Abstract: Advances in computational power have made computational fluid dynamics (CFD) simulations increasingly more accurate and efficient. It is now possible to design, analyze, and simulate aircraft using numerical techniques that greatly reduce the need for wind-tunnel testing.  One application of CFD for aircraft is in the development of stability augmentation systems and autopilots.  Here, CFD is often used to build and refine databases that serve as lookup tables for the controller design.  However, this process can become tedious and computationally expensive when many parameters are considered.  An alternative strategy explored in this work is the application of an adaptive model-predictive controller directly to the CFD simulation.  Such an approach makes no assumptions about or simplifications of the physics of the problem and instead lets the adaptive controller learn from feedback data as the simulation progresses.  We show that rapid learning of predictive cost adaptive control (PCAC, developed by Prof. Bernstein's group at the University of Michigan) is possible when the controller is directly coupled to the CFD.  Demonstration applications include studies in flight control, including flap and jet actuators, in two and three dimensions.  In addition to these demonstrations, the talk will delve into CFD-specific aspects of the coupling, including flight dynamics, actuation, and feedback.

Krzysztof J. Fidkowski is a Professor of Aerospace Engineering at the University of Michigan.  Dr. Fidkowski earned his S.B. in Physics and S.B., S.M, and Ph.D. degrees in Aeronautics and Astronautics from MIT.  Before joining the University of Michigan in 2008 as an assistant professor, he was a post-doctoral associate at the Aerospace Computational Design Laboratory at MIT.  He previously served as chair of the CFD Subcommittee of the AIAA Fluid Dynamics Technical Committee, organized fluids and CFD tracks at multiple AIAA conferences, and is an AIAA Associate Fellow.  His primary research field is in algorithmic development for computational fluid dynamics, specifically in the use of adjoint methods for numerical error estimation, mesh adaptation, and uncertainty quantification.