Preventing unwanted airflow at building doors

Air that enters through gaps in doors and windows can significantly affect a building’s energy consumption. Businesses mitigate this unwanted airflow using air curtains, which produce downward-flowing columns of air to block heat, pollutants, and insects. Simulation techniques used to improve air curtains, however, often do not accurately predict their performance. Song et al. developed a data-driven model that predicts air-curtain performance with higher accuracy.

The team used machine learning algorithms to analyze the characteristics of air-curtain infiltration. They found the combination of pressure difference and air supply velocity can quickly determine the operational performance of the air curtain, and a backpropagation neural network with only one hidden layer can predict volume flow rate with high precision. Unlike existing methods, this approach can retain three-dimensional characteristics of an air jet and other details such as operation state.

“We performed a parametric study to determine the input features,” author Xinghui Zhang said. “Then, the data used to train the model is obtained by performing simulations using a FLUENT program. The final evaluation of regression and classification performance is based on a MATLAB program.”

The team evaluated three algorithms: support vector machine (SVM), random forest (RF), and backpropagation neural network (BPNN). By calculating the accuracy evaluation index, the authors identified reasonable input features.

Zhang said future studies may involve a greater range of input features and parameters and more complex algorithms that captured nuances beyond classification and regression.

Source: “A data-driven model to determine the infiltration characteristics of air curtains at building entrances,” by Linye Song, Cong Zhang, Jing Hua, Kaijun Li, Wei Xu, Xinghui Zhang, and Chengchuan Duan, Physics of Fluids (2023). The article can be accessed at https://doi.org/10.1063/5.0173678 .