A numerical method for calculating indoor airflows using a turbulence model.

This report describes a numerical method based on a finite difference technique for simulating indoor airflows in a building using a JC-e turbulence method. The model treats three dimensional non-isothermal turbulent flows using the Boussinesq approximation for buoyancy. It solves the resulting nonlinear system of momentum, energy and turbulence equations by an explicit time marching technique to obtain a solution for either a steady state or transient flow. An upwind/central combination scheme with arbitrary specification for the switching parameter is used to approximate the convective terms. This switching parameter can be specified at each point in the flow regime allowing for different strategies in different flow regions. The switching technique includes both the central and hybrid schemes found in the literature. A pressure relaxation method is used to satisfy the Poisson equation for continuity. The model handles a variety of flow, pressure, temperature and heat flux boundary conditions' including prescribed inflows, outflows by either prescribing the flow or pressure, wall boundary conditions together with heat flux and temperature and/or heat transfer coefficients specified on the boundary. Volumetric heat sources are also included. The model has the ability of handling an arbitrary number of obstacles in the flow region. This permits the modeling of the effect of furniture and partitions on the flow field and also provides a means for modeling multi-room airflows. The predicted airflows can be used in a companion computer model for predicting the three- dimensional dispersion of contaminants in a building. The computer code for this model exists both in a vectorized version for the Cyber 205 supercomputer and in a non-vectorized version which has been successfully run on a Sun 3/260 workstation with a floating point processor board (based on a Weitek 1167) under a UNIX operating system on a Compaq 386/25 computer equipped with either an Intel 80387 or a Weitek 3167 coprocessor under an extended DOS operating system. The relative performance of these systems for the examples considered in this repon are 1 second per iteration for the Cyber 205,9 seconds for the Compaq 386/25 with a Weitek 3167 coprocessor, 30 seconds for the Compaq 386 with a 387 coprocessor and 90 seconds for the Sun 3/260 under UNIX. Isothermal simulations seem to converge in approximately 10,000 iterations and nonisothermal simulations in approximately 30,000 iterations. Several ideal and practical applications of the model are presented and the results of the simulations are compared with existing experimental data contained in the literature.