Use this URL to cite or link to this record in EThOS:
Title: Approximating computational fluid dynamics for generative design
Author: Wilkinson, S. M.
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Wind loads are a critical consideration in the early-stage design of tall buildings for mitigation of wind-induced forces through form modification. Existing research in computational fluid dynamics (CFD) development tends either towards fast-inaccurate or slow-accurate approaches; therefore offering either constrictive response time or inadequate accuracy. Novel approaches that combine both speed and accuracy are required to keep pace with developments in parametric design softwares, such as GenerativeComponents. These software tools, primarily used in early-stage generative design, allow for broad exploration and optimisation within the potential design space, which in turn requires commensurate fast-yet-accurate analysis tools. This thesis investigates the use of reduced-order models to approximate CFD simulations of wind pressure on tall buildings. It is hypothesised that: firstly, wind-induced surface pressure on tall buildings simulated by CFD can be locally approximated by geometric features; and secondly, reduced-order model predictions dominate CFD simulations in both time or accuracy and are therefore a novel non-dominated approach. Predictions are made of individual vertex pressure based on input features formed from local shape analysis. The vertex samples originate from a procedural model set which is evaluated with either steady-state Reynolds-averaged Navier-Stokes (RANS) or transient large eddy simulation (LES). An artificial neural network is used for model reduction with the training set of vertex samples; the basis methodology of which is tested on a range of study complexities. To prove the scalability of the approach, this culminates in the use of LES as the basis simulation, a test set of realistically complex building models, and an alternative approach to urban wind interference generalisation is also described, whereby a one-off large-scale context CFD simulation can be used as input to repeatable design model predictions. Furthermore, a prototype tool and an outline for its integration with an existing online analysis framework currently under development is presented. The quantitative and qualitative results of the studies show it is possible to approximate surface pressure from local shape features, thereby decoupling the prediction from the basis simulation. The reduced-order model can achieve fast-yet-accurate results, since prediction accuracy and time are invariant, or independent, of basis simulation accuracy and time; being instead solely a function of the reduced-order model performance and the geometric complexity or number of test mesh vertices. Evidence is demonstrated by the positioning of the results as a non-dominated solution in the time-accuracy objective space and the subsequent alteration of the existing Pareto frontier.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available