To evaluate the reliability of different laboratory test methods, concrete cubes were produced with the same concrete mixture as used in the prisms for the accelerated laboratory testing. The concrete cubes with 300 mm lateral length were stored at different outdoor exposure sites in Europe. The expansion and the maximum crack width were determined periodically at approximately the same temperature (15°C) in spring and autumn.

Thirteen aggregate combinations (of the 22 aggregate types included in the laboratory test program) were selected with the purpose of covering most types of reactive aggregates throughout Europe and with respect to mineralogical properties and alkali-reactivity (Table 2). In some cases the coarse fraction was tested in combination with non-reactive sand (N3 from Norway) or a fine fraction was tested with a non-reactive coarse aggregate (F2 from France) or the fine and the coarse fractions were tested together. Additionally, non-reactive reference aggregates (F2) were tested. A brief petrographic description and details about the reported reactivity in concrete structures of these aggregates are given in (2). The aggregates were grouped into three categories according to their reported field behaviour:

The former classes for “slowly” and “normally” reactive aggregates in (2) are re-named in this paper as “moderately” and “highly” reactive aggregates to comply with classes in North America (14-16).

The former RILEM standard cement CEM I 42,5 R provided by NORCEM AS, Norway was used for preparing the concrete. The total alkali content of the cement was 1.26 mass% Na₂O-equivalent.

Concrete was made with 440 kg/m³ cement, and the water to cement ratio was 0.50. The air content was approximately 1.5 vol.-%; no air-entraining agent was added to the concrete mix. In case of inappropriate workability of the concrete (slump < 20 mm) a superplasticizer was added. In accordance with the RILEM test method AAR-3 and AAR-4.1 the aggregate combination consisted of one of the following (see Table 2):

The aggregate fractions were combined in proportions of 30 mass-% fines (0 to 4 mm) and 70 mass-% coarse aggregates: 30 mass-% 4 to 10 mm and 40 mass-% 10 to 20 mm.

Two concrete cubes with 300 mm lateral length were produced for each field site and each aggregate combination (Table 3). All the cubes representing one concrete mix (i.e. one aggregate combination) were cast at one laboratory (generally in the country of origin of the aggregate) and transported to all the other laboratories (field test sites). After production, the cubes were kept in the moulds for one day, de-moulded and stored in a room at 20 ± 2°C and ≥ 95% relative humidity (or were covered with moist fabric) for 6 days before being transported to the different field sites (Figure 1).

At the different field sites, each institute that participated in this research installed two pairs of reference studs into the top surface and into the two adjacent side faces, before the cubes were exposed outdoors. Most laboratories pre-drilled holes before gluing the studs. All cubes were stored in the same direction in relation to the four cardinal points to minimize deviations between the labs resulting from different exposure to direct solar radiation (Figure 2).

During exposure, one cube was stored with its base in a tray filled with water and the other was exposed only to ambient rainfall (Figure 2). The tray was filled with water to simulate a permanently wet concrete, so that the bottom of the first cube was immersed 50 to 60 mm in water during the whole testing time. The reference points at the bottom of the first cube were always above water level enabling length change measurements. Since 2010 (6 years after exposure) the trays were refilled only by rainfall instead of a manual control of the water level due to the fact that there were no significant differences between the two conditions. The concrete cubes were stored on eight different field sites that were selected to cover all climates in Europe (2). Figure 3 gives the mean monthly temperature and precipitations for each field site.

The dimensions of the cubes at the top surface and two adjacent side faces as well as the crack width were determined periodically (first 2½ years every three months, afterwards every half year). Some laboratories have only measured once a year. The measurements were conducted at the field site at temperatures around 15°C and preferably in periods with rather stable temperatures over a 24-hour period and with limited sunshine. The mean expansions of the three side faces are presented in Figure 4.

Figure 4 shows the mean expansion of selected cubes from four field sites Milan, Italy (Mi), Düsseldorf, Germany (Du), Brevik, Norway (Br) and Trondheim, Norway (Tr). The cubes were stored “partly immersed in water”. For comparison with laboratory test results, one cube for each aggregate combination that was reliably measured up to 15 years was selected. The selection was necessary because some cubes weren’t measured constantly over the entire 15 year period. The colours used in all the figures correspond to the aggregate reactivity (see 2.2).

All highly reactive aggregates expanded within the first six years at the four field sites and showed high expansions from 0.3% to 1.6% after 15 years. In the mild and warm climates of Düsseldorf, Germany (Du) and Milan, Italy (Mi) the expansion rates decreased after some years, whereas in cold climates like Trondheim, Norway (Tr) (Figure 5) and Borås, Sweden (shown in (2)) the expansion still continued, probably due to frost that damages the concrete further once ASR has caused sufficient cracks (17).

The cubes with the moderately reactive aggregates expanded considerably slower with expansions of about 0.09% to 0.22% after 15 years.

The cubes with the non-reactive aggregate F2 neither expanded at any field site nor showed significant cracking. However, the Damage Rating Index (DRI) determined on polished sections and qualitative damage assessment performed on thin sections revealed that an alkali-silica reaction (ASR) occurred to a small extent (18). The gravel with flint F1 was considered to be potentially reactive with clear pessimum effect (2). Deviating from (19), the gravel with flint F1 is classified as non-reactive in this paper because damage in structures was not evident (2) and the cubes in Milan only showed very little expansions of 0.05 % after 9 years and no significant cracking. Thin section analysis also revealed for this concrete little ASR compared to the majority of moderately and all highly reactive aggregates (18). Based on experiences from other field site tests, a limit value of 0.050% is applied to identify non-reactive aggregates (17).

For the following laboratory-field-correlations, the mean laboratory test results of all participating laboraties were used. Some results were excluded in the case they turned out to be unreliable and if the laboratory was unexperienced with the test method.

Both versions of the AMBT (AAR-2.1 and AAR-2.2) were able to reliably distinguish between non- and highly reactive aggregates (Figure 6). It also identified the majority of the moderately reactive aggregates. The expansion of the Portuguese silicified limestone (P1) was below the acceptance limit value with both prism sizes. However, the very slowly reacting Swedish Gravel with porphyritic rhyolite (S1) behaved very differently depending on the prism size. It passed with the long thin prisms (AAR-2.1) and failed with the short fat ones (AAR-2.2).

The concrete prism test methods AAR-3 and AAR-4.1 were effective in distinguishing between the non- and highly reactive aggregates (Figure 7 and Figure 8). In general, the highly reactive aggregates showed expansion above 0.16% in AAR-3 and between 0.10% and 0.20% in AAR-4.1. An exception is the flint containing sand D2 that showed only little expansions in the laboratory tests and even passed AAR-3.

Looking at the moderately reactive aggregates, the two versions of RILEM test method AAR-4.1 were detecting its reactivity potential more reliably, if the limit value of 0.03% after 20 weeks as proposed in (3) is applied. AAR-3 failed in identifying the slowest reacting aggregates P1, S1 and IT2, whereas N4 expanded just above the proposed acceptance limit value of 0.05% after 52 weeks (3). The AAR-4.1 was far better, and displayed the reactivity potential of four of the five moderately reactive aggregates. Only P1 was close to the limit value of 0.03%. The results confirm the RILEM-proposed acceptance limit of 0.03% (3) and suggest an assessment after 20 instead of 15 weeks. Otherwise, S1 would be classified as non-reactive. P1, S1 and IT2 are the slowest reacting aggregates in this comparison (no figures).

As for RILEM AAR-3 and AAR-4.1, the German and the Norwegian concrete test methods were able to distinguish between non- and highly reactive aggregates (Figure 9). However, the German method failed to identify the moderately reactive aggregates. Even the additional 300 mm-cube didn’t show maximum crack widths ≥0.20 mm (not shown) for these aggregates. The overall expansions were lower compared with AAR-3, probably caused by a higher alkali leaching rate in the German fog chamber.

Compared with AAR-3 and the German method, the Norwegian CPT had the best match with the field performance of the cubes. It correctly displayed the alkali-reactivity potential of the three tested moderately reactive aggregates, even for the very slowly reacting S1. This can probably be attributed to the bigger prisms (100 x 100 x 450 mm³) and less alkali leaching compared to AAR-3 (75 x 75 x 250 mm³) (20).

In the Danish mortar bar test TI-B51, the expansion after 20 and 52 weeks is used for classifying aggregates into three alkali-reactivity classes. The names of the classes below are used in Figure 10 instead of the original ones described in the method:

The TI-B51 was able to show successfully the reactivity potential of the non-reactive F1 and all highly reactive aggregates, but underestimated the reactivity potential of the moderately reactive aggregates N2, P1, IT2 and S1 (Figure 10). Furthermore, the classes used in this paper compared with the classes described in the procedure of TI-B51 differ a lot. Except for B1, the four other highly reactive aggregates were classified as moderately reactive according to the limit values for TI-B51.

The result of the Danish Chatterji method is a calculated Δ-value that is shown in Figure 11. The results suggest that Δ-values of 19 and higher are indicating a potential reactivity of the aggregate. Highly reactive aggregates revealed Δ-values between 30 and 50. Exceptions are aggregates with flint like F1 and D2. The non-reactive F1 gave a very high Δ-value and the highly reactive D2 had a very low one compared to the other highly reactive aggregates.