Predictive Cost Adaptive Control of Six-Degree-of-Freedom CFD-Simulated Flight

We apply predictive cost adaptive control (PCAC) to CFD-simulated flight of a wing. PCAC is a discrete-time, indirect adaptive control technique based on model predictive control. Unlike model-based control, PCAC uses no prior model, and, unlike deep and reinforcement learning, PCAC uses no preflight data or offline learning. For rapid online multiple-input, multiple-output (MIMO) system identification, PCAC uses recursive least squares (RLS) with variable-rate forgetting. To determine control inputs, PCAC uses quadratic programming for receding horizon optimization, which allows PCAC to enforce physically meaningful magnitude- and rate-saturation constraints. PCAC is an output-feedback control technique that can be used with a reference model, and unknown disturbances may be either matched or unmatched. The present paper applies PCAC to a high-order finite-element simulation of a wing in free flight, where the rigid-body motion of the wing is subjected to aerodynamic forces and moments computed from the unsteady computational fluid dynamics (CFD) simulation. To follow a commanded position, PCAC requests control-surface deflections, which modulate the force and moment on the wing. The effect of the control-surface deflections on the motion and rotation of the wing is modeled using an arbitrary Lagrangian-Eulerian formulation. Results demonstrate the ability of PCAC to follow a commanded position of the wing using the control-surface actuators for flight with up to six degrees of freedom. With online system identification, PCAC improves its ability to follow position commands as additional commands are given. The results demonstrate the applicability of PCAC to function as a real-time adaptive autopilot, without prior knowledge of the wing aerodynamics and flight dynamics.