Detecting upset in fault tolerant control computers using data fusion techniques

Celeste M. Belcastro

doi:10.17918/00000473

Future commercial aircraft will require flight-critical systems with high reliability requirements for stability augmentation, flutter suppression, and guidance and control. Verifying integrity of the control computer in adverse operating environments is a key issue in the development and certification of critical control systems. An adverse operating environment results from electromagnetic disturbances caused by high-intensity radiated fields (HIRF). Electromagnetic fields may cause analog electrical signals to be induced on the aircraft's wiring. These signals can propagate to the onboard electronic equipment and affect performance and reliability of the control computer by causing functional error modes known as upset. This thesis considers the problem of applying distributed detection techniques to detecting upset in fault tolerant control computers. Upset in the fault tolerant controller is detected by fusing the decisions from the upset detectors for N processors. Upset in each processor is detected by fusing the decisions from M control law calculation upset detectors. The implementation of data fusion techniques in this problem requires the mathematical definition of the upset hypothesis, the characterization of complex nonlinear control law calculations, and the design of the local detectors. A mathematical definition of upset is developed for the control law calculations, the processors, and the fault tolerant controller in terms of implicit performance requirements. A model of the parametric uncertainties caused by upset phenomena is developed. These uncertainties are modeled as generalized nonhomogeneous Poisson processes. Upset in each control law calculation is sensed in terms of the residual between a measurement of the calculation (that may contain upset) and an estimate of the correct calculation for the nominal (no upset) hypothesis. Approximate linear discrete state-space models of the complex nonlinear control law calculations are defined. Models are developed for the throttle and elevator commands of the B737 Autoland control system. These models are used for designing Kalman filters to estimate the B737 throttle and elevator control law calculations under the nominal (no upset) hypothesis. The threshold for the residual-based decision rule is derived for two design cases. Performance of the distributed detection system is assessed.

Detecting upset in fault tolerant control computers using data fusion techniques

Files and links (1)

Abstract

Metrics

Details

Detecting upset in fault tolerant control computers using data fusion techniques

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media