Description: In order to fulfill the present and future aerospace needs of the nation, there has been a growing interest in adaptive systems incorporating learning algorithms. Before such adaptive systems can be adopted for use in safety-critical aerospace applications, they must be certified to meet specified reliability and safety requirements. Intelligent Automation Inc. (IAI) in collaboration with Wright State University (WSU) proposes to develop a novel systematic verification and validation framework for adaptive learning flight control systems towards real-time safety assurance and trusted autonomy. A Neural Network (NN) based adaptive controller is designed as an add-on to a previously certified baseline linear controller to enhance robustness to modeling uncertainty and fault-tolerance to system faults. Based on Lyapunov stability theory, an integrity monitoring scheme for the adaptive controller will be developed to detect potential controller malfunctions and unstable learning conditions caused by unanticipated hazardous conditions. The proposed architecture can potentially maximize the use of advanced adaptive controller with high performance capabilities, while ensuring the safety of the overall flight control system in the presence of unanticipated hazards. In Phase I, the algorithms will be demonstrated using a real-time quadrotor test environment.

Benefits: There are many potential NASA applications for this innovation, for instance, intelligent adaptive flight control systems, adaptive engine control, space exploration applications including mated flight vehicle coordination, docking, and control of autonomous robots, flyers, and satellites. The national Research Council has identified intelligent and adaptive systems as one of the five common threads for the "51 high-priority R&T challenge". Adaptive systems technologies have been identified explicitly to be the key enabler for intelligent flight controls, advanced guidance and adaptive air traffic management systems for improving safety and maintenance. Successful experimental results developed by NASA researchers have suggested the significant potential of intelligent adaptive control systems. These systems must be certified before they can be adopted for use in safety-critical aerospace applications. Conventional V&V methods are for not suitable for adaptive learning systems, and rigorous novel V&V methods must be developed before intelligent adaptive systems become part of the future.

The proposed approach can potentially be used for many safety critical applications, including military and commercial aircraft, U.S. air transportation systems, unmanned aerial vehicles, autonomous robots, nuclear power plants, etc. It will lead to benefits in the form of improved safety, survivability, and superior control performance of safety-critical systems.