Impedance spectrum analysis method for concrete using machine learning, recording medium and device for performing the analysis.

Abstract

The machine learning-based method for analyzing the impedance spectrum of concrete includes the following steps: Based on electrochemical impedance spectroscopy (EIS), current flow is assessed by measuring electrical nodes and observing the moisture and conductive ions present within the concrete; a theoretical equivalent circuit model is generated, consisting of conductive paths reflecting this current flow; impedance experiments using this theoretical equivalent circuit model are conducted to normalize the equivalent circuit reflecting the concrete microstructure; and machine learning is used to generate a predictive model that estimates the water-to-cement ratio using the parameter values ​​of the equivalent circuit. This improves the accuracy and reliability of estimating the microstructure and mix proportions of cementitious materials.

Description

Technical Field

[0001] The present invention relates to a method for analyzing the impedance spectrum of concrete using machine learning, a recording medium and apparatus for performing it, and more specifically, to a technique for improving the interpretability of impedance spectra applicable to cement-based materials by theoretically normalizing the equivalent circuit and applying machine learning to the results.

[0002] The product of this invention is developed as part of the "Development of a basic source technology for machine learning and EIS-based mix proportion prediction with an accuracy of 80% for quality management of uncured concrete at construction sites (NTIS Project No.: 1615011765)". Background Technology

[0003] Quality management at construction sites is crucial for ensuring structural performance, durability, and the safety of personnel. Therefore, the Construction Technology Promotion Act includes detailed standards for quality management and processes at Korean construction sites in its "Guidelines for Quality Management at Construction Sites." Furthermore, in accordance with smart building policies, it is necessary to introduce Fourth Industrial Revolution-related technologies applicable to construction sites.

[0004] However, on fast-paced construction sites, the practically applicable quality management techniques are very limited. In particular, quality testing of the most frequently used material—uncured concrete (including ready-mixed concrete)—remains limited to basic items such as slump, air content, and chloride content. Among these, the slump test, while a standard for judging the workability of materials, is not suitable for quality management targeting the performance and durability of structures.

[0005] Furthermore, the long-used air volume testing method is a passive technique that varies depending on the proficiency of the field technicians. There is a need to develop simple, automated field testing methods that can measure and predict the long- and short-term performance of structures for quality management purposes.

[0006] Electrochemical Impedance Spectroscopy (EIS) is a non-destructive technique that uses electrical nodes to measure the electrical current flow within a target substance by observing the moisture and conductive ions present inside. Specifically, after applying a small alternating current to the target substance, the impedance is derived based on the electrical equivalent circuit reflecting the microstructure, thereby allowing the estimation of the internal composition of the substance.

[0007] Electrochemical impedance spectroscopy has the advantage of being very simple to use, making it suitable for a variety of fields. In particular, in South Korea, it is used as a method for analyzing the body composition of the human body under the name "InBody".

[0008] Various studies using impedance spectroscopy have also been conducted on cementitious materials, with concrete as a prime example. The main focus has been on the early prediction of steel corrosion within concrete. In studies focusing solely on cementitious materials, research has been conducted on estimating compressive strength and determining setting time based on impedance changes according to mix proportions.

[0009] Despite various studies, the development of testing methods for estimating the internal composition of cementitious materials using impedance spectroscopy is very limited. This is due to two main reasons: 1) insufficient research on the equivalent circuits used for impedance derivation; and 2) insufficient research on the correlation between the resistance and capacitance that constitute the impedance and the internal composition of cementitious materials.

[0010] In reality, for the equivalent circuit applicable to the impedance spectrum of concrete, based on the interpretation of results from various studies, there are a total of 10 different circuit theories. Consequently, the derived resistance and capacitance vary significantly depending on the applicable circuit, leading to limited general interpretation of the results. Furthermore, the quantities of resistance and capacitance also differ depending on the applicable circuit, thus presenting challenges in deriving relationships such as the proportions between the constituent materials and the composition.

[0011] Existing technical documents

[0012] Patent documents

[0013] Patent Document 1: JP 2020-115146 A

[0014] Patent Document 2: JP 2003-520974 A

[0015] Patent Document 3: KR ​​10-1221684B1

[0016] Non-patent literature

[0017] Non-patent document 1: Shin, SW, Hwang, G., & Lee, CJ (2014), Electrical impedance response model of concrete in setting process, Journal of the Korean Society of Safety, 29(5), 116-122. Summary of the Invention

[0018] The problem that the invention aims to solve

[0019] Therefore, the technical problem addressed by the present invention is to provide a method for analyzing the impedance spectrum of concrete using machine learning.

[0020] Another object of the present invention is to provide a recording medium having a computer program for performing the above-described method for analyzing the impedance spectrum of concrete using machine learning.

[0021] Another object of the present invention is to provide an apparatus for performing the above-described method for analyzing the impedance spectrum of concrete using machine learning.

[0022] Solution for solving the problem

[0023] An embodiment of the impedance spectrum analysis method for concrete using machine learning for achieving the objectives of the present invention as described above includes the following steps: based on electrochemical impedance spectroscopy (EIS), using nodes for measuring electricity, current flow is grasped through water and conductive ions present inside the concrete; a theoretical equivalent circuit model is generated consisting of conductive paths reflecting the aforementioned current flow; based on impedance experiments using the aforementioned theoretical equivalent circuit model, the equivalent circuit reflecting the microstructure of the concrete is normalized; and through machine learning, a predictive model is generated that uses the parameter values ​​of the aforementioned equivalent circuit to estimate the ratio of water to cement.

[0024] In embodiments of the present invention, the microstructure of the concrete may include a cement matrix and internal pores.

[0025] In embodiments of the present invention, the aforementioned machine learning may utilize at least one of the following models: Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Decision Tree.

[0026] In embodiments of the present invention, the above parameters may include the resistance and capacitance of the equivalent circuit.

[0027] In embodiments of the present invention, a three-electrode method can be applied in the step of mastering the above-mentioned current flow, which utilizes a working electrode (WE), a counter electrode (CE), and a reference electrode.

[0028] In one embodiment of the computer-readable storage medium for achieving another object of the present invention as described above, a computer program is recorded for performing the above-described method for analyzing the impedance spectrum of concrete using machine learning.

[0029] An embodiment of the impedance spectrum analysis apparatus for concrete using machine learning for achieving another object of the present invention as described above includes: an EIS unit that, based on electrochemical impedance spectroscopy, uses electrical measurement nodes to grasp current flow through water and conductive ions present inside the concrete; an equivalent circuit unit that generates a theoretical equivalent circuit model composed of conductive paths reflecting the aforementioned current flow; a circuit normalization unit that, based on impedance experiments using the aforementioned theoretical equivalent circuit model, normalizes the equivalent circuit reflecting the microstructure of the concrete; and a prediction model unit that, through machine learning, generates a prediction model that estimates the ratio of water to cement using the parameter values ​​of the aforementioned equivalent circuit.

[0030] In embodiments of the present invention, the microstructure of the concrete may include a cement matrix and internal pores.

[0031] In embodiments of the present invention, the above-mentioned machine learning can utilize at least one model among squared exponential function Gaussian process regression, support vector regression, and decision tree.

[0032] In embodiments of the present invention, the above parameters may include the resistance and capacitance of the equivalent circuit.

[0033] In embodiments of the present invention, the EIS section described above can be applied using a three-electrode method, which utilizes a working electrode, a counter electrode, and a reference electrode.

[0034] Invention Effects

[0035] Based on the impedance spectrum analysis method for concrete using machine learning as described above, electrodes are placed on uncured cementitious materials, and electrical measurements are performed to immediately respond to changes in on-site conditions. Information such as the water-cement ratio based on this response can be derived. Therefore, this invention enables reverse engineering of material composition and prediction of durability, and can serve as foundational data for developing highly reliable quality management technologies for uncured concrete on construction sites.

[0036] Furthermore, this invention is based on machine learning, which systematizes the previous human-based quality management work, and thus has the potential to develop a simple, automated, and highly reliable field testing method. Attached Figure Description

[0037] Figure 1 This is a block diagram of an impedance spectrum analysis device for concrete using machine learning, according to an embodiment of the present invention.

[0038] Figure 2 The diagram illustrates how the experimental equivalent circuit of the present invention is normalized to a theoretical equivalent circuit reflecting the microstructure of concrete.

[0039] Figure 3This is a graph showing the Nyquist plot results of the EIS experiment based on cement slurry according to the present invention.

[0040] Figure 4 To illustrate the resistivity variation of the present invention as a parameter based on the water-cement ratio, a box plot is provided.

[0041] Figure 5 A box plot is shown to illustrate the capacitance variation of the present invention as a parameter based on the water-cement ratio.

[0042] Figure 6 This is a graph showing the water-cement ratio prediction results based on a machine learning model utilizing the squared exponential function GPR according to the present invention.

[0043] Figure 7 This is a graph showing the water-cement ratio prediction results based on the machine learning model utilizing SVR according to the present invention.

[0044] Figure 8 This is a graph showing the water-cement ratio prediction results based on a machine learning model utilizing decision trees, according to the present invention.

[0045] Figure 9 This is a flowchart of an embodiment of the present invention for analyzing the impedance spectrum of concrete using machine learning. Detailed Implementation

[0046] The detailed description of the invention described below refers to the accompanying drawings, which illustrate specific embodiments through which the invention may be practiced. These embodiments will be described in detail so that those skilled in the art can practice the invention. It should be understood that the various embodiments of the invention are different from each other, but are not mutually exclusive. For example, with respect to one embodiment, the specific shapes, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention. Furthermore, it should be understood that the position or configuration of individual components in the various disclosed embodiments may be changed without departing from the spirit and scope of the invention. Therefore, the detailed description described below is not intended to be limiting, and if properly interpreted, the scope of the invention is limited to the equivalents of those claimed in the claims and the appended claims. In the drawings, similar reference numerals refer to the same or similar functions in several respects.

[0047] The preferred embodiments of the present invention will now be described in more detail with reference to the accompanying drawings.

[0048] Figure 1 This is a block diagram of an impedance spectrum analysis device for concrete using machine learning, according to an embodiment of the present invention.

[0049] In the impedance spectrum analysis apparatus 10 for concrete using machine learning of the present invention (hereinafter referred to as the apparatus), in order to improve the interpretability of impedance spectra applicable to cement-based materials, the equivalent circuit is theoretically normalized, and machine learning is applied to the results. The applicability of a theoretical model considering the microstructure of cement-based materials allows for the normalization of circuit parameters (resistance, capacitance) to more objectively derive the influence on mix proportions.

[0050] Furthermore, machine learning can overcome the limitations of existing regression analysis, deriving correlations from a new perspective. To this end, through a literature review of existing impedance spectra, the equivalent circuit model was normalized and a database was established, and the results were experimentally verified. Specifically, the water-cement ratio (w / c ratio) of cement paste, as a fundamental stage of cementitious materials, was estimated, which can then be used for inverse calculations of material composition and durability prediction. In addition, this data can be used as foundational information for developing highly reliable quality management technologies for uncured concrete on construction sites.

[0051] Reference Figure 1 The apparatus 10 of the present invention includes an EIS unit 110, an equivalent circuit unit 130, a circuit normalization unit 150, and a prediction model unit 170.

[0052] The apparatus 10 of the present invention can be installed and run with software (application program) for performing impedance spectrum analysis of concrete using machine learning. The configuration of the EIS unit 110, the equivalent circuit unit 130, the circuit normalization unit 150 and the prediction model unit 170 can be controlled by the software. The software is used to perform the impedance spectrum analysis of concrete using machine learning running in the apparatus 10.

[0053] The aforementioned device 10 can be a standalone terminal or a module within a terminal. Furthermore, the configuration of the EIS unit 110, the equivalent circuit unit 130, the circuit normalization unit 150, and the prediction model unit 170 can be formed by integrated modules, or by more than one module. However, conversely, each configuration can also be formed by a separate module.

[0054] The aforementioned device 10 may be mobile or fixed. The aforementioned device 10 may be in the form of a server or an engine, and may also be referred to as a device, apparatus, terminal, user equipment (UE), mobile station (MS), wireless device, handheld device, or other terms.

[0055] The aforementioned device 10 can be based on an operating system (OS), that is, various software can be run or created based on the system. The aforementioned operating system is a system program that enables software to use the hardware of the device, and may include mobile computer operating systems such as Android OS, iOS, Windows Mobile OS, bada OS, Symbian OS, and BlackBerry OS, as well as computer operating systems such as Windows series, Linux series, Unix series, MAC, AIX, and HP-UX.

[0056] The aforementioned EIS section 110, based on electrochemical impedance spectroscopy, uses the nodes for measuring electricity to control the current flow through the moisture and conductive ions present inside the concrete.

[0057] For impedance spectra applicable to cement-based materials, various equivalent circuits have been proposed according to different research objectives. Specifically, circuit models for understanding the microstructure of cement-based materials based on electrical properties include: the Brick model considering the interfacial reaction between solid and liquid, circuits including all conductive paths, non-contact circuit models, and circuit models composed of continuous / discontinuous conductive paths, etc.

[0058] Circuits based on the analog proximity method include circuit models with capacitive elements having phase angles and circuit models that fit the impedance spectrum of cement slurry. Various equivalent circuits have also been proposed to address electrical changes that occur when volcanic ash, insulating particles, and fibers are added. Furthermore, a circuit has been proposed to model chloride migration phenomena with the aim of addressing steel corrosion. Circuits that fit the impedance spectrum of solidified or unsolidified concrete also exist.

[0059] In one embodiment of the invention, a model reflecting the microstructure of cementitious materials and used for uncured concrete is as follows: Figure 2 It is a circuit composed of a resistor and a capacitor connected in parallel and arranged in a continuous manner. From the Nyquist plot, it has the following characteristics: two arcs appear side by side, and each arc corresponds to a parallel circuit.

[0060] Reference Figure 2 The arc shown in the first half of the Nyquist plot is the bulk arc, corresponding to the high-frequency range, indicating the electrolyte impedance between the working electrode and the reference electrode. The arc shown afterwards is the electrode arc, which represents the reaction that occurs between the electrode and the concrete interface. The circuit with R1 and C1 connected in parallel participates in the bulk reaction, while the circuit with R2 and C2 connected in parallel participates in the electrode reaction. This circuit is mainly used to study uncured cementitious materials.

[0061] The equivalent circuit section 130 generates a theoretical equivalent circuit model consisting of conductive paths that reflect the current flow.

[0062] Focusing on the microstructure of cement-like materials, the theoretical equivalent circuit is, for example... Figure 2 When current flows through, based on the continuity within the composite material, the following distinct circuits can be replaced by parallel connection models. For example, from the perspective of current flow, the microstructure can be defined by circuits such as an insulating conductive path (ICP) that only passes through cement slurry, a continuous conductive path (CCP) that passes through the interconnected paths between micropores, and a discontinuous conductive path (DCP) that passes through discontinuous points (DP) during its passage through micropores.

[0063] When the theoretical model described above is reconstructed into an equivalent circuit, resistive elements are used to replace the circuit passing through the microvias, and capacitive elements are used to replace the circuit passing through the cement grout, thereby obtaining... Figure 3 The equivalent circuit. From the perspective of the derived impedance (Z), as shown in Equation 1 below, the relationship between the experimental equivalent circuit and the theoretical equivalent circuit can be derived.

[0064] Mathematical Formula 1

[0065]

[0066] In mathematical formula 1, ω represents each frequency, with a value of 2πf, and the unit is radians. Furthermore, R1, R2, C1, and C2 in the experimental equivalent circuit are parameters of the theoretical equivalent circuit with microstructure physical meaning, and can be replaced by the following mathematical expressions 2 to 5.

[0067] Mathematical formula 2

[0068] R CP = (R1+R2)R1 / R2

[0069] Mathematical Formula 3

[0070] R CCP =R1+R1

[0071] Mathematical expression 4

[0072] C DP = (C1+C2)[R2 / (R1+R2)] 2

[0073] Mathematical formula 5

[0074] C mat =C1C2 / (C1+C2)

[0075] In this invention, the following normalized variable (R) is derived through a theoretical equivalent circuit. CP’ R CCP’ C DP’ C mat The correlation between water and cement ratio in cement-based materials.

[0076] The circuit normalization unit 150 normalizes the equivalent circuit reflecting the concrete microstructure based on the impedance experiment using the above-mentioned theoretical equivalent circuit model.

[0077] In this invention, 140 data points were acquired to utilize machine learning to understand the impedance changes based on the water-cement ratio of the cement slurry. In the machine learning, the response variable—the water-cement ratio—as the desired outcome was designed as a continuous function ranging from 31% to 44%, with 1% differences forming 14 stages. To collect reproducibility data for each experimental subject with a single water-cement ratio, the experimental subjects were prepared by repeating the experiment 10 times, and impedance spectroscopy was applied.

[0078] The experimental sample consisted of cement slurry in a cylindrical plastic container with a diameter of 7 cm and a height of 8 cm, containing a constant volume of 300 ml. After the measured cement was poured into the container, water was added and the mixture was stirred evenly for 5 minutes, then compacted 30 times with a 150 mm tamping rod. Detailed information on the experimental sample composition is shown in Table 1 below.

[0079] Table 1

[0080]

[0081] The appropriate device for impedance spectroscopy is the Gamry PCI4 / 300, whose performance is summarized in Table 2. The impedance measurement device applies an AC voltage and performs the measurement. That is, it has the advantage that, through a specified experimental container, the results can be derived regardless of the skill level of the experimenter.

[0082] Table 2

[0083]

[0084] As an example, the three-electrode method was used as the experimental method, employing a 1cm electrode. 2 A graphite rod with a surface area of ​​1000g is used as the working electrode and counter electrode, and a saturated calomel electrode (SCE) is used as the reference electrode.

[0085] After inserting each electrode sequentially into the cement slurry with a length of 4 cm, it was connected to the PCI4 / 300 device. In the prior art, the bulk arc observed at high frequencies is not generated in the case of cement slurry. Therefore, a frequency range of 0.2 Hz to 100 kHz was set, and an AC voltage of 10 mV was applied for a total measurement of 5 minutes.

[0086] Experimental analysis of the impedance spectrum is based on Nyquist plots drawn from the established equivalent circuit. The Nyquist plot displays the real and imaginary parts of the impedance on each axis, allowing for intuitive confirmation of the parameter values. Figure 3 The Nyquist plots for each water-cement ratio are shown.

[0087] The experimental results show that, as a characteristic of cement grout without aggregate, no main arc is generated in the low-frequency range. This means that the value of C1 is very close to 0, the impedance of the circuit with the first half of the capacitor element is close to infinite, and current practically does not flow. Therefore, the value of R1, representing the main resistance, can be immediately confirmed in the high-frequency range, and an average value of approximately 5Ω can be confirmed.

[0088] Table 3 shows the results of replacing the variables obtained as experimental results with parameters equivalent to the theoretical equivalent circuit using mathematical formulas 2 to 5. Figure 4 and Figure 5 The experimental results based on the water-cement ratio were statistically presented using box plots. Figure 4 The present invention illustrates the resistance variation as a parameter based on the water-cement ratio. Figure 5 The present invention illustrates the capacitance variation as a parameter based on the water-cement ratio.

[0089] Table 3

[0090]

[0091] Box plots visually identify the maximum, minimum, median, and statistical outliers of parameters derived by repeating each water-cement ratio 10 times. In the case of statistical outliers, the data from 25% to 75% is defined as a box, and data outside 1.5 times the length of this box, i.e., outside the "whiskers," are considered outliers.

[0092] Statistical analysis revealed few outliers, thus demonstrating significant reproducibility of the experiment. However, it was difficult to observe any biases in the parameters related to the water-to-cement ratio. This is largely consistent with previous studies focusing on cement paste.

[0093] The aforementioned prediction model unit 170 generates a prediction model that estimates the ratio of water to cement using the parameter values ​​of the aforementioned equivalent circuit through machine learning.

[0094] To derive the correlation between parameters obtained from impedance spectroscopy experiments and the water-cement ratio, a predictive model was derived using machine learning. The entire dataset used for machine learning consisted of 140 sets, each representing a water-cement ratio repeated 10 times across 14 stages. Each dataset contained a normalized parameter (R0). CP’ R CCP’ C DP’ C mat The composition includes water and cement ratio. At this point, R... CP’ R CCP’ C DP’ C mat Let X be the predictor variable that can be used for prediction, and let the water-cement ratio be the response variable Y that will be predicted.

[0095] Within the entire dataset, training data and validation data within a water-to-cement ratio are randomly separated in a 7:3 ratio. Specifically, the training data is further divided into predictor variables (Xtrain) and response variables (Ytrain). The inferred response variable (Ypred) is then compared to the actual response variable (Yexp) by adding a new predictor variable (Xexp) as validation data. This process of separating training and validation data is repeated 10 times to minimize potential overfitting and underfitting issues during response variable inference.

[0096] This invention employs three different machine learning methods: Gaussian Process Regression (GPR), Support Vector Regression (SVR), and Decision Tree.

[0097] GPR is a technique that interprets measured data from a probabilistic perspective, rather than treating it as a fixed observation, based on the probabilistic distribution of a Gaussian distribution and grounded in Bayesian theorem. Specifically, it generates a probabilistic prediction model based on the measured training data, using a covariance function (k) as a foundation. The covariance function applicable to this invention is the basic squared exponential (SE) function, which is composed of the following mathematical expression 6 after optimizing the noise standard deviation (σ).

[0098] Mathematical formula 6

[0099]

[0100] The covariance function is described by the hyperparameter θ(σ). f σ f σ l The function k(x) is therefore... i x j |θ) is used to express the predictor variable data. i x j Composition. Furthermore, i refers to a point, and j refers to an observation point. σ f σ refers to the standard deviation of the signal, and σ refers to a scale of a specific length. SVR is a regression technique used when the input variables of a Support Vector Machine (SVM) are continuous variables rather than categorical variables. It can be compared with the GPR model by selecting similar covariance functions. However, there are differences in the algorithm execution based on specific data (SupportVector) compared to GPR, which uses all the input data points.

[0101] Unlike other models that are based on regression functions, decision trees generate predictive models based on regression objectives. A decision tree is a classifier that segments the entire dataset according to decision nodes, leading to the last leaf node. The predicted value is obtained by calculating the average of the data values ​​belonging to the leaf nodes.

[0102] The machine learning methods applicable to this invention are based on the commercial program MATLAB, and the code used is attached in Table 4. The time consumed to execute each machine learning model once on a dataset is approximately 123 seconds, 130 seconds, and 100 seconds for GPR(SE), SVR, and decision tree, respectively.

[0103] Table 4

[0104]

[0105] exist Figures 6 to 8 The results of each applicable machine learning model are shown in charts. Figure 6 This is a chart showing the predicted water-cement ratio based on a machine learning model using the squared exponential function GPR. Figure 7 A chart showing the predicted water-cement ratio based on a machine learning model using SVR. Figure 8 This is a graph showing the predicted water-cement ratio based on a machine learning model using a decision tree.

[0106] In the chart, the X-axis represents the actual water-cement ratio applicable to the experiment, serving as the response variable (Yexp), and the Y-axis represents the inferred response variable (Ypred) by adding a new predictor variable (Xexp) as validation data. The diagonal line of the chart is the absolute line where the predicted and actual values ​​match; the closer the data is to the line, the higher the accuracy of the prediction model.

[0107] Table 5 shows the calculated regression error generated when new data is added to the prediction model.

[0108] Table 5

[0109]

[0110] The regression errors are calculated as the mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE) according to the following mathematical formulas 7 to 9.

[0111] Mathematical Formula 7

[0112]

[0113] Mathematical formula 8

[0114]

[0115] Mathematical formula 9

[0116]

[0117] Among them, Y i Represents the i-th actual response variable. Let represent the i-th presumed response variable, and n represent the total data size. MAE, MSE, and RMSE all assume that smaller values ​​indicate better predictive performance. The characteristics of each regression error are as follows: MAE is the simple error between the predicted and actual values, providing a direct visual representation of the results. MSE is the error calculated by averaging the squared differences between the actual and predicted values; it is the most common error but sensitive to outliers. RMSE converts MSE back to an error in units similar to the actual values. Because the data is randomly partitioned, the best-performing model will change when each machine learning exercise is repeated. To compensate for reliability, the presumed response variable model (Ypred) is ultimately used, averaged after 10 repetitions.

[0118] The results from the final estimated response variable model show that the decision tree with the lowest regression error has the best predictive performance. Based on the best predictive model, impedance spectroscopy confirms that the water-to-cement ratio of the cement paste is within the error range of approximately MAE 2.10, MSE 8.59, and RMSE 2.93. However, these results were validated using cement paste with a water-to-cement ratio of 31%–44%, and can be extended to various material combinations.

[0119] Figure 9 This is a flowchart of an embodiment of the present invention for analyzing the impedance spectrum of concrete using machine learning.

[0120] The impedance spectrum analysis method for concrete using machine learning in this embodiment can be applied to... Figure 1 The device 10 is configured essentially the same as the one used in the previous device. Therefore, it is directed to... Figure 1 The same components of the apparatus 10 are given the same reference numerals, and repeated descriptions are omitted.

[0121] Furthermore, the impedance spectrum analysis method for concrete using machine learning in this embodiment can be executed by software (application), which is used to perform the impedance spectrum analysis for concrete using machine learning.

[0122] Reference Figure 9 The impedance spectrum analysis method for concrete using machine learning in this embodiment is based on electrochemical impedance spectroscopy. It uses the nodes that measure electricity to grasp the current flow through the moisture and conductive ions present inside the concrete (step S10).

[0123] In this case, the three-electrode method can be applied, which utilizes a working electrode, a counter electrode, and a reference electrode.

[0124] Generate a theoretical equivalent circuit model consisting of conductive paths that reflect the above-mentioned current flow (step S20).

[0125] Based on the impedance experiment using the above-mentioned theoretical equivalent circuit model, the equivalent circuit reflecting the microstructure of concrete is normalized (step S30). The microstructure of the concrete includes the cement matrix and internal pores.

[0126] Machine learning is used to generate a predictive model that estimates the ratio of water to cement using the parameter values ​​of the equivalent circuit described above (step S40). The parameters can be the resistance and capacitance of the equivalent circuit.

[0127] The aforementioned machine learning methods can utilize Gaussian process regression with squared exponential function, support vector regression, and decision trees, among others.

[0128] This invention establishes an algorithm for estimating the water-cement ratio of uncured cementitious materials. When the experiment is conducted with cement slurry as the object, the model with the best prediction performance has an error in predicting the water-cement ratio of approximately within MAE 2.10, MSE 8.59, and RMSE 2.93.

[0129] Electrodes are placed on uncured cement-based materials, and electrical measurements are used to immediately respond to changes in on-site conditions, and information such as the water-cement ratio can be derived based on the response.

[0130] By leveraging machine learning, the quality management work that was previously done manually can be systematized, which may lead to the development of simple, automated, and highly reliable field testing methods.

[0131] As described above, the method for analyzing the impedance spectrum of concrete using machine learning can be implemented through an application program or recorded in a computer-readable recording medium as program instructions executable by various computer components. The aforementioned computer-readable recording medium may include, individually or in combination, program instructions, data files, data structures, etc.

[0132] The program instructions recorded in the aforementioned computer-readable recording medium may be instructions specifically designed and configured for this invention, or instructions that are known to those skilled in the art of computer software and can be used.

[0133] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floppy disks; and hardware devices specially configured to store and execute program instructions, such as read-only memory (ROM), random access memory (RAM), and flash memory.

[0134] Examples of program instructions include not only machine language code generated by a compiler, but also high-level language code that can be run by a computer using an interpreter or similar means. The aforementioned hardware device can be configured to run as one or more software modules to execute the processing of this invention, and vice versa.

[0135] The above description refers to the embodiments, but it will be understood by those skilled in the art that various modifications and alterations can be made to the present invention without departing from the spirit and scope of the invention as set forth in the claims.

[0136] Industrial applicability

[0137] This invention can be used as the basis for reverse engineering of material combinations and durability prediction, and for developing highly reliable quality management technologies for uncured concrete on construction sites.

[0138] Furthermore, this invention is based on machine learning and systematizes the previously human-based quality management system, thereby potentially developing a simple, automated, and highly reliable field testing method.

[0139] Explanation of reference numerals in the attached figures

[0140] 10: An analysis device for impedance spectra of concrete using machine learning

[0141] 110: EIS Department

[0142] 130: Equivalent Circuit Section

[0143] 150: Circuit Normalization Section

[0144] 170: Predictive Models Department.

Claims

1. A method for analyzing the impedance spectrum of concrete using machine learning, characterized in that, comprises the steps of: acquiring, based on electrochemical impedance spectroscopy, current flow through water and conductive ions present inside the concrete using a node of measurement electricity; generating a theoretical equivalent circuit model composed of conductive paths reflecting the current flow; normalizing, based on impedance experiments using the theoretical equivalent circuit model, an equivalent circuit reflecting the microstructure of the concrete; and generating, through machine learning, a prediction model that estimates the ratio of water and cement using parameter values of the equivalent circuit, In the step of acquiring the current flow, a three-electrode method is used, which uses a working electrode, a counter electrode, and a reference electrode.

2. The analysis method for impedance spectroscopy of concrete using machine learning according to claim 1, wherein the microstructure of the concrete includes a cement matrix and internal pores.

3. The analysis method for impedance spectroscopy of concrete using machine learning according to claim 1, wherein the machine learning uses at least one model of a squared exponential function Gaussian process regression, a support vector regression, and a decision tree.

4. The analysis method for impedance spectroscopy of concrete using machine learning according to claim 1, wherein the parameters include resistance and capacitance of the equivalent circuit.

5. A computer-readable storage medium, wherein a computer program for executing the analysis method for impedance spectroscopy of concrete using machine learning according to claim 1 is recorded. 6.An analysis device for impedance spectra of concrete using machine learning, characterized by, comprises: an electrochemical impedance spectroscopy section that acquires, based on electrochemical impedance spectroscopy, current flow through water and conductive ions present inside the concrete using a node of measurement electricity; an equivalent circuit section that generates a theoretical equivalent circuit model composed of conductive paths reflecting the current flow; a circuit normalizing section that normalizes, based on impedance experiments using the theoretical equivalent circuit model, an equivalent circuit reflecting the microstructure of the concrete; and a prediction model section that generates, through machine learning, a prediction model that estimates the ratio of water and cement using parameter values of the equivalent circuit, the electrochemical impedance spectroscopy section uses a three-electrode method, which uses a working electrode, a counter electrode, and a reference electrode.

7. The analysis apparatus for impedance spectroscopy of concrete using machine learning according to claim 6, wherein the microstructure of the concrete includes a cement matrix and internal pores.

8. The analysis apparatus for impedance spectroscopy of concrete using machine learning according to claim 6, wherein the machine learning uses at least one model of a squared exponential function Gaussian process regression, a support vector regression, and a decision tree.

9. The analysis apparatus for impedance spectroscopy of concrete using machine learning according to claim 6, wherein the parameters include resistance and capacitance of the equivalent circuit.

Images