val-2c

Test Cell Release Experiment

Case Description

This validation problem is taken from Holland and Jalbert (1986) and was part of the validation suite of TMAP4 Longhurst et al. (1992) and TMAP7 Ambrosek and Longhurst (2008). It has been updated and extended in Simon et al. (2025). Whenever tritium is released into a fusion reactor test cell, it is crucial to clean it up to prevent exposure. This case models an experiment conducted at Los Alamos National Laboratory at the tritium systems test assembly (TSTA) to study the behavior of tritium once released in a test cell and the efficacy of the emergency tritium cleanup system (ETCS).

The experimental set up, described in greater detail in Holland and Jalbert (1986), can be summarized as such: the inner walls of an enclosure of volume are covered with an aluminum foil and then covered in paint with an average thickness , which is then in contact with the enclosure air. A given amount, T, of tritium, T, is injected in the enclosure, which initially contained ambient air, representing tritium release. A flow rate through the enclosure represents the air replacement time expected for large test cells. The purge gas is ambient air with 20% relative humidity. A fraction of that amount is diverted through the measurement system to determine the concentrations of chemical species within the enclosure.

Several phenomena are taking place and need to be captured in the model to determine the concentrations of elemental tritium (i.e., T and HT), tritiated water (i.e., HTO), and water (i.e., HO). First, The following chemical reactions occur inside the enclosure: (1) (2) mostly as a consequence of the tritium reactivity. The reaction rates of these reactions are and , respectively, where (3) and (4) Here, represents the concentration of species , and is a constant.

Second, the different species will permeate in the paint. The elemental tritium species, T and HT, have a given solubility and diffusivity , while the tritiated water, HTO, and water, HO, have a solubility and diffusivity . It is expected that the species will initially permeate into the paint and later get released as the purge gas cleans up the enclosure air.

The objectives of this case are to determine the time evolution of T and HTO concentrations in the enclosure, match the experimental data published in Holland and Jalbert (1986), and display this comparison with the appropriate error checking (see Figure 1 and Figure 2).

Model Description

To model the case described above, TMAP8 simulates a one-dimensional domain with one block to represent the air in the enclosure, and another block to represent the paint. In each block, the simulation tracks the local concentration of T, HT, HTO, and HO. Note that this case can easily be extended to a two- or three-dimensional case, but consistent with previous analyses, we will maintain the one-dimensional configuration here.

In the enclosure, to capture the purge gas and the chemical reactions, the concentrations evolve as (5) (6) (7) and (8) where is the concentration of HO in the incoming purge gas.

In the paint, TMAP8 captures species diffusion through (9) (10) (11) and (12)

At the interface between the enclosure air and the paint, sorption is captured in TMAP8 with Henry's law thanks to the ADInterfaceSorption / InterfaceSorption object: (13) where and are the concentrations of species in the enclosure and in the paint, respectively, and is the solubility (either or ). The boundary conditions are set to "no flux" since no permeation happens at the interface between the paint and the aluminum foil and the only flux leaving the enclosure is already captured by the purge gas (Holland and Jalbert, 1986).

One of the assumptions made in the original paper and TMAP4 V&V case is that the tritium is immediately added to the enclosure (Holland and Jalbert, 1986; Longhurst et al., 1992). However, this leads to an early HTO peak concentration, which does not exactly match the experimental data. In Ambrosek and Longhurst (2008), TMAP7 introduces a new enclosure to account for a slower injection of tritium. Here, we model this case with two different approaches. The first approach, like (Holland and Jalbert, 1986; Longhurst et al., 1992), assumes that the entire tritium inventory is immediately injected in the enclosure at the beginning of the experiment. The second approach assumes that the tritium inventory is being injected into the enclosure at a linear rate during a period of time until the entire tritium inventory is injected. The results of these two approaches are presented and discussed below.

Case and Model Parameters

The case and model parameters used in both approaches in TMAP8 are listed in Table 1. Some of the parameters are directly leveraged from Holland and Jalbert (1986), Longhurst et al. (1992), and Ambrosek and Longhurst (2008), but others were adapted, originally by hand (see Table 1) and then using a rigorous calibration study (see Table 2), to better match the experimental data.

Table 1: Case and model parameters values used in both immediate and delayed injection approaches in TMAP8 with , the gas constant, and , Avogadro's number, as defined in PhysicalConstants. When values are the same for both approaches, they are noted as identical. Model parameters that have been adapted from Longhurst et al. (1992) show a corrective factor in bold. Units are converted in the input file.

ParameterImmediate injection approachDelayed injection approachUnitReference
0.96IdenticalmHolland and Jalbert (1986)
0.16 (between 0.1 and 0.2)IdenticalmmHolland and Jalbert (1986)
T10IdenticalCi/mHolland and Jalbert (1986)
714IdenticalPaLonghurst et al. (1992)
0.54Identicalm/hrHolland and Jalbert (1986)
303IdenticalKHolland and Jalbert (1986)
Total time180000IdenticalsHolland and Jalbert (1986)
m/Ci/sAdapted from Longhurst et al. (1992)
4.0 Identicalm/sHolland and Jalbert (1986)
1.0 Identicalm/sHolland and Jalbert (1986)
1/m/PaAdapted from Longhurst et al. (1992)
1/m/PaAdapted from Longhurst et al. (1992)
N/A3hr

The calibration study was performed using MOOSE's stochastic tools module, and in particular the Parallel Subset Simulation (PSS) approach. The inputs and methodology provided here do not correspond to the full PSS study, but a scaled down version of it to minimize the computational costs. For this PSS study, we used 5 subsets with a subset probability of 0.1 (default) and 1000 samples per subset for a total of 5000 simulations, which were performed in parallel on 5 processors. For a full PSS study, it is common to use 10 subsets with 10000 samples per subset.

To calibrate the model against both the T and HTO concentrations in the enclosure over time, we performed a multi-objective optimization study. The penalties for the difference between the experimental data and the modeling prediction is given by

(14)

for HTO, and

(15)

for T. The metric to be optimized is then defined as the time integral of

(16)

Notably, the integral difference is defined in logarithmic space to give equal weight to all data points in the logarithmic scale during the optimization process. The complexity of the optimization metric is due to the large difference in scale for each species, as well as the discrete nature of the T measurements compared to the almost continuous nature of the HTO measurements. These differences make it challenging to optimize the fits of both species.

The comparison between the original and calibrated values of selected model parameters is summarized in Table 2.

Table 2: Calibrated model parameters values for the delayed injection case in val-2c..

ParameterNon-calibrated values (see Table 1)Calibrated values using Parallel Subset SimulationUnit
2.833 m/Ci/s
4.0 3.864 m/s
1.0 1.737 m/s
2.514 1/m/Pa
9.862 1/m/Pa
108009536s

Results and Discussion

commentnote:Update from Simon et al. (2025)

The results presented here are updated results from those presented in Simon et al. (2025). First, the initial time step was reduced from dt=60 s in Simon et al. (2025) to dt=1 s in the current case. This slightly affects the results for both the immediate and delayed injection cases. However, the results are qualitatively unchanged and conclusions remain valid. Second, the calibration approach was updated since Simon et al. (2025) with an updated multi-objective function, and new results. This improves the previous calibration results from Simon et al. (2025).

Figure 1 and Figure 2 show the comparison of the TMAP8 calculations (both with immediately injected and delayed injected T) against the experimental data for T and HTO concentration in the enclosure over time. There is reasonable agreement between the TMAP8 predictions and the experimental data. In the case of immediate T injection, the root mean square percentage errors (RMSPE) are equal to RMSPE = 58.68% for T and RMSPE = 146.23% for HTO, respectively. When accounting for a delay in T injection, the TMAP8 predictions best match the experimental data, in particular the position of the peak HTO concentration. The RMSPE values decrease to RMSPE = 89.50% for T and RMSPE = 75.66% for HTO, respectively. Note that the model parameters listed in Table 1 are somewhat different from Holland and Jalbert (1986), Longhurst et al. (1992), and Ambrosek and Longhurst (2008) to better match the experimental data. In particular, Longhurst et al. (1992) and Ambrosek and Longhurst (2008) did not validate the TMAP predictions against T concentration, which we do here in Figure 1 and in Simon et al. (2025). This affects some of the model parameters.

Figure 1: Comparison of TMAP8 calculations against the experimental data for T concentration in the enclosure over time. TMAP8 matches the experimental data well, with an improvement when T is injected over a given period rather than immediately. Calibration of the delayed injection model delivers further improvements.

Figure 2: Comparison of TMAP8 calculations against the experimental data for HTO concentration in the enclosure over time. Calibration of the delayed injection model delivers further improvements, with more accurate simulation results when T is injected over a given period rather than immediately.

As shown in the red curve in Figure 1 and Figure 2, using MOOSE's stochastic tools module notably increased the agreement between the modeling predictions and experimental data for both the T and HTO concentrations. The RMSPE for T decreases from 89.50% to 30.18% and the RMSPE for HTO decreases from 75.66% to 67.07%. Note that although the calibration approach is similar to the one presented in Simon et al. (2025), the results presented here include more simulations and the quality of the calibration is increased here (RMSPE values are further decreased here).

Figure 3 and Figure 4 show the evolution of the model parameter values and of the optimization metric (time integral of defined in Eq. (16)) as a function of the number of simulation. The calibrated model corresponds to the highest value.

Figure 3: Evolution of the model parameter values as a function of the number of simulations.

Figure 4: Evolution of the optimization metric (time integral of defined in Eq. (16)) as a function of the number of simulations. The calibrated model corresponds to the highest value.

Figure 5 and Figure 6 show the value of the calibrated parameters and the range of the data that was explored in the Parallel Subset Simulation study. Figure 5 shows the parameters that followed a normal distribution, and Figure 6 shows those that followed a uniform distribution in log scale. In both cases, the calibrated parameters are not on the extremes of the distribution, suggesting that the ranges were properly defined.

Figure 5: Calibrated parameter values compared to the normalized normal distribution used in the Parallel Subset Simulation study. None of the parameters are at the extremes of the distribution.

Figure 6: Calibrated parameter values compared to the normal distribution in the log scale used in the Parallel Subset Simulation study. None of the parameters are at the extremes of the distributions.

Input files

The input files for this case can be found at (test/tests/val-2c/val-2c_immediate_injection.i) and (test/tests/val-2c/val-2c_delay.i). Note that both input files utilize a common base file (test/tests/val-2c/val-2c_base.i) with the line !include val-2c_base.i. The base input file contains all the features and TMAP8 objects common to both cases, reducing duplication, and this allows the immediate injection and delayed injection inputs to focus on what is specific to each case. Note that both input files are also used as TMAP8 tests, outlined at (test/tests/val-2c/tests).

For the calibration study, additional input files are provided.

To run the PSS study in the terminal, users can perform:


cd ~/projects/TMAP8/test/tests/val-2c/
mpirun -np 5 ~/projects/TMAP8/tmap8-opt -i val-2c_pss_main.i

Note that this study is time consuming since a large number of simulations are being run. Modifying the PSS parameters can reduce the computational cost.

Although a very short PSS study is simulated as a test in (test/tests/val-2c/tests) to ensure these files run properly, the full calibration study is not performed regularly in tests to limit computational costs within the TMAP8 testing suite. The gold files (test/tests/val-2c/gold/val-2c_pss_results/val-2c_pss_main_out.json) and (test/tests/val-2c/gold/calibrated_parameter_values.txt) are, therefore, not continuously tested, and the calibrated model parameters used in (test/tests/val-2c/tests) are not continuously updated.

References

  1. James Ambrosek and GR Longhurst. Verification and Validation of TMAP7. Technical Report INEEL/EXT-04-01657, Idaho National Engineering and Environmental Laboratory, December 2008.[BibTeX]
  2. D F Holland and R A Jalbert. A model for tritium concentration following tritium release into a test cell and subsequent operation of an atmospheric cleanup system. In Eleventh Symposium on Fusion Engineering. IEEE Cat. No. CH2251-7, Vol 1. pp. 638-643, 11 1986.[BibTeX]
  3. GR Longhurst, SL Harms, ES Marwil, and BG Miller. Verification and Validation of TMAP4. Technical Report EGG-FSP-10347, Idaho National Engineering Laboratory, Idaho Falls, ID (United States), 1992.[BibTeX]
  4. Pierre-Clément A. Simon, Casey T. Icenhour, Gyanender Singh, Alexander D Lindsay, Chaitanya Vivek Bhave, Lin Yang, Adriaan Anthony Riet, Yifeng Che, Paul Humrickhouse, Masashi Shimada, and Pattrick Calderoni. MOOSE-based tritium migration analysis program, version 8 (TMAP8) for advanced open-source tritium transport and fuel cycle modeling. Fusion Engineering and Design, 214:114874, May 2025. doi:10.1016/j.fusengdes.2025.114874.[BibTeX]