A Controller Development Methodology Incorporating Unsteady, Coupled Aerodynamics and Flight Control Modeling for Atmospheric Entry Vehicles

Abstract

Atmospheric entry vehicle aerodynamics, flight dynamics, and control mechanisms are inherently coupled and unsteady. The state-of-the-art disciplinary models used for Mars entry vehicle simulation do not directly account for these time-dependent interactions, resulting in increased model fidelity uncertainty that can negatively affect controller performance. This can be especially detrimental given the more rigorous landing precision requirements and increased technological and volitional uncertainty expected for future missions. This work seeks to formulate and implement an entry controller tuning methodology that directly accounts for coupled, unsteady entry vehicle aerodynamic and control system behavior. The methodology uses a 6-degree-of-freedom coupled CFD-rigid body dynamics (RBD) model, extended to include flight control system modeling, for flight simulation while preserving unsteady flow history. This is capable of high-fidelity simulation to evaluate the performance of a controller, but the high cost makes it infeasible to directly use the state-of-the-art methodology for controller tuning which relies on thousands of short-duration simulations. Instead, multifidelity optimization is used. The coupled model is run to evaluate promising designs at high fidelity, while a lower-fidelity model is used to rapidly explore the design space. Crucially, each time the coupled model is executed, it produces new time-accurate trajectory and aerodynamic data that can be added to the training data for the low-fidelity aerodynamic surrogate model. A multifidelity surrogate is then constructed to provide a correction between the low- and high-fidelity results. As tuning proceeds, knowledge of the model is thus gained both by data fusion of the controller performance metrics, and by decreasing aerodynamic error in the low-fidelity surrogate. The methodology was developed through numerical experimentation with an entry vehicle equipped with a single-axis internal moving mass actuator for pitch control. A feed-forward neural network architecture with better performance than a state-of-the-art database was identified for use as the low-fidelity aerodynamic surrogate. A fusion-based multifidelity optimization method is implemented to leverage the quasi-hierarchical nature of the coupled and low-fidelity models. The methodology is demonstrated for tuning an angle of attack controller, yielding a controller that has better performance than one that is tuned using the state-of-the-art methodology.