In a repository for radioactive waste hosted in a clay formation, hydrogen and other gases may be generated due to the corrosion of metallic materials under anoxic conditions, the radioactive decay of waste and the radiolysis of water. If the gas production rate exceeds the gas diffusion rate within the pores of the clay, a discrete gas phase will form and accumulate until its pressure becomes large enough to exceed the entry pressure of the surrounding material, at which point dilatant, advective flow of gas is expected to occur. The understanding of the processes and mechanisms involved is therefore a key aspect when assessing the impact of gas flow on a repository layout and design of any future facility.
Several international projects aiming to understand the advective movement of gas through clay-rich materials have already been conducted. These include MEGAS (1991-1994), EVEGAS (1994-1996), PROGRESS (1996-1999), GAMBIT (1998-2005), NF-Pro (2002-2006) and FORGE (2009-2013), see Bond et al. (2018) for more details. However, development of new and novel numerical representations for the quantitative treatment of gas in clay-based repository systems are still required. This was the primary focus of Task A in the DECOVALEX-2019 (D-2019) project, see Tamayo-Mas and Harrington (2020), in which 8 teams attempted to model the movement of gas in plastic clays in 1D and 3D experiments performed under controlled laboratory conditions. In Task A D-2019, four types of modelling approaches were developed: (i) standard two-phase flow models incorporating a range of different mechanical deformation behaviours, (ii) enhanced two-phase flow models in which fractures are embedded within a plastic material (continuous techniques) or incorporated into the model using a rigid-body-spring network (discrete approaches), (iii) a single-phase model incorporating a creep damage function in which only gas flow is considered, and (iv) a conceptual approach used to examine the chaotic nature of gas flow. In contrast to previous international gas projects such as EVEGAS or GAMBIT, standard two-dimensional and three-dimensional two-phase flow models were capable of obtaining some good fittings with respect to experimental stress and pore pressure measurement results. However, these models did not reflect some of the important underlying physics (e.g. creation of dilatant pathways) associated with advective gas flow and were therefore unable to describe the full complexity of the processes in such low-permeability materials.
Several concerns were raised in Task A D-2019 as some key features in the modelling of advective gas were still unclear:
Parameter calibration and model constraints: marked differences were found in the calibration procedures. Indeed, both the number of the calibrated parameters (models’ degrees of freedom) and the experimental outputs used to calibrate them were significantly different. Besides, some of these parameters (e.g. tensile strength, swelling pressure…) had a clear physical meaning while others (e.g. damage smoothing coefficients, capillary spacing…) were numerical. Hence, their definition is complex and their extrapolation to other tests can be difficult. This was already observed in Task A D-2019, where different parameter values for the 1D and the 3D tests were sometimes anomalously assumed. More analysis and a better understanding are needed before using the models as a predictive tool to assess gas movement.
Heterogeneity: only two models included material heterogeneity. However, it needs to be further explored and analysed since it might provide one possible route to represent localisation of flow.
Stochasticity: the experimental data from the 1D and the 3D gas injection tests exhibited a combination of stochastic and deterministic behaviours. Both breakthrough after a period of increasing gas pressure and bulk gas flow through a main emergent pathway were seen. The instability and pathway switching observed in the 3D experiment before a main flow path was established, suggested that the precise timing of the gas breakthrough and associated gas flows could be stochastic by nature. It is therefore important for numerical modelling to understand and distinguish between the key experimental features reproducible across all experiments and those that only occur in specific experiments. Therefore, being able to analyse and model similar high-quality experimental datasets is required to help give confidence in the process of understanding.
Up-scaling: although in Task A D-2019 only experiments under controlled laboratory conditions were modelled, models that are tractable at repository scales are needed. This poses a major challenge, since accurately and efficiently including heterogeneities at small scales (which might have a significant impact on repository performance) in a field test is a complex process.
With these concerns in mind, development of new numerical representations for the quantitative treatment of gas in clay-based repository systems are therefore required, and are the primary focus of Task B under DECOVALEX-2023.
The current plan of the task includes 3 distinct stages, starting with a conceptual model development phase. The main objective of this first stage is to assess team’s modelling capabilities (those already developed within Task A D-2019 or elsewhere). Participating teams should pay special attention on how their models describe:
The development of dilatant pathways.
The permeability associated to this pathway development.
The coupling between permeability and stress.
The one-dimensional gas test data modelled in D-2019, will be made available for those participants who did not take part in the previous DECOVALEX phase. Optionally, teams will be able to test their models against this dataset and, if needed, improve their numerical procedures.
The task will be followed by a second stage, where a blind prediction test will be presented. The main objective of this second stage is to assess those models and to analyse their capabilities. Special emphasis will be placed on the fact that the purpose of this test is not to calibrate the models via fitting routines but to analyse whether the key features of the experiments are well-captured or not.
The task will move then to a third stage (a full-scale in situ test), where teams will be required to model a large-scale gas injection test (Lasgit) experiment conducted at the Äspö Hard Rock Laboratory.
A variety of high quality gas injection data have been made available for different parts of the task:
As an optional test, the one-dimensional gas experiment performed by BGS and numerically modelled in D-2019 is made again available. This test was performed on a pre-compacted Mx80 bentonite sample, where a constant volume pressure vessel was used, see Daniels and Harrington (2017) for a detailed description of the sample preparation and laboratory procedure.
The first activity is to assess existing models and codes against a blind prediction test. Teams are not encouraged to use standard two-phase flow models (see Tamayo-Mas and Harrington, 2020) but to employ enhanced approaches (either by different deformation behaviours, explicitly incorporating different pathways and/or evolving fractures). Models are not expected to replicate all aspects of experimental behaviour but to capture key features of the data. Clear details regarding the experiment, such as (i) boundary conditions, (ii) filters/volume, (iii) volume injector and (iv) compression rate will be given to remove uncertainty and make their comparison possible. Basic parameters will be provided and need to be used by all the teams. Important measured quantities that need to be reflected by the numerical outputs will be given. Special emphasis will be placed on:
Mass-balance: this needs to be satisfied and hence, gas saturations will need to be provided (time-series data).
Flow: inflow and outflow outputs will need to be provided.
Gas transfer: the conditions that allow gas to be transferred will need to be clearly reported.
Maximum gas saturations: high gas saturations suggest that either the capillary relationship governing desaturation is incorrect or too many pathways are present in the model.
Mechanical coupling.
The large-scale gas injection test (Lasgit) is a full-scale in situ test conducted in the assembly hall area in Äspö HRL at a depth of -420 m. A deposition hole, 8.5 m deep and 1.8 m in diameter, was drilled into the gallery floor. A full-scale KBS-3 canister (without heater) has been emplaced in the hole. Thirteen circular filters of varying dimensions are located on the surface of the canister to provide point sources for the injection of gas to mimic canister defects. Pre-compacted bentonite blocks with high initial water saturation have been installed in the deposition hole. The hole has been capped by a conical concrete plug retained by a reinforced steel lid capable of withstanding over 5,000 tonnes of force.
During 2013 (day 2890 – day 3255) the test programme of Lasgit concentrated on the fourth gas injection stage. This test focused on a lower canister filter (FL903). Gas injection in FL903 was initiated on day 2988 (8th April, 2013) using helium as the injection gas. Different gas pressurisation steps were applied leading to a gas breakthrough event at day 3204 (at a pressure of 6195 kPa).
For further information, please contact the task leader, Dr Jon F. Harrington.