Application Challenge 4-04 © copyright ERCOFTAC 2004
Comparison of Test data and CFD
The numerical data from the simulations undertaken for the Memorial Tunnel Test 615B have been compared with the measured values. The results illustrate qualitative agreement although large differences do exist between some of the values.
To illustrate the similarities and differences observed between the CFD simulations and the experimental tests, graphs are given of the velocities, temperatures and calculated volumetric flow rates for 5 of the instrument trees. The instrument trees selected are Loop 202 (located at 20 m from the south end of the tunnel underneath the fan room), Loop 302 (located about 66m south of the fire source), Loop 305 (located 11m north of the fire source), Loop 208 (located 189m north of the fire source, i.e. the centre of the tunnel) and Loop 214 (located 20 m from the north portal, underneath the fan room).
Below are graphs illustrating the differences between the predicted volumetric flowrates down the tunnel. The flowrates have been calculated in the CFD simulation by using an average of the velocities at each measurement location on the instrument trees.
As the graphs illustrate, the CFD simulation over-predicts the volumetric flow from the fans when the fans first start up and when they reverse near the end of the simulation. At 840 seconds (14 minutes) one of the fans switches off. The graphs clearly show the under-prediction of the volumetric flow at most of the instrument trees after this time, although when the fan is switched back on (at 1320 seconds or 22 minutes) the prediction agrees better with the measurements again. The predictions are the worst near the fire region (as the graphs for Loops 302 and 305 illustrate).
The over-prediction when the fans start up in forward and reverse mode (at the start and near the end of the simulation) may be due to the fact that the CFD simulation allowed a start-up time of 30 seconds for the fans to reach their full capacity, whereas in reality this could be longer. Note how the CFD results more closely match the measurements as the fans continue to operate after this start-up period. The discrepancy that occurs between the CFD results and the measurements when a single fan is switched off at 840 seconds may be due to a difference in the actual operating flowrate for this particular fan, i.e. this fan may have exhibited a slightly smaller operating efficiency in the test than for the others. Note the steeper gradient in the CFD curve at the moment when the fan is switched off. The CFD simulation did not allow for a gradual reduction in the flowrate to zero, as would be the case for a real fan. The discrepancy between the results for the instrument tree nearest the fire source (Loop 305) may perhaps be explained by the fact that the CFD simulation predicted a less well stratified layer of hot air and smoke from the fire than in the test.
Despite the differences noted above, the results of the simulation are promising compared to the measurements.
Graphs showing the variation in the predicted velocities with the measurements are given below. The first two sets of graphs are for three measurement levels of Loops 202 and 302:
Level A is at 1ft (0.3m) from ground level, Level D at 8 ft (2.4 m) and Level G at 13 ft (4 m). For Loop 202, the results are reasonably close to the measurements, except at measurement level G. It is not clear from the test data why the velocities are reversed at this level. A likely explanation would be the presence of a separation bubble forming just under the floor of the fan room at the south portal, which is not resolved in the CFD simulation.
The inadequacy of the CFD model in predicting the separation region is not surprising given the grid density in this region, together with the well known shortcomings of the k-ε model in this respect.
Graphs for Loops 305 and 208 are shown below:
The very high peaks appearing in the graphs for Loop 305 are due to unrealistic flow velocities predicted by the code near the fire source. The fire pan was modelled as a thin surface 0.85 m (2.8 ft) from the tunnel floor, with a source of heat and mass specified just above this surface. A gap was left underneath the pan, with a sink of mass to represent the consumption of oxygen by the fire. The high velocities were predicted through this gap in the immediate vicinity of the fire. In addition, the area of the fire pan was about 29 m2 in the model, rather than the indicated area of 45 m2 for a heat release of 100 MW. With hindsight, it would have been better to adopt a more realistic representation of the fire source than that used for this simulation.
For the purpose of comparing the predicted and measured values of temperature, contour plots are presented in addition to temperature – time graphs. Below are several sets of temperature contours illustrating the differences between the CFD prediction and the measured values:
Comparison of Measurements (top) and CFD (bottom) at 48 seconds
Comparison of Measurements (top) and CFD (bottom) at 1 min 48 seconds
The outermost temperature contours on all plots are 70°�F (21.1°�C). The next contour in each CFD plot is 140°�F (60°�C), which corresponds to the third contour on each measurements plot. The CFD contours are in steps of 70°�F (38.9°�C). As can be seen above, by 48 seconds the fire plume predicted in the CFD simulation had not spread as far as in the actual test.
At 1 min 48 seconds, although the outer contours approximately correspond in terms of the distance travelled by the hot plume down the tunnel, the hot layer predicted by the CFD simulation is less well stratified than in the test. Part of the reason for this is the flow predicted along the floor of the tunnel underneath the fire source – a consequence of the way the fire source was modelled in this simulation, as previously mentioned.
Comparison of Measurements (top) and CFD (bottom) at 7 min 48 seconds
At 7 min 48 seconds, a small amount of back-layering is predicted, as for the actual test. The temperature contours downstream of the fire reveal much more mixing across the tunnel section. The stratification apparent in the tests is not reflected in the CFD simulation results.
Comparison of Measurements (top) and CFD (bottom) at 27 min 48 seconds
At 27 min 48 seconds, the effect of the fan reversal in the CFD simulation has been to promote the movement of a ‘plug’ of hot air down the tunnel towards the north portal (left hand side). There is little or no stratification apparent. However the temperature contours roughly correspond well away from the fire source.
The contour plots have showed that despite the fairly good prediction of the longitudinal velocities and flowrates during the test, the temperature field is not so well predicted. The stratification is not adequately captured in the CFD simulation, particularly for the later times.
To enable comparison between the CFD prediction and the measurements at the instrument tree locations, graphs now follow in the same manner as for the flowrate and velocity graphs. First, for Loops 202 and 302:
The graphs clearly show the over-prediction of temperature at low level and the subsequent under-prediction at high level (for Loop 302, nearer the fire). For Loop 202, the predicted temperatures at levels D and G are however more in line with the measurements.
Graphs for Loops 305 and 214 are now shown below:
The instrument loop nearest the fire (Loop 305) clearly exhibits major differences between the measurements and the CFD prediction. Very high temperatures appear at low level and the peak points at all levels are too high. The gradually rising temperatures during the central period of the simulation reduce when the extra fan is turned on at 22 minutes (1320 seconds). The results are quite pleasing at Loop 214 (under the fan room at the north portal).
It is thought that the differences can be accounted for by the representation of the fire source in the model. In this model the following characteristics of the fire source were used:
·� A thin surface representing the fire pan, having an area of approximately 29 m2, rather than 45 m2 as indicated by the heat release per unit area property of No 2 fuel oil (for an 100 MW fire). With hindsight, the latter value would have been more appropriate.
·� A source and sink of mass was used, with a sink underneath the fire pan. In reality, as the fuel oil floats on the surface of the water, air will be drawn into the fire from the sides rather than underneath. A better practice would have been to omit the source and sink terms (being equal) and represent the fire solely as a source of heat and smoke (scalar).
·� The heat source was evenly distributed in the first layer of cells above the thin surface representing the fire pan. It may have been better to distribute the heat over a larger volume above the pan.
In summary, analysis of the results shows reasonable agreement between the measurements and the CFD simulation, particularly for the volumetric flowrates and velocities. Despite the differences apparent, the CFD calculations look promising.
It would be worthwhile to undertake a simulation with a more accurately modelled fire source. A more precise representation of the fire could well remove some of the anomalies predicted in this simulation, result in better stratification of the hot layer and lead to improved correspondence between the measured and predicted temperatures. In addition, the model could be refined to take account of radiation from the fire to the tunnel walls and the hot smoke layer. Grid refinement would also be recommended, given the limitations present at the time the study was undertaken.
© copyright ERCOFTAC 2004
Contributors: Nicholas Waterson - Mott MacDonald Ltd