Deep geological disposal is the preferred solution for radioactive waste management in many countries, including Belgium, where the Boom Clay is one of the potential candidate host formations. Over the long term, corrosion mechanisms are expected to release large amounts of gas that will rise in pressure and activate different gas transport processes in the system and the surrounding geological formation. Assessing which transfer mode prevails under which range of pressure conditions in the sound rock layers remains a major issue. This paper presents a multi-scale Hydro-Mechanical (HM) model capturing the influence of the microstructure features on the macroscopic gas flow, and especially the emergence of preferential gas-filled pathways. A detailed constitutive model for partially saturated clay materials is developed from experimental data to perform the modelling of a Representative Element Volume (REV), and integrated into a multi-scale scheme using homogenisation and localisation techniques for the transitions to the macroscopic scale. Using this tool, numerical modelling of a gas injection test in the Boom Clay is performed with the aim of improving the mechanistic understanding of gas transport processes in natural clay barriers.

Phosphate mining industry generates different types of by-products that have significant environmental impacts such as ecosystems destruction and soil contamination. To reduce their environmental footprint, these wastes were investigated as supplementary cementitious materials (SCMs). The generated by-products included a clayey material and calcareous marl which were used in the current study. Blends of the abovementioned materials with cement (ratio of 1:1) were investigated using X-Ray Fluorescence spectrometry (XRF), X-Ray Diffraction (XRD), Thermogravimetric Analysis (TDA-TG), Mercury Porosimetry, Fourier-Transform Infrared Spectroscopy (FTIR), Scanning Electron Microscopy (SEM) and Electromechanical testing device. Using these results, a learning model based on multiple linear regression (MLR) was proposed to predict the compressive strength and the specific surface area from the constituents of the material, the additives, the L/S ratio, and the hardening regime. The accuracy of the models was assessed using the correlation coefficient (R2), mean absolute error (MAE), and root mean square error (RMSE). Compressive strength results confirmed that the sample's strength improved with the amount of calcined clay. Unlike the water demand where the mixtures required more water than the OPC mixture. SEM -EDS examinations proved the existence of the C -S -H gel, responsible for the specimen strength. The used machine learning model demonstrated excellent performance and practical potential to predict both compressive strength (CS) and specific surface area (SS) by capturing both linear and nonlinear relationships. As well as time and plasticizer were the most influential factors on the properties studied (CS and SS) and their effect was positive. This sensitivity study provides important information on the critical factors influencing compressive strength and specific surface area in the different ranges considered.