On the calculation of dynamic derivatives using computational fluid dynamics

In this thesis, the exploitation of computational fluid dynamics (CFD) methods for the flight dynamics of manoeuvring aircraft is investigated. It is demonstrated that CFD can now be used in a reasonably routine fashion to generate stability and control databases. Different strategies to create CFD-derived simulation models across the flight envelope are explored, ranging from combined low-fidelity/high-fidelity methods to reduced-order modelling. For the representation of the unsteady aerodynamic loads, a model based on aerodynamic derivatives is considered. Static contributions are obtained from steady-state CFD calculations in a routine manner. To more fully account for the aircraft motion, dynamic derivatives are used to update the steady-state predictions with additional contributions. These terms are extracted from small-amplitude oscillatory tests. The numerical simulation of the flow around a moving airframe for the prediction of dynamic derivatives is a computationally expensive task. Results presented are in good agreement with available experimental data for complex geometries. A generic fighter configuration and a transonic cruiser wind tunnel model are the test cases. In the presence of aerodynamic non-linearities, dynamic derivatives exhibit significant dependency on flow and motion parameters, which cannot be reconciled with the model formulation. An approach to evaluate the sensitivity of the non-linear flight simulation model to variations in dynamic derivatives is described. The use of reduced models, based on the manipulation of the full-order model to reduce the cost of calculations, is discussed for the fast prediction of dynamic derivatives. A linearized solution of the unsteady problem, with an attendant loss of generality, is inadequate for studies of flight dynamics because the aircraft may experience large excursions from the reference point. The harmonic balance technique, which approximates the flow solution in a Fourier series sense, retains a more general validity. The model truncation, resolving only a small subset of frequencies typically restricted to include one Fourier mode at the frequency at which dynamic derivatives are desired, provides accurate predictions over a range of two- and three-dimensional test cases. While retaining the high fidelity of the full-order model, the cost of calculations is a fraction of the cost for solving the original unsteady problem. An important consideration is the limitation of the conventional model based on aerodynamic derivatives when applied to conditions of practical interest (transonic speeds and high angles of attack). There is a definite need for models with more realism to be used in flight dynamics. To address this demand, various reduced models based on system-identification methods are investigated for a model case. A non-linear model based on aerodynamic derivatives, a multi-input discrete-time Volterra model, a surrogate-based recurrence-framework model, linear indicial functions and radial basis functions trained with neural networks are evaluated. For the flow conditions considered, predictions based on the conventional model are the least accurate. While requiring similar computational resources, improved predictions are achieved using the alternative models investigated. Furthermore, an approach for the automatic generation of aerodynamic tables using CFD is described. To efficiently reduce the number of high-fidelity (physics-based) analyses required, a kriging-based surrogate model is used. The framework is applied to a variety of test cases, and it is illustrated that the approach proposed can handle changes in aircraft geometry. The aerodynamic tables can also be used in real-time to fly the aircraft through the database. This is representative of the role played by CFD simulations and the potential impact that high-fidelity analyses might have to reduce overall costs and design cycle time