Integrated failure accommodation methods

The paper presents an application of enhanced fault identification and fault tolerance applied to flight control systems. For fault tolerance, our research has shown that, in a linear system, short of total actuator failure, it is always possible to design a robust LQR controller preserving stability of the controlled system. For the case of flight control systems, stability may not be sufficient to insure passenger safety and it becomes necessary to consider trade offs between complexity of the control system and performance. The paper integrates fault detection and identification to manage a bank of robust LQR controllers designed to provide acceptable performance for a specific type and range of faults. A fault detection subsystem determines which controller should be in place at any given time. We apply the technique to design controller banks for a B747 model. The faults considered are loss of efficiency in the actuators for each of the four control surfaces: aileron, elevator, rudder and stabilizer. Good performance of the fault detection subsystem is essential and our project has researched several alternatives. Based on the information available about the system, the paper discusses alternative methods to implement fault detection ranging from methods based on detailed models of the faults to approached that rely on machine learning tools. Regardless of the theoretical performance of the proposed control system, any possible implementation must consider uncertainties and non linearity in modeling and errors in detection. For the case of the B747 we have carried out a combination of theoretical analysis and extensive simulation work. We present results describing the behavior of the controller bank under delays in detection and errors in fault identification, both in intensity and type of the fault.