Fault tolerance through direct adaptive control using neural networks

Many traditional fault tolerant Eight control systems augment a baseline controller with an algorithm to detect and then isolate faults. Often, this entails estimating a new system dynamics model and automatic synthesis of new controller parameters. An alternate approach to fault tolerant control can be utilized if the baseline controller is a direct adaptive control system. For failure modes that do not require control reconfiguration, this baseline adaptive controller demonstrates fault tolerance. This paper details the general formulation of a direct adaptive controller based on dynamic inversion with neural network adaptation. The implementation of this controller on several vehicles is also presented. In order to show the broad applicability of this approach, these vehicles are chosen from a wide range of categories, namely a tailless fighter, a reusable launch vehicle technology demonstrator, a helicopter, a ducted fan, and an acrobatic airplane. Results for nominal and fault cases for these vehicles are provided. Through this summary of fault tolerance results for each vehicle, direct adaptive control is shown to be well-suited to providing both nominal and fault tolerant control for vehicles from a wide range of categories.