A grade H35 concrete (35 MPa) was dosed according to NCh 170 (31), the minimum recommended by the Code Standards and Technical Specifications of Paving Works of the Ministry of Housing and Urban Planning for concrete pavements (32). To investigate the different effects of RAP on concrete properties, four groups of concrete mixtures with different RAP contents were prepared: 0% (control), 20%, 50% and 100% replacement of fine and coarse aggregate, by weight. These mixtures were designated R0, R20, R50, and R100, respectively. A total of four mixtures were prepared, one for each RAP content. For determining the proportions of the materials in the mixtures, the modified Faury method was used. Table 4 shows the proportions in the mixtures studied. In all mixtures, 160 kg/m³ of water was used and the water to cementitious material ratio of 0.44 was kept constant. The concrete mixtures were prepared using a mechanical mixer and compacted by a vibrating table. The samples were cured in immersion, to guarantee the correct hydration of cementitious materials until they were tested at a temperature of 23±2°C in accordance with that recommended by NCh1017 (33).

Hardened concrete was tested for compressive strength and flexural strength. Table 5 provides a summary of the tests, ages, and standards. The tests were replicated three times for the samples at each of the different curing times and different RAP contents: a total of 144 specimens tested.

The compressive strength of 150 mm cubes was measured according to NCh 1037 (34), by testing the specimens at 7, 14, and 28 days of cure at 20±2°C and applying a load rate of 0.35 N/m²/s. This research involved evaluating flexural strength using three different specimens: standard beam, modified beam, and Semi-Circular Beam (SCB). The standard specimen for flexural strength testing is normally specified as a 150 x 150 x 550 mm beam. In this case, the flexural strength was measured according to NCh 1038 (35), by applying a third point load at a load application rate of 0.016 N/m²/s. It was possible to obtain the so-called modified beam specimens from the standard beam, once tested, using the two halves resulting from the breakage. The 75 x 75 x 120 mm specimens were cut and tested in accordance with NCh 1038 (35), by applying a center point load. The semi-Circular Beams were obtained from the cutting of sliced concrete cylinders, which were cut in half, obtaining specimens that were 100 mm in diameter, 50 mm thick and 50 mm high. Flexural strength for SCB specimens was measured under EN 12697-44 (36), as recommended for asphalt mixtures. The loading application rate for the SCB specimens was 5 mm/min. Figure 1 shows the different specimens used and the configuration of each of the tests.

Figure 2 presents the results of compressive strength and flexural strength, with the relationship between them for the different curing periods of concrete mixes. Error bars indicate standard errors for 95% confidence. The results show a systematic reduction in the compressive strength of concrete mixtures made with RAP, compared to the control mix; this is consistent with the literature (13, 17, 37-39).

The reduction in strength could be due to the asphalt layer around the RAP particles being softer than the concrete and aggregate matrix. The presence of a soft binder can induce stress concentration and cause micro-cracking within the concrete matrix. Another possible reason could be the weak link between the asphalt film and the concrete matrix (17). Moreover, compressive strength increases with a longer cure time for the specimens. However, this increase gets smaller as the RAP content in the mixtures increases. Also, the incorporation of RAP into the mixture reduced the flexural strength of the concrete, which is consistent with the literature (13, 18, 21). The flexural strength reduction pattern presented the same trend as that of compressive strength. However, the rate of strength reduction in RAP mixtures was significantly lower than for compressive strength in all test modes. It was also shown that there was a systematic increase in flexural strength with specimen age, for all mixtures. Figure 2 also illustrates the compressive strength to flexural strength ratio. As the RAP content increased, there was a decrease in the strength ratio, which is due to the higher rate of reduction in the concrete’s compressive strength, compared to the reduction in its bending strength. These values tend to get closer in all three test specimens, with increasing RAP in the mixtures. For the control mix, the average values of the strength ratio for 7, 14 and 28 days’ curing of the beam, SCB and modified beam tests were 10.7, 7.4 and 4.0, respectively. This indicates that the relationship depends on the shape of the specimen and RAP content in the mixture.

Figure 3 shows the compressive and flexural strength reductions in mixtures R20, R50, and R100 with respect to the standard mixture, R0. It was observed that the reductions in the studied mixtures’ compressive strength at different testing ages were of the same order of magnitude, obtaining an average of 42.6%, 51.4%, and 69.1% for the mixtures R20, R50, and R100, respectively. This indicates that, as the amount of RAP in the mixture increased, the loss of compressive strength also increased, compared to the standard mixture. Compressive strength decreased sharply when using a 20% RAP replacement rate. Its reduction was higher than 10 MPa. As can be seen, for the same mixture at different test ages, the reduction percentage is relatively constant, which suggests that the RAP and cement matrix’s chemical bonds do not improve over time. This coincides with the findings reported by (13). A reduction in flexural strength after 28 days of curing can also be seen for all three specimens. As RAP increases, there is a higher flexural strength reduction compared to the standard mixture (0% RAP). This reduction is minor compared to that presented by the compressive strength tests, indicating that the flexural strength property is less affected by the incorporation of RAP aggregates in the mix. It is also possible to observe that the flexural strength depends on the specimen used and the test mode in the laboratory. However, a similarity was found in the flexural strength reduction values for the beam and SCB specimens.

A regression analysis of the relationship between cubic compressive strength and flexural strength was performed for three different specimens. All curing periods were taken into account for this analysis. Linear, logarithmic, and power regression models were analyzed to evaluate the prediction equations that best fit the experimental data. From the regression analysis, the equations presented in Table 6 were obtained.

The goodness of fit and the performance of the models were evaluated using statistical procedures. The coefficient of determination (R ²), Pearson correlation coefficient (r), Sum Square Error (SSE), Root Mean Square Error (RMSE), and the Mean Magnitude of the Relative Error (MMRE) of predicted flexural strength to experimental flexural strength values were calculated. Some of these parameters were estimated using Equations [1] to [3]. The lower the SSE and RMSE values, the better the model is at predicting them; values closer to zero indicate a better fit. Furthermore, a model can be considered accurate enough for most purposes when the value of the MMRE parameter is less than 0.25 (40).

where x _i is the experimental value observed for data i, x ^* _i is the estimated value for each data i and n is the total number of data.

From Table 7, it can be seen that all r values are greater than 0.8, which shows a high association strength, indicating that changes in the predictors are related to changes in the response variable and that the obtained prediction models explain much of the variability of the response. It can also be seen that all MMRE values are less than 0.25 for all prediction models, which is considered to be sufficiently accurate, and, therefore, their use is justified. From a general comparison of SSE, RMSE and MMRE, the equations that best fit the experimental data are the logarithmic and power models, considered to be the most suitable for this research. These models allow the estimation of the flexural strength of concrete from its cubic compressive strength, for the range studied in this research. It can also be seen that, for the same value of compressive strength, there is a different value of flexural strength, which indicates that this property depends on the size and shape of the sample and the type of test applied. Figure 4 presents the relationships between the compressive strength of cubes and the flexural strength of concrete using the equations in Table 1 and those obtained in this investigation, described in Table 6. Only power equations are shown for the corresponding comparison. It was observed that the proposed regression equation for the beam specimens coincided with the equation of the ACI-2002, in terms of trend and values. The SCB specimen regression equation shows a similar trend to the equations proposed by IS: 456-200 and Ahmed et al. (4), while the modified beam regression equation shows results much superior to all the equations shown.

Standardized 150 x 150 x 550 mm beams, that can reach a weight of up to 30 kg, have traditionally characterized the flexural strength property of concrete for pavement applications. The flexural strengths were measured from semicircular, smaller beam specimens (75 x 75 x 120 mm) obtained from standardized beams. A possible correlation between the flexural strength values for the three types of specimens was studied. Linear, logarithmic, and power regression models were analyzed to evaluate the prediction equations that best fit the experimental data. From the regression analysis, the equations presented in Table 8 were obtained. The goodness of fit and the performance of the models were evaluated using the same statistical parameters previously applied. Table 9 shows that all the r values are greater than 0.9, which means a very high association strength and that the prediction models obtained explain much of the variability of the response. It can also be observed that all MMRE values are less than 0.25 for all prediction models, being considered sufficiently accurate, and their use is justified. From a general comparison of SSE, RMSE, and MMRE, the equations that best fit the experimental data are the linear and power models for the SCB specimens. The logarithmic model for the modified beam specimen is the most appropriate in this research. These models allow the estimation of the flexural strength of the concrete beam from SCB or Modified Beam specimen tests. It is possible to observe that the values predicted by the linear, logarithmic, and potential models for the SCB and modified beam test specimens are close to those obtained in the laboratory for the flexural strength of a standardized beam. This indicates a relationship between the flexural strength values that is independent of the shape of the specimens tested and that allows the estimation of this property for a standard-size beam, using smaller specimens that are easily manipulated and consume less materials.