Model validation
Cases where TekRisk results are compared with external references (Yellow Book, Phast, ALOHA, the original LLNL SLAB code and SCRI), with the deviation of each one, the unresolved differences and the models that have no benchmark yet.
This page collects every case where a TekRisk result is compared with an external reference: a worked example from a book, the output of another program or the original code of a model. Each case lists the quantity compared, both values and the deviation. It also lists the differences we have not been able to explain and the models that do not have any external benchmark yet.
Status as of 30 September 2026.
1. Liquid outflow (Yellow Book, ch. 2 and §6.6.4)#
Leak flow calculator of the pool fire. Automated test.
| Case | Quantity | Yellow Book | TekRisk | Deviation |
|---|---|---|---|---|
| §2.6.4.1: acrylonitrile, 0.1 m hole | Flow at 500 s (kg/s) | 58.44 | 58.439 | 0.00 % |
| Level at 500 s (m) | 11.12 | 11.124 | +0.03 % | |
| Mass released at 500 s (kg) | 29 416 | 29 270 | −0.50 % (book erratum) | |
| §2.6.4.1, Table 2.8 | Mass released at 4698 s (kg) | 281 550 | 281 556 | 0.00 % |
| §2.6.4.2: 100 m pipe | Initial flow (kg/s) | 22.33 | 22.30 | −0.11 % |
| §6.6.4: draining, benzene | Release time (s) | 5064 | 5064.25 | 0.00 % |
| Maximum pool diameter (m) | 31.97 | 31.97 | 0.00 % |
The book does not publish the viscosity of the pipe case: with 0.337 mPa·s the flow matches exactly, and with 0.36 mPa·s it comes out −0.11 %. Table 2.8 is computed with a density of 844 kg/m³, not the 812.5 stated in the text.
2. Pool fire, Yellow Book preset (§6.6.3 and §6.6.4)#
Benzene, receptor 100 m downwind. Automated test for the equations; diagnostic for the final flux.
| Case | Quantity | Yellow Book | TekRisk | Deviation |
|---|---|---|---|---|
| §6.6.3, confined | Diameter (m) | 42.445 | 42.446 | 0.00 % |
| Flame length (m) | 46.77 | 46.78 | 0.01 % | |
| Surface emissive power (kW/m²) | 66 | 66.48 | +0.73 % (book rounding) | |
| Flux with the book's τ (kW/m²) | 4.581 | 4.583 | +0.03 % | |
| TekRisk final flux (kW/m²) | 4.581 | 6.342 | +38 % | |
| §6.6.4, unconfined | Diameter (m) | 31.97 | 31.97 | 0.00 % |
| Surface emissive power (kW/m²) | 63.4 | 63.47 | +0.10 % | |
| Flux with the book's τ (kW/m²) | 2.141 | 2.158 | +0.77 % | |
| TekRisk final flux (kW/m²) | 2.141 | 2.785 | +30 % |
Is it an error? No#
Unlike the differences in section 6, this one is fully explained.
The equations match. The "flux with the book's τ" row runs the same TekRisk equations with the example's geometry and gives 4.583 against 4.581 kW/m² (+0.03 %) and 2.158 against 2.141 kW/m² (+0.77 %, book rounding). If any formula were wrong, that row would not match.
The book's example does not apply its own method. §6.5.4 describes how the flame is built, but the worked example in §6.6.3 simplifies it in three places. TekRisk follows the method:
| Point | Book method (§6.5.4) | Worked example (§6.6.3) | TekRisk |
|---|---|---|---|
| Flame base | Step 7: wind elongates the base and "the distance between the outer flame surface and the radiated object may decrease" (fig. 6.9c) | Computes the elongated base D′ = 52.4 m but does not use it | Elliptical D′ × D base, elongated downwind from the pool edge |
| View factor | Step 12: "use the calculated flame dimensions and the distance from the flame to the radiated object" | Radius D/2 = 21.2 m and a distance of 100 m from the pool centre | Integrates over the tilted flame with its elongated base |
| Distance for τ | Steps 9 and 10: "a distance x, between the flame surface and the object" | The text says "50 m from the flame surface", but the calculation uses 100 m from the centre | Distance to the flame surface (44 m) |
Where the +38 % comes from. Starting from the example's convention, each point of the method is added one at a time. Receptor 100 m downwind, flux in kW/m²:
| Step | §6.6.3 | §6.6.4 |
|---|---|---|
| Example convention (circular base, τ over 100 m) | 4.52 | 2.14 |
| + τ measured from the flame surface | 4.80 (+6 %) | 2.25 (+5 %) |
| + base shifted downwind by the wind | 5.62 (+17 %) | 2.55 (+13 %) |
| + elongated D′ × D base | 6.34 (+13 %) | 2.79 (+9 %) |
The book's 4.581 uses the τ of eq. 6.24 instead of eq. 6.29 (4.52). That difference is 1.3 %.
Checks that rule out an error.
- No more energy is radiated than the balance allows. The emissive power comes from eq. 6.71, which assumes a straight cylinder of diameter D. TekRisk's tilted flame has 4 % less surface than that cylinder (7 322 against 7 652 m² in §6.6.3).
- The view factor is integrated correctly. On a circular base it reproduces the closed form of the book's Appendix 6.A within 0.02 %.
- Using a single optical path barely matters. TekRisk applies the τ of the nearest flame point to the whole flame. Computed element by element, the flux is only 2 % lower.
The effect depends on direction. The +38 % is the case of a receptor downwind, where the wind pushes the flame. Receptor at 100 m, §6.6.3, in kW/m²:
| Receptor | Book example | TekRisk |
|---|---|---|
| Downwind | 4.52 | 6.34 (+40 %) |
| Crosswind | 1.93 | 2.35 (+22 %) |
| Upwind | 1.30 | 1.28 (−2 %) |
In short, TekRisk uses a model that is more conservative downwind; it does not make a calculation error: it brings the flame closer to the receptor, as the book itself says. The detailed comparison with both examples is in Comparison with the Yellow Book worked examples.
3. Solid-flame jet fire#
3.1 Yellow Book example §6.6.2#
Methane at 100 bar, 100 mm hole, 30 kg/s, receptor at 150 m. Automated test.
| Quantity | Yellow Book | TekRisk | Deviation |
|---|---|---|---|
| Steps 1 to 23 (velocity, flame length and geometry, emissive power) | all within ±0.05 % | ||
| Base width (m) | 0.71691 | 0.716909 | 0.00 % |
| Flux with the book's τ (kW/m²) | 0.4172 | 0.4171 | −0.02 % |
| TekRisk final flux (kW/m²) | 0.4172 | 0.3889 | −6.8 % |
The final difference is one of method: TekRisk spreads the emissive power over the frustum area (eq. 6.56) instead of the mean cylinder (eq. 6.57), and computes transmissivity with eq. 6.29. Step 28b of the book uses an exponent of −0.08 instead of the −0.09 of eq. 6.29.
A second source, Example 3.7 of Casal (2018), is reproduced within ±1 % to ±5 % depending on the quantity.
3.2 Real DNV Phast 6.7 report#
LPG (60 % butane, 40 % propane), 11.62 kg/s, vertical release at 20 m, 6 m/s wind. Automated test (geometry ±1 %, emissive power ±2 %, radiation ±3 %).
| Quantity | Phast | TekRisk |
|---|---|---|
| Flame length in still air (m) | 45.05 | 44.88 |
| Flame length (m) | 24.16 | 24.07 |
| Flame widths / (m) | 2.59 / 8.16 | 2.58 / 8.13 |
| Radiated fraction | 0.18 | 0.179 |
| Surface emissive power (kW/m²) | 232 | 233.9 |
| Ground-level radiation, 9 points between 16.8 and 75 m | −0.4 % to +1.1 % | |
| Reach of 1.6 kW/m² (m) | 56.8 | 55.9 (−1.6 %) |
The case starts from the expanded jet that Phast prints. Only the vertical gas release has been benchmarked.
3.3 ALOHA 5.4.7: flame length#
Three runs with a 2 in pipe. Automated test (±0.5 m, because ALOHA rounds to whole metres).
| Run | ALOHA (m) | TekRisk (m) |
|---|---|---|
| Propane at 4.8 atm, 4 m/s wind | 7 | 6.94 |
| Propane at 4.7 atm, 1.5 m/s wind | 7 | 6.93 |
| Methane at 20 atm, 1.5 m/s wind | 5 | 5.36 |
The radiation zones of these same runs do not match: see section 6.
4. Fireball against SCRI Fuego#
LPG, 1000 kg BLEVE. Automated test.
| Quantity | SCRI | TekRisk | Deviation |
|---|---|---|---|
| Maximum diameter (m) | 58.0 | 58.0 | 0 % |
| Height (m) | 43.5 | 43.5 | 0 % |
| Duration (s) | 4.5 | 4.5 | 0 % |
| Flux at 10 distances (kW/m²) | 1231.36 … 6.98 | 1203.33 … 6.82 | −2.28 % constant |
| Distance to 5.05 / 12.6 / 31.5 kW/m² (m) | 205.1 / 132.4 / 85.4 | 202.9 / 131.0 / 84.5 | −1.09 % |
The −2.28 % comes from an exponent: TekRisk uses while SCRI, like CCPS eq. 2.2.41, uses .
5. SLAB dispersion against the original LLNL program#
GASPRB case (LP gas, pool evaporation), compared with the output of SLAB.exe, the Fortran program of Lawrence Livermore National Laboratory (Ermak, 1990). Diagnostic.
| Quantity | SLAB.exe | TekRisk | Deviation |
|---|---|---|---|
| Centreline concentration, 8 points between 22.7 and 1600 m | −0.6 % to +0.2 % | ||
| Ground-level concentration, 9 points | −0.8 % to −0.1 % | ||
| Distance to 100 % of the LFL (m) | 79 | 78.7 | −0.4 % |
| Distance to 50 % of the LFL (m) | 147 | 144.2 | −1.9 % |
| Distance to 25 % of the LFL (m) | 240 | 240.5 | +0.2 % |
Only the pool evaporation source has been benchmarked. The horizontal and vertical jets and the instantaneous release do not have a numerical comparison yet.
6. Unresolved differences#
6.1 Jet fire radiation against ALOHA 5.4.7#
Documented. The flame length matches (section 3.3), but the 10 / 5 / 2 kW/m² zones do not:
| Run | ALOHA (m) | TekRisk without cap (m) | TekRisk with cap (m) |
|---|---|---|---|
| Propane, 4 m/s, 2.10 kg/s | 10 / 12 / 19 | 11.4 / 15.7 / 24.5 | 7.1 / 10.5 / 16.1 |
| Propane, 1.5 m/s, 2.03 kg/s | 10 / 12 / 18 | 10.5 / 15.1 / 23.5 | 5.2 / 9.0 / 14.8 |
| Methane at 20 atm, 6.62 kg/s | 10 / 12 / 18 | 16.2 / 22.7 / 35.1 | 5.3 / 8.2 / 13.4 |
The radiated power implied by the ALOHA zones is almost the same in the three runs (≈ 9.5 MW), even though the heat released changes more than threefold, and it is lower than what the equations in ALOHA's own documentation give. The 10 kW/m² zone always comes out at 10 m. We have not found the explanation. The details and the cap option are in Black-body SEP cap.
6.2 Point-source jet fire against SCRI Fuego#
Diagnostic. LP gas release through a hole. SCRI computes 1.056 kg/s and TekRisk 0.581 kg/s (−45 %). The zone distances come out between −29 % and −53 %. SCRI's flow implies a molar mass of about 164 g/mol, which is incompatible with LP gas, so we have not been able to reconcile the two calculations.
7. Models without an external benchmark yet#
These models implement published equations and have internal tests, but have not been compared with a worked example or with another program yet:
- VCE (TNT equivalence, overpressure and impulse).
- SLAB with horizontal jet, vertical jet and instantaneous release (only pool evaporation is benchmarked).
- Individual risk (contours and average risk) and societal risk (F-N curves).
- Probit functions and fatality calculation.
- Fault trees and event trees: their tests check internal consistency.
- Warehouse fire: only the view factor is benchmarked, against Appendix 6.1 of the Yellow Book (±0.1 % for a flat radiator and ±3 % for the 10 values of Table 6.A.1).
- Pool fire with the point-source preset and the ALOHA-based preset: there are no runs of the original program.
- Substance mixtures and the line source (pipelines).
If you have a worked case for any of these models (a book example or the output of another program) and want us to benchmark it, send it to us from the feedback button in the application.