Repeatability and Reliability of New Air and Water Permeability Tests for Assessing the Durability of High-Performance Concretes

Six permeation test methods were investigated. Three different variations of air permeability setups and one high-pressure water permeability adaptation were examined to ensure that permeation properties can be determined under consistently dried and saturated conditions, respectively. These were compared with the RILEM gas permeability test (RILEM TC 116-PCD 1999) and water penetration test to BS-EN 12390-8 (BSI 2000c) to assess the reliability of the new test adaptations.

Three air permeability tests were carried out as part of this study. The first test employed a conventional, 50-mm-diameter base ring (designated as AC-A-50), the second test employed a 75-mm-diameter base ring (designated as AC-A-75), while the third test was undertaken using a modified low volume of compressed air (designated as AC-A-LV test) using a 50-mm base ring. Fig. 1 shows the instruments, further details of which are available elsewhere (Yang et al. 2010).

The three air permeability tests have the same working principle; hence, similar testing procedures were applied. A base ring was used to isolate a test area on the surface of the concrete blocks, and the test chamber was pressurized manually. When the pressure reached 0.5 bar (50 kPa), testing commenced automatically, forcing air to escape through the pores in the test specimen. As pressure levels decreased, the pressure was monitored every minute for 15 min. The natural logarithm of air pressure was plotted against time, and the slope of the last 10 data points was reported as an air permeability index (API), measured in $\ln (\mathrm{bar})/\min$ [or could be reported in $\ln (\mathrm{kPa})/\mathrm{s}$].

The test setup of the new water permeability test (AC-W-HP) is shown in Fig. 2. The test method was based on the procedure developed by Montgomery and Adams (1985), further details of which are available elsewhere (Yang et al. 2013). At the beginning of taking measurements, the cylinder supplying water to the test region was filled with water, while a second cylinder was closed by a ball valve. The test head was then clamped onto a given presaturated test specimen, and water was admitted by a syringe through the priming valve. The test system was then pressurized using compressed air. Once the pressure in the test system was slightly above 7 bar (700 kPa), initial pressurization was considered to be over and a volume reading was recorded as the initial value ($t=0\min$). As water passed through the saturated concrete under examination, pressure inside the test head decreased. To maintain the pressure at 7 bar (700 kPa), the pistons were manually advanced and the volume of water recorded every minute. After 120 min, the test was considered to have been completed. The coefficient of water permeability ($\mathrm{m}/\mathrm{s}$) was then determined based on the steady rate of flow and a calibration factor (Montgomery and Adams 1985; Yang et al. 2013).

The coefficient of gas permeability ($K_g$) was determined by following the procedure given by RILEM TC-116 PCD (1999) and using three consistently dried 50-mm-diameter cores for each mixture. Each specimen was placed in the permeability cell and nitrogen gas (${\mathrm{N}}_2$) was applied under 1.2 bar (120 kPa) of pressure. The test pressure was kept at this value and the rate of gas flow measured. The coefficient of gas permeability was calculated based on Darcy’s equation without accounting for any gas slippage effect on the rate of flow (RILEM TC-116 PCD 1999)where $K_g$ = coefficient of gas permeability (${\mathrm{m}}^2$); $\mu$ = dynamic viscosity of ${\mathrm{N}}_2$ (${\mathrm{Ns}/\mathrm{m}}^2$) at 20°C; $L$ = sample thickness (m); $A$ = section area subjected to flow (${\mathrm{m}}^2$); $Q_s$ = volume flow rate of gas (${\mathrm{m}}^3/\mathrm{s}$); and $P_i/P_o$ = inlet/outlet pressure (${\mathrm{N}/\mathrm{m}}^2$).

The coefficient of water permeability ($K_w$) was determined by carrying out the water penetration test according to BS-EN: 12390-8 (BSI 2000c). Three presaturated 100-mm-diameter cores extracted from the slab specimens were tested for each concrete mixture. A constant test pressure of 7.3 bar (730 kPa) was applied for three days at one end of the test specimen. It should be remembered that irregular waterfronts are unavoidable for all water penetration tests. Therefore, a test head with a guard ring (GR) arrangement, the details of which are as in Yang et al. (2013), was used in this study to ensure a unidirectional water penetration. At the end of the test, the specimens were split open, the depth of water penetration at the central region measured, and the permeability coefficient calculated using the following equation (Bamforth 1987):where $d$ = depth of penetration (m) at time $t$ (s); $K_w$ = water permeability coefficient ($\mathrm{m}/\mathrm{s}$); and $H$ = pressure head (m).

Based on experience gained from previous experimental work (Elahi et al. 2010; Basheer et al. 1995), five HPCs and one normal concrete (NC) were tested. Mixture proportions are reported in Table 1. Portland cement (CEM-I) conforming to BS-EN 197-1 (BSI 2000d) and fly ash conforming to BS-EN 450 (BSI 2005a) were used. The ground granulated blast-furnace slag (GGBS) and silica fume used in this research conformed to BS-EN 15167 (BSI 2006) and BS-EN 13263-1 (BSI 2009b), respectively. The superplasticizer was a polycarboxylic acid–based polymer, specifications of which complied with BS-EN 934-2 (BSI 2009a).

The fine aggregate was medium-grade natural sand, and the coarse aggregate was crushed basalt with 10 and 20 mm sizes proportioned in equal mass. The moisture condition of the aggregates was controlled by predrying in an oven at $105(\pm 5)°\mathrm{C}$ for 1 day, followed by cooling to $20(\pm 1)°\mathrm{C}$ for 24 h before using them to manufacture the concrete.

Concrete was manufactured in accordance with BS 1881, part 125 (BSI 2005b), which was followed immediately by slump and air content testing in accordance with BS-EN 12350-2 (BSI 2000a) and BS-EN 12350-7 (BSI 2000b), respectively. Three $230\times 230\times 100\text{-}\mathrm{mm}$ slabs and six 100-mm cubes were manufactured for each mixture. The molds were filled with concrete in two layers, with each layer compacted using a vibrating table until air bubbles stopped appearing on the surface. All test specimens were covered with wet hessian and placed in a constant temperature room at $18(\pm 2)°\mathrm{C}$. After 1 day, the specimens were removed from their mold and cured in a water bath at a constant temperature of $20(\pm 1)°\mathrm{C}$. The cubes were removed after 28 and 56 days and tested for compressive strength according to BS-EN 12390-3 (BSI 2009c). The fresh properties (slump and air content) and compressive strength for all concretes are given in Table 2.

The slab specimens were removed from the water bath after 3 days, wrapped in polythene sheets, and stored at a constant temperature ($20\pm 2°\mathrm{C}$) for 90 days. After this, the sides of the slabs were painted with three coats of an epoxy paint to prevent any moisture from passing through these sides. The slab specimens were then saturated in water in layers by an incremental immersion method (Yang et al. 2013; Russell et al. 2001). When the specimens were saturated, the AC-W-HP test was carried out.

Following this, the slabs were placed in a drying cabinet ($40\pm 1°\mathrm{C}$ and 35%RH) for 35 days to remove free moisture from the near-surface region. The specimens were then removed from the oven and cooled in a constant temperature ($20\pm 1°\mathrm{C}$) environment for 1 day before carrying out the three air permeability tests.

Once these tests were completed, six cores (three 50-mm-diameter and three 100-mm-diameter) were cut from the slab specimens. The 50-mm-diameter cores were dried in an oven at 40°C until they reached a constant weight and then left in a constant temperature room ($20\pm 2°\mathrm{C}$, 50%RH) to cool for 1 day. The RILEM gas permeability test was then carried out. The 100-mm-diameter cores were used for the BS-EN water penetration test. The curved surface of the cores was coated with an epoxy resin paint to ensure unidirectional water penetration during testing. The samples were then saturated by the incremental immersion as described previously (Yang et al. 2013; Russell et al. 2001) and then tested for water penetration in accordance with BS-EN 12390-8 (BSI 2000c). It should be noted that the saturation regime is mainly used to remove the influence of capillary suction on results, and it was found that this method of saturation was not sufficient to fully saturate HPCs.

The repeatability of the proposed air and water permeability test methods was assessed using values of signal-noise ratio (SNR) and discrimination ratio (DR). SNR represents the number of distinct categories that can be reliably distinguished by the measurement system and was determined using the following equation (Montgomery 2009; AIAG 2002):where ${\rho }_C$ denotes the ratio of concrete variability to total variability that is caused by samples and the test method applied, where ${\rho }_C={\sigma }_C^2/{\sigma }_T^2$; ${\sigma }_C^2$ denotes the variance due to concrete; and ${\sigma }_T^2$ denotes the total variance of results.

Another parameter widely used in statistical process control, DR, was employed. DR also represents the number of distinct categories differentiable by a measuring system and was calculated using the following equation (Montgomery 2009):where ${\rho }_C$ = ratio of concrete variability to total variability as defined previously. The measurement variance and the total variance were determined first, from which the concrete variance (${\rho }_C$) was computed by subtracting the measurement variance from the total variance (Montgomery 2009; AIAG 2002).

The reliability of the new test methods was assessed by comparing the results of the new tests with reference values determined using the RILEM gas permeability test (RILEM TC 116-PCD 1999) and the BS-EN water penetration test [BS-EN 12390 (BSI 2000c)]. Linear regression analysis (Graybill and Iyer 1994; Chatterjee and Hadi 2006) was performed to investigate the relative strength of the relationships. As data obtained from the different tests could not easily be compared directly due to their nonhomogeneous variance and different units for expressing results, values were transformed by log-function and normalized as follows:where $Z$ = normalized data; $x_i$ = log-transformed data; $x_{av}$ = average value of the specific method; and $SD$ = standard deviation of the specific method. With no physical meaning, values obtained in this way were used only to reflect relative differences in concrete permeability properties.

Reliability of the proposed water permeability test was investigated by establishing relationships between normalized results obtained ($K_{AW}$) and normalized values of permeability ($K_W$) obtained from the BS-EN water penetration test. As evaluation of $K_{AW}$ is based on flow-net theory, requiring a value of steady-state flow (Yang et al. 2013; Montgomery and Adam 1985), the as-recorded water permeability data for the six concretes are provided in Fig. 5. It can be observed that while the relationships between the volume of water flowing into concrete and time are not linear, the curvature of the plot was comparatively small after 60 min. After this point, all correlation coefficients were close to a value of 1, meaning that volume flow was proportional to time. Against this background, flow rates used to estimate permeability coefficients ($K_{AW}$) were determined using data obtained between 60 and 120 min (Yang et al. 2013).

Table 4 summarizes the subsequent permeability coefficients determined. Average water permeability coefficients ($K_{AW}$) obtained from the AC-W-HP test for all concretes were relatively low, ranging from 3.4 to $61.2\times {10}^{-16}\mathrm{m}/\mathrm{s}$. As expected, the control NC mixture achieved the highest average value ($61.2\times {10}^{-16}\mathrm{m}/\mathrm{s}$). While results from the BS-EN water penetration testing generally provided a similar trend regarding relative performance of the NC and HPC mixtures, a high degree of variance between results was immediately noticeable. Fig. 6 shows representative results obtained for the FA concrete after carrying out the BS-EN water penetration test. Clearly, the wet fronts for the three samples were irregular and corresponding values of coefficients of variance (CoV) exceeded 50%. Similar observations for this test have been reported elsewhere in the literature (Collins et al. 1986; Zhang and Gjłrv 1991; Pocock and Corrans 2007), attributable predominantly to the measuring method. Explanations include the fact that the BS-EN method uses a single value, based on visual interpretation and measurement, to compute permeability coefficients. Comparative CoV values for the AC-W-HP test, which is based on a regression analysis of the data obtained, is around 30%.

Against this background, individual, rather than average, data points for each mixture from the BS-EN test were used in subsequent reliability analysis to ensure that regression analysis was not affected by the different variances of results obtained by the two methods. Normalized $K_W$ is plotted against normalized $K_{AW}$ in Fig. 7, with 95% confidence interval (CI) limits attached. In general, the existence of a strong correlation between the two tests is evident in Fig. 7. This observation is supported by the p-values (shown in Fig. 7), which if less than 0.001 indicate statistical significance (AIAG 2002; Chatterjee and Hadi 2006). It should be noted, however, that this strong relationship deteriorates markedly if the three data points corresponding to the NC mixture are removed. This suggests that the link between the two water-based tests is strongly dependent on the type of concrete assessed.

To confirm the conclusions drawn from the regression analysis, all the hypotheses were subsequently verified by graphic analysis as advised in the literature (e.g., Graybill and Iyer 1994; Chatterjee and Hadi 2006). From the resulting diagnostic plots in Fig. 8, it can be observed that the probability plot reassembles a straight line, meaning that errors are normally distributed. The plot of residuals versus fitted values shows that the residuals are randomly scattered around zero and that no indication of inconsistent variability exists over the data range. Furthermore, there is no evidence to show dependence between residuals and fitted values. As such, the assumptions were considered to be proven and the regression analysis justified.

The strong relationship between the proposed water permeability test and the BS-EN water permeability test is perhaps unexpected, given trends previously published by the U.K. Concrete Society. In its Technical Report 31 (Concrete Society 2008), while only a general trend is shown, a weak correlation is reported between permeability coefficients determined by steady-state water permeability tests and non–steady state tests. No detailed information on the concretes or test conditions is given in the report, but a possible reason for this reported trend may be that considerable variability existed between concrete batches and specimen preconditioning history. Previous test data reported by Montgomery and Adams (1985) have shown coefficients of variation for permeability coefficients to be 30 and 50% for the same concrete batch and different batches of the same concrete, respectively. Equally, in terms of sample moisture conditioning, Hall and Hoff (2002) suggested that this is crucial for the reliability of any transport-related test technique. In contrast, the fact that samples in this investigation were taken from the same concrete batch and exposed to a predetermined saturation regime is proposed as the explanation for the strong correlation between tests observed. This observation indicates that while most permeability tests are based on sound theory, difficulty associated with cross-comparing results exists because of differences in specimens and test conditions.

The next phase of the research focused on assessing the reliability of the proposed test methods by establishing relationships with the RILEM gas permeability test. As shown in Table 4, average gas permeability coefficients obtained from the latter test for all concretes were relatively low, with the control NC mixture returning the highest average value. In comparison, the HPC mixtures achieved values that were on average around three times lower. Results of the three Autoclam air permeability tests showed a very similar trend in terms of relative performance between NC and HPC mixtures.

Reliability of the three air permeability tests was established using relationships between normalized data obtained and corresponding normalized values of permeability ($K_g$) from the RILEM gas permeability test. The results of this analysis are given in Fig. 11, with 95% confidence interval attached. For all three air tests, general positive relationships between normalized API and $K_g$ values are seen, which are further verified by the $p$-values shown in Fig. 11. All of these are significantly lower than 0.05, indicating that the relationship between the independent ($K_g$) and dependent variable (API) may be considered as statistically significant. Apparent from this analysis, however, was a weaker correlation for the AC-A-75 test than for the other two tests, with data points clearly distributing remotely from the regression line. As concrete is a nonhomogeneous material, this trend most likely reflects the larger concrete surface area employed as part of the AC-A-75 test. The conclusion from this finding is that, unless necessary, the test diameter should not be increased beyond 50 mm.

To confirm the conclusion of the regression analysis, the three hypotheses were subsequently verified by graphic analysis, as previously described. Diagnostic plots of the regression analysis for the three air tests are provided in Fig. 12, which highlights no abnormal behavior in the probability plots and the plot of residuals versus fitted values. On this basis, the conclusions drawn from regression analysis appear valid.