Machine learning (ML)-based intelligent CFD simulation for interactive design exploration of built environments

Understanding the effect of airflow in enclosed/indoor environments is of great interest due to its close relationship to occupant’s safety, thermal comfort, and energy efficiency. Optimally placed air conditioning could increase the comfort of the inhabitants. In hospitals, airflow can distribute germs and can pose a significant health hazard. Indoor airflow patterns can be very complicated, and computer simulations are an invaluable tool for understanding their characteristics during the initial design phase when designers can explore various design scenarios, which take into account multiple constraints (infection risk, energy efficiency) without compromising the critical factor of indoor air quality (IAQ). However, the complex and dynamic nature of the problem makes it challenging to perform the fluid simulations quickly to explore 100 to 1000s of design options interactively to identify the optimum.

This project will combine GPU-accelerated computational fluid dynamic (CFD) simulations of flows with machine learning (ML) algorithm to develop a novel data-driven and interactive physics-aware design optimisation method applied to the built environment. The project offers opportunities for both fundamental algorithmic developments, software development as well as working on an important application of computational engineering to climate change resilient built environment design. The student will be trained in the physics and modelling of fluid flows, in programming GPUs using C++ and CUDA as well as modern numerical algorithms of machine learning.