Validation of CFD-Based Recovery Furnace Models, 1998 International Chemical Recovery Conference Proceedings

Computational fluid dynamics (CPD) based recovery furnace models can be validated by comparisons of model predictions against experimental measurements of the predicted quantities. This paper summarizes the model validation efforts in a recently completed DOE-sponsored project to develop a new recovery boiler model carried out by the Institute of Paper Science & Technology (IPST) and the University of British Columbia (UBC). Model validation consisted of several distinct activities: l validation of isothermal flow predictions against water models of furnaces at UBC, l validation of flow predictions for cold flow in an actual recovery furnace, • validation of black liquor combustion models against laboratory data, l validation of the full recovery boiler model using data obtained from field measurements on two different recovery boilers, and • a benchmarking study of isothermal flow predictions against other CFD codes. Isothermal flow predictions obtained with the UBC code agree generally with the flow data obtained with the water models and the cold flow data obtained from an actual recovery boiler. The experimental data show that the real flows are often only quasi-stable and that there can be considerable time-dependent changes that cannot be predicted by a steady-state CPD model. Black liquor combustion models used in the overall furnace model have been based on laboratory data. However, the predictions of these models in the form which they are incorporated into the overall furnace model have not been thoroughly tested. Only limited success was obtained in the two hot furnace validation tests. Comparisons between model predictions and experimental data on gas temperatures and carryover from the first hot furnace test were not unreasonable. The model was also able to predict patterns of the black liquor sprays that had similarities to video images, but only after arbitrary specification of liquor spray initial conditions. In the second test, data were obtained on the furnace response to several different operating conditions, but only a single set of conditions was modeled and compared to data, because of the time and effort needed to set up, converge, and interpret a solution. A limited amount of gas temperature data was predicted reasonably well in this second case. A benchmarking exercise, which compared flow predictions using the UBC code and those made with the FLUENT code, showed that both codes did a reasonable job of predicting the flows and were in general agreement with each other.