Augmenting flow simulations with experimental data to improve aerodynamic efficiency

点击翻译按钮获取中文摘要

Numerical simulations play an important role in aerodynamic design since experimental measurements are typically limited and difficult to measure in all regions of the flow. Simulations can provide significantly more information than experiments, but modelling assumptions are necessary since it is not computationally tractable to simulate realistic flow conditions. In many industrial applications, for example, simulations solve the time-averaged equations using a turbulence model. The resulting simulations are then validated using the limited experimental data that are available. This project investigates a more active role for experimental data by using it as an input to simulations. Experimental measurements, which are incomplete and uncertain, are fed into a low-fidelity simulation to produce a hybrid flow field that mimics large-scale features in the experiment. This procedure, known as data-assimilation, seeks to address the deficiencies of experimental data and modelling ambiguities in simulations to produce better flow predictions. Data assimilation is of particular interest to fluid mechanics since the resources that are required for a high resolution experiment or a full-fidelity simulation are often prohibitively expensive. As such, data assimilation is the only realistic option to predict complicated flows at an affordable cost. These predictions are essential not only for design, but also for understanding how flows can be manipulated by control to reduce drag and increase aerodynamic efficiency. A rigorous framework, which can accommodate three-dimensional flows and control devices, is needed for predicting and manipulating industrial flows. The project will culminate in a tool that can convert limited experimental data around complex geometries into a fully resolved velocity field in order to improve aerodynamic design. This work will data-assimilate incomplete experimental measurements of three-dimensional velocity fields around a model vehicle to improve simulation-based predictions of mean flow quantities. The trial data will consist of a simulation that has been intentionally corrupted to resemble experimental data. In other words, the input data will be disjointed, noisy and sporadic. Moreover, these data will be systematically reduced in order to identify the minimum number of measurements that are required for successful data assimilation. Once this has been achieved, the framework will be applied to experimental data from a wind tunnel. The improved mean flow field will be used to design a control strategy that reduces drag using synthetic jets on the vehicle. Finally, the controlled flow will be data-assimilated to quantify the reduction in drag.