Learning and Adaptation for Intelligent Control of Rapidly

SSCI and Yale University (Prof. Narendra) propose to develop new practically viable intelligent control algorithms based on the Multiple Models, Switching and Tuning (MMST) methodology for rapidly-varying nonlinear systems, and apply these to the reconfigurable flight control design for Uninhibited Combat Air Vehicles (UCAV) in the presence of control effector failures and battle damage. In Phase I the work to be carried out by Yale University will be related to a theoretical study of the MMST methodology for nonlinear systems operating in rapidly-varying environments. In particular, the following tasks will be performed: (i) Develop a general methodology for creation of multiple models for a class of rapidly-varying complex uncertain nonlinear plants, (ii) Determine the criteria that can be used to choose the control law needed to compensate for rapidly varying dynamics of such plants, and (iii) Extend this methodology to general optimal control problems and stochastic control problems. In Phase I SSCI will concentrate on the application aspects of this methodology to UCAV flight control. In particular, the following tasks will be performed: (i) Develop new parametrizations for different types of control effector failures and battle damage, (ii) Develop efficient switching and/or interpolation strategies for achieving favorable performance of the overall system, (iii) Derive stability, robustness, and performance criteria for simple MMST-based reconfigurable flight controllers, (iv) Evaluate the performance obtained using the MMST-based reconfigurable control design for UCAV models, and (v) Design a nonlinear 6 DOF UCAV simulation testbed for evaluation of different intelligent control strategies. Successful completion of the above tasks is the main prerequisite for the development of a prototype of an autonomous intelligent oller light regimes will be investigated through cooperation between Yale and SSCI in Phase II.