Learning Algorithms for Preserving Safe

Project Title: Learning Algorithms for Preserving Safe Flight Envelope under Adverse Aircraft Conditions

Partnering Institutions and Investigators:
Missouri Science & Technology University (Lead Institution): Tansel Yucelen (Science PI) and S. Balakrishnan(Co-PI)
Wichita State University: Animesh Chakravarthy (Co-PI in overall project, PI at WSU) and Jim Steck (Co-PI).

Funding Agency: National Aeronautics and Space Administration (NASA), Award Number NNX15AM51A

Funding Division: Aeronautics Research Mission Directorate (ARMD)

Total Funded Amount: $749,937

Project Duration: September 1, 2015 – August 31, 2018.

Abstract: Learning algorithms have great potential for providing flight safety in the presence of adverse conditions (resulting from, e.g., degraded modes of operation, loss of control, and imperfect aircraft modeling) and reducing aircraft development costs. A major roadblock to their widespread adoption is the lack of a-priori, user-defined performance guarantees to preserve a given safe flight envelope in general and commercial aviation. Current practice relies heavily on excessive flight testing as a means of performing verification and/or development of the tools to validate existing learning algorithms. Besides the cost, the major drawback of excessive flight testing is that it only provides limited performance guarantees for what was tested; the fixed set of initial conditions, pilot commands, and failure profiles. The drawback of current tools to validate existing learning algorithms is that such tools can only provide guarantees if there exists a-priori and complete structural and behavioral knowledge regarding any and all anomalies that might occur. The proposed research will address this fundamental gap in the utilization of learning algorithms for aerospace applications by (1) establishing a new theoretical framework along with necessary and sufficient conditions for guaranteed flight control safety and resilience in the presence of aircraft adverse conditions. (2) Methods will be developed to use these algorithms effectively in the pilot decision support display of NASA Ames that indicates the proximity of the aircraft to safe flight boundaries caused by adverse conditions. As a complementary effort to flight control, a set theoretic learning approach will be utilized in estimation theory to support of pilot decision-making via providing real-time aircraft flight health and prediction. (3) The proposed algorithms will be demonstrated in flight tests using CJ-144 fly-by-wire Bonanza aircraft. The novel feature of this research is that the proposed algorithms will have the capability to preserve a given, user-defined safe flight envelope through formal analytical synthesis at the pre-design stage, instead of excessive flight testing during the post-design stage.