Back to news section
Video of fog-plume from Test 5 looking downwind, plow shape of plume from simulation can be seen (video from DNV, presented in Teknisk Ukeblad on July 3, 2020).

Published 07.07.2020.

Liquid Hydrogen Experiments Made Available by SVV – Well Reproduced by CFD Simulations

December 2019 and January 2020 the Norwegian Public Road Authorities (SVV) funded large scale liquid hydrogen experiments at DNV GL Spadeadam test site, the ambition with the tests was to generate knowledge and understanding on safety aspects handling LH2 prior to the launch of a LH2-fuelled car ferry operation in Norway next year. Norwegian Defence Research Establishment (FFI) managed the project on behalf of SVV. Draft test reports are now published on the SVV website, these will be replaced by final versions of the reports once published.

Two test series were carried out:

  • Outdoor LH2 releases (Test 5 - 0.7 kg/s vertically downwards and Test 6 - 0.8 kg/s horizontally along wind) relevant for accidental releases during bunkering. Both scenarios were repeated, in the repeated tests gas clouds were ignited downwind.
  • Closed room LH2 releases downwards inside a 24m3 container (meant to resemble a ferry tank connection space TCS with a 13m long 450mm vent stack). High release rates (0.4-0.7 kg/s) were applied filling the container with pure, very cold, gaseous hydrogen in a matter of seconds. Two experiments were ignited at stack exit. With slow air entrainment into TCS a weak explosion was seen after 30 min (Test 13), with more air entrainment a strong explosion (2 barg average, 5 barg locally) was reported in the TCS within 1-2 minutes (Test 14).
  • Olav Roald Hansen (at that time working at Lloyd’s Register) was member of a SVV appointed reference group giving feedback to the project on the draft reports in May 2020. To help interpret the results he simulated the most interesting experiments (Indoor Test 14 and Outdoor Test 5 and Test 6) with the FLACS CFD-tool using the modelling approach for LH2 releases as presented in Hansen [2020].

A brief summary of the tests and simulations follows:

Test 5 – 740 g/s downwards release from 0.32m elevation in 4 m/s wind @ 4°C

The simulation and two experiments show bifurcated plume with significant lateral spread, in simulation wind direction is constant, in experiments there is some variation in wind direction. Both in simulation and experiments the 8% plume extends around 30m downwind and 4% (LFL) plume almost 50m downwind. Minimum temperatures correlated well between simulation and observations, these were marginally higher in tests, likely due to heat transfer from ground ignored in the simulation. Maximum concentrations and minimum temperatures at arcs 30m, 50m and 100m from release observed in Test 3 and Test 5 and predicted in simulation are compared in below table.

The plume failed to ignite 24m downwind but managed to ignite and burn back to source 18m downwind. In simulation ignition at 18m did not burn back to source, while ignition at 15m distance did. Maximum explosion pressures were generally low around 15-20 mbar both in experiment and simulation.

Flammable plume (4%) from simulation of Test 5.

Gas concentrations at ground level in simulation of Test 5.

Temperatures at ground level in simulation of Test 5.

Test 6 – 830 g/s horizontal release in 2.5 m/s wind @ 4°C

Simulation and experiments show a relatively thin, concentrated plume, in the tests the plume and wind direction vary significantly with time. Both in simulation and experiments the concentration exceeds 20% at the 30m arc. In the simulation the 8% plume extends 50m downwind, in tests 2% and 6.5% are reported at 50m, but due to few sensors (only 3) in the arc the core of the plume likely misses the sensors in Test 6. Minimum temperatures correlated fairly well, somewhat higher temperatures were seen in tests, likely due to heat from ground and that the coldest part of plume failed to hit sensors. 100m arc seems mostly missed by plume, in the simulation the main part of plume lifted above ground beyond 50m downwind. Maximum concentrations and minimum temperatures at arcs 30m, 50m and 100m from observations in Test 4 and Test 6 and simulation are shown in table below.

The plume was ignited 30m downwind at concentrations around 20% and burned fast back to source generating overpressures around 10-25 mbar both in simulation and experiment.

Flammable plume (4%) from simulation of Test 6.

Gas concentrations at ground level in simulation of Test 6.

Temperatures at ground level in simulation of Test 6.

Fast flashfire/explosion Test 6 when igniting at 30m.

Test 14 – 370 g/s downwards release in closed room (TCS) with ignition at stack

This scenario was quite challenging to simulate. The stack outlet was ignited at a moment when entire TCS is filled with pure hydrogen gas at temperature below -150°C. After ignition a small panel at TCS floor level (~0.2m3) was opened to let in fresh air and the flare combined with the cold air inflow creates a chimney effect in which air is gradually entraining into the room, and gradually into the chimney. At the moment the chimney concentration reaches around 60% the flame accelerates back into the TCS and explodes violently. At this time most of the TCS is at concentrations between 40 and 60% hydrogen and at temperatures between -40 and -80°C. Almost the entire front of the TCS is equipped with a soft wall area which blows out at low overpressure, still, the hydrogen flames are fast enough to generate overpressure around 2 barg (most sensors) to 5 barg (highest) with a duration 30ms, in the simulation all sensors predicted 2 barg, higher pressure was predicted in corners.

To conclude, it is believed that the phenomena, both the air entrainment, the chimney effect and the back-burning, mixing and turbulent explosion inside the TCS are well predicted and illustrated in the CFD simulation and this way useful to explain the dynamics of the explosion. The time it takes for the air entrainment and flame to burn back is 3 times faster in the simulation, there could be many reasons for this (inaccurate temperature start condition, wrong vent area, or timing of vent area opening relative to stop of release).

Predicted concentrations in container and stack at time of ignition (upper) and at moment of flame burning back into container in Test 14.

Predicted overpressures inside and outside container in Test 14.

Reported overpressures in Test 14 (upper-from DNV GL report) and from CFD simulation (lower).

From our perspective important findings of the two test campaigns were:

  • The outdoor release tests are very valuable for confidence in modelling of bunkering release scenarios due to repeated experiments of two well described scenarios. A somewhat scarce array of sensors at 50m and 100m distance (which e.g. Test 6 greatly missed) was a minus but with the help of CFD-simulations the experiments can be properly interpreted.
  • The ignited outdoor releases showed that fast flames can be seen with ignition at concentrations around 20%, while much less intense flames will result when igniting at concentrations below 10%. No significant overpressures were reported in these tests, consistent with simulations. This could have been different of reactive plumes could accumulate e.g. between ship and jetty or below jetty.
  • Another important finding from the experiments is that no liquid hydrogen plume was observed for the outdoor tests, and barely any accumulation of solid/condensed air particles was seen.
  • For the closed room releases it was demonstrated that a very strong explosion could happen even if the concentration was very fuel-rich and there was a pressure relief vent on nearly the entire front wall. If concentration in the room had been lower (e.g. from a smaller release more likely to happen) the reactivity would be much higher, and DDT and detonations leading to severe damage should be feared. In our opinion the type of scenarios evaluated in the closed room tests, as well as significantly smaller releases, must be absolutely prevented in a TCS, thus the value of performing these tests is considered limited.

The good correlation between repeated experimental results and the CFD modelling is an indication not only that the modelling tool/approach is capable of reproducing the challenging experiments, but also that the quality of the experiments is good. It is quite impressive by DNV GL Spadeadam to perform two such demanding test series within two weeks with no prior LH2-experience.

Great thanks to Norwegian Public Road Authorities for performing and sharing these interesting experiments.

Reports can be downloaded from SVV on the following links: