6.1 PRIMARY CONTAINMENT INTEGRITY

Safety assessments for nuclear plants include the effects of Severe Accidents1 on the integrity of the reactor vessel (the primary contain­ment). Water and fast reactor vessels are potentially subject to short time-scale pressure and fluid-impact loadings from MFCI. In addition, fast reactor vessels might suffer slower pressure-induced loadings over several seconds due to conventional vaporization by larger sized corium debris, and in the United Kingdom these are termed Q*-events [202]. French and UK fast reactor designs are the pool type shown in Figure 5.10, in which a double-skinned reactor vessel houses interme­diate heat exchangers, pumps and access areas to core components. Their typical 2000 tonne sodium inventories provide enormous heat sinks, and if 50 tonne of molten corium at 5000 K were to passively equilibrate, the bulk sodium temperature would rise innocuously from 900 K at 1 bar to less than 950 K. The alternative loop-type fast reactor

1 For fast reactors, Severe Accidents are more usually termed hypothetical core disruptive accidents (HCDA). [96] design, like the German SNR300, has intermediate heat exchangers and pumps outside the reactor vessel, so the primary circuit layout is not dissimilar to that of a PWR. For obvious safety reasons code validation experiments have generally[97] [98] used water rather than molten sodium as the coolant. If it can be demonstrated that primary circuit elements can withstand the spectrum of Severe Accident loadings, then the surround­ing reinforced concrete building (the secondary containment) would provide a second, but redundant, barrier between radioactive fission products and the local population. Descriptions follow of scaled fast reactor experiments aimed at validating calculations of transient stresses in primary circuit components that would occur from MFCIs. These investigations establish the soundness of the structural analysis codes used in safety assessments for both water and fast reactor designs. Later sections outline experimental and theoretical research into the robustness of reinforced concrete containments [106,275] to potential impacts of airborne plant fragments (“missiles”) and aircraft.

Operating temperatures of around 500 °C and crowded interiors complicate replica scaling of a reactor’s interior. Models used in experiments have sought to represent isolated features (e. g., hemispher­ical dome on a cylinder [273]) or complex 1/20th scaled versions of internal structures [274,276]. The COVA series of experiments [88] on pool and loop fast reactor designs were an international collaboration between AEEW, AWRE and the European Joint Research Center Ispra. Initial results were used to validate the two-dimensional axisymmetric fluid dynamics codes ASTARTE (Lagrangian) and SEURBNUK (Eulerian). Because gross fluid motions around internal structures are more readily represented by a fixed-geometry Eulerian mesh, the latter became the focus of research activity. Structural loadings from SUERBNUK formed inputs to the mathematically decoupled structural dynamics code EURDYN, which at the time had just one-dimensional modules. Later code developments created two — and three-dimensional subroutines for analyzing the WINCON [276], STROVA and reinforced

з

concrete tests.

The rapid structural loadings from MFCIs increase the yield stress of steels by some 25% as illustrated by Figure 6.1 Because EURDYN modules did not allow for this strain-rate enhancement, scoping

500

A

B

S 200

100

calculations were necessary to span the experimental measurements by using low — and high — strain rate data. Repeatability of a load inducing detonation is patently crucial for code validation, but evidently lacking in experimental MFCIs. Accordingly chemical explosives have been used as MFCI simulants. At the Enrico Fermi Plant a TNT charge was actually detonated in a sodium-filled vessel [271], but Figure 6.2 reveals that the energy release transient from a high explosive differs materially from an MFCI event. A detonation is characterized by the shock wave spatially leading the place of energy release [203],whose release rate for an MFCI is restricted by heat diffusion within the resulting fine debris. On the other hand the far faster energy release and shock intensity from a military explosive result from a virtually concomitant rupturing of chemical bonds at the shock front. It follows that a closer match to an MFCI detonation necessitates a much smaller shock speed than in a chemical explosive. By coating easily compressed polystyrene granules with pentaerythritol tetranitrate (PETN) and then expanding the mixture in a mould, shock propagation speed is markedly reduced[99] to achieve the closer match shown in Figure 6.2. Nevertheless the low-density explosive (LDE) still releases some 20% of its energy as an over-sharp shock wave. To obtain a validation of EURDYN’s finite element modules under conditions closer to those of an MFCI, the STROVA rig [273] with a vacuum gun was used. In essence this gun consists of an

evacuated barrel with a diaphragm seal at each end. A triggered breach of the upper diaphragm drives a weight downwards[100] under atmospheric pressure so as to rupture the lower diaphragm, which then drives a piston whose impact on a hydraulic fluid loads a selected test piece.

The internationally sponsored COVA experiments [88] progres­sively added different internal components in 1/20th scale models of loop — and pool-type reactors until all the major axisymmetric features became broadly represented. An LDE charge was detonated on the axis of symmetry and each test was repeated at the previously cited research centers to eliminate systematic errors. Deforming structures were represented by thin-shell segments within SEURBNUK, and more complex or “rigid” structures via the link to EURDYN-1. Experimental pressure loadings below the liquid surface and strain patterns were generally well-represented by the two codes. However, impact loads on a model roof and the actual magnitude of strains did not meet the desired accuracy of ± 20%. Because the COVA program considered just axisymmetric structural components, discrete three-dimensional resist­ances to fluid flow had to be “smeared”. Later WINCON [276] tests

involved more realistic models of the internal structures within a pool — type fast reactor as depicted in Figure 6.3 As in the COVA series the number of different components was increased progressively through the series as illustrated by Figure 6.4, but appropriate stress-strain calculations were now performed by three-dimensional EURDYN-3 modules. Significant asymmetries in fluid flows or vessel loadings were discovered not to have been introduced by the more realistic arrange­ments. The rotating roof shield evolved as the weakest component in an early CDFR design, and impact by a moving mass of sodium from a large enough MFCI could raise it sufficiently to allow an escape of

sodium and corium into the reactor hall. Accordingly, the effectiveness of dip plates, deflector plates and crushable shields were investigated as means of protecting the roof. Though crushable material reduced peak loadings by more than a factor of 4, design difficulties settled the adoption of a simple dip plate that provided a predicted 50% reduction. With a refined version of SEURBNUK [276], sodium impacts on the model roof were found to be transformed into a series of weaker pulses

by successive recompressions of the coolant. This dynamic coupling between fluid and structural dynamics also reduced roof stresses by about 50%. As well as motivating design developments, it shows that the decoupling of SEURBNUK and EURDYN calculations is inappropriate. Ref. [276] provides a detailed account and critique of the WINCON experiments.

A more stringent validation of EURDYN modules was undertaken by the STROVA rig studies [273] on two basic types of reactor compo­nents. One set represented scaled roof elements like circular and annular plates and then progressed to a composite representation of the whole CDFR roof. The other set concerned hemispheres on cylinders which characterized portions of the fast reactor vessel itself (and clearly that of a water reactor as well). An initial test program employed aluminum models which being largely independent of strain-rate enhancement provided a more incisive test of the code. Then with ferritic steel experiments, the plate and annular models were deflected by a few thicknesses, while the composite roof was distorted by around one — quarter of its depth to take the metal into its plastic regime. Hoop strains of up to 2% were sustained by the hemisphere on cylinder models, and by repeat experiments an estimated accuracy of ±5% was achieved. A similar accuracy for other strain measurements was obtained from recordings at points of symmetry. Pressures applied to specimens were taken by tourmaline transducers to an accuracy of ±1.5%.

Strain rates up to 5/s were measured during tests on plain and annular plates, and EURDYN-1 calculations with low and high strain — rate data predicted the observed maximum deflection to within ±3% and —21% respectively. By imposing constraints to match boundary and symmetry conditions, a 45° sector of the composite roof model was sufficient for three-dimensional calculations. Because experimental strain-rates of up to 25/s were observed in the lower roof plate, calculations with the higher strain-rate data generally gave the better match. However, the predicted stress transients were too quick and the final 8 mm deflection of its inner edge was underestimated by up to 25%. Refinement of metallurgical data would clearly have enhanced predictive accuracy. Moreover, “dimpling” of this test specimen again indicates a significant interaction between fluid and structural dynam­ics. Consequently SEURBNUK and EURDYN are required to be mathematically coupled, but foreclosure of the European fast-reactor project forestalled the necessary developments.