Reduced-order modelling for high-speed aerial weapon aerodynamics

In this work a high-fidelity low-cost surrogate of a computational fluid dynamics analysis tool was developed. This computational tool is composed of general and physics- based approximation methods by which three dimensional high-speed aerodynamic flow- field predictions are made with high efficiency and an accuracy which is comparable with that of CFD. The tool makes use of reduced-basis methods that are suitable for both linear and non-linear problems, whereby the basis vectors are computed via the proper orthogonal decomposition (POD) of a training dataset or a set of observations. The surrogate model was applied to two flow problems related to high-speed weapon aerodynamics. Comparisons of surrogate model predictions with high-fidelity CFD simulations suggest that POD-based reduced-order modelling together with response surface methods provide a reliable and robust approach for efficient and accurate predictions. In contrast to the many modelling efforts reported in the literature, this surrogate model provides access to information about the whole flow-field. In an attempt to reduce the up-front cost necessary to generate the training dataset from which the surrogate model is subsequently developed, a variable-fidelity POD- based reduced-order modelling method is proposed in this work for the first time. In this model, the scalar coefficients which are obtained by projecting the solution vectors onto the basis vectors, are mapped between spaces of low and high fidelities, to achieve high- fidelity predictions with complete flow-field information. In general, this technique offers an automatic way of fusing variable-fidelity data through interpolation and extrapolation schemes together with reduced-order modelling (ROM). Furthermore, a study was undertaken to investigate the possibility of modelling the transonic flow over an aerofoil using a kernel POD–based reduced-order modelling method. By using this type of ROM it was noticed that the weak non-linear features of the transonic flow are accurately modelled using a small number of basis vectors. The strong non-linear features are only modelled accurately by using a large number of basis vectors.