COVID19 pathogens are primarily transmitted via airborne respiratory droplets expelled from infected bio-sources. However, there is a lack of simplified accurate source models that can represent the airborne release to be utilized in the safe-social distancing measures and ventilation design of buildings. Although computational fluid dynamics (CFD) can provide accurate models of airborne disease transmissions, they are computationally expensive. Thus, this study proposes an innovative framework that benefits from a series of relatively accurate CFD simulations to first generate a dataset of respiratory events and then to develop a simplified source model. The dataset has been generated based on key clinical parameters (i.e., the velocity of droplet release) and environmental factors (i.e., room temperature and relative humidity) in the droplet release modes. An Eulerian CFD model is first validated against experimental data and then interlinked with a Lagrangian CFD model to simulate trajectory and evaporation of numerous droplets in various sizes (0.1 μm-700 μm). A risk assessment model previously developed by the authors is then applied to the simulation cases to identify the horizontal and vertical spread lengths (risk cloud) of viruses in each case within an exposure time. Eventually, an artificial neural network-based model is fitted to the spread lengths to develop the simplified predictive source model. The results identify three main regimes of risk clouds, which can be fairly predicted by the ANN model.
Keywords: Airborne pathogen transmission; Artificial neural network; COVID19; Eulerian-Lagrangian-CFD; Respiratory disease; Tempo-spatial risk model.
© 2021 Elsevier Ltd. All rights reserved.