Improving simulation predictions of wind around buildings using measurements

Abstract

Wind behavior in urban areas is receiving an increasing amount of interest from city planners and architects. In the context of Singapore, knowledge of wind behavior helps to improve building ventilation and, on a larger scale, urban ventilation. Computational fluid dynamics (CFD) simulation is often employed to assess the wind behavior around buildings. However, the accuracy of CFD simulations is often unknown. Measurements can be used to help understand wind behavior around buildings more accurately. In this thesis, the additional information provided by measurements is used to estimate the set(s) of parameter values of the simulation through solutions of an inverse problem. The information content of measurement data depends on levels of measurement and modelling uncertainties at sensor locations. This thesis proposes a model-based data interpretation framework for time-variant systems, which integrates information obtained from measurements with simulation results. This framework is based on the error-domain model falsification methodology. In error-domain model falsification, multiple model instances are generated through assigning sets of parameter values to a template model. Threshold bounds are used to falsify model instances that do not explain measurements. These bounds are defined using measurement and modelling uncertainties at sensor locations. Modelling uncertainties are uncertainties associated with the template model. Error-domain model falsification has been adapted for identifying parameter values in time-variant situations such as wind behavior around buildings and for making predictions at unmeasured locations using the parameter values that have been identified. In this adaptation, measurement data and modelling uncertainties may vary with respect to time, leading to dynamic identification of parameter values as well as time-dependent predictions at unmeasured locations. At each measurement location, the model-based data interpretation framework explicitly represents modelling uncertainties. Strategies are proposed to evaluate important sources of uncertainties in CFD simulations of wind around buildings, such as uncertainties associated with turbulence and uncertainties associated with thermal processes such as convection. Results show that uncertainties, including their biases, depend on the location and the time of day. The model-based data interpretation framework is applied to several full-scale case studies. The framework successfully includes modelling and measurement uncertainties in order to provide ranges of predictions at unmeasured locations. It is concluded that the model-based data interpretation framework has the potential to identify time-dependent sets of parameter values as well as predict time-dependent ranges of predictions at unmeasured locations. Prediction ranges at unmeasured locations are reduced after measurements.