fault propagation model on the basis of its components and the system itself are established respectively, and the main failure mode fully covering the testing sample set is obtained. The test results show that the main failure of the new Air Data Sensing System results from the pressure sensor, the abnormal signal acquisition and processing circuit and pressure hole blocking, which is consistent with the expert knowledge and engineering experience conclusion.(2) to the fault detection and fault location problems for the new Air Data Sensing System, the fault detection and identification methods based on wavelet kernel principal component analysis and fault indicator vector have been explored. Using kernel principal component analysis method, the intrinsic relationship among multiple pressure tunnel and the relationship between the high-dimensional feature residual space projection and fault detection in the sample under test are analyzed, the advantage of multi-resolution analysis capacity of wavelet nuclear in the instantaneous fault detection is validated. Based on the redundant features of pressure measuring point layout, the intrinsic relationships of attack Angle and vertical pressure point and the relationships of sideslip Angle and longitudinal pressure point are analyzed respectively, the fault indicator vector library to characterization pressure channel status is built up, and that the system realizing the effectiveness of the fault source orientation by the fault indicator vector matching under the situation of the low Mach number small Angle of attack and high Mach number large Angle of attack is verified. The experimental results show that the method can realize multi-fault detection without omission. The typical fault detection rate is more than 90%, and the fault source localization rate is 100% when the total number of failure is less than three.(3) to the nonlinear fault feature extraction and fault classification problem for the new Air Data Sensing System, the fault diagnosis method based on the empirical mode decomposition and multiclass relevance vector machine has been explored. Using the signal adaptive decomposition characteristics of the empirical mode decomposition, different fault output signal types with the energy characteristics of diversity in different intrinsic mode components are analyzed, different types of fault feature vector set are established, and the modal aliasing resistance and fault feature extraction of ensemble empirical mode decomposition are validated. Using the small sample study characteristic of multiclass relevance vector machine, classification results being output by probability form, multiclass features of single model, the uncertainty relationship between the fault diagnosis and classification results is analyzed, the optimal kernel parameters selection method based on cross validation is studied, the multiple classifier model of different failure modes is built up, and the simultaneous fault type recognition advantage of multiclass relevance vector machine is validated. Compared with the traditional experience modal analysis method, this method has the obvious resistance modal aliasing advantage. The average classification probability of recognizing the corresponding fault types from the system normal work, large pressure fluctuation, pressure jump, bias and pressure constant value output is greater than 86% and 80%, and the fault classification accuracy is 100%.(4) to test and verify whether the method is effective for the fault detection, fault identification and fault diagnosis of the new AirData Sensing System, fault simulation and fault superimposed circuit to simulate all kinds of real fault based on the analysis of system failure mechanism and failure mode are designed, a new type of simulation test platform for Air Data System is devised to calibrate and test the system distributed pressure sensor, and to obtain normal test sample data and fault simulation sample set.