Investigating the Microstructure and Mechanical Behavior of Simulant ‘Lava-Like’ Fuel Containing Materials From the Chernobyl Reactor Unit 4 Meltdown

Decommissioning of the damaged Chernobyl nuclear reactor Unit 4 is a top priority for the global community. Before such operations begin, it is crucial to understand the behavior of the hazardous materials formed during the accident. Since those materials formed under extreme and mostly unquantified conditions, modelling alone is insufficient to accurately predict their physical, chemical and, predominantly, mechanical behavior. Meanwhile, knowledge of the mechanical characteristics of those materials, such as their strength, is a priority before robotic systems are employed for retrieval and the force expected from them to be exerted is one of the key design questions. In this paper we target to measurement of the standard mechanical properties of the materials formed during the accident by testing small-scale, low radioactivity simulants. A combined methodology using Hertzian indentation, synchrotron X-ray tomography and digital volume correlation (DVC), was adopted to estimate the mechanical properties. Displacement fields around the Hertzian indentation, performed in-situ in a synchrotron, were measured by analyzing tomograms with DVC. The load applied during the indentation, combined with full-field displacement measured by DVC was used to estimate the mechanical properties, such as Young’s modulus and Poisson’s ratio of these hazardous materials.


Introduction
On 26 th April 1986, a severe nuclear accident occurred at Unit 4 of the Chernobyl nuclear reactor during an experimental power failure test. The explosion and subsequent fire released a significant amount of radioactive solid and gaseous products into the surrounding environment, while the reactor core was decimated [1,2]. Temperatures far exceeded 2000 °C prior to the explosion, leading to reactions between the UO 2 fuel and the zircaloy cladding [3,4]. The fuel-cladding melt further interacted with neighbouring reactor components, including stainless steel and construction materials (concrete, sand and serpentinite) [2].
These molten materials formed a "lava-like" mixture, which dispersed within the reactor, penetrating premises of the structure, eventually solidifying in the basement levels. The solid heterogeneous mixture is known as the Chernobyl "lava" or "lava-like" fuel containing materials (LFCM).
Immediately after the accident occurred, a massive steel and concrete structure was assembled to cover the building where the Unit 4 ruined reactor was located. This structure, known as the Chernobyl nuclear power plant sarcophagus, was designed to limit the postaccident radioactive contamination. However, within the first of decade of operation the structure had severely deteriorated, allowing water to access the inside of the ruined reactor.
As a result, radioactive aerosol particles have formed, posing serious threats of airborne contamination and respirable hazards around the area [5,6]. Therefore, a decision for replacement was adopted. In November 2016, operations to replace the old sarcophagus with a new safe confinement were successfully completed.
Today, almost 35 years after the nuclear accident occurred, LFCM still resides in the vicinity of the damaged reactor Unit 4. It is estimated that the accumulated volume of the material exceeds 192 m 3 (~1250 t) [2,7]. The new safe confinement now allows for a robust plan for plant decommissioning, including the dismantling of the ruined reactor Unit 4 and the surrounding "lava". The total clean-up operations and complete decommissioning of the facilities are expected to last until at least 2065.
The extreme high-dose environment within the damaged Unit 4 reactor requires the development and manufacturing of specialised robotic systems to retrieve the LFCM.
Successful and safe decommissioning is directly linked with zero risk to material integrity during lift-out. Therefore, knowledge of the physical, chemical and, predominantly, mechanical properties of the material to be retrieved is key for appropriate design of the robotic systems.
Due to high radioactivity in the vicinity of the ruined reactor, it is next-to-impossible to carry out any standard mechanical testing on real samples. Further complications arise from their complex and heterogeneous microstructures that makes single measurements of properties carried out on safe, small-scale specimen testing non-representative of the bulk properties.
Up until today, only one successful attempt to retrieve Chernobyl "lava" has been reported. In 1990, scientists of the V.G. Khlopin Radium Institute (KRI), who took part in the investigations performed immediately after the accident, managed to manually collect samples of the Chernobyl LFCM, and successfully transport it to Leningrad (now St. Petersburg) in 1990 [2]. These samples remain the only pieces of real Chernobyl "lava" in the world that are available for experiments [2]. Thus, since real Chernobyl LFCM is inaccessible and dangerous to handle, it is necessary to examine simulant material for mechanical testing as a feasible method to indirectly determine properties of LFCM.
Successful attempts to synthesise materials resembling the formation process and microstructure of the Chernobyl "lava" have been already undertaken [7,8]. The relevant studies describe how to replicate the meltdown conditions as accurately as possible and produce a simulant material to be used for characterisation and testing. To the best of our knowledge, research focusing on mechanical testing of real Chernobyl "lava" samples or simulant materials has not yet been published until today.
The results of an experimental programme launched to investigate the mechanical properties of simulant Chernobyl LFCM samples are discussed in this paper. A methodology including synchrotron X-ray computed tomography (XCT) combined with digital volume correlation (DVC) was adopted to reveal and quantify internal microstructural changes on samples subjected to external loading. XCT is a non-destructive imaging technique, based on acquiring 2D radiographs through an object at numerous projections and then, computationally reconstructing the relevant 3D image. DVC is a powerful technique used to produce 3D full-field internal displacement and strain data from input image packages displaying a gradually-deformed object subjected to time-dependent processes [9][10][11][12]. The ability to monitor deformation in the core of the material being probed enables determination of the relationship between microstructure and mechanical behaviour at the macro-scale. This methodology has been successfully employed by researchers investigating the mechanical behaviour of mostly ductile but also quasi-brittle materials [13][14][15][16][17]. We employed this technique to evaluate the properties of LFCM-simulant materials. 3D tomograms were acquired at different stages of Hertzian indentation loading. These tomograms were used as the input datasets to employ DVC analysis and subsequently quantify a wide variety of microstructural changes occurring as load increased. Correlating the displacement calculated via DVC with the corresponding load values is a method to determine the material strength (load bearing capacity) and principal mechanical properties including Young's modulus and Poisson's ratio.

Materials and sample preparation
Investigations conducted after the Chernobyl accident revealed the presence of several different Chernobyl "lava" varieties [1,18]. Among those, black and brown LFCM are the most common. The colour differences are attributed to radiation defects and elemental composition of the materials. Brown colour is associated with the presence of zirconium in uranium phases, while the black colour is an indication of uranium dissolution in the glass matrix [2]. Detailed elemental composition of black and brown LFCM samples collected from KRI in 1990 can be found in the literature [2].
Since access to real material for mechanical testing was not feasible, LFCM-simulant samples that accurately replicate those characterised at the KRI were produced based on previously developed work, full details of the synthesis are given elsewhere [3,7,8,18,19].

Mechanical loading
Hertzian indentation was selected as the most appropriate loading technique considering the mm-scale size of the samples and based on our previous studies of brittle and quasi-brittle materials [14,20]. Steel holders were machined to retain 5 mm diameter ZrO 2 spheres, acting as indenters. A custom-designed and in-house built mechanical jig ( Figure 1) was used to retain samples during indentation and imaging.
Stepwise mechanical loading, coupled with in situ high-resolution synchrotron XCT, was conducted on beamline I12 at Diamond Light Source (DLS) [21]. Following DLS radiation safety rules and policies, the jig contained three barriers between the radioactive sample and the laboratory environment. A miniature button-type load cell was used to measure the compressive reaction force in situ with a 2 s polling interval. External force was applied to the sample manually by rotating a metallic rod attached to the indenter and aligned with the sample holder and the load cell button.
Time restrictions related to synchrotron beamtime availability only allowed investigation of two samples: one black LFCM simulant; one brown LFCM. Experiment feasibility requirements and radioactivity handling limits enforced minimising the size of the samples.

Synchrotron X-ray tomography
The DLS I12 beamline [21] was used to perform in situ XCT scans on the two samples at different stages of Hertzian indentation loading due to the combination of high flux and energies provided. The monochromatic X-ray beam was tuned at 101.97 keV. Camera optics with field-of-view: 8.0 mm x 7.0 mm and, resolution: 3.25 µm/pixel were used to encompass the entire sample volume. The exposure time was 0.4 s, with 3000 projections acquired per scan (total scan time: 20 minutes). Reconstruction of the data was performed using the opensource software Savu [22]. Data visualisation and analysis software Avizo [23] was subsequently used for further data processing including 3D visualisation, volume/length measurements and material segmentation.

Experimental procedure
The loading sequence used during stepwise mechanical loading was selected for each sample based on ex-situ trials. Immediately after the pre-planned value for the applied load was recorded, XCT scans were performed. The ZrO 2 indenter remained in contact with the sample during scanning under an applied load. The first scan prior to loading, and the last scan after sample failure, were performed with the indenter retracted. The force applied on the load cell was continuously monitored. Table 1 summarises all the actions performed in sequence for black and brown LFCM simulant samples. It is important to mention that the load values given in Table 1 refer to the initial-maximum load recorded per cycle. Immediately after reaching the targeted load value, force relaxation started, and the recorded values began to decrease. Thus, the load during the XCT scans was lower than the initial targeted value as given in Table 2.

Mechanical behaviour
The load paths followed for the two samples (black LFCM; brown LFCM), as recorded throughout the entire experimental period -including both loading and subsequent scanning processes -are presented in Figure 2.
The black LFCM simulant sample failed during the fourth loading cycle, up to 130 N. A sudden drop of the load was recorded immediately after 127 N. The brown LFCM specimen exhibited a different behaviour since it failed during scanning after the second loading cycle was completed. The "load plateau", which is observed in the graphs ( Figure 2) after each peak load condition was the period when scans were performed. Figure 3 illustrates one representative cropped 2D slice for each of the two samples tested.

Visualisation and segmentation
The multiphase nature of the samples is evident. In situ phase characterisation work was not performed due to time limitations and therefore phase identification is based on the relevant available literature. The bright spots/areas are attributed to high-density Uranium-rich phases.
Pores of various sizes are also observed. Based on the different image gray-scale ranges, segmentation of each separate phase was performed, enabling both qualitative and quantitative analysis. The threshold values separating the different phases in the gray-scale intensity range were determined using the "Line-Probe" module in Avizo [23]. A typical segmentation process is illustrated in Figure 4. Characteristic 2D XCT slices for each of the materials tested are presented, along with the corresponding post-segmentation individual phases, which are shown in white colour.
U-rich phases are assumed to include the range of (U 1-x Zr x )O 2 crystallites where the value of x is less than 0.1, while Zr-rich phases contain little or no U, being monoclinic (Zr 1-x U x )O 2 (where x < 0.1) or zircon, (Zr 1-x U x )SiO 4 (where x < 0.1). Fe-rich phases, present as small particles from the inclusion of stainless steel in the "lava" composition, are difficult to differentiate from Zr given their similar density (7.84 g cm -3 and 6.49 g cm -3 for Fe and Zr, respectively), thus they have been grouped together. It should be noted that previous analysis of these materials showed the occurrence of Fe-particles to be low, at ~ < 1 vol.%, therefore, the majority of the phases shown in Fig. 4 iv and viii are likely to be Zr-rich.
3D models were also built, as shown in Figure 5. Cuboid regions were selected from the original samples for 3D model rendering (65% to 85% of the entire volume) to reduce computational time. The segmentation enabled quantification of the contained phase volumes in both samples. Using Avizo segmentation and analysis tools, general porosity, pore size distribution, U-rich and Fe/Zr-rich phase volume fractions were determined using the preloaded datasets; results are presented in Figure 6. It is observed that the brown LFCM sample exhibited a slightly larger U-rich phase volume fraction (2.8%) than did the black LFCM (2.2%). It is also interesting to note that the brown LFCM sample porosity was almost double that of the black LFCM. A similar observation can be made for the Zr/Fe-rich medium density phase volumes. The volume fraction of Zr/Fe-rich phases contained in the brown LFCM sample (16%) is almost double that contained in the black LFCM volume (7.7%). The evolution of structural degradation was monitored as load increased up to failure. Characteristic 3D tomogram screenshots are shown in Figure 7 for the samples investigated.
The load values reported beneath each tomogram correspond to the average load value recorded during the relaxation-scanning process ( Table 2). Yellow arrows point at cracks formed due to excessive loading. The isolated crack volume network after failure is shown in Figure 8 to enable optimum visualisation. The tomograms revealed a single crack having formed in each of the investigated volumes. The width of the crack in the black LFCM sample was considerably larger than in the brown LFCM.
Segmentation of the void volume, including cracks wherever applicable, was performed on all the cropped datasets, corresponding from zero load up to failure. No differences in total void volume were identified between tomograms as load increased, except for the last scan where in both cases (black LFCM; brown LFCM) wide cracks generated due to excessive loading. This behaviour implies that both materials exhibit a similar type of brittle failure, with no signs of micro-cracking, which would reflect an increase in the total void volume, in the proximity of peak load. The above hypothesis is also corroborated from the load paths of both samples, and especially black LFCM which failed during topping up the load ( Figure 9).
The immediate drop of load beyond the peak value is typical for brittle materials.
The correlation between phase volume fractions and failure load for the two samples investigated are shown in Figure 10. As expected, sample failure occurs at a lower load when the sample contains greater porosity, and the same trend can be observed for the U-rich and Fe/Zr-rich phase volume fractions. Pore size distribution analysis, performed by combining the "Material Statistics" and "Label Analysis" Avizo modules [23], generated the histogram shown in Figure 11. Both samples exhibited almost the same total pore volume in the "small pores regime" (0 mm 3 to 0.001 mm 3) . It can be observed that the brown LFCM sample contained more pores of intermediate size (0.001 mm 3 to 0.1 mm 3 ) than did the black LFCM.
The corresponding difference is more significant when comparing the number of large pores (0.1 mm 3 to > 10 mm 3 ) present in the investigated volumes. This greater number of individual large pores present within the brown LFCM sample explains why its total porosity is almost double that of the black LFCM ( Figure 6).

Digital volume correlation
To extract the displacement and strain fields at the different loading stages, the deformed volume images were successively compared against the reference volume image set-up as the latter was determined for every sample. The Avizo "XDigital Volume Correlation" extension module [23] was employed to quantify the loading-induced deformation on the samples tested. The original tomograms were processed as follows to reduce computational time and be compatible with software recommendations. Open source software ImageJ [24] was used to convert the original 32-bit images to 8-bit images. Image noise reduction was performed by using the despeckle median and smooth filters [24]. Data-cropping was essential not only to reduce the computational cost but also to produce ideal geometric volumes. Comparisons between the DVC volumes used and the corresponding entire material volumes for both black and brown LFCM samples are shown in Figures 12 and 13. The samples were not perfect cuboids; therefore, the relevant dimensions given closely approach reality.
The Avizo "XDigital Volume Correlation" extension module is based on a double mechanism to determine the deformation occurring between two relevant image stacks. A subset-based (local) approach is initially employed to capture the large displacements within the probed sample volume. Therefore, a coarse grid is primarily implemented. The large displacement field, as calculated via the local approach, is subsequently used as input to initialise a more robust finite element (FE)-based (global) DVC algorithm. The corresponding global-approach mesh is significantly finer, allowing displacement field calculations of greater precision.
The 3D tomograms related to the scans conducted after the sample failure were not considered for analysis. The significant width of the cracks formed during the last loading cycle would generate unreliable and erroneous displacement and strain fields [25]. The scans performed at the first loading cycle for each sample were adopted as the initial reference volumes. This approach was based on the considerable sample movement and rotation occurred after immediate contact with the indenter. As a result, erroneous results were produced when the corresponding DVC analysis was performed for the transition from zero load to the first loading cycle. After applying the initial small load, the sample was more stable and the transition to higher loads provided more reliable results for comparison. The concept of using scans associated with small initial loading as the reference data for DVC analysis has been commonly followed in the literature [14].
Prior to performing DVC on the real data, a virtual displacement method was employed to test the reliability of the image registration technique, which is used to correlate the different image datasets. On all occasions, the reference tomogram (corresponding to the first loading cycle) was subjected to certain prescribed displacements across the three axes. The test strain values imposed across the three axes to evaluate the reliability of the method are shown in Table 3 for both samples. The reference and deformed volumes were then analysed to evaluate if the DVC outputs (displacement; strain field) were close to the initial, pre-defined deformation values.
Metric and transform ports were set to "correlation" and "translation and rotation" modes, respectively [23]. Correlation threshold was set to 0.5. After running the local approach, a displacement field and a strain tensor were created along with a metric map. The displacement field was subsequently passed as an input file to run the global algorithm.
Maximum number of algorithm solver iterations was set to 30 while the convergence criterion was maintained steady at 0.001 for all analyses. Various sizes of the FE mesh implemented were tested in the form of a sensitivity analysis to determine which mesh best described the pre-defined deformation state. The ideal sub-volume sizes (as determined from the local approach) and mesh sizes (as determined from the global approach) were used as input DVC parameters for analysis of the real XCT data corresponding to successive loads.  Table 4.
The reference and deformed volume datasets, which were used in pairs to monitor the sample internal deformation as load increased are given in Table 5 along with the correlation coefficient of the DVC runs. The corresponding sizes of all the final FE-based tetrahedral meshes implemented are also included. The results of the DVC analysis provided quantitative information about the deformation occurred within the investigated sample volumes. More specifically, displacements at every node of the implemented model as well as strains at each element of the generated mesh were extracted on all occasions across the three axes. In total, nine 3D visualised datasets were created per DVC run, including three displacement maps (u x , u y and u z , reflecting the nodal displacements across the x, y and z axes, respectively), three axial strain maps/tensors (ε xx ; ε yy ; ε zz ) and three shear strain maps/tensors (ε xy ; ε xz ; ε yz ).
The relevant 3D displacement and strain maps for the black LFCM sample are illustrated in Even though only the transition from the first to the second loading cycle was analysed using DVC, certain remarks can be made. The black LFCM sample did not exhibit any significant variations in displacement and strain values during the same period (from the first to the second loading cycle). However, a slightly different behaviour was observed for the brown LFCM. The displacement profile ( Figure 19) reveals strongly non-uniform deformation across the top surface of the analysed volume. The diagonal stripe of larger u z values shown in Figure 19c seems to be in good agreement with the localised increased axial strains noted predominantly across the y-axis (ε yy in Figure 20b) and, to a lesser extent, across the z-axis (ε zz in Figure 20c). With regards to the displacements reported in the legends of Figures 16   and 19, it is important to note that the relevant values are given in µm.

Mechanical properties
The DVC process enabled quantitative analysis, in addition to the qualitative comparison described in the previous section. The displacement and strain data, as extracted for all the nodes and elements of the 3D meshes implemented, were used as inputs to examine the prospect of determining the mechanical properties of the materials tested. Alternative approaches were employed to estimate Young's modulus and Poisson's ratio values. The first method was based on the displacement data.  Based on the schematic shown in Figure 22, the sample deformation can also be calculated based on equation 2.
Combining the equations 1 and 2 enables determining the vertical strain ε zz based on the average displacements U 1 and U 2 , as determined via the DVC analysis, and the sample initial height (Equation 3).
The above process was followed to determine the vertical strain ε zz at all the load-transition phases. A similar workflow process to estimate the mechanical properties of quasi-brittle, heterogeneous materials has been available in the literature [16]. Based on the vertical strain ε zz values determined at each loading step, Young's modulus was subsequently calculated based on Hook's law according to the following equation: Where E is the Young's modulus (measured in MPa), σ zz * is the stress equivalent of a uniaxial compression state, ΔF is the difference in load between two successive loading stages (measured in N) and A is the sample surface (measured in mm 2 ). The above formula has been adopted as a simplification of the stress state developed in the sample volumes during loading. Hertzian indentation does not generate a uniform stress state (e.g. uniaxial tension/compression), and therefore, equation 4 may not be appropriate for use. However, the sample volumes were significantly limited and cannot be considered as infinite plates, as in a classical indentation problem. In addition, the strain profiles measured by DVC analysis showed a relatively uniform vertical axial strain distribution during the early loading stages, resembling a uniaxial compression behaviour. Since the Poisson ratio was unknown and its value could not be predicted due to sample heterogeneity, and load-penetration depth curve during unloading was not available, analytical related to indentation loading expressions could not be used to determine the Young's modulus. Therefore, our method, even though a simplification, was considered to provide at least quantitatively, a way of comparing the values of Poisson ratio and Young's modulus between different types of simulants.
A specific Young's modulus value was calculated for every separate load combination and, consequently, DVC analysis performed per sample. In the case of black LFCM, where two different load transition scenarios were investigated using DVC, an average Young's modulus value was calculated. The corresponding results obtained after applying the displacement method are given in Table 6. To corroborate the results obtained from the displacement method and to monitor the uniformity of material stiffness within the entire investigated volume, similar calculations were made using the available vertical strain ε zz data (Figures 17c, 20c).
The average strain was determined on each occasion, considering the values exported from all the mesh elements. Subsequently, the Young's modulus was calculated using the Equation 4. The results are provided in Table 7. A good agreement between the two methods can be observed.
The relatively non-uniform vertical strain distributions, as shown in the relative DVC maps (Figures 17c, 20c), could imply inconsistent material stiffnesses. Therefore, in addition to calculating the average strain values considering the entire investigated volumes, potential variations of the Young's modulus within the samples were also evaluated. The xy planar average vertical strains were determined across the heights of the DVC volumes generated.
Subsequently, xy planar Young's moduli were estimated based on the relative compressive stresses (σ zz *).
The alteration of stiffness across the height of the inspected DVC volume per sample is illustrated in Figure 23.  Figure 24. Noticeably, a significant variation in Poisson's ratio can be observed for all the samples tested.

Finite Element Modelling
The hypothesis of a uniaxial compression stress state, as expressed via Equation 4, was necessary to provide numerical solutions of the problem discussed in this paper. However, potential effect of this simplification on the calculated Young's modulus value was also investigated. Finite element (FE) analyses were performed using the software package ABAQUS 6.14 [26] to simulate the indentation tests performed on the materials. The geometry of the samples and the indenter was accurately reproduced in the FE models implemented. The results of the FE analyses were used to reproduce the ratio shown in Equation 4, which was adopted to derive the Young's modulus based on the DVC analyses.
The Young's modulus test value, used as input for the FE analyses, was compared against the corresponding ratio of Equation 4. Potential variation between the two values accounts for the effect of the uniaxial compression stress state hypothesis (instead of indentation) on the value of the material Young's modulus.

FE models
Two principal FE models were produced to simulate both black and brown LFCM indentation tests. The geometries adopted were identical to those described in Figures 12 and   13. The cuboids created to simulate the "lava" samples were partitioned to enable visualisation of the corresponding DVC-associated interior volumes. Realistic representation of the sample microstructure was beyond the scopes of this FE modelling work package.
Therefore, both samples were simulated as homogeneous solids.

FE type, mesh, and material modelling
Three-dimensional eight-node hexagonal solid elements (C3D8) with hourglass control and reduced integration were used to build the meshes of the "lava" samples. The indenter was modelled as a 3D discrete rigid body. Mesh convergence studies were conducted to determine the optimum element size, providing an accurate solution combined with minimum computational time. Therefore, the approximate global element size for both meshes was 0.3 mm. Since the investigation of the sample failure state was not part of the FE modelling work, both "lava" materials and the ZrO 2 Hertzian indenter were modelled as isotropic, elastic materials. The Young's modulus and Poisson ratio for ZrO 2 were set equal to 210 GPa and 0.22, respectively. Since the material properties of elasticity for both black and brown LFCM were unknown, theoretical values were adopted to implement the FE analyses.
Therefore, test values of 6 GPa and 0.3 were used as input data for the Young's modulus and Poisson ratio of both materials.

Boundary conditions, interaction properties and load application
The The normal behaviour of the system was modelled with a hard contact pressure-overclosure relationship, allowing surface separation under tension, and restricting node penetration under compression. A penalty contact algorithm with a friction coefficient equal to 0.5 was employed to describe the tangential behaviour of the contact properties. A displacementcontrolled method was selected to simulate the loading process. Therefore, a prescribed vertical displacement of 0.5 mm was implemented at the reference point, ensuring the indenter moves as a rigid body only across the z-axis.

FE results
Processing of the FE modelling results followed the methodology employed to analyse the DVC data (strain method). The first step of the analysis included calculating the average vertical axial strain ε zz at each xy plane of the generated mesh. Consequently, planar average strain values were determined across the z-axis. These strain values were subsequently averaged to determine the total average ε zz strain, in compliance with the process followed for the DVC analysis. This workflow was repeated for every increment of the analysis. The goal of the FE modelling work was to reproduce the ratio given in Equation 4. Therefore, the hypothetical vertical stress σ zz * was determined, assuming uniaxial compression conditions, as explained in the section 3.4. The following formula was employed at each increment of the analysis: * = where RF is the reaction force (N) applied at the reference point, and A is the surface area (mm 2 ) of the "lava" samples. Average vertical stress σ zz at the top (σ zz-top ) and the bottom (σ zz-bottom ) surface of the samples as well as average Von Mises stress (σ VM ) were also calculated for the sake of comparison. The evolution of all stresses with loading is given in  Table 7) to estimate the corrected stiffness of the "lava" materials.

Discussion
Performing indentation loading with successive XCT and DVC on the samples tested on a synchrotron beamline is an effective method to determine the mechanical behaviour and the microstructure of materials. The radioactive nature of our samples hindered the possibility to use standard methods of mechanical testing (e.g. use of strain gauges), since a greater material volume would be required. The tomograms revealed the multiphase nature of the materials and the in-situ indentation their near brittle fracture behaviour.
Assuming that these surrogate, laboratory-manufactured materials behave similarly to the real Chernobyl LFCM inventory, a very rigorous decommissioning process needs to be followed Based on the phase volume analysis ( Figure 6) and the pore size distribution histogram ( Figure 11), it was expected that brown LFCM would prove to be more brittle than black illustrating the full size of the sample, is given in Figure 27. The corner lump present across the bottom surface of the sample has certainly affected its mechanical response, leading to a premature yielding during scanning. Therefore, it is believed that the load bearing capacity of the brown LFCM sample would be considerably higher. The fact that it is both more porous and stiffer compared to black LFCM hinders the possibility to estimate with confidence which of the two materials has a greater yield strength. Quasi-brittle materials differ from brittle ones since they display a softening behaviour after peak load instead of a sudden complete loss of load bearing capacity. Among common quasibrittle materials, concrete and graphite are probably the most broadly investigated. A typical range of Young's modulus values for concrete could lie between 14 GPa and 41 GPa. Based on its composition, graphite stiffness is slightly lower (4 GPa to 28 GPa). Generally, the Young's modulus values calculated for black LFCM and brown LFCM, using the DVC data lie in good agreement with typical corresponding values reported for other brittle and quasibrittle materials. Based on the sample loading behaviour (Figure 2) it seems that there is no softening after the peak load. Thus, it could be safely confirmed that the two samples tested on the DLS I12 beamline are classifiable as brittle materials of slightly increased stiffness compared to the average. The displacement and strain maps created after performing the DVC analysis revealed zones of elevated deformation in locations where cracks formed after sample failure, as shown by the corresponding tomograms. The DVC maps and the tomograms related to the pre-failed status of the samples were combined to investigate potential correlation between the microstructure and the crack network formed after sample failure. Figure 28 presents tomograms related to the first loading stage (11 N and 15 N for the black LFCM and the brown LFCM, respectively) superimposed with corresponding DVC strain maps (ε xy and ε yy strain for the black LFCM and the brown LFCM, respectively).
A narrow crack present within the black LFCM sample, which has formed during sample preparation (yellow arrow in Figure 28a) and prior to loading, is very close and almost perfectly parallel to the zone of elevated strain, as shown from the corresponding DVC map.
It is expected that this crack facilitated interior sample deformation. The orientation of the elevated strain zone (black arrows) around the upper surface of the cuboid in Figure 28a shows that the presence of large pores has acted as a preference path for further deformation and potentially crack propagation. The brown LFCM sample volume (Figure 28b) exhibited no obvious material discontinuities apart from pores being present. The elevated strain zone, as demonstrated by a red colour strip (black arrows), contains several small pores along the length. It is interesting to note that no pores exist at the sides of this elevated strain zone, while a solid Si-rich matrix can be observed.

Conclusions
A combined investigation of the microstructure and mechanical behaviour of hazardous materials requiring retrieval from Chernobyl Unit 4 nuclear reactor power plant has been presented, using simulant samples. The methodology included a series of stepwise Hertzian indentation loading performed on surrogate materials, coupled with successive synchrotron XCT, and subsequently analysed using DVC. The goal was predominantly to determine the morphology, load bearing capacity and stiffness of the samples investigated. The key findings are summarised as follows:  The combined Hertzian indentation-XCT-DVC methodology proved to be a successfully defined process to determine the mechanical behaviour of hazardous materials with sample volume restrictions during testing.
 Sample testing showed that the "lava" simulant materials can be classified as brittle with no evident softening behaviour.
 Brown LFCM exhibited almost two times greater general porosity than black LFCM (~12% compared to ~6%)  DVC-based displacement and strain maps clearly revealed the zones of forthcoming excessive cracking before those appeared in the relevant tomograms.
 The range of Young's moduli calculated for the two samples tested on the DLS I12 beamline was approximately 7 GPa to 24 GPa.
 The different failure mechanisms noted for the two materials tested hinders the possibility to directly compare their load-bearing capacities.
 The brown LFCM sample stiffness was considerably higher than that of the black LFCM sample.