Method for estimating the compressive strength of concrete

Abstract

To provide a concrete compression strength estimation method using a laser-induced breakdown spectrometry that is hardly affected by self-absorption.SOLUTION: The compression strength of concrete is estimated by irradiating a mortar part of concrete with a laser pulse to induce a plasma and measuring an emission spectrum of the plasma that includes ion and atom beams of the same element whose self-absorption is weak, subtracting the average value of spectrum intensity from the measured spectrum and then performing a partial least square regression in which the spectrum intensity divided by a standard deviation is adopted as an explanatory variable and the average value of the compression strength determined by a compression strength test is adopted as an objective variable, and multiplying the coefficient determined by this partial least square regression to the measured spectrum.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present invention relates to a method for estimating the compressive strength of concrete. More specifically, it relates to a method for estimating the compressive strength of concrete using laser-induced breakdown spectroscopy.

Background Art

[0002] As one method for non-destructively estimating the compressive strength of a solidified body, a method for estimating the compressive strength using laser-induced breakdown spectroscopy (LIBS) has been proposed (for example, Patent Document 1). In this estimation method, a laser pulse is irradiated onto a solidified body with known components to induce plasma, and the strength of the solidified body is measured based on the emission intensity ratio between the neutral atomic line and the ionic line of a specific component element (specifically, calcium) in the emission spectrum of the plasma. That is, by focusing on the fact that the ratio of the spectral intensities of two calciums is proportional to the compressive strength, the compressive strength is estimated.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Non-Patent Documents

[0004]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, regarding the compressive strength estimation method using LIBS described in Patent Document 1, the present inventors conducted experiments using concrete and found that element-specific emission lines generated inside the plasma were absorbed by atoms of the same element in the peripheral area of ​​the plasma, resulting in a phenomenon called self-absorption where the spectral intensity appears to decrease. This made it impossible to appropriately estimate the compressive strength using the emission intensity ratio.

[0006] Generally, it is said that when laser light is shone onto concrete, the resulting shock wave heats the particles under laser ablation, generating plasma. The formation of the shock wave depends on whether the particles emitted from the sample surface can obtain sufficient reaction force. In other words, it depends on the hardness of the surface. From this, it can be seen that the hardness of the surface correlates with the temperature of the plasma. Since the temperature of the plasma correlates with the ratio of ion beams to atomic beams of the same element (emission intensity ratio), it was thought that by using the emission intensity ratio of neutral atomic beams to ion beams of the same element, a correlation with the strength of the material could be obtained.

[0007] However, by fabricating concrete test specimens with different compressive strengths and determining the emission intensity ratio of the neutral atomic line and ionic line of calcium, a major component element of cement, as described in the embodiment of Patent Document 1, it was found that the emission line of calcium becomes constant with respect to the compressive strength of the concrete due to self-absorption. In other words, it was found that when self-absorption occurs in the emission line of calcium, the correlation between the emission intensity ratio of the calcium atomic line and ionic line used for compressive strength estimation and the compressive strength is weak. Prior art document 1 reports that a correlation between the emission intensity ratio and compressive strength was obtained by calculating a correction coefficient for self-absorption and correcting the emission intensity. However, this experiment revealed for the first time that even when the same correction was performed, no correlation was obtained between the emission intensity ratio and the compressive strength.

[0008] The present invention aims to provide a method for estimating the compressive strength of concrete using laser-induced breakdown spectroscopy, which is less susceptible to the effects of self-absorption. [Means for solving the problem]

[0009] To achieve this objective, the concrete compressive strength estimation method of the present invention involves irradiating the mortar portion of the concrete with a laser pulse to induce a plasma, and in the emission spectrum of the plasma... Either magnesium or iron The method involves measuring the spectrum containing both ion and atomic lines, subtracting the average spectral intensity from the measured spectrum, and then dividing by the standard deviation of the spectral intensity to obtain the explanatory variable. A partial least-squares regression is then performed using the average compressive strength obtained from a compressive strength test as the dependent variable. The coefficient obtained from the partial least-squares regression is then multiplied by the measured spectrum to estimate the compressive strength of the concrete.

[0011] Furthermore, it is preferable to perform partial least-squares regression using the spectra classified as mortar, obtained by classifying materials using multivariate analysis with spectral data in the ultraviolet wavelength range containing magnesium, silicon, and calcium. [Effects of the Invention]

[0012] According to the concrete compressive strength estimation method of the present invention, the compressive strength of concrete can be estimated using laser-induced breakdown spectroscopy.

[0013] Furthermore, when performing PLS regression using spectra classified as mortar after material classification, it becomes possible to estimate the compressive strength of concrete more accurately. [Brief explanation of the drawing]

[0014] [Figure 1] This is a schematic diagram showing an example of an experimental setup for performing LIBS on concrete test specimens. [Figure 2]These are emission spectra obtained by laser irradiation in the wavelength range of 393-425 nm, where (a) represents mortar, (b) represents limestone, (c) represents mountain sand, (d) represents the first principal component coefficient, and (e) represents the second principal component coefficient. [Figure 3] This is a score plot obtained using a simple Bayesian classifier to determine the classification region and principal component analysis using spectra in the 393-425 nm wavelength range measured for each material. The square, circular, and triangular plots represent the principal components obtained from the spectra of mortar, limestone, and mountain sand, respectively, and the solid lines indicate the classification boundaries. [Figure 4] This figure shows the compression intensity dependence of the emission intensity ratio of calcium ion beams (396.8 nm) and atomic beams (422.7 nm), as well as the emission intensity ratio corrected for the effect of self-absorption. (a) shows the emission intensity ratio of calcium ion beams and atomic beams, and (b) shows the correlation with the corrected emission intensity ratio. [Figure 5] These are spectral diagrams in the 245-315 nm wavelength range, where trace element ion and atomic beams are observed, obtained by irradiating the mortar portion of concrete with laser light using a single laser. (a) shows the mortar, (b) shows the limestone, (c) shows the mountain sand, (d) shows the first principal component coefficient, and (e) shows the second principal component coefficient. [Figure 6] This score plot shows the results of classifying principal components obtained by principal component analysis using spectra in the wavelength range of 245-315 nm using a simple Bayesian classifier. (a) shows the results for laser irradiation 1-50, and (b) shows the results for laser irradiation 51-200. The square, circular, and triangular plots represent the principal components obtained from the spectra of mortar, limestone, and mountain sand, respectively, and the solid lines indicate the classification boundaries. [Figure 7] This graph shows the dependence of the emission intensity ratio of magnesium ion beams and atomic beams on the compression intensity, with (a) representing single-pulse irradiation and (b) representing double-pulse irradiation. [Figure 8] This figure shows the relationship between the estimated mean squared error and the number of components, obtained through 10 tolerance checks to determine the optimal number of components for partial least squares regression. (a) shows the case of single-pulse irradiation, and (b) shows the case of double-pulse irradiation. [Figure 9] It is a diagram showing the correlation between the estimated result and the measured result of the compressive strength when six partial least squares components are used.

Embodiments for Carrying Out the Invention

[0015] Hereinafter, the configuration of the present invention will be described in detail based on embodiments.

[0016] The method for estimating the compressive strength of concrete according to the present invention irradiates a laser pulse on the mortar part of the concrete to induce plasma, measures a spectrum including ion lines and atomic lines of the same element with weak self-absorption in the emission spectrum of the plasma, subtracts the average value of the spectrum intensity from the measured spectrum, and then divides by the standard deviation of the spectrum intensity to obtain an explanatory variable. Partial least squares regression is performed with the average value of the compressive strength obtained in the compressive strength test as the target variable, and the coefficient obtained by the partial least squares regression is multiplied by the measured spectrum to estimate the compressive strength of the concrete.

[0017] Concrete is composed of mortar and coarse aggregate. Further, the mortar is composed of cement paste in which water and cement have reacted and fine aggregate. Generally, the compressive strength of concrete is determined by the hardness of the cement paste, which has the lowest hardness among the materials constituting the concrete. Therefore, in order to estimate the compressive strength without being affected by the distribution of the coarse aggregate, it is considered that a method of estimating the hardness using the spectrum measured for the cement paste part is effective. However, since the fine aggregate used in concrete includes materials with a particle size on the order of micrometers, no matter how small the laser is focused, it is impossible to irradiate only the cement paste part, and the mortar composed of the fine aggregate and the cement paste will be ablated.

[0018] Therefore, a laser beam is irradiated onto the mortar portion to induce plasma. The irradiation of the laser beam is performed, for example, visually, toward the mortar portion (a portion where no large aggregate exists). When the pulsed laser beam is focused on the surface of the object, ablation causes the target material particles to be released from the material surface. The released particles interact with the shock wave generated by ablation, resulting in adiabatic compression and ionization of some of the particles. The intensity of the shock wave generated at this time depends on the reaction force obtained from the surface of the object, and the harder the object, the stronger the shock wave is formed. Therefore, the harder the object, the more adiabatic compression of the target material particles occurs, increasing the temperature and electron density of the plasma. As a result, the emission intensity of the spectrum observed in the plasma changes. That is, the emission intensity ratio of the ion line and the atomic line of the same element is proportional to the excitation temperature of the plasma, and it is considered that the compressive strength of concrete, which is correlated with hardness, can be estimated from the emission intensity of the plasma.

[0019] On the other hand, the inventors fabricated concrete test specimens with different compressive strengths and investigated the spectral characteristics when estimating the compressive strength of concrete using the emission intensity ratio of calcium ion lines and atomic lines, which is the main component of concrete, as specifically mentioned as an example in Patent Document 1. As a result, they revealed that the emission intensity ratio of calcium ion lines and atomic lines becomes constant with respect to the compressive strength of the concrete (the emission intensity ratio of calcium saturates, and a discrepancy occurs between the estimated compressive strength and the actual compressive strength). Furthermore, it was found that the spectral width of calcium increased at this time, indicating that calcium emission was being self-absorbed. In other words, experiments using concrete revealed that when element-specific emission lines generated inside the plasma are absorbed by atoms of the same element in the peripheral part of the plasma, and the spectral intensity appears to decrease (self-absorption), the correlation between the emission intensity ratio of ion lines and atomic lines of a specific element and the compressive strength disappears, making it impossible to appropriately estimate the appropriate compressive strength using the emission intensity ratio. Moreover, self-absorption was observed more significantly as the concentration of the element being measured increased, or as the plasma temperature increased. Here, the ratio of emission intensities of atomic and ionic lines of the same element is proportional to the plasma excitation temperature. Furthermore, the compressive strength of concrete, which correlates with hardness, is also proportional to the plasma excitation temperature. From this, it can be seen that the higher the compressive strength of concrete, the more susceptible it becomes to the effects of self-absorption. From these findings, it was discovered that when self-absorption occurs in the emission lines, the correlation between the ratio of emission intensities of calcium atomic and ionic lines used to estimate compressive strength and the compressive strength is weak.

[0020] Therefore, we attempted to solve this problem by correcting the luminescence intensity to account for self-absorption. However, it became clear that in the case of concrete with a high calcium concentration, the self-absorption of calcium is significant, making it impossible to accurately estimate the compressive strength.

[0021] Furthermore, as an alternative to the self-absorption correction described above, methods such as using a tunable laser to excite atoms located in the lower energy levels that cause self-absorption and thereby reduce their number density, or utilizing de-excitation processes such as Penning ionization, can be considered. These are excellent methods for experimentally suppressing self-absorption and are effective for laboratory measurements. On the other hand, the experimental setups for these methods are complex and not suitable for on-site measurements.

[0022] Furthermore, we sought elements that are less prone to self-absorption and in which both ionic and atomic beams can be observed, and attempted to estimate compressive strength without complicating the experimental method by using these elements. For example, ordinary Portland cement, the most commonly used, is rich in calcium and silicon, and contains trace amounts of aluminum, iron, titanium, sodium, magnesium, potassium, and sulfur. Spectra originating from these elements are observed in the ultraviolet and near-infrared wavelength ranges. Therefore, we irradiated the mortar portion of concrete with laser light to induce a plasma and investigated whether trace element ionic and atomic beams could be observed in the plasma emission spectrum in the range of 250 to 900 nm. As a result, as shown in Figure 5, magnesium ionic beams (279.55 nm, 280.27 nm) and atomic beams (285.21 nm) were observed. The upper energy levels of the two ionic beams are very close during their excitation process, and self-absorption occurs, changing the emission intensity ratio and full width at half maximum of the two ionic beams. In this experiment, the emission intensity ratio and full width at half maximum of the two ion lines did not change under measurement conditions such as the test specimen and laser energy. These spectra are resonance lines, but it is thought that no self-absorption occurred due to the low magnesium concentrations in the water and cement. In addition, iron ion lines (263.10 nm) and atomic lines (302.58 nm) were also observed. These iron ion and atomic lines, like the magnesium emission lines, are correlated with compressive strength and can be considered emission lines with little self-absorption.

[0023] Therefore, the inventors measured the spectra in the ultraviolet wavelength range where magnesium ion and atomic beams are observed, and attempted to estimate the compressive strength using the emission intensity ratio of the ion and atomic beams. However, contrary to initial predictions, experimental results showed that the emission intensity ratio of magnesium remained constant within the range of standard deviation with respect to the compressive strength. Generally, the emission intensity of each element increases linearly with respect to the concentration of that element, so the emission intensity ratio of that element does not depend on the concentration. Therefore, the dependence of the emission intensity ratio on compressive strength should not change whether the magnesium concentration of each concrete test specimen is the same or different. However, in cases where mountain sand and cement paste of different hardness are ablated simultaneously, it is conceivable that the atomic excitation temperature of the plasma changes depending on the ratio of the amount of sand and cement paste. In other words, when estimating the compressive strength of composite materials such as concrete, it is thought that the emission intensity ratio depends not only on its mechanical properties but also on the amount ratio of the materials being ablated (in other words, the atomic excitation temperature of the plasma). On the other hand, since there are mountain sand particles with a particle size smaller than the focusing diameter of the laser beam, this amount ratio cannot be controlled experimentally. Furthermore, while it is possible to classify mortar and sand with relatively large particle sizes in material classification, it is difficult to classify sand with particle sizes smaller than the focusing diameter of the laser beam. For this reason, although magnesium is thought to have low self-absorption and that the emission intensity ratio correlates with compressive strength, the experimental results show that compressive strength cannot be appropriately estimated using only the emission intensity ratio.

[0024] Based on these results, we found that the compressive strength of concrete cannot be appropriately estimated using the emission intensity ratio of ion and atomic beams of specific elemental components in the plasma emission spectrum.

[0025] Based on this knowledge, the inventors have found that they can appropriately estimate the compressive strength of concrete by irradiating concrete with laser pulses to induce plasma, measuring the emission spectrum of the plasma which contains emission lines of elements that correlate with the compressive strength of concrete and have low self-absorption, and performing data analysis using partial least squares (PLS) regression, which takes into account not only the peaks of the spectrum but also its overall shape, i.e., not only the emission intensity of magnesium but also the spectral information of other elements.

[0026] Here, the emission lines of elements that correlate with the compressive strength of concrete and have low self-absorption are elements that are unlikely to undergo self-absorption and for which both ionic and atomic lines are observed. These are not limited to specific elements, but as mentioned above, they are elements that are present in trace amounts in the mortar components of concrete and for which both ionic and atomic lines are observed. For example, to give just one example, in ordinary Portland cement, which is the most commonly used, these are magnesium or iron.

[0027] In the case of concrete using ordinary Portland cement, both atomic and ionic lines of magnesium and iron are clearly observed in the plasma emission spectrum, in addition to the main component calcium. These magnesium and iron are present in Portland cement only in trace amounts. Therefore, the self-absorption of the emission lines of magnesium and iron is weak, and it is thought that they can be used to estimate compressive strength.

[0028] Therefore, for example, when focusing on magnesium, a laser beam is visually shone onto the mortar area (an area without large aggregates) to induce plasma. The spectrum in the ultraviolet wavelength range where both magnesium ion and atomic lines are observed in the plasma emission spectrum is measured, and data analysis is performed using PLS regression to appropriately estimate the compressive strength of the concrete.

[0029] In the measurement, the same location on the concrete mortar is irradiated with laser light, for example, 200 times. Irradiation methods include using a single laser (single-pulse measurement) and using two lasers with a time difference to irradiate with two laser beams (double-pulse measurement). The emission spectra from the 51st to 200th laser irradiations are then measured. This is because, in the initial stages of laser irradiation, for example, the 1st to 50th irradiations, the emission intensity of magnesium and silicon fluctuates significantly. Therefore, it is preferable to use the emission spectra from the 51st to 200th irradiations, after which surface contamination has disappeared, no longer affecting laser ablation, and the atomic excitation temperature of the plasma has stabilized. In other words, using the data from the 51st to 200th laser irradiations from the spectral dataset obtained for each measurement location is preferable for obtaining a stable dataset. Furthermore, from the mean squared error (MSE) of the compressive strength estimates in the PLS regression described later, and from the comparison of the estimated and measured values, it was evaluated that there is no significant difference in compressive strength estimation between single-pulse and double-pulse measurements. Therefore, whether it is single-pulse or double-pulse measurement is not a particularly essential element in this invention.

[0030] Then, the spectra in the ultraviolet wavelength range where magnesium ion and atomic lines are observed in the plasma emission spectrum are measured. The mean and standard deviation of the spectral intensity are calculated for the measured spectra, and the value obtained by subtracting the mean from each spectrum and dividing by the standard deviation (hereinafter referred to as autoscaling) is used as the explanatory variable. A PLS regression is then performed with the mean value of the compressive strength obtained from the compressive strength test as the dependent variable, and the coefficients obtained from the PLS regression (i.e., PLS variables) are multiplied by the measured spectra to estimate the compressive strength of the concrete.

[0031] The formula for estimating compressive strength using PLS regression is expressed as follows (1). Estimated compressive strength = coefficient matrix × spectrum …(1) Here, the coefficient matrix is ​​a matrix (1 row, n columns, where n is the number of pixels in the spectral data) where the coefficient A and the emission intensity x of a certain emission line are combined. The coefficient matrix is ​​expressed as a linear sum of the features (principal components) obtained by PLS, and wavelength components with larger coefficients contribute more to the estimated value.

[0032] Here, PLS regression may be performed using all datasets measured from the 51st to the 200th time, but it is preferable to perform the analysis using only the data from when the cement paste was ablated more. This is not an essential element of the present invention, and in some cases, the analysis may be performed using all measured datasets. In other words, compressive strength estimation by PLS regression can be performed regardless of whether material classification is performed, but concrete compressive strength can be estimated more appropriately by performing material classification and using datasets classified as mortar.

[0033] Material classification can be achieved, for example, by measuring the spectrum containing characteristic emission lines for each material, performing multivariate analysis using this spectrum, calculating feature quantities that are weighted by the intensity of the characteristic emission lines and are independent of each other (orthogonal), and then comparing the relative magnitudes of these feature quantities.

[0034] Furthermore, while this embodiment uses Principal Component Analysis (PCA) as the multivariate analysis for calculating features, it is not limited to this, and other methods such as PLS or Support Vector Machines (SVM) may also be used. In addition, a learner is used to set the threshold required for classification when classifying materials based on the relative sizes of the features. For example, this embodiment uses a Naive Bayes classifier (NBC), but other suitable methods include Support Vector Machines and Decision Trees.

[0035] In this embodiment, material classification is performed using PCA and NBC. The analysis uses spectra obtained by continuously irradiating a single point with laser light for mortar, limestone, and mountain sand. PCA performs autoscaling as a pretreatment. For example, for each laser irradiation spectrum measured for the test specimen, the lowest observed spectral intensity in the wavelength range is subtracted from the spectral intensity as a baseline, or a baseline obtained by fitting a cubic polynomial is subtracted from the spectral intensity. The mean and standard deviation of the spectral intensity are then calculated for each material's spectral dataset, and the value obtained by subtracting the mean from each spectrum is divided by the standard deviation, i.e., autoscaling is performed. PCA uses the entire observed spectrum as multiple variables to classify materials. This makes it possible to appropriately classify materials with similar compositions. Here, it is convenient to use a classifier to automatically determine thresholds and classify the spectra for each material.

[0036] When performing material classification, one example is to perform autoscaling after accumulating and averaging the spectra classified as mortar using the NBC model, provided that the number of spectral data classified as mortar using the NBC model is 80% or more of the total number of data used at each measurement location. In this embodiment, if the proportion of data classified as mortar to the total data is less than 80%, it is considered undesirable to use it for PLS regression because the small number of cumulative averages is thought to affect spectral noise in the PLS regression, although this is not the only example. On the other hand, when material classification is not performed, autoscaling is performed after accumulating and averaging all measured spectral datasets at each measurement location. Furthermore, the goodness of fit, i.e., generalization performance, for data with unknown variables was evaluated using MSE, which shows the difference between predicted and measured values ​​when cross-validation is performed 10 times in PLS regression, and it was shown that there were no problems.

[0037] In data analysis, by using a simple Bayesian classifier with the principal components obtained by PCA as variables, each spectrum can be appropriately classified into three materials. For example, using a spectral dataset measured for each material, the measured spectra can be classified into mortar, limestone, and mountain sand.

[0038] According to verification results by the inventors, when compressive strength estimation was performed using PLS regression with variables deemed optimal for spectral data of elements in which no self-absorption was observed and ion and atomic beams were observed, the estimated compressive strength values ​​were in general agreement with the values ​​obtained in the compressive strength test. From these results, it was found that by using PLS regression, compressive strength estimation can be performed for concrete without being affected by the self-absorption of the elements used in the estimation. Moreover, since PLS regression is robust to spectral noise, it is considered applicable to on-site measurements. Here, "using variables deemed optimal" means "how many variables, which are coefficients obtained by PLS regression, to select." This selection method is, for example, to select the number of PLS ​​variables that result in the smallest MSE value, and is a generally known method for performing PLS regression.

[0039] The above-described embodiment is merely one example of a preferred embodiment of the present invention, and is not limited thereto. Various modifications can be made without departing from the spirit of the invention. For example, in the above-described embodiment, material classification and concrete compressive strength estimation are performed sequentially using spectra containing the same elements. For example, material classification is performed using spectral data in the ultraviolet wavelength range containing magnesium, silicon, and calcium, and compressive strength estimation is performed by PLS regression using the dataset classified as mortar. However, this is not an essential element of the present invention, and is not particularly limited to using spectra containing the same elements or processing sequentially. As long as the measurement locations are the same, spectral data containing different elements suitable for material classification and concrete compressive strength estimation may be used, or spectral data for compressive strength estimation may be taken after spectral data for material classification. In other words, to estimate the compressive strength of concrete, it is sufficient to measure the spectrum containing ion and atomic lines of the same element with weak self-absorption within the plasma emission spectrum. Measuring spectra containing combinations of iron and silicon, calcium and silicon, or silicon and aluminum, where only atomic lines can be observed, which are effective in making material classification more accurate, is not a particularly essential element in this invention.

[0040] Furthermore, in the above-described embodiment, data analysis is performed using PLS with a material-classified dataset to estimate the compressive strength of concrete. However, the method is not limited to this, and compressive strength may be estimated using only the mortar dataset obtained by material classification and a partial least squares method. Moreover, compressive strength may be estimated using PLS with all spectral data without performing material classification. In this case as well, the compressive strength of concrete can be estimated appropriately. [Examples]

[0041] We verified the method for estimating the compressive strength of concrete using the laser-induced breakdown spectroscopy method of the present invention.

[0042] [Production of test specimens] Five mortar materials with different compressive strengths were prepared by varying the ratio of water to cement. These were poured into rectangular molds and cured for one day. After demolding, the mortar was cured in a water tank maintained at 20°C. Seven days or more after placement, the concrete was removed from the water tank and hollowed out to create multiple cylindrical mortar test specimens (concrete cores) with different compressive strengths. Each mortar test specimen was 100 mm in diameter and 200 mm in length. The mortar test specimens were stored indoors, and compressive strength tests were conducted two years after placement. The concrete was cured in water for at least 48 hours prior to the compressive strength test and removed immediately before the test. For the compressive strength test, a universal testing machine (Marui Corporation Ltd., MS-100-BC) was used to measure the maximum load, and the compressive strength was determined from the dimensions and maximum load of the test specimen. Compressive strength tests were performed using two test specimens prepared for each mix, and the average value was used as the compressive strength of the concrete for that mix. The concrete mixes for the test specimens are shown in Table 1. Tap water was used for the water, ordinary Portland cement for the cement, mountain sand for the fine aggregate, and limestone with a maximum size of 20 mm for the coarse aggregate. In addition, an air-entraining (AE) water-reducing agent and an AE agent were used to adjust the air content of each test specimen to 3.6-4.5%.

[0043] [Table 1]

[0044] A concrete test specimen (hereinafter referred to as Core 1) was placed between two rollers (arranged in parallel) 2 that rotated by a motor drive of the experimental apparatus 4 shown in Figure 1. While the specimen was rotating, an Nd:YAG laser (532 nm, 10-50 mJ) 8 was focused onto Core 1 using a plano-convex lens 6 with a focal length of 150 mm, and the emission of plasma was measured with a spectrometer 12 and ICCD 14. This made it possible to irradiate a curved surface with laser light while maintaining a constant distance from the objective lens 6 to the laser focusing position. Here, the lens used for focusing the laser 8 and the plasma was common, and a coaxial optical system was used to separate the laser light and the emission of plasma using a dichroic filter 7 that reflects the laser light. In the figure, 3 is a motor, 9 is a mirror, 10 is a polarizing beam splitter, 11 is an optical fiber, 12 is a spectrometer, 13 is a computer for apparatus control, etc., 15 is a lens, and 16 is a delayed pulse generator.

[0045] Experimental apparatus 4 allows for both single-pulse measurement (one laser pulse) and double-pulse measurement (two laser pulses with a time difference). For example, the laser energy is set to 30 mJ for single-pulse measurement and 10 mJ for the first pulse and 30 mJ for the second pulse for double-pulse measurement. The laser pulse interval is set to 1 μs. Multiple spectra were obtained by repeatedly irradiating each location with laser light and measuring the spectrum after each pulse. When measuring spectra in the visible and ultraviolet wavelength ranges, the mortar area was irradiated 50 and 200 times respectively, and the spectrum was measured after each pulse.

[0046] For measurements in the visible and ultraviolet wavelength ranges, diffraction gratings with 1200 grooves / mm and 600 grooves / mm were used, respectively. A pulse delay generator (Stanford Research, DG645) 16 was used to repeatedly generate laser 8 at 10 Hz, and an external trigger signal was input to ICCD 14 to synchronize laser 8 and ICCD 14. The gate delay time of ICCD 14 was set to 0.5 u, and the gate width to 5 u.

[0047] For data analysis, numerical analysis software (MATLAB® R2021b) was used to perform baseline correction and fitting of spectra, and multivariate analysis using built-in functions. Because concrete has a non-uniform distribution of coarse aggregate and mortar, even when irradiating the same mortar area with laser light, the aggregate may also be irradiated. Furthermore, since mortar and aggregate have different hardnesses, it was considered necessary to use only the spectra measured for mortar in the data analysis. Therefore, in this experiment, PLS analysis was performed using only mortar data obtained by material classification of the measured spectral data.

[0048] PCA and NBC were used for material classification. The analysis used spectra obtained by continuously irradiating a single point with laser light for mortar, limestone, and mountain sand. As a pretreatment, for each laser irradiation of specimen No. 2, the lowest spectral intensity in the observed wavelength range was subtracted from the spectral intensity as a baseline, or a baseline obtained by fitting a cubic polynomial was subtracted from the spectral intensity. Then, for each material's spectral dataset, the mean and standard deviation of the spectral intensity were calculated, and the value obtained by subtracting the mean from each spectrum was divided by the standard deviation, i.e., autoscaling was performed. PCA uses the entire observed spectrum as multiple variables for material classification, making it possible to appropriately classify materials with similar compositions. While it is possible to classify mortar and aggregate using the principal components obtained by PCA as a threshold, this threshold changes depending on the experimental conditions. Here, NBC, which classifies data into multiple groups using explanatory variables, was used to automatically determine the threshold and classify the spectra for each material. Since the analysis in NBC assumes that the variables used are independent of each other, the first and second principal components, which are orthogonal to each other, were used as variables. Under the experimental conditions in this study, the sum of the contribution rates of the first and second principal components exceeded 90%, so only these two variables were used.

[0049] Figures 2(a)-(c) show the spectra obtained by irradiating mortar, limestone, and mountain sand with laser light. Limestone, being mainly composed of calcium carbonate, has a very high calcium concentration and is characterized by a large full width at half maximum in its calcium spectrum. Mountain sand is characterized by the clear observation of iron and silicon spectra. Iron and silicon spectra are also observed when measuring mortar, but their intensity is lower compared to the calcium spectrum.

[0050] Furthermore, Figures 2(d) and (e) show the coefficients of the first and second principal components when PCA was performed using spectral datasets obtained by irradiating mortar, limestone, and mountain sand with laser light. The contribution rates of the first to third principal components were 88%, 7.7%, and 0.99%, respectively. In other words, it was found that the materials could be roughly classified into three categories by using the first principal component. In the first principal component, the coefficients are high in the wavelength range where the spectra of iron, silicon, and calcium are observed, and the calcium spectrum has a shape that appears to be self-inverted. This is thought to be a result of the difference in the full width at half maximum of the calcium spectrum depending on the material.

[0051] Figure 3 shows the classification region by NBC and the score plot obtained when PCA was performed using the spectra measured for each material. The first principal component shows the greatest difference between mortar and mountain sand, indicating that the first principal component mainly classifies mountain sand and mortar. Similarly, the second principal component shows the greatest difference between mortar and limestone, indicating that the second principal component classifies mortar and limestone. In other words, the positions of the score plots for each material are far apart, and it can be evaluated that the three materials can be appropriately classified using the first and second principal components.

[0052] Next, for comparison, we examined the concrete compressive strength estimation method described in Patent Document 1. Figure 4 shows the compressive strength dependence of the emission intensity ratio of calcium ion beams (396.85 nm) and atomic beams (422.67 nm). The error bars in the figure indicate the standard deviation of the emission intensity ratio obtained when measurements were taken at 10 locations for each test specimen. From these results, the emission intensity ratio is estimated to be between 25 and 70 N / mm².2 The standard error remained constant within the specified range. This is thought to be because, due to the high calcium concentration, element-specific emission lines generated within the plasma are absorbed by atoms of the same element in the peripheral region of the plasma, resulting in a phenomenon called self-absorption, where the spectral intensity appears to decrease, and the emission intensity ratio saturates. This was also the case when the emission intensity ratio was corrected for the effect of self-absorption.

[0053] Therefore, the plasma emission spectrum, which includes ion and atomic lines of the same element with weak self-absorption, was measured. The measured spectrum was autoscaled and used as the explanatory variable. A PLS regression was then performed, with the average compressive strength obtained from the compressive strength test as the dependent variable. The compressive strength of the concrete was estimated by multiplying the variable obtained from the PLS regression by the measured spectrum.

[0054] In PLS regression, a linear regression model is created to predict the dependent variable by calculating a variable that maximizes the variance between the explanatory and dependent variables. In this example, PLS regression was performed using auto-scaled measured spectra as explanatory variables and the average compressive strength obtained from compressive strength tests as the dependent variable. When material classification was not performed, all measured spectral datasets were averaged for each measurement location before auto-scaling. When material classification was performed, if the number of spectral data classified as mortar using a model constructed with a simple Bayesian classifier was 80% or more of the number of data used for each measurement location, the spectra classified as mortar were averaged before auto-scaling. If the proportion of data classified as mortar to the total data was 80% or less, it was considered that the small number of cumulative averages would affect the spectral noise in the PLS regression, so it was not used for PLS regression. To evaluate the goodness of fit of the PLS regression model to unknown data, i.e., the generalization performance, the MSE obtained after 10 cross-validations was used.

[0055] To estimate compressive strength, ion and atomic beams that are less prone to self-absorption were used. The most commonly used ordinary Portland cement is rich in calcium and silicon, and contains trace amounts of aluminum, iron, titanium, sodium, magnesium, potassium, and sulfur. Spectra originating from these elements are observed in the ultraviolet and near-infrared wavelength ranges. Therefore, the mortar portion of the concrete was measured, and the spectra of trace element ion and atomic beams were investigated in the 250-900 nm range. As a result, as shown in Figure 5, magnesium ion beams (279.55 nm, 280.27 nm) and atomic beam (285.21 nm) were observed. The two ion beams have very similar upper energy levels during their excitation process, and self-absorption occurs, changing the emission intensity ratio and full width at half maximum (FMAX) of the two ion beams. In this experiment, the emission intensity ratio and FMAX of the two ion beams did not change under measurement conditions such as the test specimen and laser energy. These spectra are resonance lines, but since the magnesium concentrations in water and cement are low, it is thought that no self-absorption is occurring. Therefore, the compressive strength was estimated using spectra in the ultraviolet wavelength range where magnesium ion and atomic lines are observed.

[0056] Figures 5(a) to (c) show that atomic lines of calcium and aluminum were also observed in the 300-315 nm range. Although limestone is composed of highly pure calcium carbonate, it also contains trace amounts of magnesium, so magnesium ion and atomic lines were observed. Unlike the spectra in the visible wavelength range, the intensity of the mountain sand spectrum was the highest across the entire observed wavelength range, compared to mortar and limestone. This is because the spectra derived from magnesium, silicon, and iron, which are relatively abundant in mountain sand, are observed more strongly in the ultraviolet wavelength range than in the visible wavelength range.

[0057] Figures 5(d) and (e) show the principal components obtained when PCA was performed using the spectra obtained from the 51st to 200th laser irradiations in a dataset of spectra obtained when mortar, limestone, and mountain sand were irradiated with laser light. In the first principal component, it can be seen that the coefficients are high in the wavelength range in which the spectra of silicon, calcium, and aluminum are observed. This is because the emission intensity of calcium is high only when limestone is measured, and the emission intensity of silicon and aluminum differs depending on the material, so the spectra derived from calcium and silicon are useful indicators when classifying materials. The contribution rates of the first and second principal components were 92% and 3.8% for single-pulse measurements and 75% and 15% for double-pulse measurements, indicating that the first principal component can almost completely classify the three materials. Therefore, NBC was performed using the first and second principal components as variables. By using a simple Bayesian classifier with the principal components obtained by PCA as variables, the measured spectra were classified into mortar, limestone, and mountain sand. Classification results using spectral datasets measured for each material revealed that each spectrum could be appropriately classified into one of three materials.

[0058] Figure 6 shows the score plots obtained when PCA was performed using the classification region by NBC and the spectra measured for each material. The first principal component shows the greatest difference between mortar and mountain sand, indicating that the first principal component mainly classifies mountain sand and mortar. Similarly, the second principal component shows the greatest difference between mortar and limestone, indicating that the second principal component classifies mortar and limestone. The shape of the coefficient of the second principal component, which appears as if self-inversion of the spectrum has occurred in the wavelength range where the calcium spectrum is observed, is thought to reflect the difference in the full width at half maximum of the calcium spectrum when measuring mortar and limestone. The positions of the score plots for each material are far apart from each other, and it was evaluated that the three materials can be appropriately classified using the first and second principal components. If mountain sand or limestone with small particle sizes is present at the laser irradiation site, it is expected that mortar and mountain sand will be ablated simultaneously, and a mixture of spectra of mortar and limestone or mortar and mountain sand will be measured. In such cases, the score plots are thought to be located near the boundary of the region classified by NBC.

[0059] Because the spectrum of limestone differs significantly in shape from that of mortar and mountain sand, the spectral scores obtained from measuring limestone were plotted at a different position than those obtained from measuring mortar and mountain sand. The scores for mortar and mountain sand differed in the coefficients of the second principal component. In the second principal component, the coefficients were high in the wavelength range where magnesium ion beams are observed, suggesting that the difference in magnesium emission intensity when measuring mortar and mountain sand is reflected in the coefficients of the principal component.

[0060] When using the dataset from the 1st to 50th laser irradiations (Figure 6(a)), the spectral scores for mortar and mountain sand were distributed close together. Therefore, the classification results differed between the dataset from the 1st to 50th laser irradiations and the dataset from the 51st to 100th laser irradiations (Figure 6(b)). Although not shown, the classification results for the datasets from the 100th to 150th laser irradiations and the 151st to 200th laser irradiations were the same as those for the dataset from the 51st to 100th laser irradiations. In the classification results using the dataset from the 1st to 50th laser irradiations, the classification boundary between mountain sand and limestone disappeared, and the range classified as mortar widened. Since the spectra of mortar and mountain sand have similar shapes, it is considered inappropriate for the classification boundary between mountain sand and limestone to disappear. Thus, the reason why the classification results differed depending on the number of laser irradiations is that the emission intensity of magnesium and silicon fluctuated greatly during the 1st to 50th laser irradiations. This is thought to be because surface contamination affected the laser ablation, causing the plasma atomic excitation temperature to be unstable. Therefore, we decided to classify the material and calculate the emission intensity using the data from the 51st to 200th laser irradiations among the spectral datasets obtained for each measurement location.

[0061] Figure 7 shows the compressive strength dependence of magnesium ion and atomic beam intensities in single-pulse and double-pulse measurements. When material classification was not performed, the emission intensity ratio was calculated from the cumulative average of all spectral data for each measurement location. When material classification was performed, the emission intensity ratio was calculated from the cumulative average of spectra classified as mortar for each measurement location. The error bars in the plot indicate the standard deviation of the emission intensity ratio obtained at each measurement location. Regardless of whether material classification was performed or not, the magnesium emission intensity ratio remained constant within the standard deviation range with respect to compressive strength in both single-pulse and double-pulse measurements. Although it was thought that the magnesium emission intensity ratio would correlate with compressive strength, this experimental result showed that compressive strength cannot be estimated using the magnesium emission intensity ratio. The emission intensity of each element increases linearly with respect to the concentration of that element, so the emission intensity ratio of that element generally does not depend on the concentration. Therefore, the compressive strength dependence of the emission intensity ratio should not change whether the magnesium concentration is the same or different for each concrete test specimen. However, when mountain sand and cement paste of different hardnesses are ablated simultaneously, the atomic excitation temperature of the plasma may change depending on the ratio of mountain sand to cement paste. In other words, when estimating the compressive strength of composite materials such as concrete, the emission intensity ratio is thought to depend not only on its mechanical properties but also on the ratio of the materials being ablated. On the other hand, since there are mountain sand particles with a particle size smaller than the focusing diameter of the laser beam, this ratio cannot be controlled experimentally. In material classification, it is possible to classify mortar and mountain sand with relatively large particle sizes, but it is difficult to classify mountain sand with a particle size smaller than the focusing diameter of the laser beam. Therefore, we attempted to estimate the compressive strength by taking into account not only the emission intensity of magnesium but also the spectral information of other elements using multivariate analysis. That is, we came up with the idea of ​​performing data analysis using not only the peaks of the spectrum but also their entire shape.

[0062] To determine the optimal number of components for PLS regression, the component number dependence of MSE was calculated through 10 cross-validation trials. As shown in Figure 8, in single-pulse measurements, MSE decreased with the number of components used in PLS regression, reaching its minimum at around 11 components when no material classification was performed, and around 6 components when material classification was performed. Beyond this minimum, MSE remained roughly constant. While MSE sometimes increased or decreased with a large number of components, this depended on the dataset split during cross-validation, and this result changed with each cross-validation trial. The minimum MSE was around 0.6 regardless of whether material classification was performed or not. In double-pulse measurements, without material classification, MSE increased with the number of components before decreasing. On the other hand, when material classification was performed, MSE decreased monotonically and remained roughly constant above 16 components. The minimum MSE was slightly lower with material classification, at around 0.25. Generally, MSE decreases with the number of components. The temporary increase in MSE when material classification was not performed is thought to be because the emission intensity of the spectrum changed significantly when laser light was irradiated onto aggregate and when laser light was irradiated onto mortar, and this fluctuation could not be explained by the PLS component. Compared to single-pulse measurement, double-pulse measurement showed a tendency for the emission intensity of each element to vary with each irradiation. It is thought that performing material classification has the effect of suppressing such variations in emission intensity.

[0063] Figure 9 shows the results of estimating compressive strength using the first to sixth variables of PLS. In the case of double-pulse measurement, the number of components that yields the lowest MSE is 16 or more, but since there is no significant difference in MSE for 6 or more components, there is a possibility that the estimation results are overfitting. Therefore, the compressive strength was estimated with 6 components, which yields the minimum MSE. When the estimated compressive strength was linearly approximated, parameters such as the determination error and slope of the approximation line did not differ significantly between single-pulse and double-pulse measurements. This is thought to be because the MSE of the PLS regression in single-pulse and double-pulse measurements was roughly the same. The slope of the line when the estimated compressive strength was linearly approximated was less than 1. This is thought to be partly due to the use of the compressive strength obtained from the compressive strength test as the objective variable. In PLS regression, since the results of the compressive strength test were used as the objective variable, the analysis was performed based on the assumption that the compressive strength is constant regardless of the measurement location. However, the hardness of mortar, which is generally correlated with the compressive strength of concrete, has large local variations. Therefore, if we obtain data on the hardness of the mortar at each measurement point using LIBS and set this as the target variable, the slope of the approximate line is expected to approach 1. In this analysis, we created a single regression model to estimate the compressive strength, but it is expected that the estimation accuracy will be further improved by measuring the compressive strength of many test specimens with different compressive strengths and compositions using LIBS and then using multiple regression models to estimate the compressive strength of that dataset.

[0064] A comparison of estimated and measured MSE and compressive strength suggests that there is no significant advantage between single-pulse and double-pulse measurements for estimating compressive strength. Therefore, the dilution effect in double-pulse measurements is not considered to contribute significantly to the estimation of mechanical properties such as compressive strength and hardness.

[0065] As described above, in the data analysis, the measured spectra were classified into mortar, limestone, and mountain sand by using NBC as a variable, with the principal components obtained by PCA being used. From the classification results using the spectral dataset measured for each material, it was found that each spectrum could be appropriately classified into the three materials.

[0066] Furthermore, focusing on the spectrum of magnesium as an example, where no self-absorption is observed and both ion and atomic lines are observed, we performed material classification using spectral data in the ultraviolet wavelength range containing magnesium, silicon, and calcium. Using the dataset classified as mortar and the variables deemed optimal, we performed compressive strength estimation using PLS regression. The estimated compressive strength was in general agreement with the value obtained from the compressive strength test.

[0067] These results demonstrate that by using multivariate analysis (PLS regression) with the entire shape of the plasma emission spectrum, including not only the peaks containing ion and atomic lines of the same element with weak self-absorption, it is possible to estimate the compressive strength of concrete non-destructively and without being affected by the distribution of coarse aggregate. [Explanation of symbols]

[0068] 1. Core (concrete test specimen) 2 Laura 3 motors 4. Experimental apparatus 5 Automatic Stage 6. Objective lens 7 Dichroic filters 8 lasers 9. Miller, 10 Polarizing Beam Splitter 11 Optical Fiber 12 Spectrometer 13. Computers for device control, etc. 15 lenses 16. Delayed pulse generator

Claims

1. A method for estimating the compressive strength of concrete, characterized by irradiating the mortar portion of concrete with a laser pulse to induce plasma, measuring the spectrum containing ion and atomic lines of either magnesium or iron in the emission spectrum of the plasma, subtracting the average value of the spectral intensity from the measured spectrum, and then dividing by the standard deviation of the spectral intensity to obtain the explanatory variable, and performing partial least squares regression with the average value of the compressive strength obtained from a compressive strength test as the dependent variable, and multiplying the measured spectrum by the coefficient obtained from the partial least squares regression to estimate the compressive strength of the concrete.

2. The method for estimating the compressive strength of concrete according to claim 1, characterized in that a spectrum containing characteristic emission lines suitable for material classification is classified as a material using multivariate analysis, and the spectral data classified as mortar is used to perform the partial least squares regression.

3. The method for estimating the compressive strength of concrete according to claim 2, characterized in that the spectrum used when classifying mortar, limestone, and mountain sand is magnesium, silicon, and calcium.

Images