Spatiotemporal distribution prediction of coughing airflow at mouth based on machine learning - Part I: Study on boundary conditions at mouth in numerical simulation of cough airflow

Abstract

In the post epidemic era, the movement and distribution of pathogenic airflow and droplets produced by cough in the building space have been widely studied. Due to the limitations of research methods, there are few detailed research data on the temporal and spatial distribution of boundary conditions during cough, which is the basis of research and the key boundary conditions of computer simulation. Previous experiments have obtained cough airflow velocity distribution away from the mouth. This study aims to infer detailed data at mouth for CFD boundary conditions based on these experimental data. This is the first part of the research. Based on experiments, the types of parameters contained in the boundary conditions near the mouth during coughing are discussed. The main parameters are determined, including the maximum velocity of the mouth air flow, and the distribution function of the ejected air flow, among others, and the approximate value range. Different parameter combinations are used as boundary conditions for simulation, and with various combinations, database of conditions are obtained. Preliminary machine learning is performed on these databases, and boundary condition data consistent with experimental results are inferred. The study demonstrates that when the velocity distribution of the air flow at mouth satisfies the normal distribution function on the central vertical two-dimensional profile, the maximum velocity of the mouth air flow is 15m/s. Part 2 will use the complex neural network model to fit and infer more accurate boundary condition. The findings of this study can provide more accurate boundary conditions for simulating pathogenic airflow, as well as a supplementary database for epidemiological research.