| A theory to detect and identify multiple, simultaneous failures of actuators and sensors in practical control systems is presented. The type of failures that are addressed include zero output, biases, stuck actuators and sensors, hard-over failures, and reversal of control actuation. Using only the measurements from sensors, operator and computer commands, and the knowledge of the mathematical model of the system, an error metric (residual) is developed that contains all the information about the failures. An important feature of the error metric is that it linearly parameterizes the effect of failures, thus enabling detection and identification of multiple failures through a real-time, recursive identification method.; Starting with designs based on the linearization of a nonlinear system, the theory is extended to linear-parameter-varying nonlinear systems. The nonlinear error metric is robust to changes in the operating condition while retaining linear relationship to failure parameters. As control systems with redundancies pose special challenges during failure, a systematic procedure to detect and identify failures in redundant actuators and sensors is presented. When there is insufficient knowledge of the control system dynamics, or when the disturbance dynamics is complicated, a method to generate robust input-output maps that can be used as failure detection metrics is developed.; The theory is validated though simulations conducted on linear and nonlinear models of commercial aircraft. Realistic models for atmospheric turbulence and measurement noise are used. Multiple failures in actuators and sensors are identified in their source and magnitude. The results demonstrate that failure detection and identification is possible without altering the existing control system and the methods can be used in a retrofittable arrangement to improve aircraft safety. |
