Research On Failure Detection And Fault Management Techniques Of Flush Air Data Sensing System

Accurate measurements of air data parameters are important for both flight testing and control of aircraft. Air data measurements are typically performed using intrusive booms that extend beyond the local boundary layer. These booms have been found to be excellent at making steady-state measurements at low to intermediate angles of attack. However, the performance of these instruments deteriorates during high angles of attack and highly dynamic maneuvers. They are also sensitive to vibration and alignment error, and are susceptible to damage during both flight and maintenance. Flush air data sensing (FADS) systems were developed in response to problems associated with intrusive booms. These instruments infer the air data parameters from pressure measurements taken with an array of ports that are flush to the surface of the aircraft, and are completely nonintrusive. However, because the locations of the pressure measurements are on the outer surface of the aircraft, locally induced flowfields can seriously complicate the calibration of these devices. Additionally, the semiempirical models that have typically been used to process the FADS pressure signals have experienced numerical instabilities, which resulted in momentary degradations in the system performance.Failure management is one of the most important factors for FADS. In the failure condition, some of the orifices may be blocked or some of the sensors or other data acquisition instruments may fail to function, resulting in erroneous values of computed air data parameters, corresponding to the erroneous pressure values. Many attempts have been made to address the failure management problems arising out of measurement uncertainty and to develop fault tolerant algorithms for FADS. These are aimed at evaluating the effects of various error sources on the overall uncertainty using error simulation and statistical error estimation methods. Neural network methods have been used to detect and compensate for lost input signals.In this paper, several aspects of the problem of FADS fault detection and isolation are investigated, including fault diagnosis based onχ2 analysis, fault diagnosis based on parity analysis, fault diagnosis based on neural network.1. Improved the principle ofχ2 analysis, perfected theχ2 detection and processing flow, presented regional correlation method aiming at the weak link of fault isolation;2. Improved and developed fault diagnosis based on parity analysis, refined and improved the fault detection and management flow, making the method more practical; 3. A new FADS neural network algorithm was proposed based on RBF network, the principle and process of RBF algorithm was expounded, a standard failure indicator vector scheme method was proposed based on RBF algorithm, the detail of fault detection and process flow was explained.