Citation

Nilsen AR, Log T. Fire Safety J. 2009; 44(1): 33-49.

Copyright

(Copyright © 2009, Elsevier Publishing)

DOI

10.1016/j.firesaf.2008.03.001

PMID

unavailable

Abstract

aStord/Haugesund University College, Bjørnsonsgate, 5501 Haugesund, Norway Information from full-scale fire tests are gathered and systemised. The knowledge from these tests is used as input to three different models, ranging from a simple spreadsheet model to advanced computational fluid dynamics (CFD) modelling, for calculating the temperature in the smoke layer. The deviation between the fire tests and the computed results is described and an evaluation of how this may influence the use of the models is discussed from the point of view of risk analysis. For experiments with small fires, i.e., one to five cars, one bus or one truck without cargo, the calculated temperature time curves from all models comply well with the measurements from the full-scale fire tests. For the larger fires, more deviation was found. The computational results were, however, on the same “order of severity” as the test fires, and thereby useful for hazard calculations. In contradiction to the general belief, the simplest models gave, in the same way as the CFD codes, results close to the values of severity recorded in the test fires with respect to HRR and smoke layer temperatures. The simple models may therefore be a good tool in risk analysis for not too complex tunnel structures. Keywords: Tunnel fire; Fire modelling; CFD; Hand models The Alpine tunnels mentioned were all equipped with transverse ventilation systems. It was expected that these systems would be able to exhaust the smoke from a fire to enable the escape from the close vicinity of the fire. All these tunnels had earlier experienced several HGV fires where none of these fires developed to a CTF. In the Alpine tunnels, however, the ventilation systems were not able to remove the smoke from the fire scene. Investigation reports [1], [2], [3], [4], [5] and [6] describe the systems as under dimensioned, based on preaccepted solutions instead of realistic fire scenarios. Preaccepted values based on small vehicle fires, buses or small HGV fires without fire spreading to other vehicles were probably found in the literature, such as PIARC [13] and other regulations, with typical heat release rate (HRR) of about 20–30 MW [13], [14], [15], [16] and [17]. Dimensioning fires in tunnels in all risk analysis should indeed be based on realistically described fire scenarios rather than some arbitrary table values [18]. Today, HRR from 100 to 350 MW may therefore also be described as possible tunnel fire scenarios [19], [20], [21], [22], [23], [24] and [25] and should be considered as representative HRR's in a tunnel fire risk analysis. In general, the models for evaluating fire severity in tunnels may be based on “hand calculations” using spreadsheets or more advanced CFD models. The models investigated in the present work were the hand calculation model based on a collocation done by Ingason et al. [19], [20] and [25] refined in spreadsheets [11], the CFD code SOLVENT [26] and [27] specially developed for tunnel fires with the post processing software Tec plot [28] and the powerful Kamelon FireEx [29], [30], [31], [32], [33] and [34] (kfx) including a combustion sub model. All the three models may be used for calculating temperatures at different locations in the tunnel with increasing precision and complexity from the spreadsheet version to the kfx model. The HRR input to the spreadsheet model is based on the t-squared fire [19], [20], [25], [35] and [36]. The hand calculation model and the SOLVENT CFD model do not have a combustion sub model. This means that the HRR must be a fixed input to the model. The calculated temperatures will then be based on the HRR and the cooling from the cold air entering the fire area and the heat loss to the tunnel structure. In the kfx model, the combustion model allows for continuous calculation of HRR based on the actual fire development and the interaction with the fire environment. kfx is a pool model, and there are no possibilities to simulate fires in 3D fuels, i.e., vehicles. A model for simulating fire on PMMA exists in the kfx model and a model for fires in cellulose materials are under development. A pool fire may be affected different than 3D cellulose fires when changes are made in the boundaries (back radiation) or the ventilation [37] and [38]. Also the base of the fire may be differently. In the pool fire, the flame starts at the pool surface at the ground while in the 3D fires the flame may start at higher elevations. The flames may therefore reach the ceiling faster when it comes from a 3D fire than a pool fire. Flames from the pool fires may in fact not reach the ceiling if the ventilation velocity is high enough. Combinations of 3D and pool fires in one model may give more realistic simulations. The objective of the present study was then to evaluate how these models, representative of the span in complexity and precision of fire modelling, would comply with a representative span of full-scale fire tests. The point of interest to be calculated is 10 m downstream the fire. This is near the fire and there are uncertainties in the correctness of all models so close to the combustion zone. This is the only point where all the included tests have recordings. This may be considered as a conservative place to compare models. Through the last 15 years, several full-scale fire tests in road tunnels have been carried out. The EUREKA EU 499—fire in tunnels project in 1990–1992 [21] and [22] in Repparfjord (Hammerfest, Norway) and the Runehamar Fire Tests in 2003 [23] revealed information about HRR (in the range 100–200 MW) and temperatures in tunnel fires. Several minor tests, in the Byfjord—and Bømlafjord tunnels in 1998 and 1999 [8], Hanekleiv—and Banehei tunnels [39] and [40] in 2000 and 2001, Ømbesvik tunnel over the years 1997–2001 [41], [42], [43] and [44] has been carried out and several more like them world wide [7]. In Norway, such tests are used for evaluating the contingency plan and tunnel safety systems. Measurements of temperature, CO2 and NOx, ventilation velocity (and smoke movement), fuel and of course traffic load and tunnel geometry are usually presented in the reports. Large pool fire experiments, such as the Memorial tunnel fire tests [27] and [45] may also be a basis for data collection. HRR and temperature data from these full-scale experiments are collected and systemised in the present work. In the case of lacking HRR data, the HRR had to be estimated based on knowledge of the fire scenarios, material data (especially the calorific potential) and the measured temperatures from the experiments. The calculated temperatures from all three models to be evaluated in the present work were compared to recorded temperatures in the full-scale tests. In Repparfjord tunnel fire test, two tests involving single cars were carried out [21]. Temperatures and smoke movement were recorded. HRR was estimated and the most probable HRR curves were proposed. Fires in two cars were carried out in the Norwegian tunnels Bømlafjord, Byfjord, Banehei and Hanekleiv [8], [39] and [40]. Two of these, Hanekleiv and Banehei, are two tube tunnels of high traffic loads (20,000 and 33,000 vehicles per day, respectively). The Byfjord and Bømlafjord tunnels are long sub sea tunnels (lengths of 5800 m (232 m below the sea) and 7800 m (260 m below the sea)) with moderate traffic loads (5,000 and 2,500 vehicles per day, respectively). Temperature curves from several of these fires showed remarkable resemblance, as shown in Fig. 1. The tunnels Byfjord, Bømlafjord and Banehei are all tunnels with large tunnel cross-section area. They have three lanes and their tunnel cross-section areas are 76 m2 (T11.5 profile [14]), the ventilation velocity was 3 m s−1. Two cars from the 1980s were set on fire. The recorded temperatures did not exceed 50 °C and the HRR in these fires was proposed as 5–10 MW. In the Hanekleiv tunnel, the tunnel cross-section area was 50 m2 (T8.5 profile [14]), and in the Repparfjord tunnel the cross-section was 30 m2. In the Hanekleiv tunnel, the HRR may have been larger than in the others experiments. The cars were arranged in the same way but one of the cars was a van and may have contributed both with more fuel and higher flame base. The ventilation velocity in the Hanekleiv tunnel (2–2.5 m s−1) was higher than in the C21 fire test in Repparfjord (0.3 m s−1). For the other car, C11, the ventilation velocity was not given. The full-scale tunnel fire test in the Repparfjord tunnel, width 6 m, height 5 m (cross-section area 30–40 m2) and length 2000 m involved larger vehicles, such as HGV's and a public bus. Estimates of HRR in the HGV and bus experiments were 124 and 30 MW, respectively (see Fig. 2a). Concentration of CO, CO2, O2, ventilation velocity and temperatures were recorded. Two steel freight containers were placed as a flow obstruction upstream of the fire in the HGV experiment. For the HGV fire, temperatures downstream the fire front at +10 m were recorded (see Fig. 2b). The temperatures reached 700–800 °C after 13 min. After 5–10 min, the fire spread from the driver's cabin to the cargo. The ventilation velocity was 6 m s−1 for the first 13.5 min. The fans were then stopped. This drop in ventilation velocity had a major impact on the HRR. It dropped from 120 MW to about 60 MW (see Fig. 2a). The temperature downstream dropped about 100 °C (see Fig. 2b). The fans were restarted again at 16.5 min giving a ventilation velocity of 3 m s−1 and temperatures 850 °C and HRR 128 MW. In the bus fire experiment with a low ventilation velocity, estimated to be 0.3 m s−1, the recorded temperature was about 600 °C, i.e., 200 °C less than the maximum temperature recorded in the HGV fire test. The Memorial tunnel is 850 m long with a tunnel cross-section area of 60 m2. Four of the 98 fire tests involved large diesel pool fires giving HRR about 100 MW. The variation in temperature and HRR in one of these fires, 621A, (longitudinal ventilation), is shown in Fig. 3a and b, respectively. The HRR was slightly larger than the pre-calculated values in the start (2 min, 130 MW) and at the end (24 min, 170 MW) of the experiment. In between, the HRR varied from about 50 to 90 MW. For the 621A, the average temperature was about 600 °C at +10 m downstream from the fire and 0.5 m beneath the ceiling. Through these experiments the ventilation velocity was varied by switching fans on/off giving a ventilation rate about 2.5–3.0 m s−1. Four tests simulating HGV fires in a longitudinal ventilated tunnel were carried out in the Runehamar tunnel (cross-section area 47–50 m2). In these tests, the tunnel ceiling and walls were protected against the fire by thermal insulation resulting in a reduced cross-section area (32 m2). This insulation may have reduced the heat loss to the walls and ceiling downstream the fire resulting in higher temperatures. The burning materials simulated a trailer cargo load of wood pallets and plastics in the T1 test. The total calorific potential in these materials was about 240 GJ. The measures of the load were: length 10 m, width 2.9 m and height 3.3 m (starting 1.1 m above the driving lane). The average heat of combustion of these materials may have been around 20 MJ kg−1 and the ventilation velocity was 3 m s−1. The HRR was reported to reach a maximum of 200 MW and temperatures above 1300 °C (see Fig. 2a and b, respectively). Three principle types of models are regularly used for calculating consequences in fires. These are hand models, two-zone models and field models (CFD) [7], [35] and [36]. CFD models are based on the finite volume method to calculate the governing equations. Using a numerical method, the model solves the 3D, time-dependent equations (field equations) describing the laws of conservation of mass, momentum, energy, turbulence parameters and species, subject to the specific boundary conditions of the problem, throughout the domain of interest. However, all input to the models depend on a qualitative scenario description. In this scenario description, the fire size and growth rate may be estimated based on different models. The scenario may be a description of earlier tunnel fires or settings that may occur in the future. From these HRR data calculations of temperature, toxicity, smoke movement and sight, etc., may be carried out with the three principle consequence models. Several factors may contribute to the HRR: The overall HRR in fires may generally be expressed as [46] Over some years, Ingason et al. [19] have gathered information on heat release rates from real fires, small-scale experiments and full-scale tunnel fire tests. From this work, models for predicting future heat release rates to be used in emergency preparedness work may be expressed by [20] and [47]: The SOLVENT model is a CFD tool verified by a 98 full-scale fire tests in the Memorial Tunnel Fire Ventilation Test Program with pool fires in the range of 10–100 MW. SOLVENT is based on the general purpose CFD code (COMPACT-3D) with buoyancy-augmented k–ε turbulence modelling and includes component models for representing jet fans, the fire region, radiant heat transfer from the fire, smoke movement, wall roughness and conjugated heat transfer. The model may be used for fire and ventilation calculations. It is easy to use and have a low entry level. The calculations are based on the HRR data set up for the scenario. This is done by determining the mass flux of volatiles in a defined volume. HRR is calculated from the mass flux and is not coupled to any change in tunnel ventilation or radiation from flames or hot smoke. Data from the SOLVENT model is easily post processed in the Tec plot software. Temperatures at different locations, ventilation velocity, pressure, smoke density and visibility as a function of time may then be presented for scenario evaluation and risk analysis. More advanced simulations were done in the Kamelon FireEx CFD (kfx) model. It is capable of calculating heavy and light gas dispersion and hydrocarbon spray and pool fires in the open or in a tunnel enclosure. The model has been validated against a range of fires from small-scale laboratory fires to large-scale jet and pool fires. Recreating the HRR calculated based on 3D fire tests, for example fires in vehicles or wood cribs, may not be possible in this code, or at least difficult. Evaporation rates in pool fires may be affected differently by changes in the surroundings due to, say, flame behaviour, radiation from other burning items or the surrounding or the orientation of the fuel and. The code uses a Cartesian finite volume technique to solve the basic equations. The effect of turbulent transport is modelled by the k–ε turbulence model. The Eddy Dissipation Concept (EDC) is used for combustion modelling and the Eddy Dissipation Soot model and CHEMKIN is used for the calculations of thermodynamic properties. A Discrete Transfer Model is used for Radiation modelling. The HRR is determined from the mass flux of volatiles caused by the evaporation of fuel. The HRR is thereby back coupled to any change in tunnel pressure, ventilation rate, local temperatures and radiation from flames and hot smoke, as principally described in Eq. (3). The HRR curve for a tunnel fire is a good description of the fire scenario. In order to use the hand model, Eqs. (4), (5) and (6), in the prediction of the HRR curve, some features of the vehicle must be known, such as the fuel mass, heat of combustion and time to maximum HRR. A trial-and-error-method has been used to assess data for the tests were no HRR was available. For the small car fires used in this work, except the C21 fire in the Repparfjord experiments [21], no calculations of HRR based on measurements from the full-scale tests were done. The HRR for the fires in the Byfjord tunnel, the Bømlafjord tunnel, the Hanekleiv tunnel and the Banehei tunnel were calculated based on data in Table 1, using Eqs. (4), (5) and (6). Values of α and β were in the present work estimated based on available information from the fire tests, such as recorded temperatures and fire development. Two cars (1000 kg each) with estimated 20–25% combustibles (30 MJ kg−1) give about 14 GJ. For the experiment in the Hanekleiv tunnel, the fire was a car and a van weighing a total of 3000 kg giving 18 GJ. For the C11 experiment in the Repparfjord tunnel, the total calorific potential was proposed to be 6 GJ [21]. Ventilation velocity was not reported in the EUREKA experiments. In the present work a ventilation velocity of 0.4 m s−1 was estimated for the EUREKA experiments based on general experience with unventilated tunnels. In the other tests ventilation velocities of 3 m s−1 were recorded except for the Hanekleiv tunnel where it varied from 1.5 to 2.5 m s−1. The time to maximum HRR was calculated in the EUREKA C21 experiment. In the other experiments, this time was judged after the temperature peaks. Maximum HRR was estimated based on previous experience with different fire sizes, as given in Table 1. Growth rates, α, for cars and buses are proposed in the literature [19]. It may however, be desirable to use different values for every scenario described. Two cars catching fire after a collision may start burning at the same time, or with some delay, causing the one car fire affecting the growth rate of the other. Two different growth rates may therefore be used for similar objects in the same fire scenario. If the HRR is known as in this work, α complying with the fire tests may be used. The α values proposed for car fires in the Byfjord tunnel, Table 1, are in the order of medium to fast growth. For the Byfjord and Bømlafjord tunnels the upstream car was ignited and the fire spread from this to the car downstream. Two cars affecting another may achieve higher growth rate than single cars. Single cars may have fast or medium growth rate while two cars burning in close vicinity of each other, affecting each others growth rate by external radiation, may be even higher. Another parameter is the way the initial fire is arranged. The use of accelerants, (gasoline in a small pool, diameter=0.15 m or some clothing soaked in the accelerant) may affect the fire development significantly resulting in ultra fast growth rate (Banehei and Bømlafjord tunnels). For the calculations of HRR in the Hanekleiv tunnel it was decided to use fast growth rate due to the larger vehicle and partly from the knowledge gained from the tests where the fire growth was rapid. Using the values proposed by Ingason for the first car (medium growth) and a larger growth rate on the other car (fast growth) gave a good match between recorded temperatures and calculated temperatures. In the Byfjord, Bømlafjord and Banehei tunnels (all with cross-section areas=76 m2) (see Fig. 4 and Fig. 5) the maximum HRR may have been as high as 11 MW due to the two cars involved in the fire burning at the same time. The calculated HRR should be equal for all three fires with some adjustment made for the time elapsed until flashover. In the Hanekleiv tunnel test and the EUREKA tests in the Repparfjord tunnel, the cross-section areas were smaller (50 and 35 m2, respectively). The calculated HRR curves are shown in Fig. 5. There are some differences due to a slightly lower ventilation velocity (2.5 m s−1) and maybe some larger HRR compared to the three tunnels described above. The calculated HRR for the Repparfjord C21 experiment showed some difference between the calculated HRR and the one reported after the experiments (see Fig. 1 and Fig. 5). Therefore the HRR was adjusted to give a development similar to the original HRR. This was done by altering α and β as two peaks for the HRR instead of one. For the C21 experiment the adjusted HRR matched the original curve. The curve from the experiments are based on enthalpy flows and mass flow of the combustion products CO and CO2 measured +20 and +30 m downstream the fire [21]. Calculations done with Eqs. (4), (5) and (6) showed good agreement with the proposed HRR curve from the EUREKA C11 fire tests (see Fig. 1 and Fig. 5). For simulations in kfx, the pool may contain the same calorific potential as the vehicles, but this will not assure that the calculated HRR will be the same for the two situations. To find the right pool area combined with ventilation velocity, geometry and thermal properties of tunnel surfaces requires some pre-calculations, testing or iteration. Heptane was used for pool fires inside constructed vehicles (vehicles of about same size as those used in the fire tests) to simulate combustible materials. kfx calculations in the Byfjord tunnel, given the pool area Af=2.71 m2, gave a maximum mass flux of volatiles of 0.18 kg s−1 at 10 s and a maximum of 0.3 kg s−1 after 600 s. This gives a HRR the range of 10–12 MW, which is similar to the maximum HRR as calculated with Eqs. (4), (5) and (6) (see Fig. 4). Calculating the HRR in the Hanekleiv tunnel simulation (pool area Af=2.75 m2) the maximum mass flux of volatiles was 0.38 kg s−1 after 600 s giving a HRR in the range of 16–17 MW, i.e., similar to the calculations using Eq. (4), (5) and (6) (see Fig. 5). kfx calculated values were 1–2 MW higher than in the hand model calculation for the Byfjord tunnel and 5–6 MW too low at the 800 s peak in the Hanekleiv tunnel (see Fig. 5). From all the HRR curves presented in Fig. 4 and Fig. 5, temperatures were calculated by the hand model, Eqs. (7), (8) and (9). The calculated temperatures complied well with the measured temperatures, as seen in Fig. 6, Fig. 7, Fig. 8 and Fig. 9. The full-scale tests in the Byfjord tunnel and Hanekleiv tunnel (cross-section areas of 76 and 50 m2, respectively; see Fig. 6 and Fig. 9) served as references for comparing temperature recordings and calculations by the hand model, the SOLVENT and kfx CFD codes. The Input for the calculations on the hand model and SOLVENT was HRR curves calculated with Eq. (4), (5) and (6) (see Fig. 4 and Fig. 5). In the SOLVENT model default values and recommendations in the literature were used [26], [27], [28] and [45]. The properties of the surroundings were set to: density 2400 kg m−3, heat capacity 880 J kg−1 K−1 and thermal conductivity 1.7 W m−1 K−1. The results from the hand model and the kfx model compared very well to the recorded temperatures in the Byfjord tunnel test, except that the kfx model reached the maximum temperature very early. The pool fire fuel model gives a rapid raise in temperature followed by a steady temperature of about 35 °C (see Fig. 6). The SOLVENT model gave about 25 °C too high. This deviation may be due to the choice of flame volume in the model. The results from the hand model and the SOLVENT model compared very well with the recorded temperatures in the Hanekleiv tunnel test, as shown in Fig. 9. This gives a rapid rise in temperature followed by a steady temperature of about 35 °C (see Fig. 6). Also in this case the kfx model gave high temperatures from the start, but it also gave temperatures about 20 °C too high after the peak in HRR. It is possible that the liquid pool fire model gives too great a response in evaporation due to the increased back radiation as a result of internal flashover in the larger car involved, i.e., all the added heat flux results in evaporation. In reality however, solid combustion in high irradiance levels would normally lead to increased charring rate. The reason for the higher recorded temperatures in the Hanekleiv tunnel compared to the Byfjord tunnel is mostly due to the reduced cross-section area, light weight concrete covering the polyurethane frost protection foam and the larger fuel load (van and car instead of two cars). For the other tunnels with tunnel cross-section area of 76 m2 (Bømlafjord and Banehei tunnel; Fig. 7 and Fig. 8, respectively), temperatures calculated with the hand model compare very well with the recorded temperatures (see Fig. 7 and Fig. 8). The divergence between the calculated and recorded temperatures in the Bømlafjord tunnel after about 24 min was due to the sudden rupture of a gasoline tank which caused the back door of the car to open, causing flames to burst out and reach the ceiling in a short time period. The fire in the Banehei tunnel was unfortunately extinguished by the fire department after 10 min but may be comparable before this point. For the EUREKA C11 tests involving a single car fire, the temperatures calculated by the hand model compared well with the recorded temperatures (see Fig. 10). This may indicate that given proper knowledge about recorded temperatures, tunnel data and the size and type of car(s) involved, the hand model may give results comparable to both experimental data and CFD calculations. There may of course be uncertainties and they could, and should, always be described using for example probability distributions [18]. In the EUREKA C21 tests, there were two maxima in the HRR causing similar temperature maxima. For the calculation by the hand model, an adjusted HRR curve was used (see Fig. 5). The calculated temperatures show similar time dependence as the HRR. Adjusting for the changing HRR, the calculated temperatures compare well with the recorded temperatures. Not knowing anything about the actual HRR, the calculated temperatures based solely on tunnel data and fuel information, are still acceptable, as shown in see Fig. 11. The values of α and β given in Table 1 and Table 2, are proposed in such a way that the calculated HRR curves based on these gives temperature curves that may be comparable to the temperatures recorded in fire tests and those calculated with other models based on the same HRR. Most of the values are proposed by the authors. HRR curves from the larger experiments are replotted to serve as references for the calculations. In the present work, the HRR was calculated using Eqs. (4), (5) and (6) for the hand model, the recorded HRR was used in the SOLVENT model while the kfx model calculated the HRR based on its pool fire model. This HRR was compared to the HRR calculated based on measurements during the tests. For the bus fire test in the Repparfjord tunnel, the ventilation velocity was kept constant. A maximum HRR of 28 MW was achieved after 7 min (see Fig. 12). Temperatures in the hot smoke under the ceiling were recorded 10 m downstream the fire and a maximum of 700 °C were reached after about 12 min. The calorific potential in this test was estimated to 40 GJ. The mass of wooden cribs used to simulate passenger's luggage was 170 kg. The total mass involved was about 2000 kg. The heat of combustion for this load was estimated as 20 MJ kg−1 [21]. The HRR was calculated by Eqs. (4), (5) and (6) (using values from Table 2) but the time to achieve the maximum HRR had to be delayed 5 min so that the temperatures (see Fig. 13) calculated using Eqs. (7), (8) and (9) match the recorded temperatures +10 m downstream (see Fig. 12) where the HRR from the model are compared to the test results. The maximum HRR in the EUREKA tests [21] was achieved 5 min before the maximum temperature recorded, Tmax=620 °C. Usually, the temperatures follow the HRR more closely. Temperatures calculated with the hand model gave the nearest result at 540 °C. The temperatures calculated with the SOLVENT model, using the original HRR as input, gave temperatures reaching 520 °C, i.e., 100 °C lower than recorded maximum in the test. In kfx, the bus fire was modelled as a pool fire of heptane with a surface area of Af=9.88 m2 inside a bus model. There are differences between pool fires and vehicle fires in the growth rate and how the plume behaves. Flames from pool fires tend to stay on the ground and lean towards the floor when ventilation velocities exceed a certain level, while flames from vehicles leave the evaporating zone at a higher elevation and may therefore move towards the ceiling. This may affect the near fire temperatures and the comparison of temperatures may be better further from the fire. Using Eq. (1), with a combustion efficiency at 0.7 (low ventilation velocity), the HRR was calculated to about 30 MW, which is close to the one proposed from the fire test and the one calculated using Eq. (4), (5) and (6). Temperatures calculated with kfx were about 150 °C lower, at the peak, than the temperatures recorded in the test (see Fig. 13). Full-scale HGV experiments were also done in the Repparfjord tunnel. A truck with a semi trailer was loaded with 2000 kg of furniture and set on fire. The total calorific potential was calculated to 87 GJ. Upstream of the truck, two freight containers simulated the locomotive in the Channel tunnel fire [21]. Measurements of mass loss and gas analysis were performed as basis for calculating HRR. Several different HRR curves were proposed, one extreme reaching peaks of 200 MW. It was, however, impossible to make the calculated temperatures by all the three calculation models of the present work, fit to the temperatures recorded in this fire accurately. Using simulated fans to create the ventilation changes may have been possible in the SOLVENT and kfx. Another HRR curve proposed [21] (see Fig. 14) which had a form that complied better with the recorded temperatures (see Fig. 15) was therefore used in the present work. The ventilation velocity was changed from 6 to 0 m s−1 after 13.5 min and up again after 16.5 min with 3 m s−1 during the test causing a drop in the HRR from 120 MW to about 40 MW and a temperature fall from 800 to 600 °C. The effects of ventilation on fires were demonstrated in this test. Carvel et al. [37] discussed this phenomenon in some length and argued that an increase in ventilation velocity would significantly increase the HRR. Reproducing these effects in the calculation of HRR may be possible in a CFD model where solid combustion and 3D fires are implemented. The calculated HRR curves from both Eqs. (4), (5) and (6) and kfx gave results that may be compared to a curve proposed in the EUREKA 499 project [21] (see Fig. 14). It is the maximum HRR that is comparable. The pool fire in kfx was placed under a hollow HGV model same sizes as the HGV in the experiments and the pool size was about 26 m2 (13 m×2.5 m). This configuration may be more correct than just a pool. In all calculations the ventilation velocity was fixed to 6 m s−1. Most of the variations are already introduced through the fixed heat release rate in the hand model and in the SOLVENT model. The changes in HRR from pool fires during changes in ventilation velocities may not be so severe. These assumptions may make the choice bearable. The pool fire used in kfx had much faster development than both the other calculations and the test, the maximum HRR was held at its maximum for a longer time period. Back radiation from the surroundings had no effect due to the location of the pool inside the HGV. The temperatures calculated with all models compared well to the one recorded (see Fig. 15). The Memorial tunnel fire tests were carried out with different ventilation velocities (by turning fans on an off), about 2.5–3.5 m s−1, during the experiment. One of these experiments was a 100 MW diesel pool fire. Significant variation in HRR was recorded due to variations in ventilation velocity. Using fans to create the ventilation changes would have been possible in the SOLVENT and kfx models, but to compare with the hand model, an average ventilation velocity of 2.5 m s−1 was chosen in the calculations. This may give adequate results. Calculations of HRR using Eqs. (4), (5) and (6) showed that using a maximum HRR of 120 MW and Eqs. (7), (8) and (9) for temperature calculations gave temperatures of around 500 °C (see Fig. 16 and Fig. 17, respectively). In the SOLVENT model, the HRR calculated based on measurement was used as input. In the kfx model a 21 m2 rectangular pool fire, width=3.5 m and length=6 m, of heptane was used. This gave a HRR at about 100 MW and calculated temperatures of 700 °C 10 m downstream the fire (see Fig. 16 and Fig. 17, respectively). The correct fuel to use could have been distillate fuel oil No. 2 (No. 1 is Kerosene) and the original pool size should have been used but these types of fuels may change their composition during a fire and complicate the calculations. The heptane pool fire used gave a fair HRR. The calculated temperature from both CFD models complied well with the recorded temperature from the fire test, while the hand model gave temperatures 2–300 °C lower even though the HRR was higher (see Fig. 16 and Fig. 17, respectively). The hand model results are sensitive to the ventilation velocity. By using a ventilation velocity of 1.5 m s−1 the calculated temperatures were about 800 °C. The fire test T1 in the Runehamar tunnel simulated a large vehicle fire. The test setup was a cargo of wooden materials and plastics (a total of 10,911 kg and a calorific potential 240 GJ) [23], 10.45×2.9×4.5 m3. The ventilation velocity was kept at 3 m s−1. In the test T1, a pulsing phenomenon occurred, due to a thermal counter flow caused by a downward slope in the tunnel downstream the fire, and may have significantly affected the ventilation velocity. Differences in velocity upstream and downstream of the fire may have occurred due to thermal effects [44]. The recorded temperatures reached a maximum of 1100 °C after 10 min. In the same experiment, the maximum HRR of 200 MW occurred after 20 min. The tunnel was thermally insulated to protect the real tunnel thus giving less heat loss and a smaller tunnel cross-section area in the vicinity of the fire. Based on the HRR calculated with Eq. (4), (5) and (6), the temperatures were calculated with Eqs. (7), (8) and(9) (see Fig. 18). The HRR was in this calculation built up as items burning side by side with different input values according to Table 2. The HRR then complied well with the original HRR from the test and the temperatures calculated based upon this HRR reached about 1100 °C after 15 min. Using another set of values (α=0.5, β=0.0007 t0=5 min, Etot=240 GJ and HRRmax=200 MW) may give the same temperature after 15 min. The form or shape of the HRR curve may deviate but it reached a maximum HRR after 15 min. Input to the SOLVENT model for temperature calculations was the original HRR from the tests. The HRR input was given in intervals of one minute though a finer solution may give better results. The temperatures reached about 1100 °C for this model at 22 min. The temperatures calculated by kfx were based on the kfx HRR pool fire HRR model (see Fig. 18). In the kfx modelling, the 3D wooden pallet fire was simulated by pool fires at different elevations. At 1.1 m above the tunnel floor, a longitudinal pool of 10 m×1 m represented the lowest fire level, on top of which, two 0.5 m wide pools of 10 m length separated by 1 m was placed 3 m above the tunnel floor. An additional pool was added at the upstream end of the highest level. This pool had an area for 4.8 m2. All together, 3300 kg of heptane filled the pools with a fuel surface area of 19.8 m2. kfx calculated a mass flux of 1.7 kg s−1 after 60 s and 4.3 kg s−1 after 471 s giving 90 and 180 MW, respectively. HRR curves calculated with the Eqs. (4), (5) and (6) reached 200 MW as the original HRR from the tests. It is interesting to see how the mass flux of volatiles in the kfx model increases as a function of back radiation from the hot smoke layer and surroundings (Eq. (3)). The growth of a pure liquid fire is usually faster and reaching maximum values much earlier than 3D solid fires. The calculated temperatures for all models reached the same level as the recorded temperatures, i.e., 1100 °C. The major difference in results is the time needed to achieve maximum temperature, where especially the kfx with its pool fires show high temperatures early, as expected. The temperatures calculated with the hand model and the SOLVENT model show similar shape as the HRR curve and reach maximums at the same time as the HRR has its maximum (Figs. 18 and 19). All models were able to give information on the fire development and temperatures, 10 m downstream the fire; which was comparable to the recorded temperatures in the full-scale experiments. Real fires have shown that the HRR may have a maximum between 50 and 350 MW [19] and that many people may die in fire scenarios that did not look very harmful beforehand, such as a fire in a HGV loaded with margarine and flour due to fast fire development and fire spread to other vehicles. Full-scale tests have shown that several types of cargo and materials in thermally insulated trailers may rapidly give severe fire scenarios. In spite of this knowledge, current codes often do not include fire scenarios more severe than 20–50 MW. It may be in due time to ask why the codes for evaluating tunnel fire safety are really not performance based in line with general modern building codes? Using performance based building codes the risk analyst should always consider the maximum fire in an enclosure, among others. Maximum HRR in a tunnel may be calculated using the volumetric flow rate of air ( m3 s−1), the air density (ρ=1.2 kg m−3) and the heat of combustion (3×103 kJ kg−1 (air)). For a tunnel having 3 m s−1 ventilation velocity and 30 m2 tunnel cross-section area this should give a maximum HRR=320 MW. The simplicity of this calculation is striking, and may lead to an in depth analysis checking if there may ever be enough fuel involved or if it may burn fast enough to achieve such a HRR in the actual tunnel. The major fires in the Alpine tunnels have shown that up to 20–40 large vehicles have been involved in the same fire. These kinds of fires may have a large HRR (300 MW?) and durations from 5 to 50 h [1], [3], [22], [23], [24] and [50]. Fire in a single large vehicle is examined in the full-scale fire tests. Fires in large vehicles may have more than ultra fast growth rates, depending on the scenario, causing rapid temperature increase in the hot smoke layer forming under the ceiling. Typical scenarios may be fires starting before vehicles stops or ignition of fuel leaks. This may result in fire spread to other HGVs due to radiation from the hot smoke gases or direct contact with flames and hot smoke gases. When performing risk analysis of fires in tunnels, the fire scenarios should include large HGV fires or other fires likely to ignite other cars with high calorific potential. In all scenarios the HGV density should be discussed even in tunnels where this density is normally low. The Alpine tunnels that experienced CTF had all the same amount of large vehicle traffic (about 2–3000 per day). The transport of dangerous cargo is not included because this traffic is regulated. A discussion on what types of cargo that should be regarded as dangerous goods may give surprising answers. With increasing traffic load, the probability of a fire, as well as the potential fire load (potential consequence) and number of people at risk all increase at the same time. This risk is not well assessed based on single vehicles fire scenarios. The pre-accepted approach to risk reduction seems to be to increase the number of hand held fire extinguishers and emergency telephones. At severe risks, automatic red warning lights and closing of the tunnels may occur. These measures may not help much for the people trapped inside the tunnel in a fire scenario. When assessing the risk associated with tunnel fires, the cost/benefit of this analysis is important. If this work is too costly, it may be neglected. CFD models are ranked very highly with respect to the detailed analysis that can be performed with such tools and may be used when there are complex tunnel structures. But they are still time consuming, 1–3 days per scenario, pre- and post-processing of the results, and are therefore expensive. They are often used by experts (the user should always be experts both in fire dynamics and the codes), with high cost rates. Due to the threshold level in knowledge needed in computer technology and fire dynamics to be able to use such models, they are not commonly used by tunnel engineers and fire brigades and they should not be used unless the expertise are presence. Since there are uncertainties about a fire that may occur in the future, in the way that such a fire may have a wide range in location, fuel load, etc. would it then be necessary to calculate a given selection of fire scenarios to a very high accuracy? Does the model need to be significantly more accurate in the predictions than the variation of the potential fire scenarios? If temperatures at a certain point were calculated for a fire scenario with the CFD model SOLVENT and the hand model, the outcome should be relatively close. The uncertainty around the future scenarios used to predict the HRR used in the models may be larger and of greater importance than the deviance between the outcomes from the two models. Uncomplicated models may reduce the threshold level for getting started on the fire consequence analysis. A HRR curve based on a time period to established fire (t0), followed by an acceleration period up to a given HRR value until burn out may be close enough to describe a future fire scenario as proposed by Lacroix [47]: Easy, fast to use, and inexpensive models allow for a higher number of scenario simulations. Using simple spreadsheet models force the user to understand the basics of the model used. It also makes it easier to perform the uncertainty analysis of any future possible fire. This uncertainty may be used as a risk quantity of future tunnel fires [18]. In the present work, it has been shown that a simple hand model in a spreadsheet gives values of temperatures in the near field of the fire that comply at a certain degree with full-scale measurements and CFD-modeling. From the hand model, time curves can be made shoving HRR and temperature together with other quantities. The SOLVENT model may give 2D or 3D iso-contour plots of many quantities and the kfx model may give an opportunity to watch the development on the screen as animation or plots in 2D or 3D and the possibility to make animation films in the post-processor. The more advanced models may give more detailed results when looking into temperature differences near the fire or the fluid dynamics around corners. With the hand model, the temperatures in a given zone can be presented while the CFD results can be presented as detailed temperature profiles in all points. These features may be used in the decision process. For all models, even the ones that are simple to use, there is need for competence in their use, in judging the input to the model, the uncertainty in the input and of course understanding the output, particularly if the scenarios are more complex than the ones described in this paper. In risk analysis, the future possible fire should be taken into account. As many parameters of this future fire are unknown, performing risk analysis on a range of fire scenarios to gain knowledge of potential fire hazards seems to be a good analysis strategy. The large uncertainties in predicting any future fire is assumed to be much higher than the deviation in the models when predicting the fire severity of known fire loads. It is demonstrated that given a known HRR, all three models comply well with measured temperatures. Both the t-squared model for estimating HRR and the kfx-combustion model gives HRR comparable to those measured in the tests. The deviation between the models is very small compared to the uncertainties in predicting any unknown future tunnel fire. It therefore turns out that the simple hand based model should be recognized as good enough for assessing the temperatures in a given tunnel fire. Though other models might have a little better accuracy, testing a larger number of models would not alter this conclusion. Real tunnel fires may grow to involve a large number of vehicles. The simple hand based model has not been tested for predicting the fire severity in such circumstances. It does, however, give temperatures downstream of the fire which may be used to predict the likelihood of fire spread to other vehicles. If the fire is assumed to spread, a new simulation with increased HRR should be done. Fires in vehicles show a variety of fire growth rates up to the maximum HRR. Describing the HRR of the fire as insignificant (but still very dangerous) in an incubation period followed by a sudden increase in HRR to its maximum may describe the fire severity sufficiently well, but with a little less satisfactory-looking temperature curve. The attempts to model the growth rate of a completely unknown future fire with high degrees of precision may not be worth the attention it is given today for use in risk judgements, as the focus should rather be shifted towards better predictions about any future fire scenarios described by the uncertainty around these predictions.