Using reduced basis surrogates for inverse identification of thermo-hydro-mechanical parameters of Boom clay in an in-situ heater test

In Belgium, an underground research facility (URF) named HADES (High Activity Disposal Experimental Site) was constructed in the Boom Clay layer to study the potential of poorly indurated clays to host a deep geological disposal for radioactive waste. The thermo-hydro-mechanical (THM) perturbation of the Boom Clay in response to the heat-emitting HLW (High Level Waste) has been studied in HADES URF through in-situ heating tests at various scales. The small-scale ATLAS heater test, that ran in four phases (ATLAS I, II, III, and IV) between 1993 and 2012, is one of the most extensive experiment carried out in this URF. Each of these ATLAS test phases are performed to confirm and refine the THM properties of the Boom Clay estimated from laboratory experiments.Mathematical models to evaluate and interpret THM couplings are generally very efficient, however they are also often characterized by high dimensionality in which the associated computational cost becomes critical when dealing with inverse problems. To efficiently carry out inverse analysis of large THM models, a surrogate model built from model order reduction technique is proposed in this study to substitute the high-fidelity THM model describing the in-situ heater tests. Aimed at reducing the dimensionality of the THM model, the high-fidelity model is projected to a low dimensional reduced basis (RB) subspace. The RB space, spanned by so-called ‘snapshot’ solutions of the governing THM equations, is constructed using a greedy adaptive procedure guided by an a-posteriori error estimator that selects the optimal snapshot points within a given parametric space.In the present work, the reduced basis method is used to generate surrogate models for solving inverse problems related to coupled processes in ATLAS III Heater test. A two-step parameter identification procedure is performed by first identifying the unknown thermal parameters which are subsequently used as inputs to the coupled THM model followed by identifying the unknown hydro-mechanical parameters in the second stage.The accuracy of the reduced basis model is validated by comparing its generated solutions against high-fidelity full order solutions. The substantial system size reduction and the associated computational gain result in a surrogate model highly suitable for parameter identification procedures for material properties of the Boom Clay material. The effectiveness of the proposed strategy is demonstrated by performing inverse analysis based on direct-search and genetic algorithm (GA) optimization supported by real sensor measurement data where up to 800 times faster computational speed-up was achieved.