气流
人工神经网络
色散(光学)
湍流
计算流体力学
流量(数学)
模拟
计算机科学
人工智能
工程类
数学
机械
机械工程
物理
几何学
航空航天工程
光学
作者
Ting Dai,Sumei Liu,Junjie Liu,Nan Jiang,Qingyan Chen
标识
DOI:10.1016/j.buildenv.2023.110624
摘要
Fast fluid dynamics (FFD) with the dynamic Smagorinsky model can provide accurate predictions of outdoor airflow and pollutant dispersion. However, it is a time-consuming process because the model coefficient is calculated through a formula consisting of subgrid stress difference Lij and strain rate tensor difference Nij, which are obtained by filtering the flow field twice. To reduce the computing time, this investigation proposed a new dynamic Smagorinsky model which calculated Lij and Nij by means of an artificial neural network (ANN). The method was implemented in an open-source program, OpenFOAM. A single-building case was selected as the training set for the ANN. Other cases of varying complexity were used as testing sets to verify the model's generalizability. The dynamic model coefficient has the same distribution characteristics and similar values in different outdoor cases. The ANN model can capture those distribution characteristics and predict the dynamic model coefficient accurately. This study compared the accuracy and computing efficiency of FFD with the new dynamic Smagorinsky model, the standard Smagorinsky model, and the original dynamic Smagorinsky model. Compared with the original dynamic Smagorinsky model, the new dynamic Smagorinsky model saved the computing time by more than 60% when calculating turbulent viscosity, thus reducing the overall computing time by more than 20%, while maintaining similar accuracy. The use of FFD with the new dynamic Smagorinsky model is recommended for outdoor airflow and pollutant dispersion studies.
科研通智能强力驱动
Strongly Powered by AbleSci AI