Deep learning for CFD analysis in built environment applications

a review

Abstract

The study and control of the airflow in indoor environment is of great importance since it directly affects human daily life primarily in terms of health and comfort. Fast and accurate airflow predictions are therefore desirable when it comes to built environment applications of inverse design, system control, evaluation, and management. Computational fluid dynamics (CFD) enables detailed predictions through numerical flow simulations and it has been consistently used to simulate airflow motion, heat transfer, and contaminant transport in indoor environment. However, CFD still faces many challenges mainly in terms of computational expensiveness and accuracy. With digitization, recent interest is posed on new data driven tools to either substitute CFD typically for faster predictions or aid the CFD simulation for improved accuracy. More specifically, the abilities of deep learning and artificial neural networks (ANN) as universal non-linear approximator, handling of high dimensionality fields, and computational inexpensiveness are very appealing. This work reviews current deep learning applications in built environment research, which are only limited to surrogate modeling as replacement for expensive CFD simulation. ANN enables fast and sometimes even real-time prediction, but usually at a cost of a degraded accuracy. For this reason, we also critically review what it is done and presented in fluid mechanics simulations research in general, to propose and inform about different techniques other than surrogate modeling for built environment applications and possibly improve the predictions quality as well. More precisely, ANNs can enhance the turbulence model in various way for coupled CFD simulations of higher accuracy, improve the efficiency of POD decompositions methods, leverage crucial physical properties and information with physics informed deep learning modeling, and even unlock new advanced methods for flow analysis such as super-resolution techniques. All these methods are very promising and largely yet to be explored in the built environment scene. Together with promising advancements, deep learning methods come with challenges to overcome, such as the availability of consistent large flow databases, the extrapolation task problem, and over-fitting, etc.