Measuring Chloride Concentrations for Various Cement-Based Materials Systems

(ProQuest: ... denotes formulae omitted.)

INTRODUCTION

An important factor affecting the durability and service life of cement-based systems containing metals is the corrosion of these metals. Although the alkaline environment in concrete is reported to form a protective passive layer on the metal system, chloride concentrations above a certain level (chloride threshold value) disrupt this protective layer and cause the metallic material to corrode in the presence of moisture and oxygen.1 Because a significant number of structures are exhibiting corrosion of the embedded metal, many practitioners and researchers commonly need to determine the chloride concentration in these cement-based materials. The standard ASTM method is time consuming and often expensive and requires larger amounts of cementbased materials. A faster, more economical method (compared with the ASTM method) to determine the chloride concentration of cement-based materials that requires smaller quantities and uses readily available equipment would be beneficial for both research and the industry.

RESEARCH SIGNIFICANCE

Chloride concentrations of ground cement-based material samples, collected from various cement-based systems containing metallic materials exposed to chlorides, often need to be analyzed for the assessment of current conditions and for determining the expected service life of these systems. Assessment of the chloride concentration in various parts of the system with sufficient statistical certainty requires the collection and analysis of a large number of samples. A rapid method of chloride concentration analysis that produces well-correlated results more quickly than the standard, time-consuming, and more labor-intensive potentiometric-titration methods can save significant time and money. A rapid method that is robust for various cement-based systems can also allow for more frequent inspections and the analysis of more samples at each inspection for greater statistical certainty. Obtaining samples from smaller depths can also provide more accurate results. More accurate and statistically valid results at lower costs can results in significant benefits to the concrete industry.

BACKGROUND

In cement-based materials, both free and bound chlorides are present; the free chlorides are in the pore solution or loosely bound in the hydrated cement paste products. Although it is generally believed that the free chlorides are the main contributors to the corrosion of metallic materials, chloride threshold values are generally reported as acidsoluble, total chloride concentrations, that is, as both free and bound chlorides.2 The standard test methods used to determine the acid-soluble, total chloride concentration are AASHTO T-2603 and the ASTM C1152.4 Both methods use potentiometric titration to determine the concentration of chlorides that are extracted from ground concrete samples using a nitric acid solution. A standard silver nitrate solution is used as a titrant together with a chloride or silver ion-specific electrode to monitor the potential of the extraction solution, which is then correlated to the chloride concentration. The major difference between the two methods is that the ASTM method requires the use of a 10 g (0.353 oz) sample passing through a No. 20 sieve and the AASTHO method requires the use of a 3.0 g (0.106 oz) sample passing through a No. 50 sieve.

Because both standard methods used to determine the acid-soluble total chlorides concentration are time consuming and labor intensive, research has been performed to develop quicker alternative methods. One such rapid acid-soluble chloride concentration determination method was developed under the Strategic Highway Research Program (SHRP).5 In this method, ground concrete samples are digested in an acetic acid and isopropyl alcohol solution. After the addition of a dilute (3.75 mg/L [5 × 10^sup -4^ oz/gal.]) sodium chloride solution, the potential of the solution is measured using a chloride-specific electrode and multi-meter combination. The addition of a dilute sodium chloride solution is performed to ensure that the ion concentration of the solution is higher than the lowest detectable chloride concentration of the electrode and multi-meter combination. The addition of this dilute solution also reduces the temperature of the digested concrete sample and eliminates the need for correction of the results for temperature. The concentration of the chlorides in the solution is then calculated using the measured potential of the solution and a calibration curve. The calibration curve is obtained through the measurement of potential values of solutions with known chloride concentrations using the electrode and multi-meter combination. The calibration curve shows the linear change in the measured potential values with increasing concentrations of chlorides similar to the Nernst equation (Eq. (1)) that shows the linear change in measured potential with changing chemical composition.

... (1)

where E^sup 0^^sub X,X^sup x+1^^is the standard electrode potential; R is the ideal gas constant; T is the temperature in degrees Kelvin; F is Faraday's constant; x is the number of ions transferred per mole of reaction; and a^sub X^sup x+1^^ is analogous to the reaction quotient.

A study performed by Weyers et al.6 compared the acidsoluble chloride concentrations determined using the rapid method and the ASTM method. In this study, chloride concentrations of laboratory cast concrete samples were evaluated in addition to chloride concentrations of 12 concrete samples that were collected from the field. The study reported that the results were well correlated for both laboratory cast samples and field samples and established regression models to convert rapid test method results to ASTM method results. For the laboratory cast samples, regression models were developed based on 3.0 and 9.0 g (0.106 and 0.317 oz) samples, and the study indicated that the intercept and slope values of the two models were almost the same. The authors concluded that little was gained by increasing the sample size from 3 to 9 g (0.106 to 0.317 oz). All laboratory cast samples were prepared using the same coarse and fine aggregates. Based on the location where the samples were collected, four regression models were established for the field samples with different intercept and slope values. The authors attributed the differences in intercept and slope values to different aggregate types used in the samples from different locations.

Another study performed by Weyers et al.7 examined 80 concrete samples collected from four bridges using the rapid method and the AASHTO method and reported a good correlation. The study, however, also reported that the variance of the results appeared to increase with increasing chloride concentration values.

Khan8 also measured the chloride concentrations of 47 concrete samples using the rapid method and the potentiometric titration method and suggested coefficients for a regression model to estimate the potentiometric titration method results from the rapid method results. Khan8 also suggested that the coefficients of the regression analysis can be refined by testing of concrete samples with different cement contents, aggregate types, and meter-electrode combinations.9 The study presented in this paper measured chloride concentrations of grout, low-strength flowable cementbased mixtures (LSFC) with nonconventional aggregates, and normal-strength concrete using the rapid method and the standard potentiometric titration method. Samples of 1.5 and 3.0 g (0.053 and 0.106 oz) were tested for the grout and normal-strength concrete samples.

In post-tensioned concrete bridge members, the ducts containing the tension strands are grouted to protect the strands against corrosion. Breached ducts and faulty joints at anchorage locations, however, may allow chlorides and water to be transported into the grout and to corrode the strands. Because even a relatively small corrosion pit can lead to the failure of strands that are under large tension loads, correct determination of the chloride concentration of grout adjacent to the strand is critical. Limited work, however, has been performed to determine if this rapid test method is statistically valid for these grouts. In this study, the applicability of the rapid method to the grout samples is investigated.

LSFC mixtures with high amounts of supplementary cementitious materials (SCMs) and recycled, nonconventional aggregates are being used for backfilling of metallic pipes instead of soils. These mixtures commonly have a compressive strength of less than 8.3 MPa (1200 psi) at the age of 28 days. Because of their high flowability in their fresh state, these mixtures surround the pipes completely and provide a good, constant-density support in addition to protection against corrosion. The transport of chlorides from external sources into these LSFC mixtures can cause the corrosion of embedded metallic pipes. Therefore, the testing of these LSFC mixtures for chloride concentrations and the applicability of the rapid method for these mixtures are also important for predicting long-term performance. Similar to the grouts, limited work has been performed on assessing the applicability of the rapid method for the LSFC mixtures.

The rapid method measurements were performed with 1.5 and 3.0 g (0.053 and 0.106 oz) samples. Because the critical chloride threshold is the chloride concentration of the cement-based material immediately adjacent to the metallic material, smaller quantities of ground cement-based material (1.5 g [0.053 oz]) obtained from the close proximity of the corroding material will provide for a more representative sample for determining the critical chloride threshold. This can be better explained using a commonly observed chloride profile curve as shown in Fig. 1. Assuming that the quantity of the ground cement-based materials sample collected is directly correlated to the depth from which grinding is started. The quantity of a ground cement-based material sample collected from a Distance A from the steel interface will be larger compared to the quantity collected from a Distance B. The average chloride concentration obtained from the larger sample (C^sub A^), however, will also be higher than the average chloride concentration obtained from the smaller sample (C^sub B^) that can lead to incorrect chloride concentrations and underestimates of the service life of the cement-based system. Correlation of the results obtained using different methods, different cement-based systems, and different quantities have been examined in this paper.

EXPERIMENTAL DESIGN AND MATERIALS

This study examined three cement-based systems for the applicability of the rapid method: normal-strength concrete, grout, and LSFC mixtures. Two grout mixtures with 3 and 6% by weight sodium chloride concentration in regard to the total mixture weight were prepared using a repair grout for post-tensioned bridges. The mixtures had a water-cementitious material ratio (w/cm) of 0.24 and were prepared by mixing 11.34 kg (25 lb) of grout with 2.72 kg (6 lb) of laboratory tap water that contained 0.42 and 0.84 kg (0.93 and 1.85 lb) of dissolved sodium chloride for the 3 and 6% mixtures, respectively. The repair grout for post-tensioned bridges is a cement-based grout that compensates for shrinkage in both the plastic and hardened states and contains no chlorides or fine aggregates (sand free). Mixtures prepared with the repair grout are usually used for grouting tight clearances, such as within ducts of post-tensioning strands for corrosion protection. A total of 12 cylindrical samples, 75 x 150 mm (3 x 6 in.) in size, were prepared for chloride concentration determination.

Normal-strength concrete mixtures with two different w/cm (0.4 and 0.6) and two different sodium chloride concentrations (3 and 6% by weight in regard to the total mixture weight) were prepared. Table 1 shows the concrete mixture proportions. All mixtures were prepared using ASTM C150 Type I portland cement. Sodium chloride was mixed with the batch water and completely diluted by vigorous stirring prior to placing into the mixer. Six cylindrical samples, 100 x 200 mm (4 x 8 in.) in size, were prepared from each mixture for chloride concentration determination.

Crushed limestone and sand were used as coarse and fine aggregates for the normal-strength concrete. Figures 2 and 3 show the particle size distribution of coarse and fine aggregates determined following ASTM C136.10 The fineness modulus of the fine aggregate was 3.12.

The bulk specific gravity and absorption capacity of the fine aggregate were determined to be 2.61 and 1.50% following ASTM C128.11 The bulk specific gravity and absorption capacity of the coarse aggregate were determined to be 2.63 and 1.07% following ASTM C127.12

In addition to the grout and normal-strength concrete samples, 15 cylinders, 75 x 150 mm (3 x 6 in.) in size, cast using LSFC mixtures, were evaluated for chloride concentration. LSFC mixtures contained low amounts of cement, high amounts of Type F and C fly ash, and nonconventional aggregates, such as foundry sand, bottom ash, and highcarbon fly ash. These 15 samples were randomly selected from a large group of LSFC mixture samples that were exposed to a 3% chloride solution for 26 months. Table 2 shows the mixture proportions of the randomly selected LSFC samples and the number of samples from each mixture. All LSFC samples were prepared using ASTM C150 Type I portland cement. Tables 3 and 4 show the chemical compositions and the physical properties of the materials used in the LSFC samples. The unconfined compressive strength values of the selected samples at 28 days ranged from a minimum of 0.12 MPa (17.40 psi) to a maximum of 1.09 MPa (158.09 psi).

All grout and concrete samples were cured for 28 days following ASTM C192/C192M-02.13 Grout and concrete samples with 3 and 6% by weight sodium chloride concentrations were ground using a rotary grinder. Because the LSFC samples were too weak for grinding, they were cut, and the sample collected from their center was ground using a ceramic mortar and pestle. All ground samples were passed through a No. 20 sieve and their chloride concentrations were determined using the potentiometric titration method and the rapid method. The rapid method was performed using 1.5 and 3.0 g (0.053 and 0.106 oz) samples and the potentiometric titration test was performed using 10 g (0.35 oz) samples following ASTM C114.14 The rapid testing of the LSFC samples was performed using only the 3.0 g (0.106 oz) samples. An ion-selective electrode with a filling solution suitable for measuring chloride solution concentrations up to 355 ppm and a multimeter were used for all testing.

Statistical analysis and results

The correlations between the chloride concentrations measured by the rapid method from the 1.5 and 3.0 g (0.053 and 0.106 oz) normal-strength concrete and grout samples and the chloride concentrations measured with the titration method are shown in Table 5. The table shows that all the results were highly correlated and all the correlations were statistically significant.

A multiple linear regression analysis was performed using the chloride concentrations of the grout and normal-strength concrete mixtures measured with the titration method as the dependent variable. The chloride concentration measured with the rapid method using 3.0 g (0.106 oz) samples and the mixture type were used as independent variables. Grout, normal-strength concrete with a w/cm of 0.4, and normalstrength concrete with a w/cm of 0.6 were the three different values of the mixture type variable. Analysis of variance indicated that although the overall model was significant with a coefficient of determination (R^sup 2^) of 0.98, the mixture type (that is, grout or concrete) and the w/cm of the concrete were not significant factors affecting the relationship between the measurements obtained with the rapid method and the titration method. A similar multiple regression analysis was performed using the chloride concentration of grout and concrete mixtures measured with the rapid method using 1.5 g (0.053 ounce) samples and the mixture type as independent variables. Results were similar, indicating that the overall model was significant with an R^sup 2^ of 0.98 and the mixture type was an insignificant variable.

Based on these results, two simple linear regression analyses were performed between the chloride concentration measurements obtained with the titration method and measurements obtained with the rapid method. The first analysis used the rapid method results obtained with the 3.0 g (0.106 ounce) samples and the second analysis used the rapid method results obtained with the 1.5 g (0.053 ounce) samples as the independent variables. Analysis of variance for both models indicated that the overall models were significant with an R^sup 2^ value of 0.98. However, examination of the residuals indicated that the assumption of residuals with constant variance independent of the measured chloride concentration values was not satisfied. Figures 4 and 5 show the residuals of the simple regression models plotted against the chloride concentrations determined with the rapid method using 3.0 and 1.5 g (0.106 and 0.053 oz) samples. The lines shown on the figures are visual aids to show the increase in the spread of residuals with the increase in chloride concentration. Weyers et al.6 also observed the increase of variance when the results of titration method were compared with the results obtained from the AASHTO method in their study.

If the variance of the residuals is directly proportional to the chloride concentration of the samples, then fitting a transformed model, as shown in Eq. (2), would be more appropriate than trying to fit a simple linear regression model between the measurements obtained with the titration method and the measurements obtained with the rapid method.

... (2)

where [Cl^sup -^]^sub titration^ is the chloride concentration measured with titration method; [Cl^sup -^]^sub rapid^ is the chloride concentration measured with the rapid method; and a, b are the regression coefficients.

If the direct proportionality assumption is correct, the residuals of the transformed model would be randomly distributed and the relation between the chloride concentration measurements using the two different methods could be described as shown in Eq. (3) using the same coefficients a and b.

[C^sup -^]^sub titration^ = b + a[Cl^sup -^]^sub rapid^ (3)

Figures 6 and 7 show the residuals of the transformed models using the rapid chloride concentration measurements with the 3.0 and 1.5 g (0.106 and 0.053 oz) samples, respectively. Clearly, the residuals are randomly distributed with a constant variance satisfying the assumption of the linear regression analyses. Therefore, the relation between the measurements of the two methods can be shown using the coefficient of the transformed models as shown in Eq. (4) and (5) for the 3.0 and 1.5 g (0.106 and 0.053 oz) samples, respectively.

[C^sup -^]^sub titration^ = 0.03 + 1.00 × [Cl^sup -^]^sub rapid, 3 g^ (4)

where [Cl^sup -^]^sub rapid, 3 g^ is the rapid method result using 3.0 g (0.106 oz) samples, and

[Cl^sup -^]^sub titration^ = 0.10 + 0.98 × [C^sup -^]^sub rapid, 1.5 g^ (5)

where [C^sup -^]^sub rapid, 1.5 g^ is the rapid method result using 1.5 g (0.053 oz) samples.

Tables 6 and 7 show the analysis of variance tables for the two models shown in Eq. (4) and (5), respectively. These tables show the results with the statistical test of hypothesis that the intercept and slope values are not different than zero. If the probability value shown in the last column of Tables 6 and 7 is smaller than 0.05, this indicates that for a 95% probability, the hypothesis that the intercept or the slope value is not different from zero cannot be rejected, that is, the value is statistically not significant. Results in Table 6 indicate that the model established for the 3.0 g (0.106 oz) samples had a statistically significant slope of unity and a statistically insignificant intercept. Results in Table 7 indicate the model established for the 1.5 g (0.053 oz) samples had statistically significant intercept and slope values. The slope value of the model for the 1.5 g (0.053 oz) samples is also very close to unity. These slope values indicate that, for both models, the results of both methods increased the same amount for the same amount of increase of the chloride concentration. For the results obtained using 1.5 g (0.053 oz) samples, the addition of a very small intercept is necessary to convert the rapid method results to the ASTM results. The R^sup 2^ values of both models are 0.98, that is, 98% of the variation of the chloride concentration values measured with the titration method can be explained with the chloride concentration values obtained with the rapid method. The mean square error values for both models are approximately 0.08. Figures 8 and 9 show the fitted regression models and the 95% confidence intervals on the mean response at each data point for the fitted models. Investigation of Cook's distance measure for both models also indicate that there were no influential observations, that is, taking out any of the data points from the data would not affect the model significantly.

Figure 10 shows the box plots of chloride concentrations measured with the titration method and the rapid method using 1.5 and 3.0 g (0.053 and 0.106 oz) samples from the grout and normal-strength concrete samples. For 3 and 6% sodium chloride concentrations, the total expected chloride concentrations to be measured are 1.82 and 3.64%, respectively (shown in the figure). The box plots show that the mean results obtained from both methods were below the expected percentages. This could be a result of the acids used for both methods (nitric acid and acetic acid)-the ASTM standard and the rapid method-were not able to free all the available chlorides.

Because the intercept and slope values obtained for the 1.5 and 3 g (0.053 and 0.106 oz) samples were very similar, the data obtained from both sample sizes was combined and a new multiple linear regression analysis was performed using the sample size as a classification variable. The analysis of variance indicated that the sample size was not a significant factor affecting the relation between the readings of the two methods. Therefore, a simple linear regression analysis was performed using the combined data from both sample sizes between the results of the titration method and the rapid method. A similar procedure used in the earlier models was used to determine the coefficients of the model and eliminate the increase of the spread of the residuals with increasing chloride concentrations. Equation (6) shows the model developed for the combined data. Table 8 shows the analysis of variance results for the model, and the R^sup 2^ value of the model is 0.98.

[Cl-]titration = 0.07 + 0.99 × [Cl-]rapid (6)

Figure 11 shows the fitted regression model and the 95% confidence interval on the mean response at each data point for the fitted model.

Figure 12 shows the fit of the developed model in Eq. (6) to the data provided earlier in the literature by Weyers et al.6 and by Khan.8 The data from Weyers et al.,6 shown in Fig. 12, were collected only from laboratory cast samples. The rapid method data obtained from the study performed by Khan8 was corrected for a small error made in the calculation of the data that was later reported by Khan.9 The R^sup 2^ value of the model shown in Eq. (6) for the combined data in Fig. 12 is 0.99. The figure also shows the linear model with typical coefficient values (0.8984 intersection and 1.165 slope) proposed by Weyers et al.6 that was included in the AASHTO T260-94. It should be noted that the proposed values were for chloride concentrations expressed in lb/yd3 units, and to show the model on Fig. 12 in percent chloride units, the proposed intersection value was divided by 39.15 (typical weight in lb for yd3 of concrete used by Weyers et al.6). The slope will not change with unit conversion. The figure also shows the model developed using the data reported by Khan after the correction of the rapid method data.8 Figure 12 shows that the model developed in this study fits to the combined data well and almost all of the data points fall into the 95% prediction interval of the model. The models proposed by Weyers et al.6 and by Khan8 seem to overestimate and underestimate the percent chloride values obtained by the titration method at high chloride concentrations, respectively.

A simple linear regression analysis performed between the chloride concentrations of the LSFC samples measured with the rapid method and the titration method indicated that the model was statistically significant with an R^sup 2^ value of 0.95. The rapid chloride concentration measurement method was used for evaluating 3.0 g (0.106 oz) samples. Table 8 shows the parameter estimates and their significance obtained through regression analysis between the titration method and rapid method results. The intercept was not statistically significantly different from zero and the slope was statistically significantly different and lower than unity. This indicates that the model passes through the origin and the increase in the results obtained by the titration method was smaller compared to the increase in the results obtained by the rapid method for the same amount of chloride increase, whereas the increase in the results of both methods were almost equal for the concrete and grout samples (slope close to 1).

Figure 13 shows the fitted regression model and the 95% confidence interval on the mean response at each data point for the fitted model. Investigation of the residuals indicated that they were independent from the chloride concentration values. Although Fig. 13 shows that one of the data points had a high leverage, that is, was away from the rest of the data on the x-axis, examination of the Cook's distance measures indicated that this point was not an influential data point, that is, removal of this point would not change the coefficients of the model considerably. As noted previously, the LSFC samples were randomly selected from a larger group of samples. Although one of the samples exhibited a higher chloride concentration compared to the rest of the samples, the difference between the results obtained using the two different methods was not significantly different from the difference observed for the rest of the samples.

Figure 14 shows the box plots of the chloride concentrations measured using the rapid method and the titration method from the LSFC samples. The more distinct difference between the measured chloride concentration values of LSFC samples compared to the concrete and grout samples using the two methods is likely due to the inherent differences of these LSFC mixtures. In an earlier study, Clemeña and Apusen15 suggested that the use of solutions with known chloride concentrations for electrode calibration purposes in the rapid method could lead to erroneous results if the tested solution contains other ions that can affect the milivolt readings. Additional ions coming from the high amounts of supplementary cementitious materials used in the LSFC mixtures may be affecting the calibration of the electrode in the rapid chloride measurements.

It is an interesting observation that although the titration method uses a much stronger acid (nitric acid) compared with the rapid method, which uses a weak acid (acetic acid), the chloride concentration values obtained from the LSFC samples using the rapid method were higher compared to the values obtained using the titration method. This is contrary to the case of the grout and normal-strength concrete samples.

SUMMARY AND CONCLUSIONS

A fast, reliable method to determine the chloride concentration in cement-based materials can save significant time and money. Engineers and researchers responsible for testing and monitoring cement-based materials for corrosion performance can economically determine the chloride concentration of a larger number of samples in a relatively shorter time if correlations between the faster and slower methods can be determined. The rapid chloride concentration determination method developed under the SHRP is a good candidate for such a quick and reliable method for various cement-based material systems.

The results obtained from this study indicate that for grout and normal-strength concrete samples, the measurements obtained from the titration method and the rapid method with 1.5 and 3.0 g (0.053 and 0.106 oz) samples are highly correlated. Results indicated that using a grout mixture or a regular concrete mixture with different water-cement ratios did not significantly affect the results. Developed linear regression models had intercept and slope values close to zero and one, respectively. This indicates that, for concretes and grouts produced from similar materials, the rapid method can be used as an economical alternative to the titration method to reliably measure chloride concentrations in much shorter time periods. The model developed in this study and shown in Eq. (6) is a good fit for the data provided in the literature.

The statistically insignificant effect of sample size (1.5 versus 3 g [0.053 versus 0.106 oz]) justifies the use of a sample as small as 1.5 g (0.053 oz) in cases when collection of larger samples is not practical or when it is desired to determine the chloride concentration in a thin layer of cement-based material with greater accuracy, for example, in the close proximity of metallic material. This can provide better estimates of the critical chloride threshold level. It should be noted that care must be taken that a large portion of the sample is not from ground aggregate.

Results of LSFC samples also show that there was a good correlation with a very high R^sup 2^ value between the measurement methods. The slope of the fitted model, however, was significantly different and lower than 1, unlike the grout and normal-strength concrete samples. The authors believe that the chloride concentration readings obtained from the rapid method and the titration method for the LSFC mixtures are different due to the inherent differences in these mixtures, such as low cement content, high amounts of SCMs, the use of nonconventional aggregates, and type of introduction of chlorides. The conversion factor for the specific samples tested in this study was 0.707. Because LSFC mixtures vary widely in the type and amount of their ingredients for different required fresh and hardened characteristics, however, testing a sufficiently large number of samples using both methods to establish a regression model for conversion is recommended.

Results from this research clearly indicate that the SHRP method for assessing the chloride concentration is applicable for various cement-based systems. Future research should examine modifying the calibration procedure of the rapid method to take into account the ions coming from unconventional aggregates, as suggested in the literature, to decrease the difference between the results of the two methods.

ACKNOWLEDGMENTS

The authors would like to thank R. Dotson for his assistance with sample preparation and testing.