- Email: [email protected]
Image analysis for determination of cement content in concrete to improve accuracy of chloride analyses
Cement and Concrete Research 99 (2017) 1–7
Contents lists available at ScienceDirect
Cement and Concrete Research journal homepage: www.elsevier.com/locate/cemconres
Image analysis for determination of cement content in concrete to improve accuracy of chloride analyses
MARK
Carolina Boschmann Käthlera,⁎, Ueli M. Angsta,b, Matthias Wagnerc, Bernhard Elsenera,d a
ETH Zurich, Institute for Building Materials (IfB), ETH Hönggerberg, CH-8093 Zurich, Switzerland Swiss Society for Corrosion Protection, Technoparkstrasse 1, CH-8005 Zurich, Switzerland c Tecnotest AG, Alemannenweg 4, CH-8803 Rüschlikon, Switzerland d University of Cagliari, Department of Chemical and Geological Science, I-09100 Monserrato, CA, Italy b
A R T I C L E I N F O
A B S T R A C T
Keywords: Image analysis Durability Aggregate Cement paste Chloride
The chloride content Ccl expressed as %Cl by weight of cement is important in condition assessment of reinforced concrete structures. Whereas standardized procedures determine Ccl in concrete powder, the cement content Cm is generally assumed equal to the mix design or an experience-based constant value. This work shows in concrete with maximal aggregate diameter 32 mm, Cm exhibits significant variability in 50 mm diameter cores, because the specimens are too small to be representative of bulk concrete. In such specimens, Cm might differ from the bulk cement content by a factor of up to 2. Thus, a reliable determination of Ccl in terms of %Cl by weight of cement requires the analysis of both Ccl and Cm in a concrete specimen. A procedure based on colouring the cement paste, scanning the specimen surface, and image analysis allows the practically non-destructive determination of Cm with good accuracy.
1. Introduction Chloride ions are known to impair the service life of reinforced concrete structures, as they can cause chloride-induced reinforcement corrosion. In marine environments or when using de-icing salts chloride-induced corrosion is the most common deterioration mechanism for reinforced concrete structures [1]. Therefore, measuring the chloride concentration in concrete (CCl) is common in condition assessment of existing reinforced concrete structures. Comparing the measured CCl with the so-called critical chloride content (Ccrit), i.e. the chloride threshold for corrosion initiation [2], is the widely accepted procedure to assess the risk of chloride-induced corrosion. Standardized methods [3–6] for the determination of CCl include three principal steps: 1) taking a sample, 2) extraction of the chloride ions, and 3) analysis of the chloride concentration. The result is a value of the chloride concentration referred to the dry mass of concrete. Each of the steps 1–3 are performed using different procedures, thus the size of the sample, the way of extraction and the analytical method can lead to errors in the final result. Several round robin tests [7,8] in the past have shown that on a homogeneous, reasonably fine concrete powder the total chloride content CCl can be determined with good accuracy. In accredited laboratories neither the extraction method nor the method of analytical chemistry used leads to significant errors.
⁎
Corresponding author. E-mail address: [email protected] (C. Boschmann Käthler).
http://dx.doi.org/10.1016/j.cemconres.2017.04.007 Received 2 November 2016; Received in revised form 13 April 2017; Accepted 18 April 2017 0008-8846/ © 2017 Elsevier Ltd. All rights reserved.
In engineering practice and in durability standards it is often preferred to relate CCl to the mass of cement. This representation is considered the best way to include both the aggressivity of the chloride ions and the corrosion inhibitive properties of the cement matrix at the steel reinforcement [2]. Usually, however, the cement content in the concrete is not known and may even vary between different parts in a structure. Thus, an assumption has to be made and commonly 300 kg cement per m3 concrete as a bulk cement content is considered adequate. In reality, concrete is not homogeneous, but a composite material consisting of three phases: cement paste, aggregates, and air voids. The aggregate volume fraction of bulk concrete typically is 60–80% of the concrete. Single coarse aggregate particles are likely of diameter ~ 30 mm, which thus have a volume of 14 cm3 (spherical shape assumed). Concrete specimens taken from structures for measuring CCl are typically drilling cores of diameter (ddc) in the range of 30 to 50 mm. These cores are typically cut or ground into slices of 5 to 10 mm thickness, which corresponds to specimen volumes of 3 to 10 cm3 – a volume comparable to a single coarse aggregate particle. It is thus expected that the volume fraction of cement paste can vary strongly from one concrete specimen to another, depending on the presence or absence of coarse aggregate particles. As is shown in Fig. 1, the actual cement content of a concrete specimen (Cspecimen) can differ from the
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
Sampling (e.g. core drilling / slicing)
measuring cement content in specimen
Cref
deviation of sample
cement content in bulk cement
Cspecimen
error of method: ∆C & ∂C
actual cement content in specimen
Cm
measured cement content in specimen
Fig. 1. Definition of error and deviation: Cref is equal to the cement content according to the mix proportion (bulk concrete), whereas Cspecimen is equal to the actual cement content within a specific concrete specimen. The measured CCl within this specimen needs to be referred to Cspecimen. Cm is the cement content on the specimen surface, measured with the suggested procedure including image analysis. All abbreviations are listed in Table 3.
distinguished (aggregates, cement paste, air voids). Thus, some methods [21–23] include etching a layer of cement paste followed by colouring the cement paste and aggregates with gypsum and black paint respectively to enhance contrast between both features. This method is laborious and therefore not convenient for practice. Ozen and Guler [20] simply used a desktop flatbed scanner for image acquisition. However, it is worth mentioning that the cement paste used by Ozen and Guler [20] is of whitish colour and therefore comparatively straightforward to differentiate from greyish aggregates without further image processing. The problem for implementation of this procedure in practice is that cement paste is itself often greyish. To account for this, Hammer [19] describes a method of staining the cement paste deep purple with Alizarinred S, without staining the aggregates [24]. After image processing, the optimal threshold is readily defined for determination of cement content in hardened concrete. In this paper, we describe and propose a method for the reliable and almost non-destructive determination of the cement content in specimens of hardened concrete – containing partly limestone aggregates in the concrete. The method includes colouring the concrete specimen, image acquisition by a flatbed scanner and image analysis. The method will be assessed in terms of applicability in research and practice; furthermore, it is applied to concrete specimens to quantify the variability in cement content in specimens of hardened concrete; the
cement content of the bulk concrete (Cref). Furthermore, all methods to determine the cement content of a concrete specimen exhibit an error due to limited measurement precision. As depicted in Fig. 1, the difference between Cspecimen and the measured cement content, Cm, is termed “error of the method” (ΔC, δC). Fig. 2 illustrates an example of two drilling cores taken from the same concrete. There are clear differences in cement paste volume fraction. To anticipate here the results of this work, the cement content in Fig. 2A is 17.5 M-% and in Fig. 2B is 5.5 M-%, thus differing by a factor > 3. When expressing the chloride concentration in concrete as CCl by weight of cement we believe that it is crucial to determine, in addition to the chloride concentration, the cement content in each specimen. Ideally, methods to measure the cement content in hardened concrete should be reliable, quick, almost non-destructive, and applicable to a wide variety of concretes. There exists a number of methods [9–11]; some are implemented in codes [10,11]. Some methods are based on determining the mass of the filter cake after acid digestion of the concrete specimen to quantify the aggregate mass fraction [9,12,13]. However, these are only applicable to lime-free aggregates [14]. Other methods use image analysis, either of electron microscope pictures [15] or of macro scale pictures [16–21]. Image analysis benefits from the contrast in brightness or colour of the features to be
A
B
Fig. 2. Two drilling core slices (ddc: 50 mm) from the same concrete mix dmax 32. Section A shows few coarse aggregates and has a higher cement content (17.5 M-%). Section B has more coarse aggregates and has therefore a lower cement content (5.5 M-%).
2
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
2.2.3. Conversion In this work, area fractions of the phases (aggregates, cement paste, voids) were considered equal to the respective volume fractions. This can be justified by the fact that the thickness of the slices was much smaller than their lateral dimensions (analysed surface > slice thickness). The volume fraction of cement paste was converted to mass percentage with the assumption for densities of cement paste ρcp and aggregates ρagg. The density ρagg was assumed to be 2.6 g/cm3 whereas the density of cement paste ρcp was calculated with Powers' model [27]:
Table 1 Nominal concrete mix proportions. Ingredients [kg/m3]
Mixture 1 “dmax 32”
Mixture 2 “dmax 16”
Portland cement Water (wc-ratio: 0.5) Aggregate (0–4 mm) Aggregate (4–8 mm) Aggregate (8–16 mm) Aggregate (16–32 mm) Cref [M-%] Curing
330 330 165 165 473 756 378 567 378 567 662 0 13.83 13.83 4 years wrapped in plastic sheeting at 20 °C (climate chamber)
ρcp =
1 + 0.23 × α 0.317 + w c
(1)
In the present work, the w/c ratio was 0.5 and the degree of hydration α was assumed to be 0.95. Therefore, ρcp = 1.49 g/cm3. The mass of cement paste in a specimen mcp was converted to mass of cement mc with the following formula (according to Powers' model [27]):
results and implications for chloride analysis are discussed. 2. Experimental methods
mc =
2.1. Materials
mcp (0.23 × α) + 1
(2)
The calculated mass percentage of cement is equal to Cm.
Concrete cubes (150 × 150 × 150 mm3) with two different mixtures were cast according to the mix proportion given in Table 1. The aggregates were gravel, containing a variable amount of calcium carbonate (of the order of 20–30%). For curing and storage, the cubes were wrapped in plastic and stored in ambient temperature 20 °C, i.e. left for self-desiccation. At the time of testing, the concrete was four years old.
2.2.4. Experiments 2.2.4.1. Assessment of suggested procedure. In the first step, for each tested concrete type, two slices of ca. 15 cm × 15 cm × 2 cm were analysed to determine the error of the method. To rule out any wall effects, all sides of the analysed surface area were reduced by 1 cm in length. The analysed surfaces had therefore an area of 160 cm2, which is 20 times larger than the surface of a 32 mm aggregate (assumption: circular shape cut at centre). The analysed surface is therefore assumed to be representative for the bulk concrete. This means that Cspecimen equals Cref (Fig. 1). The difference between Cm and Cspecimen corresponds to the error of the image analysis method. This error was expressed in absolute and relative values, ΔC and δC, respectively, which are defined in Eqs. (3) and (4).
2.2. Procedure 2.2.1. Sampling The concrete specimens were cut with a water-cooled diamond saw into slices of approximately 20 mm thickness. The slices were polished with a water-cooled diamond grinder and washed under running tap water. Subsequently, surface water was immediately removed with compressed air. Each slice was then placed in a freshly made solution of Alizarinred S (0.1 g to 100 ml H2O) for 3 min. This leads to a colouring of the cement paste (deep purple) of the slice surface exposed to the solution, because a precipitate is formed [19,24,25]. After colouring, image acquisition was conducted with a flatbed scanner. A resolution of 300 dpi was found to be a good compromise between time for image processing and accurate image resolution. This yields a pixel size of 85 μm, which determines the minimum particle size that may be identified in image analysis.
ΔC = Cm − Cref
(3)
ΔC Cref
(4)
δC =
2.2.4.2. Quantification of variability in cement content in concrete specimens. In the second step, only small fragments of the entire picture were subjected to image analysis. The four images of the representative specimen surfaces were cut with a grid into specimen surfaces of sizes 6 cm2, 18 cm2, and 40 cm2. The sizes 6 cm2 and 18 cm2 are comparable to practical specimen dimensions (drilling core diameter (ddc) 30 and 50 mm, respectively). A high number of cases, i.e. 50 cases for specimen size 6 cm2, 18 cases for size 18 cm2, and 8 cases for 40 cm2, were studied to individually assess the deviations of the cement content and the influence of parameters such as maximal aggregate diameter (dmax) and ddc.
2.2.2. Image processing The image processing and analysis was conducted in the software “Fiji” (“ImageJ”) and the corresponding plugin “trainable WEKA segmentation”. The standard settings were chosen as training features [26]. In “trainable WEKA segmentation”, three classes were defined, one for each phase: aggregate, cement paste, and air voids. For each class, representative areas were first manually selected on an image (two or more per class). The areas included various aggregates (different colours, textures, and sizes) and cement paste (lighter and darker areas). After the first training of the classifier, the classifier was improved with further area selections until the result of the overlay looked precise at visual check-up. From the final result (a classified image) the area fraction of aggregates and cement paste was computed (as the number of pixels in each class). Fig. 3 depicts an example of the original and segmented picture of a representative specimen surface. Since the judgement of the classified image depends on the operator performing the image analysis (the person training the classifier), we let 3 different people individually and independently conduct the image analysis for one selected image to assess the uncertainty related to the personnel.
3. Results and discussion 3.1. Feasibility of the suggested procedure Fig. 4 and Table 2 show the accuracy of the method in detecting the actual cement content in a concrete specimen, i.e. the difference between the measured (Cm) and known (Cref) cement content. The data is based on the analysis of all 160 cm2 large surfaces, thus assumed as representative of the bulk concrete. These results demonstrate that the proposed method for the determination of the cement content in a concrete specimen delivers accurate results. The relative error δC is always < 2% (Table 2), i.e. negligibly low. When applied to specimens taken from concrete structures in 3
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
A
B
Fig. 3. The left image A shows the original, scanned image with the stained cement paste; the right image B is the classified image after image analysis (aggregates are shown in red, cement paste in green, and air voids in purple). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
practice, the method may exhibit a larger uncertainty due to a number of factors. For the conversion from area percentage of cement paste to mass percentage of cement, assumptions are necessary (Eqs. (1) and (2)), i.e. for the density of the aggregates, the degree of hydration, and the w/c ratio. While in our laboratory tests, these were largely known; accurate information is often scarce for concrete taken from engineering structures. To give a first-hand estimate of the influence of such uncertainties, the conversion was also done for an extreme deviation in w/c ratio, namely for w/c = 0.4 instead of 0.5, and for a degree of hydration α = 1 instead of 0.95. With such a combination of wrong assumptions, the maximal absolute deviation of the cement content ΔC increased from 0.3 M-% to 1.9 M-%, the maximal relative deviation of the cement content δC from 2.0% to 13.4%. Similarly, also the assumption for the density of the aggregates was varied (from 2.3 to 2.9 g/cm3). The maximal absolute deviation of the cement content ΔC increased from 0.3 M-% to 1.5 M-%, the maximal relative deviation of the cement content δC from 2.0% to 10.6%. This uncertainty of approx. 10–15% is still by far better than the common way of assuming a unique cement content such as 300 kg/m3, which may affect the result of chloride analyses by a factor roughly 2 (see next section). A further influence may be related to the person carrying out the image analysis. This is to some extent subjective because the classifier for image analysis needs to be trained and the result is assessed by the person by means of visual judgement of the classified image. A trial with 3 test persons, independently carrying out the analysis for the
13.7
Table 2 Measured Cm at sample surfaces representative of bulk concrete (Cref 13.8 M-%). Specimen
Cm [M-%]
ΔC [M-%]
δC [%]
dmax 32 – 1 dmax 32 – 2 dmax 16 – 1 dmax 16 – 2 Average dmax 32 Average dmax 16
13.7 14.1 13.9 13.8 13.9 13.85
− 0.14 0.27 0.07 − 0.03 0.07 0.02
− 0.98 1.95 0.48 − 0.18 0.49 0.15
same image, revealed that ΔC was always smaller than 0.5 M-% and δC was always smaller than 3.5%. Thus, the procedure is relatively independent of personnel. Summarizing, compared to the variability arising from specimen dimensions with respect to aggregate diameter (see next section), the uncertainties related to assumptions inherent to the method and to personnel can be considered negligible. Thus, the suggested procedure is feasible to accurately determine the cement content in small specimens of hardened concrete, even with aggregates partially containing limestone (20–30%). 3.2. Influence of specimen size and coarse aggregate diameter dmax Fig. 5 shows all measured cement contents, Cm, as a function of the
14.1
13.9
13.8
Cref
Fig. 4. Suggested procedure applied to sufficiently large specimens, i.e. specimens representative of bulk concrete. Two analyses per concrete mix. The results show negligible deviations from the reference cement content (Cref = 13.8%).
4
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
specimen surface area. Depending on investigated surface area, up to 50 individual values were obtained per dmax. To give insight into the statistical distribution of the data, probability density functions were fitted for the obtained Cm at each investigated surface area and for each dmax. The type of distribution was assumed to be normal; standard deviation σ and mean μ were fitted with the method of the least squares. The results are shown in Fig. 6. Both, from Figs. 5 and 6, it is apparent that the variability in Cm decreased as expected with increasing specimen surface area. Furthermore, for a given surface area, the variability in Cm was larger for concrete with dmax = 32 mm than for dmax = 16 mm. These results confirm the expected influence of coarse aggregates. The two vertical lines in Fig. 5 correspond to specimen sizes comparable to ddc 30 and 50 mm, respectively. Clearly, concrete disks cut from cores of common diameters will exhibit significant variability in terms of cement volume fraction in each specimen. The actual cement content may be both lower and higher than in the bulk concrete, namely roughly by a factor of 2 (for dmax = 32 mm and ddc = 50 mm). Cm ranged from ca. 8 M-% to 25 M-%, while Cref was 13.8 M-%. Thus, concrete cores of diameters 30–50 mm drilled from concrete with dmax = 32 mm are not large enough to yield specimens that are – in terms of phases present – representative of the bulk concrete. Another strategy (alternative to image analysis) to reduce the deviation of cement content in a specimen may be increasing the specimen size (area). However, to reduce the deviation to an acceptable level, a marked increase of ddc is needed. As can be seen in Fig. 5, even an area of 41 cm2 (approximately corresponding to ddc = 70 mm) provides cement contents from 10 to 20 M-%, thus deviating from the reference value (13.8 M-%) by a factor of ± 50%. Additionally, increasing the sample size until it becomes representative for the bulk concrete is not feasible for practice, because condition assessment should be as less destructive as possible. Fig. 5 may give some guidance also for the drilling method of sampling sometimes used: The surface area needed to reduce the uncertainty in measured chloride content to below ± 50% is ca. 60 mm2 (for dmax 32 mm). With a ddc of 20 mm this would correspond to ca. 19 drilling holes, which is considered a high number.
Table 3 Abbreviations. Abbreviation
Meaning
α ΔC δC ρcp ρagg CCl Ccrit Cm
Degree of hydration Absolute error of the proposed method Relative error of the proposed method Density of cement paste Density of aggregates Chloride content Critical Chloride content Cement content measured with the proposed method in a concrete specimen Cement content in the bulk concrete Actual cement content in a concrete specimen Drilling core diameter Maximal aggregate diameter Mass of cement paste
Cref Cspecimen ddc dmax mcp
dmax 16 mm dmax 32 mm AVG dmax 32 mm AVG dmax 32 mm
Cref
ddc 50 mm
3.3. Implication for chloride analysis in concrete ddc 30 mm
In the previous sections it was shown that the actual cement content in a small concrete specimen can vary strongly. As a consequence, also the chloride content, CCl, present in the cement paste volume fraction, can be very different in different specimens (taken from concrete with supposedly identical bulk chloride content). Based on the variability in cement content, which was quantified in this work (see Figs. 5 & 6), the resulting variability in CCl referred to mass of cement has been
Fig. 5. Measured cement contents in different specimens, Cm, as a function of the specimen surface area. The variability in Cm increases strongly with decreasing specimen surface area. The effect is more pronounced the larger the coarse aggregates in the mix. Dots = individual measurements; lines = max. observed variability; vertical lines = common specimen dimensions (core diameter 30 and 50 mm).
A
B Cref σ = 0.8
surface: 40 cm2 surface: 18 cm2 surface: 06 cm2
surface: 40 cm2 surface: 18 cm2 surface: 06 cm2
Cref
σ = 1.3 σ = 3.4 σ = 3.4 σ = 4.4
σ = 6.0
Fig. 6. Probability density functions fitted to experimental data shown in Fig. 5. With increasing specimen surface area the standard deviation decreases.
5
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
A
B
chloride content referred to mass of cement, where corrosion initiation is probable range of common Ccrit in Codes
Fig. 7. For 4 assumed chloride contents in the bulk concrete (0.05, 0.10, 0.15 and 0.3 M-%) the distribution of possible chloride contents referred to mass of cement for ddc = 50 mm are depicted as a box plots. The light grey shaded area indicates the range of chloride threshold values for corrosion initiation in standards, the darker grey shaded area indicates the range of chloride contents where the likelihood for initiation is high. The horizontal lines of the boxes represent the 25%, 50%, and 75%-percentiles, and the whiskers are the extreme values.
concrete. According to international standards and common engineering practice, typical specimens for chloride analyses are 5–10 mm thick slices from 30 to 50 mm diameter concrete cores – obtained by drilling, cutting, powdering, etc. For concretes with dmax of the coarse aggregates = 32 mm, the actual cement content in such specimens might differ from the bulk content by a factor of up to 2. For 90% of the cases, this factor is < 1.5. 2. Whenever the conversion of chloride by mass of concrete (as measured) to chloride by mass of cement is based on assuming a constant cement content – either according to the mix proportion or a constant value such as 300 kg/m3 – the deviation mentioned in conclusion 1 directly affects the chloride content. Therefore, the actual chloride content in the concrete may be both under- and overestimated by a factor of up to 2 when referred to mass of cement. This may lead to misjudgements in condition assessments of existing structures and negatively impact planning of repair works – both in terms of cost (too much/early repair work devised) and safety (not enough/too late repair devised). Furthermore, we believe that this is one possible reason for the high variability observed in literature data on chloride threshold values for corrosion initiation. 3. To counteract the negative influence of the presence of coarse aggregates in concrete specimens for chloride analysis, we suggest to implement the here presented procedure for the determination of the cement content in individual specimens before analysing the chloride content. The uncertainty of the method is approx. 10–15%. Given the variability in cement concent (conclusion 2), applying this procedure permits thus considerable improvements in chloride measurements in concrete. 4. Further research should investigate the applicability of the method to cement pastes made with supplementary cementitious materials, to other aggregate types (e.g. with higher calcium carbonate contents, i.e. more than the presently investigated 20–30%, or coloured aggregates), or to carbonated concrete. Finally, the application of pattern recognition might be a possibility to improve further the method of image analysis, as the patterns of aggregates and cement paste are often similar.
computed, as shown in Fig. 7. As expected, chloride analyses performed with cores of common diameters (50 mm) [5] may be associated with a marked variability in CCl referred to mass of cement. This variability is real, and is not a measurement uncertainty. However, if a constant cement content as e.g. 300 kg/m3 is assumed to convert the measured CCl (in the concrete) to CCl by mass of cement, the actual chloride content may differ significantly from the converted one. This effect increases with increasing chloride contents. The chloride content where corrosion will probably initiate, is often assumed to be between 0.4 M-% and 1 M-% referred to mass of cement [28–30]. For concrete with dmax 32 mm no meaningful statement concerning corrosion initiation can be made for chloride contents above 0.5% related to concrete mass, because the possible actual values related to cement may be both below and above common threshold values for corrosion initiation (Fig. 7). Only if the cement content of the actual sample is determined the risk of corrosion can be assessed reliably. For chloride contents higher than ca. 2% by mass of concrete, on the other hand, accurate information about the actual cement content is not crucial because the likelihood for corrosion is high in any case. As a conclusion, common and standardized dimensions of specimens for chloride analysis appear to be too small to represent the bulk concrete, particularly not for dmax 32 mm. In this case, the actual chloride content in the concrete may be both under- and overestimated by a factor of up to 2. This may explain part of the huge scatter related to Ccrit reported in the literature because the cement content was in many studies not separately determined [2,9,12]. Additionally, due to the uncertainties in chloride measurements, engineers sometimes tend to add a “safety margin” to the results, which leads to more conservative condition assessments, and ultimately to higher costs in infrastructure management. 4. Conclusions and suggestions for further research work In this work, we suggested a procedure to quantify the cement content in disk shaped concrete specimens. The procedure includes colouring the concrete specimen surface, image acquisition with a flatbed scanner, and computer assisted image analysis. The procedure is practically non-destructive, relatively fast, and provides accurate results (max. 15% measurement error). From application of the procedure to concrete specimens, the following conclusions are drawn:
References [1] fib - Task Group 5.8, Condition Control and Assessment of Reinforced Concrete Structures: Exposed to Corrosive Environments (Carbonation/Chlorides), Fédération Internationale du Béton fib/International Federation for Structural Concrete, Lausanne, 2011. [2] U. Angst, B. Elsener, C.K. Larsen, Ø. Vennesland, Critical chloride content in reinforced concrete—a review, Cem. Concr. Res. 39 (2009) 1122–1138. [3] ASTM International, Standard Test Method for Acid-soluble Chloride in Mortar and Concrete, ASTM International, West Conshohocken, PA (USA), 2003.
1. The actual cement content in specimens for chloride analyses of common size exhibits significant variability. This is because common specimens are too small to be representative of the bulk 6
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
[17] A.-S. Dequiedt, M. Coster, L. Chermant, J.-L. Chermant, Study of phase dispersion in concrete by image analysis, Cem. Concr. Compos. 23 (2001) 215–226. [18] M.G.D. Gutierrez-Padilla, A. Bielefeldt, S. Ovtchinnikov, J. Pellegrino, J. Silverstein, Simple scanner-based image analysis for corrosion testing: concrete application, J. Mater. Process. Technol. 209 (2009) 51–57. [19] M. Hammer, Entwicklung mineralogischer Färbetechniken und ihre Anwendung auf spezifische Betonphasen zur Analyse der Zusammensetzung von zementgebundenen Baustoffen, Mathematisch-Naturwissenschaftlich-Technische Fakultät, MartinLuther-Universität Halle-Wittenberg, Halle (Saale, D), 2007, p. 196. [20] M. Ozen, M. Guler, Assessment of optimum threshold and particle shape parameter for the image analysis of aggregate size distribution of concrete sections, Opt. Lasers Eng. 53 (2014) 122–132. [21] S.L. Chan, Application of stereological and image analytical methods for concrete testing - part 2: the practical use of Quantimet 900 automatic image analyse system for the post-determination of the grading curves for hardened concrete, Mater. Tech. 13 (1985) 109–112. [22] S.L. Chan, Application of stereological and image analytical methods for concrete testing - part 1: review on some Stereological methods; applications for determining the grading curves of hardened concretes, Mater. Tech. 10 (1982) 181–190. [23] H. Gudmundsson, S. Chatterji, A.D. Jensen, N. Thaulow, P. Christensen, The measurement of paste content in hardened concrete using automatic image analyzing technique, Cem. Concr. Res. 9 (1979) 607–612. [24] G.M. Friedman, Identification of carbonate minerals by staining methods, J. Sediment. Petrol. 29 (1959) 87–97. [25] L. Wu, Surface Complexation at Calcium Mineral-water Interfaces, Division of Inorganic Chemistry, Luleå University of Technology, Luleå, 1994, p. 148. [26] ImageJ, Trainable Weka Segmentation, (2015). [27] T.C. Powers, T.L. Brownyard, Studies of physical properties of hardened portland cement paste part 1–9, J. Am. Concr. Inst. 43 (1946–1947) (101–132, 249–336, 469–504, 549–602, 669–712, 846–864, 865–880, 933–969, 971–992). [28] SIA, SIA 269/2, Erhaltung von Tragwerken - Betonbau, Schweizerischer Ingenieurund Architektenverein, Zürich, 2011. [29] fib - The International Federation for Structural Concrete, Model Code for Service Life Design, International Federation for Structural Concrete (fib), Lausanne, 2006. [30] P. Vassie, Reinforcement Corrosion and the Durability of Concrete Bridges, Proceedings of the Institution of Civil Engineers, London, 1984, pp. 713–723.
[4] DIN, EN 14629, Produkte und Systeme für den Schutz und die Instandsetzung von Betontragwerken - Prüfverfahren - Bestimmung des Chloridgehalts von Festbeton, Deutsches Institut für Normung, Berlin, 2007. [5] SIA, SIA EN 14629, Produkte und Systeme für den Schutz und die Instandsetzung von Betontragwerken - Prüfverfahren - Bestimmung des Chloridgehalts von Festbeton, Schweizerischer Ingenieur- und Architektenverband, Zürich, 2007. [6] RILEM TC 178-TMC, “Testing and modelling chloride penetration in concrete”, analysis of total chloride content in concrete—recommendation, Mater. Struct. 35 (2002) 583–585. [7] F. Hunkeler, H. Ungricht, F. Deillon, Untersuchungen zur Chloridbestimmung im Beton und Durchführung eines 2-stufigen Ringversuchs, ASTRA Bundesamt für Strassen, Bern, 2000. [8] M. Castellote, C. Andrade, Round-robin test on chloride analysis in concrete. I. Analysis of total chloride content, Mater. Struct. 34 (2001) 532–556. [9] J. Gulikers, Reliability of laboratory techniques for chloride analysis of reinforced concrete structures, in: C. Andrade, J. Kropp (Eds.), Testing and Modelling the Chloride Ingress into Concrete RILEM Proceedings PRO 19, RILEM Publications, 2000, pp. 439–450. [10] ASTM International, Standard Test Method for Portland-cement Content of Hardened Hydraulic-cement Concrete, ASTM International, West Conshohocken, PA (USA), 2013, p. 6. [11] DIN, DIN 52170 Teil 4, Bestimmung der Zusammensetzung von erhärtetem Beton, Deutsches Institut für Normung, Berlin, 1980. [12] U.M. Angst, R. Polder, Spatial variability of chloride in concrete within homogeneously exposed areas, Cem. Concr. Res. 56 (2014) 40–51. [13] B. Elsener, U. Angst, M. Wagner, Schlussbericht ASTRA AGB 2012/010: Methode zur Bestimmung des kritischen Chloridgehalts an bestehenden Stahlbetonbauwerken, ASTRA, 2016. [14] C. Boschmann, Genauere Bestimmung des Chloridgehalts in Beton Berücksichtigung des Anteils an Gesteinskörnern, Institut für Baustoffe, ETH Zürich, Zürich, 2014. [15] J. Stark, D. Ehrhardt, CEM II/B-S Zementsysteme im Betonstrassenbau, Wirtschaftsverlag NW, Bremerhaven (D), 2010. [16] A. Abbas, G. Fathifazl, B. Fournier, O.B. Isgor, R. Zavadil, A.G. Razaqpur, S. Foo, Quantification of the residual mortar content in recycled concrete aggregates by image analysis, Mater. Charact. 60 (2009) 716–728.
7
Contents lists available at ScienceDirect
Cement and Concrete Research journal homepage: www.elsevier.com/locate/cemconres
Image analysis for determination of cement content in concrete to improve accuracy of chloride analyses
MARK
Carolina Boschmann Käthlera,⁎, Ueli M. Angsta,b, Matthias Wagnerc, Bernhard Elsenera,d a
ETH Zurich, Institute for Building Materials (IfB), ETH Hönggerberg, CH-8093 Zurich, Switzerland Swiss Society for Corrosion Protection, Technoparkstrasse 1, CH-8005 Zurich, Switzerland c Tecnotest AG, Alemannenweg 4, CH-8803 Rüschlikon, Switzerland d University of Cagliari, Department of Chemical and Geological Science, I-09100 Monserrato, CA, Italy b
A R T I C L E I N F O
A B S T R A C T
Keywords: Image analysis Durability Aggregate Cement paste Chloride
The chloride content Ccl expressed as %Cl by weight of cement is important in condition assessment of reinforced concrete structures. Whereas standardized procedures determine Ccl in concrete powder, the cement content Cm is generally assumed equal to the mix design or an experience-based constant value. This work shows in concrete with maximal aggregate diameter 32 mm, Cm exhibits significant variability in 50 mm diameter cores, because the specimens are too small to be representative of bulk concrete. In such specimens, Cm might differ from the bulk cement content by a factor of up to 2. Thus, a reliable determination of Ccl in terms of %Cl by weight of cement requires the analysis of both Ccl and Cm in a concrete specimen. A procedure based on colouring the cement paste, scanning the specimen surface, and image analysis allows the practically non-destructive determination of Cm with good accuracy.
1. Introduction Chloride ions are known to impair the service life of reinforced concrete structures, as they can cause chloride-induced reinforcement corrosion. In marine environments or when using de-icing salts chloride-induced corrosion is the most common deterioration mechanism for reinforced concrete structures [1]. Therefore, measuring the chloride concentration in concrete (CCl) is common in condition assessment of existing reinforced concrete structures. Comparing the measured CCl with the so-called critical chloride content (Ccrit), i.e. the chloride threshold for corrosion initiation [2], is the widely accepted procedure to assess the risk of chloride-induced corrosion. Standardized methods [3–6] for the determination of CCl include three principal steps: 1) taking a sample, 2) extraction of the chloride ions, and 3) analysis of the chloride concentration. The result is a value of the chloride concentration referred to the dry mass of concrete. Each of the steps 1–3 are performed using different procedures, thus the size of the sample, the way of extraction and the analytical method can lead to errors in the final result. Several round robin tests [7,8] in the past have shown that on a homogeneous, reasonably fine concrete powder the total chloride content CCl can be determined with good accuracy. In accredited laboratories neither the extraction method nor the method of analytical chemistry used leads to significant errors.
⁎
Corresponding author. E-mail address: [email protected] (C. Boschmann Käthler).
http://dx.doi.org/10.1016/j.cemconres.2017.04.007 Received 2 November 2016; Received in revised form 13 April 2017; Accepted 18 April 2017 0008-8846/ © 2017 Elsevier Ltd. All rights reserved.
In engineering practice and in durability standards it is often preferred to relate CCl to the mass of cement. This representation is considered the best way to include both the aggressivity of the chloride ions and the corrosion inhibitive properties of the cement matrix at the steel reinforcement [2]. Usually, however, the cement content in the concrete is not known and may even vary between different parts in a structure. Thus, an assumption has to be made and commonly 300 kg cement per m3 concrete as a bulk cement content is considered adequate. In reality, concrete is not homogeneous, but a composite material consisting of three phases: cement paste, aggregates, and air voids. The aggregate volume fraction of bulk concrete typically is 60–80% of the concrete. Single coarse aggregate particles are likely of diameter ~ 30 mm, which thus have a volume of 14 cm3 (spherical shape assumed). Concrete specimens taken from structures for measuring CCl are typically drilling cores of diameter (ddc) in the range of 30 to 50 mm. These cores are typically cut or ground into slices of 5 to 10 mm thickness, which corresponds to specimen volumes of 3 to 10 cm3 – a volume comparable to a single coarse aggregate particle. It is thus expected that the volume fraction of cement paste can vary strongly from one concrete specimen to another, depending on the presence or absence of coarse aggregate particles. As is shown in Fig. 1, the actual cement content of a concrete specimen (Cspecimen) can differ from the
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
Sampling (e.g. core drilling / slicing)
measuring cement content in specimen
Cref
deviation of sample
cement content in bulk cement
Cspecimen
error of method: ∆C & ∂C
actual cement content in specimen
Cm
measured cement content in specimen
Fig. 1. Definition of error and deviation: Cref is equal to the cement content according to the mix proportion (bulk concrete), whereas Cspecimen is equal to the actual cement content within a specific concrete specimen. The measured CCl within this specimen needs to be referred to Cspecimen. Cm is the cement content on the specimen surface, measured with the suggested procedure including image analysis. All abbreviations are listed in Table 3.
distinguished (aggregates, cement paste, air voids). Thus, some methods [21–23] include etching a layer of cement paste followed by colouring the cement paste and aggregates with gypsum and black paint respectively to enhance contrast between both features. This method is laborious and therefore not convenient for practice. Ozen and Guler [20] simply used a desktop flatbed scanner for image acquisition. However, it is worth mentioning that the cement paste used by Ozen and Guler [20] is of whitish colour and therefore comparatively straightforward to differentiate from greyish aggregates without further image processing. The problem for implementation of this procedure in practice is that cement paste is itself often greyish. To account for this, Hammer [19] describes a method of staining the cement paste deep purple with Alizarinred S, without staining the aggregates [24]. After image processing, the optimal threshold is readily defined for determination of cement content in hardened concrete. In this paper, we describe and propose a method for the reliable and almost non-destructive determination of the cement content in specimens of hardened concrete – containing partly limestone aggregates in the concrete. The method includes colouring the concrete specimen, image acquisition by a flatbed scanner and image analysis. The method will be assessed in terms of applicability in research and practice; furthermore, it is applied to concrete specimens to quantify the variability in cement content in specimens of hardened concrete; the
cement content of the bulk concrete (Cref). Furthermore, all methods to determine the cement content of a concrete specimen exhibit an error due to limited measurement precision. As depicted in Fig. 1, the difference between Cspecimen and the measured cement content, Cm, is termed “error of the method” (ΔC, δC). Fig. 2 illustrates an example of two drilling cores taken from the same concrete. There are clear differences in cement paste volume fraction. To anticipate here the results of this work, the cement content in Fig. 2A is 17.5 M-% and in Fig. 2B is 5.5 M-%, thus differing by a factor > 3. When expressing the chloride concentration in concrete as CCl by weight of cement we believe that it is crucial to determine, in addition to the chloride concentration, the cement content in each specimen. Ideally, methods to measure the cement content in hardened concrete should be reliable, quick, almost non-destructive, and applicable to a wide variety of concretes. There exists a number of methods [9–11]; some are implemented in codes [10,11]. Some methods are based on determining the mass of the filter cake after acid digestion of the concrete specimen to quantify the aggregate mass fraction [9,12,13]. However, these are only applicable to lime-free aggregates [14]. Other methods use image analysis, either of electron microscope pictures [15] or of macro scale pictures [16–21]. Image analysis benefits from the contrast in brightness or colour of the features to be
A
B
Fig. 2. Two drilling core slices (ddc: 50 mm) from the same concrete mix dmax 32. Section A shows few coarse aggregates and has a higher cement content (17.5 M-%). Section B has more coarse aggregates and has therefore a lower cement content (5.5 M-%).
2
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
2.2.3. Conversion In this work, area fractions of the phases (aggregates, cement paste, voids) were considered equal to the respective volume fractions. This can be justified by the fact that the thickness of the slices was much smaller than their lateral dimensions (analysed surface > slice thickness). The volume fraction of cement paste was converted to mass percentage with the assumption for densities of cement paste ρcp and aggregates ρagg. The density ρagg was assumed to be 2.6 g/cm3 whereas the density of cement paste ρcp was calculated with Powers' model [27]:
Table 1 Nominal concrete mix proportions. Ingredients [kg/m3]
Mixture 1 “dmax 32”
Mixture 2 “dmax 16”
Portland cement Water (wc-ratio: 0.5) Aggregate (0–4 mm) Aggregate (4–8 mm) Aggregate (8–16 mm) Aggregate (16–32 mm) Cref [M-%] Curing
330 330 165 165 473 756 378 567 378 567 662 0 13.83 13.83 4 years wrapped in plastic sheeting at 20 °C (climate chamber)
ρcp =
1 + 0.23 × α 0.317 + w c
(1)
In the present work, the w/c ratio was 0.5 and the degree of hydration α was assumed to be 0.95. Therefore, ρcp = 1.49 g/cm3. The mass of cement paste in a specimen mcp was converted to mass of cement mc with the following formula (according to Powers' model [27]):
results and implications for chloride analysis are discussed. 2. Experimental methods
mc =
2.1. Materials
mcp (0.23 × α) + 1
(2)
The calculated mass percentage of cement is equal to Cm.
Concrete cubes (150 × 150 × 150 mm3) with two different mixtures were cast according to the mix proportion given in Table 1. The aggregates were gravel, containing a variable amount of calcium carbonate (of the order of 20–30%). For curing and storage, the cubes were wrapped in plastic and stored in ambient temperature 20 °C, i.e. left for self-desiccation. At the time of testing, the concrete was four years old.
2.2.4. Experiments 2.2.4.1. Assessment of suggested procedure. In the first step, for each tested concrete type, two slices of ca. 15 cm × 15 cm × 2 cm were analysed to determine the error of the method. To rule out any wall effects, all sides of the analysed surface area were reduced by 1 cm in length. The analysed surfaces had therefore an area of 160 cm2, which is 20 times larger than the surface of a 32 mm aggregate (assumption: circular shape cut at centre). The analysed surface is therefore assumed to be representative for the bulk concrete. This means that Cspecimen equals Cref (Fig. 1). The difference between Cm and Cspecimen corresponds to the error of the image analysis method. This error was expressed in absolute and relative values, ΔC and δC, respectively, which are defined in Eqs. (3) and (4).
2.2. Procedure 2.2.1. Sampling The concrete specimens were cut with a water-cooled diamond saw into slices of approximately 20 mm thickness. The slices were polished with a water-cooled diamond grinder and washed under running tap water. Subsequently, surface water was immediately removed with compressed air. Each slice was then placed in a freshly made solution of Alizarinred S (0.1 g to 100 ml H2O) for 3 min. This leads to a colouring of the cement paste (deep purple) of the slice surface exposed to the solution, because a precipitate is formed [19,24,25]. After colouring, image acquisition was conducted with a flatbed scanner. A resolution of 300 dpi was found to be a good compromise between time for image processing and accurate image resolution. This yields a pixel size of 85 μm, which determines the minimum particle size that may be identified in image analysis.
ΔC = Cm − Cref
(3)
ΔC Cref
(4)
δC =
2.2.4.2. Quantification of variability in cement content in concrete specimens. In the second step, only small fragments of the entire picture were subjected to image analysis. The four images of the representative specimen surfaces were cut with a grid into specimen surfaces of sizes 6 cm2, 18 cm2, and 40 cm2. The sizes 6 cm2 and 18 cm2 are comparable to practical specimen dimensions (drilling core diameter (ddc) 30 and 50 mm, respectively). A high number of cases, i.e. 50 cases for specimen size 6 cm2, 18 cases for size 18 cm2, and 8 cases for 40 cm2, were studied to individually assess the deviations of the cement content and the influence of parameters such as maximal aggregate diameter (dmax) and ddc.
2.2.2. Image processing The image processing and analysis was conducted in the software “Fiji” (“ImageJ”) and the corresponding plugin “trainable WEKA segmentation”. The standard settings were chosen as training features [26]. In “trainable WEKA segmentation”, three classes were defined, one for each phase: aggregate, cement paste, and air voids. For each class, representative areas were first manually selected on an image (two or more per class). The areas included various aggregates (different colours, textures, and sizes) and cement paste (lighter and darker areas). After the first training of the classifier, the classifier was improved with further area selections until the result of the overlay looked precise at visual check-up. From the final result (a classified image) the area fraction of aggregates and cement paste was computed (as the number of pixels in each class). Fig. 3 depicts an example of the original and segmented picture of a representative specimen surface. Since the judgement of the classified image depends on the operator performing the image analysis (the person training the classifier), we let 3 different people individually and independently conduct the image analysis for one selected image to assess the uncertainty related to the personnel.
3. Results and discussion 3.1. Feasibility of the suggested procedure Fig. 4 and Table 2 show the accuracy of the method in detecting the actual cement content in a concrete specimen, i.e. the difference between the measured (Cm) and known (Cref) cement content. The data is based on the analysis of all 160 cm2 large surfaces, thus assumed as representative of the bulk concrete. These results demonstrate that the proposed method for the determination of the cement content in a concrete specimen delivers accurate results. The relative error δC is always < 2% (Table 2), i.e. negligibly low. When applied to specimens taken from concrete structures in 3
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
A
B
Fig. 3. The left image A shows the original, scanned image with the stained cement paste; the right image B is the classified image after image analysis (aggregates are shown in red, cement paste in green, and air voids in purple). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
practice, the method may exhibit a larger uncertainty due to a number of factors. For the conversion from area percentage of cement paste to mass percentage of cement, assumptions are necessary (Eqs. (1) and (2)), i.e. for the density of the aggregates, the degree of hydration, and the w/c ratio. While in our laboratory tests, these were largely known; accurate information is often scarce for concrete taken from engineering structures. To give a first-hand estimate of the influence of such uncertainties, the conversion was also done for an extreme deviation in w/c ratio, namely for w/c = 0.4 instead of 0.5, and for a degree of hydration α = 1 instead of 0.95. With such a combination of wrong assumptions, the maximal absolute deviation of the cement content ΔC increased from 0.3 M-% to 1.9 M-%, the maximal relative deviation of the cement content δC from 2.0% to 13.4%. Similarly, also the assumption for the density of the aggregates was varied (from 2.3 to 2.9 g/cm3). The maximal absolute deviation of the cement content ΔC increased from 0.3 M-% to 1.5 M-%, the maximal relative deviation of the cement content δC from 2.0% to 10.6%. This uncertainty of approx. 10–15% is still by far better than the common way of assuming a unique cement content such as 300 kg/m3, which may affect the result of chloride analyses by a factor roughly 2 (see next section). A further influence may be related to the person carrying out the image analysis. This is to some extent subjective because the classifier for image analysis needs to be trained and the result is assessed by the person by means of visual judgement of the classified image. A trial with 3 test persons, independently carrying out the analysis for the
13.7
Table 2 Measured Cm at sample surfaces representative of bulk concrete (Cref 13.8 M-%). Specimen
Cm [M-%]
ΔC [M-%]
δC [%]
dmax 32 – 1 dmax 32 – 2 dmax 16 – 1 dmax 16 – 2 Average dmax 32 Average dmax 16
13.7 14.1 13.9 13.8 13.9 13.85
− 0.14 0.27 0.07 − 0.03 0.07 0.02
− 0.98 1.95 0.48 − 0.18 0.49 0.15
same image, revealed that ΔC was always smaller than 0.5 M-% and δC was always smaller than 3.5%. Thus, the procedure is relatively independent of personnel. Summarizing, compared to the variability arising from specimen dimensions with respect to aggregate diameter (see next section), the uncertainties related to assumptions inherent to the method and to personnel can be considered negligible. Thus, the suggested procedure is feasible to accurately determine the cement content in small specimens of hardened concrete, even with aggregates partially containing limestone (20–30%). 3.2. Influence of specimen size and coarse aggregate diameter dmax Fig. 5 shows all measured cement contents, Cm, as a function of the
14.1
13.9
13.8
Cref
Fig. 4. Suggested procedure applied to sufficiently large specimens, i.e. specimens representative of bulk concrete. Two analyses per concrete mix. The results show negligible deviations from the reference cement content (Cref = 13.8%).
4
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
specimen surface area. Depending on investigated surface area, up to 50 individual values were obtained per dmax. To give insight into the statistical distribution of the data, probability density functions were fitted for the obtained Cm at each investigated surface area and for each dmax. The type of distribution was assumed to be normal; standard deviation σ and mean μ were fitted with the method of the least squares. The results are shown in Fig. 6. Both, from Figs. 5 and 6, it is apparent that the variability in Cm decreased as expected with increasing specimen surface area. Furthermore, for a given surface area, the variability in Cm was larger for concrete with dmax = 32 mm than for dmax = 16 mm. These results confirm the expected influence of coarse aggregates. The two vertical lines in Fig. 5 correspond to specimen sizes comparable to ddc 30 and 50 mm, respectively. Clearly, concrete disks cut from cores of common diameters will exhibit significant variability in terms of cement volume fraction in each specimen. The actual cement content may be both lower and higher than in the bulk concrete, namely roughly by a factor of 2 (for dmax = 32 mm and ddc = 50 mm). Cm ranged from ca. 8 M-% to 25 M-%, while Cref was 13.8 M-%. Thus, concrete cores of diameters 30–50 mm drilled from concrete with dmax = 32 mm are not large enough to yield specimens that are – in terms of phases present – representative of the bulk concrete. Another strategy (alternative to image analysis) to reduce the deviation of cement content in a specimen may be increasing the specimen size (area). However, to reduce the deviation to an acceptable level, a marked increase of ddc is needed. As can be seen in Fig. 5, even an area of 41 cm2 (approximately corresponding to ddc = 70 mm) provides cement contents from 10 to 20 M-%, thus deviating from the reference value (13.8 M-%) by a factor of ± 50%. Additionally, increasing the sample size until it becomes representative for the bulk concrete is not feasible for practice, because condition assessment should be as less destructive as possible. Fig. 5 may give some guidance also for the drilling method of sampling sometimes used: The surface area needed to reduce the uncertainty in measured chloride content to below ± 50% is ca. 60 mm2 (for dmax 32 mm). With a ddc of 20 mm this would correspond to ca. 19 drilling holes, which is considered a high number.
Table 3 Abbreviations. Abbreviation
Meaning
α ΔC δC ρcp ρagg CCl Ccrit Cm
Degree of hydration Absolute error of the proposed method Relative error of the proposed method Density of cement paste Density of aggregates Chloride content Critical Chloride content Cement content measured with the proposed method in a concrete specimen Cement content in the bulk concrete Actual cement content in a concrete specimen Drilling core diameter Maximal aggregate diameter Mass of cement paste
Cref Cspecimen ddc dmax mcp
dmax 16 mm dmax 32 mm AVG dmax 32 mm AVG dmax 32 mm
Cref
ddc 50 mm
3.3. Implication for chloride analysis in concrete ddc 30 mm
In the previous sections it was shown that the actual cement content in a small concrete specimen can vary strongly. As a consequence, also the chloride content, CCl, present in the cement paste volume fraction, can be very different in different specimens (taken from concrete with supposedly identical bulk chloride content). Based on the variability in cement content, which was quantified in this work (see Figs. 5 & 6), the resulting variability in CCl referred to mass of cement has been
Fig. 5. Measured cement contents in different specimens, Cm, as a function of the specimen surface area. The variability in Cm increases strongly with decreasing specimen surface area. The effect is more pronounced the larger the coarse aggregates in the mix. Dots = individual measurements; lines = max. observed variability; vertical lines = common specimen dimensions (core diameter 30 and 50 mm).
A
B Cref σ = 0.8
surface: 40 cm2 surface: 18 cm2 surface: 06 cm2
surface: 40 cm2 surface: 18 cm2 surface: 06 cm2
Cref
σ = 1.3 σ = 3.4 σ = 3.4 σ = 4.4
σ = 6.0
Fig. 6. Probability density functions fitted to experimental data shown in Fig. 5. With increasing specimen surface area the standard deviation decreases.
5
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
A
B
chloride content referred to mass of cement, where corrosion initiation is probable range of common Ccrit in Codes
Fig. 7. For 4 assumed chloride contents in the bulk concrete (0.05, 0.10, 0.15 and 0.3 M-%) the distribution of possible chloride contents referred to mass of cement for ddc = 50 mm are depicted as a box plots. The light grey shaded area indicates the range of chloride threshold values for corrosion initiation in standards, the darker grey shaded area indicates the range of chloride contents where the likelihood for initiation is high. The horizontal lines of the boxes represent the 25%, 50%, and 75%-percentiles, and the whiskers are the extreme values.
concrete. According to international standards and common engineering practice, typical specimens for chloride analyses are 5–10 mm thick slices from 30 to 50 mm diameter concrete cores – obtained by drilling, cutting, powdering, etc. For concretes with dmax of the coarse aggregates = 32 mm, the actual cement content in such specimens might differ from the bulk content by a factor of up to 2. For 90% of the cases, this factor is < 1.5. 2. Whenever the conversion of chloride by mass of concrete (as measured) to chloride by mass of cement is based on assuming a constant cement content – either according to the mix proportion or a constant value such as 300 kg/m3 – the deviation mentioned in conclusion 1 directly affects the chloride content. Therefore, the actual chloride content in the concrete may be both under- and overestimated by a factor of up to 2 when referred to mass of cement. This may lead to misjudgements in condition assessments of existing structures and negatively impact planning of repair works – both in terms of cost (too much/early repair work devised) and safety (not enough/too late repair devised). Furthermore, we believe that this is one possible reason for the high variability observed in literature data on chloride threshold values for corrosion initiation. 3. To counteract the negative influence of the presence of coarse aggregates in concrete specimens for chloride analysis, we suggest to implement the here presented procedure for the determination of the cement content in individual specimens before analysing the chloride content. The uncertainty of the method is approx. 10–15%. Given the variability in cement concent (conclusion 2), applying this procedure permits thus considerable improvements in chloride measurements in concrete. 4. Further research should investigate the applicability of the method to cement pastes made with supplementary cementitious materials, to other aggregate types (e.g. with higher calcium carbonate contents, i.e. more than the presently investigated 20–30%, or coloured aggregates), or to carbonated concrete. Finally, the application of pattern recognition might be a possibility to improve further the method of image analysis, as the patterns of aggregates and cement paste are often similar.
computed, as shown in Fig. 7. As expected, chloride analyses performed with cores of common diameters (50 mm) [5] may be associated with a marked variability in CCl referred to mass of cement. This variability is real, and is not a measurement uncertainty. However, if a constant cement content as e.g. 300 kg/m3 is assumed to convert the measured CCl (in the concrete) to CCl by mass of cement, the actual chloride content may differ significantly from the converted one. This effect increases with increasing chloride contents. The chloride content where corrosion will probably initiate, is often assumed to be between 0.4 M-% and 1 M-% referred to mass of cement [28–30]. For concrete with dmax 32 mm no meaningful statement concerning corrosion initiation can be made for chloride contents above 0.5% related to concrete mass, because the possible actual values related to cement may be both below and above common threshold values for corrosion initiation (Fig. 7). Only if the cement content of the actual sample is determined the risk of corrosion can be assessed reliably. For chloride contents higher than ca. 2% by mass of concrete, on the other hand, accurate information about the actual cement content is not crucial because the likelihood for corrosion is high in any case. As a conclusion, common and standardized dimensions of specimens for chloride analysis appear to be too small to represent the bulk concrete, particularly not for dmax 32 mm. In this case, the actual chloride content in the concrete may be both under- and overestimated by a factor of up to 2. This may explain part of the huge scatter related to Ccrit reported in the literature because the cement content was in many studies not separately determined [2,9,12]. Additionally, due to the uncertainties in chloride measurements, engineers sometimes tend to add a “safety margin” to the results, which leads to more conservative condition assessments, and ultimately to higher costs in infrastructure management. 4. Conclusions and suggestions for further research work In this work, we suggested a procedure to quantify the cement content in disk shaped concrete specimens. The procedure includes colouring the concrete specimen surface, image acquisition with a flatbed scanner, and computer assisted image analysis. The procedure is practically non-destructive, relatively fast, and provides accurate results (max. 15% measurement error). From application of the procedure to concrete specimens, the following conclusions are drawn:
References [1] fib - Task Group 5.8, Condition Control and Assessment of Reinforced Concrete Structures: Exposed to Corrosive Environments (Carbonation/Chlorides), Fédération Internationale du Béton fib/International Federation for Structural Concrete, Lausanne, 2011. [2] U. Angst, B. Elsener, C.K. Larsen, Ø. Vennesland, Critical chloride content in reinforced concrete—a review, Cem. Concr. Res. 39 (2009) 1122–1138. [3] ASTM International, Standard Test Method for Acid-soluble Chloride in Mortar and Concrete, ASTM International, West Conshohocken, PA (USA), 2003.
1. The actual cement content in specimens for chloride analyses of common size exhibits significant variability. This is because common specimens are too small to be representative of the bulk 6
Cement and Concrete Research 99 (2017) 1–7
C. Boschmann Käthler et al.
[17] A.-S. Dequiedt, M. Coster, L. Chermant, J.-L. Chermant, Study of phase dispersion in concrete by image analysis, Cem. Concr. Compos. 23 (2001) 215–226. [18] M.G.D. Gutierrez-Padilla, A. Bielefeldt, S. Ovtchinnikov, J. Pellegrino, J. Silverstein, Simple scanner-based image analysis for corrosion testing: concrete application, J. Mater. Process. Technol. 209 (2009) 51–57. [19] M. Hammer, Entwicklung mineralogischer Färbetechniken und ihre Anwendung auf spezifische Betonphasen zur Analyse der Zusammensetzung von zementgebundenen Baustoffen, Mathematisch-Naturwissenschaftlich-Technische Fakultät, MartinLuther-Universität Halle-Wittenberg, Halle (Saale, D), 2007, p. 196. [20] M. Ozen, M. Guler, Assessment of optimum threshold and particle shape parameter for the image analysis of aggregate size distribution of concrete sections, Opt. Lasers Eng. 53 (2014) 122–132. [21] S.L. Chan, Application of stereological and image analytical methods for concrete testing - part 2: the practical use of Quantimet 900 automatic image analyse system for the post-determination of the grading curves for hardened concrete, Mater. Tech. 13 (1985) 109–112. [22] S.L. Chan, Application of stereological and image analytical methods for concrete testing - part 1: review on some Stereological methods; applications for determining the grading curves of hardened concretes, Mater. Tech. 10 (1982) 181–190. [23] H. Gudmundsson, S. Chatterji, A.D. Jensen, N. Thaulow, P. Christensen, The measurement of paste content in hardened concrete using automatic image analyzing technique, Cem. Concr. Res. 9 (1979) 607–612. [24] G.M. Friedman, Identification of carbonate minerals by staining methods, J. Sediment. Petrol. 29 (1959) 87–97. [25] L. Wu, Surface Complexation at Calcium Mineral-water Interfaces, Division of Inorganic Chemistry, Luleå University of Technology, Luleå, 1994, p. 148. [26] ImageJ, Trainable Weka Segmentation, (2015). [27] T.C. Powers, T.L. Brownyard, Studies of physical properties of hardened portland cement paste part 1–9, J. Am. Concr. Inst. 43 (1946–1947) (101–132, 249–336, 469–504, 549–602, 669–712, 846–864, 865–880, 933–969, 971–992). [28] SIA, SIA 269/2, Erhaltung von Tragwerken - Betonbau, Schweizerischer Ingenieurund Architektenverein, Zürich, 2011. [29] fib - The International Federation for Structural Concrete, Model Code for Service Life Design, International Federation for Structural Concrete (fib), Lausanne, 2006. [30] P. Vassie, Reinforcement Corrosion and the Durability of Concrete Bridges, Proceedings of the Institution of Civil Engineers, London, 1984, pp. 713–723.
[4] DIN, EN 14629, Produkte und Systeme für den Schutz und die Instandsetzung von Betontragwerken - Prüfverfahren - Bestimmung des Chloridgehalts von Festbeton, Deutsches Institut für Normung, Berlin, 2007. [5] SIA, SIA EN 14629, Produkte und Systeme für den Schutz und die Instandsetzung von Betontragwerken - Prüfverfahren - Bestimmung des Chloridgehalts von Festbeton, Schweizerischer Ingenieur- und Architektenverband, Zürich, 2007. [6] RILEM TC 178-TMC, “Testing and modelling chloride penetration in concrete”, analysis of total chloride content in concrete—recommendation, Mater. Struct. 35 (2002) 583–585. [7] F. Hunkeler, H. Ungricht, F. Deillon, Untersuchungen zur Chloridbestimmung im Beton und Durchführung eines 2-stufigen Ringversuchs, ASTRA Bundesamt für Strassen, Bern, 2000. [8] M. Castellote, C. Andrade, Round-robin test on chloride analysis in concrete. I. Analysis of total chloride content, Mater. Struct. 34 (2001) 532–556. [9] J. Gulikers, Reliability of laboratory techniques for chloride analysis of reinforced concrete structures, in: C. Andrade, J. Kropp (Eds.), Testing and Modelling the Chloride Ingress into Concrete RILEM Proceedings PRO 19, RILEM Publications, 2000, pp. 439–450. [10] ASTM International, Standard Test Method for Portland-cement Content of Hardened Hydraulic-cement Concrete, ASTM International, West Conshohocken, PA (USA), 2013, p. 6. [11] DIN, DIN 52170 Teil 4, Bestimmung der Zusammensetzung von erhärtetem Beton, Deutsches Institut für Normung, Berlin, 1980. [12] U.M. Angst, R. Polder, Spatial variability of chloride in concrete within homogeneously exposed areas, Cem. Concr. Res. 56 (2014) 40–51. [13] B. Elsener, U. Angst, M. Wagner, Schlussbericht ASTRA AGB 2012/010: Methode zur Bestimmung des kritischen Chloridgehalts an bestehenden Stahlbetonbauwerken, ASTRA, 2016. [14] C. Boschmann, Genauere Bestimmung des Chloridgehalts in Beton Berücksichtigung des Anteils an Gesteinskörnern, Institut für Baustoffe, ETH Zürich, Zürich, 2014. [15] J. Stark, D. Ehrhardt, CEM II/B-S Zementsysteme im Betonstrassenbau, Wirtschaftsverlag NW, Bremerhaven (D), 2010. [16] A. Abbas, G. Fathifazl, B. Fournier, O.B. Isgor, R. Zavadil, A.G. Razaqpur, S. Foo, Quantification of the residual mortar content in recycled concrete aggregates by image analysis, Mater. Charact. 60 (2009) 716–728.
7