# Tunnel fire

## Best Practice Advice for the AC

1. Key Fluid Mechanics

The Memorial Tunnel Fire Ventilation Test Programme consists of a series of full-scale fire tests carried out in a disused road tunnel in West Virginia in the United States from September 1993 to March 1995 [4]. The purpose of the tests was to establish the effectiveness of various forms of emergency ventilation in controlling smoke spread in the tunnel for different fire sizes. The results are summarized in an extensive database which can be used for validation purposes as well as design guidance.

A number of different ventilation arrangements were considered including fully transverse, partially transverse, longitudinal and natural. For this application challenge the longitudinal test 615B has been selected.

The tunnel is 854m in length, ascending from the south to north portals with a slope of 3.2%. The main tunnel section is rectangular, surmounted by a semi-circular ceiling. The width of the tunnel is 8.76m and the centre-line height 7.86m. At each portal there is a length of horizontal ceiling, 19.2m long at the south side and 18.5m at the north side. The fire source is located 238m from the south portal with a heat release rate rising to a maximum of just under 100 MW.

During the first two minutes of the test, the flow was allowed to develop without any forced ventilation and the smoke was observed, as expected, to migrate in the direction of increasing grade. Thereafter a number of jet fans were activated retarding and then reversing the uphill flow. Later in the test some of these fans were deactivated and then reactivated in order to observe the effect of this on the flow.

The general physical characteristics of the case may be summarized as follows:

• Transient period followed by steady-state

• Three-dimensional

• Heat and mass transfer

• Non-Boussinesq buoyancy effects (variable density)

• Turbulence

• Combustion

Key parameters are the heat release rate of the fire (as a function of time) in units of MW and the Froude number, defined for example as:

${\displaystyle Fr^{2}}$ = inertial forces / buoyancy forces = ${\displaystyle \rho V^{2}/(\Delta \rho gH)}$

where ${\displaystyle V}$ is the mass-averaged tunnel velocity far upstream of the fire, ${\displaystyle \rho }$ is the unheated air density, ${\displaystyle \Delta \rho }$ is the difference between ${\displaystyle \rho }$ and density of the hot gases from the fire, ${\displaystyle g}$ is the acceleration due to gravity and ${\displaystyle H}$ is the tunnel height.

The design or assessment parameters (DOAPs) relevant in this case are:

• Mass flow rate of air through tunnel

• Volume flow rate at different locations

• Critical velocity (at which back-layering of smoke in controlled)

• Time-varying velocity distributions

• Time-varying temperature distributions

• Time-varying carbon monoxide distributions

Of particular interest for the designer is the critical velocity and the extent of any back-layering of smoke opposite to the main flow direction.

The relevant UFRs for this AC are:

1.Buoyant flows in simple cavities (4-09)

2.Confined buoyant plume (4-08)

though neither is perfectly suited to it.

The best practice advice is based on two sets of CFD calculations for this AC, summarised respectively in [2] and [3-5]. Reference is also made to calculations for another set of fire tests in the Ofenegg tunnel described in [6]. The simulations referred to for this AC use a quite simple practical approach which captures the qualitative features of the flow and quantitatively some of the DOAPs but is not adequate for resolving some details of the flow, especially in the vicinity of the fire. It is therefore recommended that further simulations are carried out to determine what level of modelling is required to provide more accurate and detailed results.

2. Application Uncertainities

The key application uncertainties are:

1.Specification of wall roughness;

2.Specification of blockage effects and distributed resistance;

3.Lack of detailed information on external wind conditions;

4.Instrumentation accuracy.

For the temperature measurements, the thermocouples have an accuracy of ± 0.75%.

Due to the heavy layers of insulation which were applied to the instrument tubing and wiring, along with the instrument trees supporting the instrumentation, the thickness of this insulation greatly magnified the physical dimensions of the equipment it was meant to protect. The consequence of this was a reduction in the cross-sectional area of the tunnel, hence affecting the accurate measurement of air velocity.

In order to minimise the impact of the insulation on the measurements obtained from the instrumentation, cold airflow tests were undertaken, involving the use of vane anemometers positioned on a test rake at the same lateral and elevational coordinates as the pitot tubes on the instrument trees. Longitudinal airflow past the test rake and instrument trees was generated using the tunnel jet fans. The measurements obtained from the test rake and instrument trees were then compared, in order to obtain a suitable ratio for each pitot tube on the instrument trees. These ratios were then used in order to correct the measured velocity data obtained during the fire tests.

Computational Domain and Boundary Conditions

The computational domain should ideally include the full length of the tunnel between the two portals (this was the case in both reference calculations). In this case the losses associated with inflows and outflows to atmosphere need to be included, as discussed below. While it is possible to extend the domain outside the tunnel portals there is evidence from [6] to suggest that this is not necessary in general.

At both portals fixed uniform static pressure boundary conditions can be applied supplemented by appropriate loss terms. For the outflow condition, the imposition of a uniform static pressure implies the correct loss of one dynamic head. For the inflow case 1.5 dynamic heads should be added to represent the loss coefficient of 0.5 and the acceleration of the flow from rest. The two pressures can made different if wind effects are thought to be present.

Rough wall functions should be applied on all walls along with appropriate thermal boundary conditions. In both reference calculations, a fixed wall temperature was applied either 5 or 10°C above ambient, which is probably sufficient for the time period considered. A more complete approach would be to model heat conduction into the walls.

Jet fans can be represented either as momentum sources or as a paired inlets/outlet arrangement.

Discretisation and Grid Resolution

The case is most naturally solved using a 3D body-fitted hexahedral mesh which will give the best combination of accuracy and computational efficiency. (Note: the two reference calculations used respectively Cartesian and body-fitted hexahedral meshes.) As with most tunnel flows, the case is most sensitive to grid resolution in the transverse direction, across the tunnel width, with at least 750 cells recommended in any transverse plane. The grid should be clustered towards the solid surfaces though respecting any limitations placed by wall boundary conditions, in particular wall functions. A lower density can be used in the axial direction, for example 350 cells along the tunnel length, clustered towards any significant locations, especially the fire location.

Both reference calculations used first-order upwind schemes for convection discretisation which is probably sufficient for the present case which is dominated by axial flow generated by the jet fans. In general however the use of first-order convection schemes cannot be recommended for the modelling of tunnel fires, in particular if the details of the tunnel near-field and smoke stratification are to be predicted, when appropriate schemes of at least second order should be used.

Physical Modelling

5.1 Density variation and buoyancy

In both reference calculations the fluid is assumed to be an ideal gas with the thermodynamic properties of air. Both assume “weakly compressible” behaviour i.e. the density variations depend on the local temperature (and not pressure). The Boussinesq approximation for buoyancy is not suitable due to the large temperature variations.

5.1 Turbulence model

Both reference cases use the standard high-Reynolds-number k-ε model with wall-function boundary conditions and one of them, that in [2], uses basic buoyancy modifications. This appears to give sufficient accuracy for prediction of the general qualitative features of the flow for this case in which stratification does not play a major role. In tunnel fire cases in which there is more significant stratification there may be advantages in using more advanced turbulence models, for example models which take into account the anisotropic effects of buoyancy upon the turbulence field.

5.2 Fire source

A key aspect in the modelling of tunnel fires is the definition of the fire source. In both of the reference calculations, an inert approach was used, i.e. the fire was represented as a source of heat without the use of any combustion model. In this approach the fire-development curve (heat release rate versus time) must be prespecified, in this case based on the experimental data. This approach appears to be quite satisfactory for the case considered, i.e. with strong longitudinal ventilation, but may yield less satisfactory results in other cases, for example with natural ventilation, where lack of oxygen may control the fire size in a manner which is hard to specify a priori. For this reason it may be advisable to represent the fire source using a full combustion model in many cases of practical interest.

In one reference calculation the fire was represented as a volume source while in the other it was represented in the form of an inlet of heated smoke/air and an outlet removing an appropriate amount of unheated air.

Neither of the reference calculations attempt to model radiation directly using a radiation model. Instead the fire heat release rate is reduced by an approximate “radiative fraction” which is that proportion of the energy released which is assumed to be radiated directly away from the fire, typically taken to fall in the range 0.2 to 0.4 of the total. This approach appears to be adequate for the case considered and is common industrial practice, in particular where the detail of the near-field behaviour around the fire is of only secondary interest. However it is unlikely to be sufficient in more complex situations, for example a naturally-ventilated case with stratification or where structural heating effects are of interest. In this case it would be expected that the use of a radiation model, such as the discrete-transfer or Monte-Carlo models, may improve the detail of the predictions, however no data are presented here to confirm this.

Recommendations for Future Work

The following further work is recommended in order to establish more detailed best practice advice for the modelling of tunnel fires in general:

1. Further computations using more refined meshes.

2. Comparison of results with and without radiation modelling.

3. Comparison of results with different fire-source models, especially between the use of combustion modelling and inert “heat source” models.

4. Investigation of the effect of more advanced turbulence models, especially anisotropic models.

5. Investigation of more advanced wall treatments e.g use of conjugate heat transfer model into solid walls.

6. A general parametric investigation to determine which aspects of the physical modelling are most significant and which less significant.

## 7. References

1.Memorial Tunnel Fire Ventilation Test Program, Test Report, Bechtel/Parsons Brinckerhoff for the Massachusetts Highway Department, Nov 1995, with CD-ROM. Further details at: www.tunnelfire.com and http://www.fhwa.dot.gov/bridge/tunnel/tunres2.htm..

2.Karki, K.C., Patankar, S.V., Rosenbluth, E.M., Levy, S.S., CFD model for jet fan ventilation systems. Proceedings of the 10th International Symposium on the Aerodynamics and Ventilation of Vehicle Tunnels, Boston, USA, November 2000.

3.Castro, J D, Else, K, Rhodes, N: CFD Modelling of Memorial Fires, Mott MacDonald Report 50838/01/B, February 1998. (Work carried out for the Centre d’Etudes des Tunnels, France.)

4.CETU Report: Evaluation of Memorial Tunnel CFD Simulations, March 1999.

5.Rhodes, N: CFD Modelling of Tunnel Fires, World Road Association (PIARC) World Road Congress, Kuala Lumpur, 1999.

6.Biollay, H., Chassé, P.: Validating and optimizing 2D and 3D computer simulations for the Ofenegg Tunnel fire tests, 8th International Symposium on the Aerodynamics and Ventilation of Vehicle Tunnels, Liverpool, U.K., pp 357-379, July 1994.