Comparison of mathematical models for exposure assessment with computational fluid dynamic simulation.

For many years exposure to airborne contaminants has been estimated by air or biological monitoring. In occupational settings, mathematical models increasingly are employed as adjuncts to monitoring, for instance, during process design or in retrospective epidemiological studies. Models can make predictions in a wide variety of scenarios, can be used for rapid screening, and may reduce the need for monitoring in exposure assessment. However, models make simplifying assumptions regarding air flow and contaminant transport. The errors resulting from these assumptions have not been systematically evaluated. Here we compare exposure estimates from the single-zone completely mixed (CM-1), two-zone completely mixed (CM-2), and uniform diffusivity (UD) models with workroom concentration fields predicted by computational fluid dynamics (CFD). The room air flow, concentration fields, and the breathing zone concentration of a stationary worker were computed using Fluent V4.3 for factorial combinations of three source locations, three dilution air flow rates and two emission rate profiles, constant and time-varying. These numerical experiments were used to generate plausible concentration fields, not to simulate exactly the processes in a real workroom. Thus, "error" is defined here as difference between model and CFD predictions. For both constant and time-varying emission sources, exposure estimates depended on receptor and source location. For the constant source case, ventilation rate was shown to be inconsequential to CM-1 model error. CM-1, CM-2, and UD models differed in their agreement with CFD. UD was closest to CFD for estimating concentration in the simulated breathing zone (BZ) near the source, although large errors resulted when the model was applied to the plane of possible breathing zones. CM-1 performed better for this plane but underestimated the near-source BZ exposure. For the near-source BZ location, CM-2 replicated CFD predictions more closely than CM-1 did, but less closely than UD did. Error in CM-1 model estimation of short-term average exposure to a time-varying source was highly dependent on ventilation rate. Error decreased as ventilation rate increased.