Non-destructive screening methodology based on handheld XRF for the classification of concrete: cement type-driven separation

The utilization of locally available concrete waste for producing recycled concrete aggregates is recognized as one of the most sustainable ways of satisfying the growing demand for concrete production. However, the quality of concrete waste depends on its origin and it may significantly differ from one concrete structure to another. Knowing the chemical composition of the parent concrete is crucial for determining or verifying the origin of the raw materials. For this reason, pre-demolition concrete waste streams need to be characterized and classified. Therefore, a new non-destructive method for determining the cement and aggregate type in hardened concrete using handheld X-ray fluorescence (hXRF) analyser is presented in this paper. The method was tested on different raw powders and on concretes containing different types of cements including CEM I 42.5 N (Portland cement), CEM II/B-V 42.5 N (Portland-fly ash cement), CEM III/B 42.5 N (GGBFS cement). Combined desktop XRF and Energy-dispersive X-ray Spectroscopy (EDS) measurements were used for the purpose of validation. The results revealed that the curing of concrete affects the results: a dried concrete surface condition was optimal for measurements since it limits the impact of the concrete surface moisture and efflorescence on characteristic element oxides, such as CaO. The effective measurement duration was 30 s. A CEM III/B 42.5 N (GGBFS)-based concrete surface was distinguished from other concretes using Al2O3, MgO and Fe2O3 as characteristic oxides. The inner layers of concrete were rich in SiO2, the oxide characteristic for the aggregate composition tested in this study. This shows that hXRF is suitable for use in concrete, provided that the concrete surface is dried and the characteristic elements are defined to ensure a distinction between different cement and aggregate types. Direct adoption of such characterization, however, requires field testing across a wide range of concrete compositions and in situ conditions.

testing Á Handheld X-ray fluorescence Á Mineral resources Á Sustainability

Introduction
Over the past decade, extensive research has been carried out on the utilization of recycled concrete aggregates (RCA) and the effect of their properties on the physical and mechanical properties, durability and structural behaviour of new concrete. The variations in physical and chemical properties of RCA lead to a wide range of mechanical and durability properties of mortars and concretes with RCA [1]. A primary concern for the application of RCA in concrete plants is the unknown origin of RCA: the concrete mix design needs to be modified with each new batch of RCA.
Despite extensive research activities in this field, RCA are mostly used in low-grade applications such as road construction [2]. In order to improve the quality of RCA, mixing different parent concrete waste streams should be avoided [3] and more efficient comminution technologies should be used [4]. Preselection of concrete streams requires the characterization and quality assessment of the parent concrete. Knowing the origin and the properties of parent concrete(s) can help to improve predictions of the impact of RCA on the properties of new concrete.
Over the past decades, demolition of concrete structures has principally been carried out without significant preselection and, in particular, without any preselection of concrete. Recently, a step-by-step procedure has been proposed that can be followed to characterize concrete using non-destructive test methods before its demolition [3]; assuming that no records about the parent concrete are available. The classification of concrete is based on strength and composition. The compressive strength of the parent concrete has been shown to exert a strong influence on the compressive strength of recycled aggregate concrete [3] and on the energy consumption during the crushing of concrete [5]. Visual inspection of a concrete surface can lead to direct classification of a concrete element based on characteristic signs of concrete deterioration.
Undeteriorated concrete can be the source of highquality RCA. Characterization techniques for determining the chemical composition of RCA are needed to further enable the separation process of undeteriorated concrete structural members. A stateof-the art non-destructive technique for determining the material chemical composition, is handheld X-ray fluorescence (hXRF), owing to the compatibility with desktop XRF for cement and concrete characterization, easy installation, maintenance and large-scale applications. Furthermore, hXRF is capable of non-destructive analysis and detection of a large range of elements with no sample preparation, making it preferable for in situ testing of concrete structural members. Chemical characterization of different materials with hXRF has been presented in a comprehensive literature survey [3]. Although research has been carried out on a wide range of scientific and industrial applications of hXRF in archaeology, metals and historical mortars characterization and classification [6,7,8,9,10,11,12,13,14,15,16], only two studies have attempted to investigate content of chlorides and sulphates specifically on concrete surfaces using hXRF [17,18]. No studies have investigated the concrete composition with hXRF in a systematic way. However, there are studies on a non-destructive micro-analysis technique called micro X-ray fluorescence (lXRF) for concrete element mapping.
The lXRF technique was used to obtain images of chloride profiles in concrete [19,20,21], to detect the presence of silane coatings in concrete [22] and to determine the content of cement paste and aggregates in concrete [23]. The technique is similar to bulk X-ray fluorescence (XRF): it contains polycapillary optics for focusing X-rays to a size of approximately 50 lm in diameter, whereas bulk XRF investigates samples with a size of approximately 10 mm in diameter [20]. Although lXRF is capable of non-destructive X-ray imaging, the instrument chamber geometry limits the sample geometry for analysis. Therefore, it cannot be applied for in situ analysis of large vertical and horizontal surfaces of concrete structures surfaces. Recent developments in the field of handheld XRF instruments allow for real time data to be collected allowing for real time measurements and decisions to be made assuming that the hXRF is used correctly [24].
Testing the composition of parent concrete before demolition can provide two types of information: (1) the composition of major components in concrete (cement and aggregates), and (2) potential minor elements and contamination. Knowing the cement type can assist in deciding on the upcoming application for the recycled material. For example, based on the cement type in the parent concrete, one can decide whether the fine RCA would be valuable as an alternative raw material for clinker manufacturing [25][26][27], as a pozzolanic material [28] or as a filler [29][30][31][32]. When the goal of using RCA is CO 2 sequestration, the most suitable cement type in parent concrete, would be ordinary Portland cement, containing the highest CaO and calcium hydroxide (Ca(OH) 2 ) that can render them the ability to react with CO 2 . In the context of selective demolition, research has not yet investigated the chemical composition of parent concrete before its demolition.
With this in mind, this study differentiates itself from others by the fact that it aims for pre-demolition non-destructive identification of concrete waste streams. The vast majority of studies have been focused on characterization of RCA after concrete recycling, when concrete streams have been mixed and when it is impossible to trace back the origin of the parent concrete streams. The objective of pre-demolition concrete identification is a systematic classification of the concrete structural members in order to provide groups of concrete members with similar mechanical and chemical properties. In this way, the differences in properties between recycled concrete materials and natural aggregates could be substantially reduced (in terms of composition, strength, contamination, water absorption).
For this purpose, in this research, a new method to determine the composition of concrete in an automated way was developed. The study focuses on nondestructive characterization of the chemical composition of cement and aggregates in three concrete mix designs using an hXRF analyser. Kreijger [33] showed that the skin of a concrete structure consists of three layers, the cement skin (* 0.1 mm thick), the mortar skin (* 5 mm) and the concrete skin (* 30 mm). Therefore, the research hypothesis of this study is that testing the chemical composition of the surface (first concrete skin layer) will represent the cement type, whereas testing the chemical composition of the inner layers will yield the aggregate type in a relatively homogeneous environmental setting and, therefore, can be predicted from the direct element readings of hXRF. Additionally, the chemical composition of powder materials (cement and aggregates) was measured. Combined desktop XRF and energy-dispersive X-ray spectroscopy (EDS) measurements were used for validation.

Experimental program
The aim of this study was to assess the applicability of hXRF for characterizing the chemical composition of: (1) powders and (2) hardened concrete for the purpose of concrete sorting based on cement type. Therefore, the experimental program consisted of three stages, (i) testing chemical composition of different cements, supplementary cementitious materials and aggregates with hXRF, (ii) selecting the most-used cement types in order to design concrete mixtures and (iii) testing the chemical composition of the concretes.

Materials
Handheld X-ray fluorescence analyses of 5 cement powders (up to 90% passing 75 lm), two secondary by-products (ground granulated blast furnace slag and fly ash), limestone filler, crushed and round river sands, were performed in this study. The raw materials used in experiments, their suppliers and densities are listed in Table 1.

Mixture design
Three types of cement were used: CEM I 42.5 N (Portland cement), CEM II/B-V 42.5 N (Portland-fly ash cement), CEM III/B 42.5 N (GGBFS cement) ( Table 2). All mixtures had an identical water-tocement (w/c) ratio of 0.50 and a constant sand-tocement and gravel-to-sand ratio. The ratios were kept constant for all mixtures in order to examine only the effect of cement type on the chemical composition of concrete surfaces.

Mixing sequence and curing
A 40 L mixer was used for mixing the concrete batches. The mixing procedure was as follows: gravel, river sand and cement were dry-mixed for 1 min; thereafter, water was added followed by another 2 min of mixing. No chemical admixtures were used to preclude additional effects of these materials on the strength development and to enable examining only the effect of cement type on the chemical composition. After mixing, the cement concretes were cast into 150 9 150 9 150 mm 3 cubic moulds and compacted with a compaction table for 15 s. Subsequently, 24 h after casting, the samples were demoulded and stored in a curing chamber at a temperature of 20°C and relative humility 99% until the age of testing. At a curing age of 90 days, the concrete samples were split into three groups to investigate the effect of surface state on chemical composition: (1) Water saturated concrete surface: additional 10 days in the curing chamber; (2) Air dried concrete surface: additional 10 days in open air; (3) Oven dried concrete surface: additional 10 days in an oven at 75°C.

Bulk element composition testing methods
Supplementary analyser The samples were analysed using a Bruker S1 TITAN 800 handheld Energy Dispersive X-ray Fluorescence analyser ( Fig. 1, left), equipped with a rhodium tube from which X-rays are emitted, and a high-resolution Silicon Drift (SDD) detector. The instrument produces an X-ray beam at a 45°angle from the centre of the analyser's tip ( Fig. 1, right). Regarding filter use, it was not used for calibration method and it was shown that the signal on historical mortars, with composition similar to concrete, was the best when no filter was used [9], hence, this approach was followed in this study as well.
The hXRF spectrometer must always be in contact with sample ( Fig. 1, left), not at a distance from it [35]. X-rays have a limited penetration depth of few lm to several mm, depending on the nature of the sample matrix. For that reason, the measured composition by hXRF will be representative to the outmost layers [36]. Samples were at least 10 mm thick across the beam profile to fully ensure that only the desired material was analysed, and thus, a reliable comparison between samples was obtained. The instrument was turned on at least 10 min prior to the initial measurement, to allow the detector to fully cool and reach a stable temperature. The instrument's X-ray tube was turned on  These settings are in accordance with calibration settings. The calibration was performed by measuring certified reference materials including different cement types. The Bruker TM calibration method for cement was used, being created with samples that have a similar chemistry to the samples being analysed, therefore producing a much closer correlation with the chemical composition and enabling more accurate data to be collected in real time. Supplementary  Fig. S1 shows hXRF spectra of quartz sand and limestone filler and their characteristic element peaks. The characteristic of quartz sand was the high Si concentration (95.4 wt.% ± 0.99), and that of limestone was the high Ca concentration (93.94 wt.% ± 0.15). The quartz and limestone were used to monitor for low-level contamination on the analysis range or in the detector. It should be noted that for the characterization of materials such as steel reinforcement or steel fibres in concrete, a new calibration is needed, since the calibration method for cement includes elements and their analysis ranges which are characteristic for cementitious materials (MgO, CaO, Al 2 O 3 , SiO 2 , P 2 O 5 , SO 3 , K 2 O, TiO 2 , MnO, Fe 2 O 3 ).

Desktop
X-ray fluorescence (XRF) spectrometer The chemical composition of 5 cement powders, two secondary by-products (ground granulated blast furnace slag and fly ash), limestone filler, crushed and round river sands, were also analyzed with desktop X-ray fluorescence (XRF) spectrometer. The desktop XRF measurements were performed in Heidelberg Cement Benelux laboratories in Maastricht, the Netherlands. The XRF spectra of the studied materials were recorded with a PANalytical AXIOSmAX Advanced 4 kW Rh 60 kV LiF220 Ge111-c PX1 spectrometer on pressed pellets. Regarding sample preparation for desktop XRF, loose powders were mixed with a binding agent, Boreox (C 12 H 22 O 11 ) from Fluxana. Powders were then pressed and prepared in a POLAB APM automatic sample preparation module machine with final appearance as shown in Fig. 2. The powder fineness of 80% \ 16 lm was required for analyses.

Environmental scanning electron microscope and energy-dispersive X-ray Spectroscopy (ESEM-EDS)
The Thermo Scientific TM UltraDry EDS Detector was used for measurements. Semiquantitative analysis was carried out with Thermo Pathfinder v1.3.22 using the standardless method. The accelerating voltage for quantitative EDS analyses is recommended to be at least 2-3 times the voltage of the element absorption edge. For example, Al, Si, Ca and Fe have the absorption edge voltage of 1.560,  1.840, 4.037 and 7.110 keV respectively, implying that the accelerating voltage should be at least 8.074 (4.037 9 2) keV for Ca and 14.220 (7.110 9 2) keV for Fe, respectively [37]. Considering the main oxides expected to occur in concrete, e.g., CaO, SiO 2 , Al 2 O 3 , Fe 2 O 3 , MgO, SO 3 , the accelerating voltage was 15 keV to enable quantification of iron (to access K lines of iron). Previous work showed that at such a voltage, 95% of the characteristic X-rays escaping cementitious materials were generated within a depth of * 900-1700 nm (depending on the phase, it's density and on the measured element) [38]. The accelerating voltage of 15 keV was also used during hXRF measurements.

Sample preparation for powders
Regarding sample preparation for hXRF and EDS, standard powder cups covered with a foil and pressed powders were prepared [35], however, they were not providing satisfactory results for hXRF analyzed samples and the cups could also not be used for EDS analysis. Hence, a new method was proposed for testing powders, using epoxy impregnation (Fig. 3), since the epoxy can attenuate the X-rays generated for low tube voltages (\ 15 keV). The samples were epoxy impregnated to also keep the particles fixed and to obtain a perfectly flat and smooth surface for hXRF contact measurements, which was not possible with the standard sample preparation method (a sample cup using 4 lm prolene or ultralene film). An additional advantage of using epoxy impregnated samples is the elimination of any air pockets which might have affected the penetration and escape of X-rays. To prepare the powder-epoxy sample, the Conpox Resin BY 158 (liquid) was mixed with the Conpox Hardener HY 2996 (Polyamine liquid) and then powder was added into it. The ratio of epoxy resin to hardener used was 3:1. The density of the epoxy resin and hardener mix was 1.16 g/cm 3 . The mixing of epoxy resin, hardener and the powder was done through gentle stirring for 10 min to ensure dispersion of the powder and to minimize the formation of air bubbles in the sample. The mixture was then cast in a 40 mm round silicon rubber mold with a thickness of 10 mm. It was left for 24 h to harden at room temperature to ensure that the powder particles would settle on the bottom of the mould. Subsequently, the samples were ground following the procedure described in Supplementary  Table S2, to obtain a very flat and smooth (''mirrorlike'') surface. In this study, the sample processing procedure is shown in Fig. 3. The prepared samples for element analysis are shown in Supplementary  Fig. S2. It should be noted that the individual particles in powders may have different properties (e.g. size, composition, density and shape). The effect of particles properties on segregation of powder mixtures during epoxy-impregnation was not considered since powders have none or limited amount of heavy particles (e.g. particles that contain only Fe 2 O 3 ). In addition, the goal of the hXRF and EDS analyses is the bulk chemical composition of powders. When powder bulk is analyzed, the aforementioned differences are averaged out and more reliable results can be obtained [39].

Preparation of the concrete specimens
The production of concrete specimens (mix design, mixing, geometry, curing) is detailed in Sect. 2.2. Based on Kreijger's study on the concrete skin [33], determining the chemical composition of the outer surface (first concrete skin layer) would charecterize the cement type. Therefore, prior to testing, it was essential to make sure that the concrete surface is flat and free of dust or any other foreign materials. All the surfaces, except the concrete trowelled surface, were characterized using hXRF. The trowelled surface was not characterized in order to avoid localized effects such as inconsistent trowelling and trowel burn, which may affect the tip of the hXRF. The outer surfaces were cleaned with ethanol and dried with an air gun without any surface grinding, as this could have caused loss of material that is crucial for the characterization of the paste-only zone. The prepared concrete surfaces for hXRF tests are shown in Fig. 4. In order to obtain representative results on each sampling area and each concrete type, at least nine measurements were taken. The grid of nine points can be seen on the samples surfaces in Fig. 4. The grid was also used in order to evaluate the repeatability of the results, i.e. to compare the obtained results with previous measurements made at the same position. The inner layers were accessed using a diamond abrasive tool with a grit size of 60 for grinding and afterwards cleaning the concrete surface with ethanol and drying it with an air gun ( Supplementary Fig. S3).
The criteria for selecting measurement duration were based on convenience and accuracy. Usually, measurement lasts from 10 s to 10 min [40]. Shorter duration (10-30 s) is used more for qualitative scanning and sorting. Major elements in percentage concentrations can be analysed in a minute or less, whereas minor elements at part per million (PPM) concentrations may need to be analysed for 3-10 min or longer. The other and ultimately more important criterion is precision. A fluorescence spectrum in this study were recorded for 30, 60, 90 and 120 s live time on three concrete types (SC1, SC2, SC3). Figure 5 shows the preparation and placement of a typical concrete block (50 9 50 9 20 mm 3 ) inside the electron microscope for EDS surface measurements. Cutting of a concrete cube required water. Subsequently, concrete blocks were placed in an oven at 60°C for drying until the testing. The microscope and EDS settings were kept constant for all the samples.
The imaging resolution was 512 9 384 pixels. The image resolution was 1 lm per pixel. An example of a composite ESEM-BSE image of nine fields is presented in Fig. 5. The matrix 3 9 3 (00-22) was selected so that the chemical composition with a 1-lm resolution over an area larger than a single field could be determined [39].

Bulk composition of powders with hXRF
The chemical compositions of powders obtained with hXRF are shown in Table 3. The chemical composition of the sands and limestone filler was based on a very few elements. The chemical content of sands was mainly siliceous ([ 95% SiO 2 ). The limestone filler is characterized by its high CaO content. Cement powders are heterogeneous with respect to chemistry as a result of their production processes. The chemical composition of the cements can vary within wide limits, depending on the percentage of secondary byproducts (fly ash and GGBFS) mixed with ordinary Portland cement. The ordinary Portland cement consists mainly of CaO and SiO 2 , and the minor components are Al 2 O 3 , Fe 2 O 3 , MgO, SO 3 . The CaO content of most of the cements ranged from 50 to 64%, and that of SiO 2 ranged from 20 to 43%. The hXRF determines precision as the margin of error or the standard deviation by analyzing X-ray counting statistics for each measurement [41]. Standard deviation can be recorded as 1r, 2r or 3r. The confidence intervals are 68.27% for one standard deviation, 95.45% for two standard deviations and 99.73% for three standard deviations. The standard deviation was recorded as 3r. Analysis showed that statistical tests revealed the expected relationship between precision and concentration of the sample where increasing concentration is followed by increasing precision (Fig. 6) as also found by Durance et al. [41]. Si, Mg and K measurement precision is more concentration dependant than of the other main elements.  could not be identified. A high moisture content of the sample leads to the underestimation of the target element concentrations as also observed for soils [42].

Effect of concrete surface condition on hXRF measurements
The replacement of air with water in pore-spaces increases photoelectric absorption in addition to scattering since the mass attenuation coefficients are higher for water than air [43]. The photoelectric absorption is the main photon interaction process at low energies (below 100 keV) [44]. In addition, an increase in pore saturation increases the density of the sample and so increases the photoelectric absorption effect. The fluorescence from light elements will be absorbed and scattered to a greater extent in water than fluorescence from heavy elements [45] such as for example Fe 2 O 3 . Drying improves the accuracy of the results, decreasing the effect of water saturation on the recordings.
b Fig. 5 Cutting and placement of concrete block inside the electron microscope for EDS surface measurements. Blue and yellow images are the area of interests (3 9 3 matrix (fields are labelled 00-22), with an individual SI field comprising 512 9 384 pixels) for concrete surface and concrete inner characterization. All nine fields of the area were analysed under the same analytical conditions

SO 3
Water saturated surface Air dried surface Dried surface Concrete (water saturated) surfaces were enriched in CaO, which suggests Ca leaching to the surface and carbonation during curing leading to precipitation of CaCO 3 (white precipitates on the SC1, SC2, SC3 surfaces, Fig. 4). Carbonation is favored by CO 2 ions in the liquid near the concrete surface. Nevertheless, the distinction of cement types in concretes SC1 and SC2 was evident with a decrease of CaO and an increase in SiO 2 and Al 2 O 3 in SC2 concrete compared with concrete SC1. Regarding the composition of SC2 and SC3 concrete surfaces, distinct concentrations can be seen among MgO, Al 2 O 3 and Fe 2 O 3 . Although the oxide concentrations at concrete surface deviate from the original composition of powder cements, they are within the broader cement classification region [46]. These oxides (CaO, SiO 2 , MgO, Al 2 O 3 and Fe 2 O 3 ) can be defined as characteristic oxides which can be used for concrete identification driven by cement type to facilitate fast screening of a large number of concrete elements where bulk composition of cement and aggregates is sought. For instance, cement type CEM III/B used for SC3 concrete production was recognized by the characteristic MgO concentration (which is relatively higher than that of SC1 and SC2 concretes) and a CaO:SiO 2 ratio of nearly 1. Figure 8 shows the effect of different measurement durations on the element concentration on the dried concrete surfaces. The surface dried state was selected for measurements based on results from Sect. 3.2.1. The chemical composition of concrete surfaces was measured for 30, 60, 90 and 120 s. The longer the measurement duration, the higher the received X-ray radiation dose. The results show that for all concretes and their elements, 30 s was sufficient and there was no significant change in element concentrations with longer measurement durations. This finding is in line with the study by Bourke and Ross [47]. Optimal beam time for routine in situ logging was c. 30 s per beam, both in mining plus mode [48] and in soil mode [47] for lithological discrimination of rock samples, ranging in composition from gabbro to rhyolite. Similarly, the study by García-Florentino et al. [15] showed that almost all the elements present in the certified reference material and historical mortars can be detected using the lowest real acquisition time (50 s), thus considering the low improvement of the level of detection increasing the real acquisition time. Table 3 shows element concentrations for SC1 dried concrete surface for three spots. The number of spots was initially nine, however, recordings were very similar (Supplementary Table S3). The goal was to evaluate the ability of hXRF to obtain consistent results when performing multiple measurements at different spots on the same sample. Each spot was scanned nine times and the average was reported. As a result, the agreement was considered acceptable for most elements among the spots, though for CaO and SiO 2 , large standard deviations were observed. Sometimes a problem with accuracy will be indicated by a high standard deviation (counting statistics) reported by the instrument [13]. These standard deviations are considered acceptable because the instrumental standard deviation for each measurement is low, in the order of c. ± 10 wt.%. The standard deviation in Table 4 was reported as a deviation for replicate analyses which was not instrumental standard deviation. Since they were \ 10 wt.%, with regard to the factory reported instrumental standard deviation, these deviations were found acceptable. Furthermore, the coefficient of variation ranging from 5.5% (Fe 2 O 3 ) to 33.5% (SiO 2 ) was as sensitive to the changes in the composition from one spot to other as it was the standard deviation.

Concrete's layer scanning
When the composition data from the surface and inner layers (matrix below the surface and single aggregate with size of 16 mm) were compared for SC1 dried concrete surface, Fig. 9, it was clear that the measurements were influenced by the aggregate vol.% and composition, more than at the surface. The surface was rich in CaO compared with the inner layers, suggesting the dominant presence of cement paste at the concrete surface (as the aggregates were silicious). This confirms Kreijger's theory and our hypothesis that b Fig. 7 Effect of surface state on the chemical composition of concrete surfaces the outer concrete skin is represented by the chemical composition of the cement paste [33].
Besides the chemical characterisation of the surface and inner layers, deeper inner layers have important characteristics. Variation of CaO and SiO 2 in inner concrete layers was due to a larger analysis spot size (5 mm X-ray beam spot size) used where it was likely to scan a mix of fine aggregates and cement paste, this effect being larger at deeper layers. Supplementary  Fig. S4 shows BSE images of top concrete surfaces SC1, SC2 and SC3 (Fig. S4 a, c, e) and a close-up view of the surfaces (Fig. S4 b, d, f). Additionally, it shows the appearance of sections at 100 and 200 lm below the surface of concrete SC1 where aggregate particles

SC3
30 sec 60 sec 90 sec 120 sec Fig. 8 Effect of measurement duration on the chemical composition of dried concrete surfaces become more dominant (Fig. S4 g, h). Compared with the surface layer, a doubled content of SiO 2 was already measured at 100 lm below the concrete surface. Further study in this region (0-100 lm) is required to identify the exact depth at which the composition of aggregates has a critical influence on the measurements by taking different particle size distributions (e.g. 0/8 mm, 0/16 mm, 0/32 mm) and wall effects (related to w/c ratio) into account.

Powder chemical composition: hXRF versus desktop XRF results
In this study non-destructive handheld XRF data were compared to data obtained by commonly used laboratory techniques as a way of evaluating the ability of the non-destructive technique to discriminate among different powdered materials and hardened concretes. The element concentration obtained by hXRF for powders was compared to data obtained by desktop XRF and EDS measurement techniques under laboratory conditions (Supplementary Table S4, Supplementary Fig. S5). Comparable concentration values within ± 5% were observed for all elements, particularly in the case of sand, limestone powder and CEM I. The results for the other powder materials are shown in Supplementary Tables S5, S6, S7. The results from hXRF and desktop XRF are presented side by side combined in Fig. 10 for CaO wt.% and SiO 2 wt.% to illustrate the accuracy for high elements concentrations. A comparison of the data showed the variation expressed as a standard error of the two techniques to be similar. This is expected since powders pass procedures such as drying, crushing, grinding, sieving and homogenising such that the composition analysis is fully comparable among different analytical techniques. Previous studies of Chinchón-Payá et al. [17,18] evaluating chlorides content on concrete surfaces using hXRF also observed consistent results between hXRF and standardised analytical acidsoluble titration method. Similarly, Hunt and Speakman [7] showed that the hXRF clay/sediment calibration calculates highly accurate concentrations for the mid-Z trace elements performing almost identical to the desktop XRF.

Concrete chemical composition: hXRF versus EDS results
Compared to materials of natural origin such as soils, rocks, clays, sediments, whose composition is strongly dependent on the result of the erosion of the earth's crust over a vast timespan [49], concrete production has always been man-controlled under standard specifications such as given by EN 206: 2013 [50]. In such standards, concrete ingredients need to satisfy the requirements given by the product norms for cement [46] and aggregates [51]. An example of such requirements is the composition of these ingredients, which must be certified in order to be used in concrete. This makes concrete chemical characterization relatively less complex compared with natural materials, considering large geochemical variations in natural materials. However, as the original project documents of the aged concrete structures are often not available, applying forensic materials engineering protocols for concrete characterization is essential. Non-destructive chemical composition characterization of concrete using hXRF was evaluated in this study since it can be performed in situ, unlike other destructive methods which have been employed to characterize and control the chemical compositions of cement materials and products in the laboratory conditions (desktop XRF, EDS, electron probe microanalysis, inductively coupled plasma optical emission spectrometer). The hXRF technique may speed up parent concrete identification in a structure prior to its demolition, and make it possible to classify concrete elements with different chemical composition or potential contamination.
Based on EN 197-1:2011 [46], the cement classes contain 34 clinker systems, 27 products in the family of common cements and 7 products in the family of sulfate resisting common cements. These cement classes are defined based on the content of the main constituents: clinker, blast furnace slag, silica fume, pozzolana, fly ash, burnt shale and limestone. These constituents (when present in a concrete mix as a blended cement) cannot be individually identified with hXRF. Nevertheless, the bulk cement chemical composition can be analysed with hXRF and typical ranges for oxides can be defined using the minimum and maximum values for cement types' compositions as shown in Supplementary Table S8.  This classification broadly supports the work of other studies on the identification of different materials based on their major element oxides. Since there are still no studies on systematic concrete composition analysis with hXRF available in the literature, examples of rocks close to concrete origin are used. Such an example is limestone, the main source of CaO to make clinker for cement manufacturing [52]. Different limestones (calcarenite, sandy or bioclastic limestones) were distinguished based on their element concentrations such as of Ca, K, Rb, Sr, Ti and Zr measured with hXRF [53]. The in situ use of hXRF allowed for the identification of Normandy Chalkstone sources (dedolomitized, quartz-glauconite, calcarenite, bryozoan and grainy chalkstone) preserved in archaeological sites based on concentrations of CaO, SiO 2 and Sr [54]. The hXRF has been used for quality control and suitability of quartz for an industrial lot at a ferrosilicon plant [55,56]. Critical elements (contaminants) such as iron, titanium and aluminum were analysed with hXRF in situ on a block of quartz (10cm-side) with an accuracy of 13% for Fe 2 O 3 , TiO 2 and CaO (accuracy for Al 2 O 3 was 24%) [55]. Classification of obsidian sources from volcanic glasses, rocks, artificial glasses or slags was also done based on the major element composition with hXRF [57].
Note that concrete characterization is generally more challenging compared with powder analysis. This is because powders are usually dried and homogeneous whereas the concrete specimens are much more heterogeneous (causing more sampling uncertainty). Concrete contains random assemblage of phases of different compositions and of varying shapes and sizes. Furthermore, concrete contains free and structural water attracting CO 2 as shown in Sect. 3.2.1 unlike powders. Preparation of samples may also introduce water. For example, the comparison of hXRF and EDS data ( Table 5) has shown that the concentrations of elements measured with hXRF are not always in line with those measured with EDS. It shows also that the results of hXRF were not systematically higher or lower. The discrepancy between hXRF and EDS measurements of the concrete surface can be attributed to several reasons. To prepare a concrete surface for EDS analysis it is necessary to cut a block that can fit EDS stage (Fig. 5). Cutting requires wetting. This step might attract CO 2 and further precipitation of carbonates on the concrete surface. Drying concrete surface eliminates matrix issues related to water; however, structural water and carbonates (CaCO 3 ) may remain. Fine grinding of the concrete surface with ultrafine sandpaper (P-grade 2000) of grit size smaller than 10 lm can be done for improved sample preparation (e.g. removing surface CaCO 3 ). Use of coarser sandpaper may lead to the characterization of the cement mortar rather than the cement paste since a double content of SiO 2 is measured at 100 lm below the concrete surface (Fig. 9). The two techniques, hXRF and EDS, have also different settings such as, more controlled working conditions during EDS analysis (vacuum vs air). Moreover, having used the standardless method instead of standards-based microanalysis could lead to a large difference especially for the elements having low concentrations [58]. Further experiments, using a broader range of cements, would help to establish a greater degree of accuracy on this matter.
Regarding concrete sampling uncertainty in the laboratory conditions, in order to produce accurate element concentrations, hXRF analysis has been performed on flat concrete surfaces and multiple measurements were done at different spots on the same sample (Sect. 3.2.3) to obtain good quality estimates similarly to rocks bulk analysis [59,60]. Considering the concrete surface preparation of existing concrete structures exposed for years under different environmental conditions in the field, a proper cleaning of the surface (free from dust, salts, loose material, surface coatings, membranes, repair products, injection materials) and surface preparation without removal of concrete skin (cement paste) are key tasks for future studies.

Conclusions
In this experimental study, non-destructive characterization of cementitious powders and concrete compositions using handheld XRF was performed. The effect of concrete surface condition (water-saturated, airdried and oven-dried), measurement duration (30, 60, 90, 120 s), repeatability on concrete surface and concrete's layer scanning were studied. The following conclusions were drawn: (1) Selected powders were tested using hXRF and BRUKER calibration method for cement. The hXRF technique shows an accuracy at a level of 5% or less for the determination of high element concentrations. (4) The curing of concrete affects the results. The water-saturated state shows that the CaO content is overestimated, specifically for concrete SC1 with CEM I (Portland cement). The Ca leaching to the concrete surface and subsequent carbonation are the main reason for this behaviour. This behavior is less obvious in concrete specimens SC2 and SC3 because of their lower CaO content compared with concrete SC1. Drying of surfaces prior to hXRF testing is shown to be more appropriate for chemical composition measurements. (5) CEM III/B (GGBFS cement) used for SC3 concrete production is recognized by a characteristic MgO concentration (which is higher than for SC1 and SC2 concretes). (6) A clear distinction can be made between concrete surface (represented by cement paste) and inner layers (represented by the cement mortar and aggregates) using SiO 2 and CaO as characteristic element oxides. There is a clear difference with regard to the high content of SiO 2 at the inner sections compared with the surface, confirming the siliceous composition of the aggregates used for production of concretes.
(7) Testing the composition of concrete surfaces using calibration method for cementitious powders is shown to be feasible. Although results for repeatability showed high precision, development of calibration method for typical concrete compositions and surface preparation protocols will improve the concrete surface measurements.
Finally, it can be concluded that the hXRF shows considerable potential as a non-destructive technique with minimal sample preparation for rapidly assessing chemical composition and quality of concrete. This would be helpful for ensuring identification of concrete at the source, improving the demolition efficiency and guiding the quality of recycled concrete. A follow-up study is designed to identify exact depth at which the composition of aggregates has a critical influence on the measurements taking into account different aggregate content, particle size distributions (e.g. 0/8 mm, 0/16 mm, 0/32 mm) and the wall effects (w/c ratio), as this is the case in practice for existing structures.