Prediction system, prediction device, and program

Abstract

By using machine learning to predict and display the changes in compressive strength at multiple ages, this system supports concrete process management and contributes to quality assurance. [Solution] The prediction system 1 for predicting the compressive strength of concrete comprises a model generation unit 120 that generates a prediction model 300 by machine learning using a training dataset 300 in which age is the explanatory variable and compressive strength is the objective variable; a prediction unit 220 that predicts the compressive strength at an input target age using the prediction model 300; and a result display unit 230 that displays the trend of compressive strength corresponding to a plurality of target ages predicted by the prediction unit 220.

Description

[Technical Field]

[0001] This invention relates to a technology for predicting the compressive strength of concrete using machine learning. [Background technology]

[0002] In building and civil engineering projects, managing the compressive strength of concrete is crucial for ensuring structural quality and project schedule. Concrete hardens and develops strength over time after placement, meaning that the next stage of construction cannot proceed until the required strength is achieved. Therefore, accurately predicting the strength development is a critical project management challenge. Furthermore, failing to meet the design strength necessitates re-placing the concrete, significantly impacting the project schedule. Thus, it is desirable to be able to predict potential strength deficiencies early on.

[0003] Techniques for predicting the compressive strength of concrete have been studied for some time. For example, methods have been proposed to analyze the relationship between compressive strength and materials, mix design, and manufacturing conditions using simple or multiple regression analysis and derive estimation formulas (Non-Patent Literature 1). Using such methods, it is possible to predict the compressive strength at 28 days of age during the mix design stage. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Toshiyasu Toyofuku, "A Study on a Method for Estimating the Compressive Strength of Concrete Using Data Analysis," Transactions of the Japan Society of Civil Engineers, No. 420 / V-13, pp. 125-134, August 1990. [Overview of the project] [Problems that the invention aims to solve]

[0005] On the other hand, from a process control perspective, it is desirable to be able to understand not only the predicted strength at a specific age, but also the strength progression at multiple ages. Visually displaying the strength progression allows users to intuitively grasp when a predetermined strength will be reached, enabling them to create efficient process plans. Furthermore, by grasping the trend of strength development early, it is possible to determine the possibility of insufficient strength and take countermeasures at the appropriate time.

[0006] However, conventional strength prediction techniques, such as those described in Non-Patent Document 1, can predict values ​​at a specific age, but they are not designed to predict compressive strength over multiple ages and to track its progression. In particular, from the standpoint of process control and quality control, it is desirable to be able to visually grasp the changes in strength at each age, and there is room for further improvement in this regard.

[0007] This invention has been made in view of the above-mentioned problems, and aims to provide a prediction system that supports concrete process management and contributes to quality assurance by predicting and displaying the changes in compressive strength at multiple ages using machine learning. [Means for solving the problem]

[0008] The first invention for achieving the aforementioned objective is a prediction system for predicting the compressive strength of concrete, comprising: a generation unit that generates a prediction model by machine learning using a training dataset in which age is the explanatory variable and compressive strength is the objective variable; a prediction unit that predicts the compressive strength at an input target age using the prediction model; and a display unit that displays the progression of compressive strength corresponding to a plurality of target ages predicted by the prediction unit.

[0010] The 2The present invention is a prediction device for predicting the compressive strength of concrete, comprising: a storage unit that stores a prediction model learned by machine learning to predict compressive strength using explanatory variables including age as input; a prediction unit that uses the prediction model to predict the compressive strength at a target age; and a display unit that displays the progression of compressive strength corresponding to a plurality of target ages predicted by the prediction unit.

[0012] The 3 The invention of the computer made the computer the first 2 This is a program designed to function as a prediction device for the invention. [Effects of the Invention]

[0013] The present invention provides a prediction system that supports concrete process management and contributes to quality assurance by predicting and displaying the changes in compressive strength at multiple ages. [Brief explanation of the drawing]

[0014] [Figure 1] This is a block diagram showing an example of the overall configuration of a prediction system according to an embodiment of the present invention. [Figure 2] This figure shows an example of the hardware configuration of a learning device and a prediction device. [Figure 3] This block diagram shows an example of the functional configuration of the prediction system according to the first embodiment. [Figure 4] This figure shows an example of the structure of training data in the first embodiment. [Figure 5] This flowchart shows an example of the learning process flow in the first embodiment. [Figure 6] This flowchart shows an example of the prediction processing flow in the first embodiment. [Figure 7] This figure shows an example of how the prediction results are displayed in the first embodiment. [Figure 8] This figure shows another example of displaying the prediction results in the first embodiment (with a prediction interval). [Figure 9]It is a block diagram showing an example of the functional configuration of the prediction system according to the second embodiment. [Figure 10] It is a diagram showing an example of the structure of learning data in the second embodiment. [Figure 11] It is a flowchart showing an example of the flow of learning processing in the second embodiment. [Figure 12] It is a flowchart showing an example of the flow of prediction processing in the second embodiment. [Figure 13] It is a diagram showing an example of the display of prediction results in the second embodiment. [Figure 14] It is a diagram showing another example of the display of prediction results (with prediction intervals) in the second embodiment. [Figure 15] It is a diagram showing an example of the output of a test score sheet. [Mode for Carrying Out the Invention]

[0015] [Overview] Hereinafter, embodiments of the present invention (hereinafter referred to as the present embodiment) will be described in detail based on the drawings. The prediction system according to the present embodiment is a system that predicts the compressive strength of concrete, and uses a prediction model learned by machine learning based on a learning dataset with the age of the material as an explanatory variable and the compressive strength as a target variable to display the transition of the compressive strength corresponding to a plurality of prediction target ages.

[0016] This embodiment includes a form (first embodiment) in which prediction is performed before the placement of concrete and a form (second embodiment) in which prediction is performed using test results after placement. These embodiments all have the above basic configuration, and the configuration of the learning data and the input data at the time of prediction are different. Hereinafter, each embodiment will be described in detail.

[0017] [Overall System Configuration] Figure 1 is a block diagram showing the overall configuration of the prediction system 1 according to this embodiment. As shown in Figure 1, the prediction system 1 includes a learning device 100 and a prediction device 200. The learning device 100 is a device that generates a prediction model 300 using past concrete test data. The prediction device 200 is a device that predicts the compressive strength of concrete using the prediction model 300 generated by the learning device 100.

[0018] The prediction model 300 generated by the learning device 100 is provided to the prediction device 200, for example, via a network. The learning device 100 and the prediction device 200 may be configured as a single device or as separate devices. Furthermore, each device may be configured as a single computer or as a distributed processing system composed of multiple computers. In addition, the prediction device 200 can be configured as a cloud server and made accessible from a user's terminal device via a network. For the sake of clarity, in the following description, the learning device 100 and the prediction device 200 will be described as separate, standalone devices.

[0019] [Hardware configuration] Figure 2 shows an example of the hardware configuration of the learning device 100 and the prediction device 200. Since the learning device 100 and the prediction device 200 have the same basic hardware configuration, the following explanation will use the prediction device 200 as an example.

[0020] The prediction device 200 is configured using, for example, a general-purpose personal computer or server computer. As shown in Figure 2, the prediction device 200 consists of a control unit 201, a storage unit 202, a communication unit 203, an input unit 204, a display unit 205, a peripheral device I / F 206 unit, etc., all connected via a bus 208. Note that the configuration shown in Figure 2 is just one example, and the prediction device 200 can take various configurations depending on the application and purpose.

[0021] The control unit 201 includes a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), and a ROM (Read Only). It has a memory (RAM), RAM (Random Access Memory), etc. The CPU calls programs stored in the memory unit 202 or recording media such as ROM into the work memory area on RAM and executes them, and the GPU is used to execute a large amount of computational processing at high speed. The control unit 201 drives and controls each part connected via the bus 208 and uses the CPU and / or GPU to realize the various processes of the prediction device 200 described later.

[0022] ROM is non-volatile memory that permanently stores programs and data such as the computer's boot program and BIOS. RAM is volatile memory that temporarily stores programs and data loaded from storage media such as the memory unit 202 and ROM, and also includes a work area used by the control unit 201 for various processing.

[0023] The storage unit 202 is an HDD (Hard Disk Drive), SSD (Solid State Drive), flash memory, etc., and stores the program executed by the control unit 201, the data necessary for program execution, the OS (Operating System), etc. In particular, in this embodiment, the application program for causing the prediction device 200 to execute the prediction process described later, and the prediction model 300 learned by machine learning are stored in the storage unit 202.

[0024] The communication unit 203 includes a communication interface and a communication control circuit that mediate communication between the prediction device 200 and controls communication over the network. The network includes LAN (Local Area Network), WAN (Wide Area Network), the Internet, etc., and can be wired or wireless. The communication unit 203 may, for example, receive a prediction model 300 from the learning device 100 or acquire temperature data, etc., from an external database.

[0025] The input unit 204 includes an input device such as a keyboard, mouse, or touch panel. The input unit 204 transmits data and operation instructions entered by the user to the control unit 201. For example, the user inputs prediction conditions (cement type, curing method, nominal strength, planned sampling date, etc.) via the input unit 204.

[0026] The display unit 205 includes a liquid crystal display, an organic EL display, etc., and displays data such as images and text on the display according to instructions from the control unit 201. In particular, in this embodiment, the display unit 205 displays the changes in compressive strength corresponding to multiple predicted target ages as a graph.

[0027] The peripheral device interface (I / F) section 206 is a port for connecting peripheral devices, and includes short-range wireless communication such as USB and Bluetooth (registered trademark). The control unit 201 transmits and receives data with peripheral devices via the peripheral device interface section 206.

[0028] The learning device 100 has the same hardware configuration as the prediction device 200. The learning device 100 includes a control unit 101, a storage unit 102, a communication unit 103, etc. (not shown). The storage unit 102 of the learning device 100 stores a training dataset (past concrete test data) and a program for performing machine learning. The control unit 101 generates a prediction model 300 by executing the program stored in the storage unit 102.

[0029] [First Embodiment] [overview] The prediction system 1a according to the first embodiment predicts the compressive strength at the target age without using test results before concrete placement. This embodiment is useful for process management before concrete placement. For example, it can be used to check in advance whether a predetermined strength is likely to be obtained, or to understand the progression of strength development in advance in order to create the shortest possible process.

[0030] [Functional Configuration] Figure 3 is a block diagram showing an example of the functional configuration of the prediction system 1a. As shown in Figure 3, the learning device 100 includes a learning data acquisition unit 111 and a model generation unit 112. The prediction device 200 includes a model storage unit 210, a prediction unit 220, and a result display unit 230. The control unit of each device reads and executes the program stored in the storage unit to realize these functions.

[0031] The learning data acquisition unit 110 generates and acquires a learning dataset 400a from past concrete test data. The learning dataset 400a consists of multiple learning data sets, with age as the explanatory variable and compressive strength as the target variable. In addition to age, the learning data can include information on the sampling date, temperature information, curing method, cement type, nominal strength, etc., as explanatory variables. Details of the learning dataset 400a will be described later.

[0032] The model generation unit 120 generates a prediction model 300a (pre-construction model) using machine learning with the training dataset 400a acquired by the training data acquisition unit 110. In this embodiment, the model generation unit 112 trains a regression model using LightGBM (Light Gradient Boosting Machine). LightGBM is a decision tree-based machine learning algorithm that enables fast and highly accurate predictions.

[0033] Furthermore, the model generation unit 120 generates a quantile regression model 301a in order to estimate the prediction interval. Quantile regression is a regression method that estimates conditional quantiles and can quantitatively evaluate the variability of predicted values. For example, when displaying the 80th percentile prediction interval, two models are trained: one that predicts the 90th percentile and another that predicts the 10th percentile. This allows the system to estimate the interval in which the predicted value is contained with an 80% probability.

[0034] The prediction model 300a and quantile regression model 301a generated by the model generation unit 120 are stored in the storage unit 102 and provided to the prediction device 200. If the prediction device 200 and the learning device 100 are configured as a single unit, the models generated by the model generation unit 112 are directly stored in the storage unit 202 of the prediction device 200.

[0035] The model storage unit 210 of the prediction device 200 stores the prediction model 300a, which has been learned by machine learning, in the storage unit 202. The storage unit 202 stores the prediction model 300a (a model that outputs predicted values ​​of compression intensity) and the quantile regression model 301a (a model for estimating prediction intervals).

[0036] The prediction unit 220 uses the prediction model 300a stored in the memory unit 202 to predict the compressive strength at the input target age. Since the prediction model 300a is a regression model, predictions can be made at any age. Specifically, the prediction unit 212 predicts the compressive strength for each age from age 1 to 28, for example, based on the prediction conditions (cement type, curing method, nominal strength, planned sampling date, etc.) input by the user. The prediction unit 212 also calculates the prediction interval for each age using the quantile regression model 301a.

[0037] The results display unit 230 displays the changes in compressive strength corresponding to multiple predicted ages predicted by the prediction unit 220 on the display unit 205. For example, the results display unit 230 displays a graph with age on the horizontal axis and compressive strength on the vertical axis. The graph displays the predicted values ​​of compressive strength at each age, and the prediction interval is also displayed as needed. In addition, a baseline indicating the target strength (nominal strength) set by the user may be displayed on the graph.

[0038] [Structure of training data] Next, the configuration of the training dataset 400a in the first embodiment will be described. The training dataset 400a is constructed from past concrete test data. In this embodiment, by using a large amount of test data accumulated by the testing institution, it is possible to train a highly accurate predictive model 300a.

[0039] Figure 4 shows an example of the structure of the training data in the first embodiment. Each training data includes information on age, sampling date, temperature, curing method, cement type, nominal strength, etc., as explanatory variables, and compressive strength corresponding to the age as the objective variable. Here, information on the sampling date refers to information obtained based on the sampling date, such as numerical information of the sampling date or information on the month to which the sampling date belongs. In other words, one training data represents the test result of compressive strength at a specific (single) age, along with various condition information.

[0040] The details of the explanatory variables included in each training dataset are explained below. (1) Age of the material (test age) The age of the concrete refers to the number of days elapsed from the day the concrete sample is taken until the compression test is conducted. For example, compression tests are performed on concrete at various ages, such as 1 day, 3 days, 7 days, 14 days, and 28 days. Generally, the longer the age of the concrete, the higher its strength tends to be. (2) Information that quantifies the date of collection The sampling date (the date the concrete sample was collected) can be quantified. Specifically, the number of days elapsed from a predetermined reference date (for example, January 1, 1 AD) can be used. This quantification allows for the representation of the chronological order and time intervals between sampling dates, enabling the predictive model 300a to learn annual technological trends (such as improvements in cement manufacturing technology).

[0041] (3) The month in which the collection date falls The month in which the sample was collected (January to December) can be included as an explanatory variable. This monthly information allows the prediction model 300a to learn seasonal trends (e.g., higher temperatures and earlier intensity development in summer, lower temperatures and later intensity development in winter). (4) Temperature information related to the date of collection Since the strength development of concrete is greatly affected by temperature, temperature information can be included as an explanatory variable. In this embodiment, temperature information for a predetermined period prior to the sampling date is used. For example, the average temperature at the test site for 7 days prior to the sampling date (the week before the sampling date) can be used as an explanatory variable. For standard curing, the temperature of a temperature-controlled room can be used, and for on-site underwater curing, the outdoor temperature can be used.

[0042] A key feature of this embodiment is the use of temperature information from "prior" the sampling date. In general concrete strength prediction, methods using cumulative temperature (the accumulated temperature after concrete placement) are known. However, using cumulative temperature requires predicting future temperatures up to the target concrete age. For example, to predict the strength at 28 days, it is necessary to predict the temperature for the next 28 days, which may result in a decrease in prediction accuracy.

[0043] In contrast, this embodiment uses temperature information from before the sampling date (past), eliminating the need to predict future temperatures. This eliminates the uncertainty associated with temperature prediction.

[0044] Furthermore, the difference between the average temperature for the past week prior to the sampling date and the average temperature for the same period one year prior can also be included as an explanatory variable. This allows the prediction model 300a to learn annual temperature trends (such as temperature increases due to global warming).

[0045] (5)Curing method Curing methods refer to methods for curing concrete test specimens. Curing methods include standard curing, on-site underwater curing, air curing, and sealed curing. Standard curing is a method of curing in a temperature-controlled water tank indoors at the testing facility. On-site underwater curing is performed in a water tank outdoors at the testing facility to simulate conditions close to those of actual on-site temperature changes. Air curing is a method of curing the test specimen in the air, and sealed curing is a method of curing the test specimen by covering it with a sealed container or the like to prevent evaporation of moisture.

[0046] (6) Cement type Examples of cement types include ordinary Portland cement (N) and blast furnace cement type B (BB). Ordinary Portland cement is mainly used for building construction, while blast furnace cement type B is mainly used for civil engineering. The strength development characteristics differ depending on the type of cement. (7) Nominal strength The nominal strength is the design strength of concrete. For example, a nominal strength of 27 N / mm². 2 , 36 N / mm 2 These are some examples. The higher the nominal strength, the higher the actual compressive strength tends to be.

[0047] The target variable included in each training dataset is the compressive strength of the test results. Compressive strength is the value obtained by dividing the maximum load at which a concrete test specimen fails when a compressive load is applied by the cross-sectional area of ​​the specimen, and the unit is N / mm². 2 Alternatively, it is expressed in MPa.

[0048] Data number 1, shown in Figure 4, is training data corresponding to the results of a test conducted on a certain concrete sample at 7 days old under standard curing conditions. Data number 2 is training data corresponding to the results of a test conducted on a different concrete sample at 28 days old under standard curing conditions. Although data numbers 1 and 2 were collected on different dates, both were collected in October and are samples from the same season. Data number 3 is training data corresponding to the results of a test conducted on yet another concrete sample at 7 days old under in-situ water curing, and was collected in December. Thus, the training dataset 400a includes data on compressive strength for various combinations of conditions such as collection date, age, temperature conditions, curing method, cement type, and nominal strength.

[0049] Furthermore, the training dataset 400a includes data from different curing methods, which compensates for imbalances in the number of data points and biases in the age distribution, contributing to improved prediction accuracy. For example, underwater curing in the field has fewer data points and a greater bias in age distribution compared to standard curing. On the other hand, standard curing has abundant data, including data across various age ranges. By combining and learning from this data from different curing methods, the prediction model 300a learns the differences in strength development characteristics for each curing method and enables highly accurate predictions even for curing methods with limited data.

[0050] [Learning Process] Next, the flow of the learning process performed by the learning device 100 will be explained with reference to the flowchart in Figure 5. The control unit 101 of the learning device 100 executes the processing of each step in Figure 5, thereby generating the prediction model 300a.

[0051] The control unit 101 of the learning device 100 acquires past concrete test data from the memory unit (step S101). The test data includes information such as the sampling date, age, curing method, cement type, nominal strength, and compressive strength for each concrete sample.

[0052] Next, the control unit 101 generates a training dataset 400a from the acquired test data (step S102). For each test data, training data is generated with the age of the material, the quantified sampling date, the month of the sampling date, the average temperature for the past week before the sampling date, the temperature difference from one year ago, the curing method, the type of cement, and the nominal strength as explanatory variables, and the compressive strength as the objective variable. As described above, the generated training dataset 400a includes training data for various combinations of conditions.

[0053] Next, the control unit 101 generates a prediction model 300a (pre-casting model) using machine learning with the training dataset 400a (step S103). In this embodiment, a regression model is trained using LightGBM. The control unit 101 trains the prediction model 300a with the explanatory variables of the training dataset 400a as input and the target variable (compressive strength) as output. The trained prediction model 300a can output predicted values ​​of compressive strength for any combination of explanatory variables. In particular, since the prediction model 300a is a regression model, it is possible to continuously predict the compressive strength at any age, not limited to a specific age included in the training data.

[0054] Furthermore, the control unit 101 trains a quantile regression model 301a using the same training dataset 400a to estimate the prediction interval (step S104). The quantile regression model 301a is used to quantitatively evaluate the variability of the predicted values. Specifically, the control unit 101 trains two models: one to predict the upper quantiles and another to predict the lower quantiles. For example, when displaying the 80th percentile prediction interval, it trains a model to predict the 90th percentile and a model to predict the 10th percentile. This makes it possible to estimate the interval in which the predicted value is contained with an 80% probability.

[0055] The control unit 101 stores the learned prediction model 300a and quantile regression model 301a in the storage unit 102 (step S105). The stored models are provided to the prediction device 200 and used for prediction processing.

[0056] [Predictive processing] Next, the flow of the prediction process performed by the prediction device 200 will be explained with reference to the flowchart in Figure 6. The control unit 201 of the prediction device 200 performs the processing of each step in Figure 6, thereby predicting the change in the compressive strength of the concrete.

[0057] The control unit 201 of the prediction device 200 accepts prediction conditions from the user before concrete placement (step S201). Prediction conditions include cement type, curing method, nominal strength, and planned sampling date. The user inputs these prediction conditions via the input unit 204. For example, the cement type is N (ordinary Portland cement), the curing method is standard curing, and the nominal strength is 27 N / mm². 2 Enter a date such as December 1, 2025, as the planned collection date.

[0058] Next, the control unit 201 generates prediction parameters based on the prediction conditions (step S202). The prediction parameters are the values ​​of explanatory variables to be input to the prediction model 300a, and include information on the age of the material to be predicted, the sampling date, temperature information, cement type, curing method, and nominal strength. For example, the control unit 201 performs the following processing.

[0059] First, the control unit 201 calculates the numerical value of the sampling date (for example, the number of days elapsed since January 1, 1 AD) and the month of the sampling date from the scheduled sampling date. Next, the control unit 201 obtains temperature information from the storage unit 202 or an external database via the communication unit 203. As temperature information, we will obtain the average temperature for the week prior to the scheduled sampling date, and the difference between that average temperature and the average temperature for the same period one year ago. If actual measurement data is available, we will use that data; otherwise, we will substitute it with temperature data from the same period one year ago. For example, if the planned sampling date is December 1, 2025, and temperature data for November 24 to November 30, 2025 does not yet exist, the first feature (average temperature for the week prior to the planned sampling date) will be the average temperature from November 24 to November 30, 2024. Additionally, the second feature (difference from one year ago) will be calculated by determining the difference between the average temperature from November 24 to November 30, 2024, and the average temperature from November 24 to November 30, 2023. The cement type, curing method, and nominal strength entered by the user are included directly in the prediction parameters. The age of the material to be predicted is set in the prediction parameters while being sequentially changed in step S203, which will be described later.

[0060] Next, the control unit 201 uses prediction parameters to predict the compressive strength at multiple target ages (step S203). For example, the control unit 201 changes the target age from 1 to 28 in 1-day increments, and repeatedly inputs prediction parameters into the prediction model 300a stored in the memory unit 202 to obtain predicted values ​​for compressive strength. In this process, values ​​other than the target age included in the prediction parameters (digitized sampling date, month of sampling, temperature information, cement type, curing method, nominal strength) are fixed, and only the target age is changed, thereby obtaining predicted values ​​for compressive strength at each age of 1, 2, 3, ..., 28 days.

[0061] Furthermore, the control unit 201 calculates the prediction interval for each age using the quantile regression model 301a (step S204). Specifically, the control unit 201 inputs prediction parameters into the upper quantile model to obtain the upper quantile values, and inputs prediction parameters into the lower quantile model to obtain the lower quantile values. This calculates the prediction interval (e.g., 80% prediction interval) for each age.

[0062] The control unit 201 displays the changes in compressive strength at multiple predicted target ages as a graph on the display unit 205 (step S205). The horizontal axis of the graph represents age (days), and the vertical axis represents compressive strength (N / mm²). 2 The graph displays the predicted values ​​for each age group, and, if necessary, the prediction intervals.

[0063] [Example Display] Figure 7 shows an example of displaying the prediction results in the first embodiment. This example shows the display of the prediction results after inputting the prediction conditions before concrete placement. The horizontal axis represents the age (days), and the vertical axis represents the compressive strength (N / mm²). 2 The white circles (〇) indicate the predicted values ​​for each age of the wood. By displaying predicted values ​​for multiple ages, it is possible to understand the changes in compressive strength as the wood ages. In this example, the changes are displayed more visually by connecting each predicted value with a solid line in a line graph.

[0064] As shown in Figure 7, the change in compressive strength from age 1 to 28 is displayed. The control unit 201 may also display a reference line on the graph indicating the user-set target strength (nominal strength). In the example in Figure 7, the dashed line represents the target strength of 27 N / mm². 2 This shows the progression of strength at multiple ages, allowing users to visually understand when the target strength will be reached. By displaying the strength progression at multiple ages in this way, users can predict in advance when the concrete will reach a predetermined strength and plan the shortest possible construction schedule before concrete placement.

[0065] Figure 8 shows another example of displaying the prediction results in the first embodiment (with a prediction interval). This example also shows the same prediction results as Figure 7, but in addition to the predicted value, the prediction interval is displayed. The horizontal axis is the age (days), and the vertical axis is the compressive strength (N / mm²). 2 ) The white circles (〇) indicate the predicted values, and the shaded areas indicate the 80% prediction interval.

[0066] As shown in Figure 8, prediction intervals are displayed above and below the solid line connecting the white circles (〇) that represent the predicted values. The prediction intervals allow for decisions that take into account the variability of the predictions. For example, by checking when the lower limit of the prediction interval exceeds the target intensity (dashed line), it is possible to determine when the target intensity will be reached with a high probability. Displaying prediction intervals in this way allows users to make more reliable process control decisions. In addition, the degree of uncertainty in the prediction can be understood from the width of the prediction intervals.

[0067] As described above, according to the first embodiment, it is possible to predict the changes in compressive strength at multiple ages without using test results before concrete placement and display them as a graph. This allows users to consider process management at the stage before concrete placement and to understand in advance the likelihood of reaching a predetermined strength and the shortest possible process.

[0068] [Second Embodiment] [overview] The second embodiment predicts the compressive strength at a target age after concrete placement, using test results. This embodiment is particularly useful for concrete process management on site. By using test results at a specific age obtained after placement to predict the strength at a later age, it is possible to determine the time when the predetermined strength will be reached early and optimize the timing of transition to the next process. Furthermore, if the prediction based on the test results indicates a risk of not reaching the predetermined strength, a decision can be made early without waiting for a judgment period such as 28 days of age, and countermeasures such as re-placing can be considered.

[0069] The second embodiment differs from the first embodiment in that it uses test results as input to the prediction model 300b (post-placement model). The compressive strength obtained as test results is an actual measured value that shows the actual strength development status of the sampled concrete, and by making predictions based on this, it is possible to make predictions with higher accuracy than the prediction in the first embodiment.

[0070] [Functional Configuration] Figure 9 is a block diagram showing an example of the functional configuration of the prediction system 1b according to the second embodiment. The basic functional configuration of the prediction system 1b according to the second embodiment is the same as that of the first embodiment, and as shown in Figure 3, it comprises the functional parts of the learning device 100 and the prediction device 200. The differences from the first embodiment will be described below.

[0071] The learning data acquisition unit 110 of the learning device 100 generates and acquires a learning dataset 400b from past concrete test data. The learning dataset 400b consists of multiple learning data sets, with a first age, the compressive strength corresponding to the first age, and a second age greater than the first age as explanatory variables, and the compressive strength corresponding to the second age as the objective variable. Each learning data set is composed of a combination of test data for multiple ages taken from the same concrete. In addition to the first age, the compressive strength corresponding to the first age, and the second age, the learning data may also include information on the sampling date, curing method, cement type, nominal strength, etc., as explanatory variables. Details of the learning dataset 400b will be described later.

[0072] The model generation unit 120 generates a predictive model 300b (post-placement model) using machine learning. In this embodiment as well, LightGBM is used to train the predictive model 300b and the quantile regression model 301b. However, the configuration of the explanatory variables differs from that of the first embodiment.

[0073] The prediction unit 220 of the prediction device 200 takes the age of the material from the test results, the compressive strength from the test results, and the age of the material to be predicted as inputs, and predicts the compressive strength at the age of the material to be predicted. Specifically, the prediction unit 220 inputs the age of the material from the test results as the value corresponding to the first age during learning, the compressive strength from the test results as the value corresponding to the compressive strength corresponding to the first age, and the age of the material to be predicted as the value corresponding to the second age, into the prediction model 300b, and obtains the compressive strength at the age of the material to be predicted as the compressive strength corresponding to the second age.

[0074] The prediction unit 220 calculates the strength progression by sequentially making predictions for each age, from an age greater than the age of the test results up to 28 days. For example, when using test results from 7 days of age, it predicts the compressive strength for each age from 8 to 28 days. The prediction unit 220 also calculates the prediction interval for each age using the quantile regression model 301b.

[0075] The results display unit 230 displays the changes in compressive strength corresponding to multiple predicted ages, similar to the first embodiment. However, in the second embodiment, test results are also displayed. This allows the user to grasp the already obtained test results and the prediction of future strength development in an integrated manner. Note that the graph display of test results is not mandatory, and only the changes in compressive strength at the predicted age may be displayed.

[0076] [Structure of training data] Next, the configuration of the training dataset 400b in the second embodiment will be described. The training dataset 400b in the second embodiment is constructed from past concrete test data, similar to the first embodiment. However, the configuration of the training data differs from that of the first embodiment.

[0077] Figure 10 shows an example of the structure of the learning data in the second embodiment. As shown in Figure 10, each learning data includes, as explanatory variables, a first age, the compressive strength corresponding to the first age, a second age greater than the first age, information on the sampling date, curing method, cement type, nominal strength, etc., and as the objective variable, the compressive strength corresponding to the second age. Here, information on the sampling date refers to information obtained based on the sampling date, such as numerical information of the sampling date or information on the month to which the sampling date belongs. In other words, one learning data represents the relationship of "what happens to the compressive strength of older ages when the compressive strength of a certain age is known," along with various conditional information.

[0078] Let's explain in detail how training data is generated. First, data from compression tests conducted at multiple ages on concrete samples taken from the same pouring location and the same mixer truck is extracted. For example, if a single piece of concrete has undergone compression tests at both 7 days and 28 days of age, these two sets of data are combined to generate training data.

[0079] Specifically, the test results at 7 days of age are used as the first age and the corresponding compressive strength, the test results at 28 days of age are used as the second age, and the test results at 28 days of age are used as the compressive strength corresponding to the second age (target variable) to generate one training data set. In this way, by combining test results from multiple ages obtained from the same concrete, the relationship between the compressive strength at a specific age and the compressive strength at a later age can be represented as training data.

[0080] If a single concrete sample is tested at three or more different ages, multiple training data sets can be generated. For example, if tests are conducted at 7, 14, and 28 days of age, three training data sets can be generated from the three combinations: "7 days → 14 days", "7 days → 28 days", and "14 days → 28 days".

[0081] The details of the explanatory variables included in each training dataset are explained below. (1) 1st age The first age refers to the number of days elapsed from the day the concrete sample was taken until the compression test was conducted. In the training dataset 400b, tests were conducted on the same concrete sample at multiple ages, and the shorter of these ages is considered the first age. (2) Compressive strength corresponding to the first age The compressive strength corresponding to the first age is the result of the compression test at the first age.

[0082] (3) 2nd age The second age refers to the age at which a compression test was conducted on the same concrete at an age greater than the first age. For example, if a concrete sample is tested at both 7 days and 28 days, the first age would be 7 days and the second age at 28 days. In the training data, the result of the compression test at the second age becomes the target variable. (4) Information that quantifies the date of collection Similar to the first embodiment, numerical information representing the sampling date (the date the concrete sample was collected) can be used. Specifically, the number of days elapsed from a predetermined reference date can be used. (5) The month in which the collection date falls Similar to the first embodiment, the month in which the sampling date falls (January to December) can be included as an explanatory variable.

[0083] (6)Curing method Similar to the first embodiment, the curing method can be included as an explanatory variable. (7) Cement type Similar to the first embodiment, cement type (N, BB, etc.) can be included as an explanatory variable. (8) Nominal strength Similar to the first embodiment, the call intensity can be included as an explanatory variable.

[0084] The target variable included in each training dataset is the compressive strength corresponding to the second age of the wood. In other words, it is the value obtained from an actual compression test conducted at the second age.

[0085] Data number 1 shown in FIG. 10 is learning data combining the compressive strength (30.2 N / mm 2 ) obtained by performing a compression test on a certain concrete sample at the first age of 7 days and the results of a compression test performed on another specimen taken from the same sample at the second age of 28 days. Data number 2 is learning data generated from the same concrete sample (same sampling date) as data number 1, and the compressive strength (30.2 N / mm 2 ) is the same as that of data number 1, but the second age is 14 days. Thus, by performing tests on one concrete sample at multiple ages, a plurality of learning data with different second ages can be generated. Data number 3 is learning data for a concrete sample with a sampling date (December) different from those of data numbers 1 and 2, with the first age of 3 days, the second age of 28 days, and the curing method being field underwater curing.

[0086] The learning data set 400b includes learning data generated from concrete under conditions such as various sampling dates, curing methods, cement types, specified strengths, etc. Also, it is desirable to have a rich variety of combinations of the first age and the second age. For example, it includes learning data of various combinations such as "age 1 day → age 7 days", "age 1 day → age 28 days", "age 3 days → age 28 days", "age 7 days → age 14 days", "age 7 days → age 28 days", etc.

[0087] Note that the learning data set 400b includes data on different curing methods, which can complement the bias in the number of data and the age distribution. For example, although the number of data for field underwater curing is less compared to standard curing, by combining and learning the data of standard curing, the prediction model 300b can learn the characteristics of strength development due to the difference in curing methods and can perform highly accurate predictions for curing methods with less data.

[0088] [Learning process] Next, with reference to the flowchart in Figure 11, the flow of the learning process performed by the learning device 100 in the second embodiment will be explained. The control unit 101 of the learning device 100 performs the processing of each step in Figure 11, thereby generating the prediction model 300b.

[0089] The control unit 101 of the learning device 100 acquires past concrete test data from the memory unit (step S301). The test data includes information for each concrete, such as information identifying the placement location (placement location ID, mixer truck ID, etc.), sampling date, age, curing method, cement type, nominal strength, and compressive strength.

[0090] Next, the control unit 101 generates a training dataset 400b (step S302). Specifically, the control unit 101 identifies test data collected from the same concrete based on the pouring location ID, mixer truck ID, etc. If tests have been conducted on the same concrete at multiple ages, the control unit 101 combines this data to generate training data.

[0091] For example, if a concrete sample is tested at both 7 days and 28 days of age, the control unit 101 uses the data from 7 days of age as the first age and the corresponding compressive strength, the data from 28 days of age as the second age, and generates training data with the compressive strength at 28 days of age as the target variable. If a single concrete sample is tested at three or more ages, multiple training data sets are generated from all possible age combinations.

[0092] The training dataset 400b generated in this manner includes, as explanatory variables, the first age, the compressive strength corresponding to the first age, the second age, numerical information on the sampling date, the month of the sampling date, curing method, cement type, and nominal strength, and as the dependent variable, it includes the compressive strength corresponding to the second age.

[0093] Next, the control unit 101 generates a predictive model 300b using machine learning with the generated training dataset 400b (step S303). In this embodiment as well, LightGBM is used to train a regression model, similar to the first embodiment. The control unit 101 takes the explanatory variables of the training dataset 400b as input and the target variable (compressive strength corresponding to the second age) as output to train the predictive model 300b. The trained predictive model 300b can output a predicted value of the compressive strength at the target age for a combination of explanatory variables including test results and the target age. In particular, since the predictive model 300b is a regression model, it is possible to continuously predict the compressive strength at any age, not limited to a specific age included in the training data.

[0094] Furthermore, the control unit 101 trains a quantile regression model 301b using the same training dataset 400b to estimate the prediction interval (step S304). Similar to the first embodiment, two models are trained: one to predict the upper quantiles and another to predict the lower quantiles.

[0095] The control unit 101 stores the learned prediction model 300b and quantile regression model 301b in the storage unit 102 (step S305). The stored models are provided to the prediction device 200 and used for prediction processing.

[0096] [Predictive processing] Next, with reference to the flowchart in Figure 12, the flow of the prediction process performed by the prediction device 200 in the second embodiment will be explained. The control unit 201 of the prediction device 200 performs the processing of each step in Figure 12, thereby predicting the change in the compressive strength of the concrete.

[0097] The control unit 201 of the prediction device 200 receives input from the user regarding test results and prediction conditions after concrete placement (step S401). The test results are the results of compression tests already conducted, and include the age of the material and the compressive strength of the test results. The prediction conditions include cement type, curing method, nominal strength, sampling date, etc.

[0098] For example, a user conducted a compression test at 7 days of age, and the compressive strength was 30.2 N / mm². 2 Enter the test results, stating that the results were as follows: Also, specify the cement type as ordinary Portland cement (N), the curing method as standard curing, and the nominal strength as 27 N / mm². 2 Enter a date such as October 1, 2025, as the collection date.

[0099] Next, the control unit 201 generates prediction parameters based on the input information (step S402). The prediction parameters are the values ​​of explanatory variables input to the prediction model 300b, and include the age of the material to be predicted, the age of the material in the test results, the compressive strength of the test results, numerical information of the sampling date, the month of the sampling date, the curing method, the type of cement, and the nominal strength. The numerical information of the sampling date and the month of the sampling date are calculated from the sampling date. The age of the material to be predicted is set as a prediction parameter while being sequentially changed in step S403, which will be described later.

[0100] Next, the control unit 201 predicts the compressive strength at multiple target ages using prediction parameters (step S403). Specifically, the control unit 201 sequentially makes predictions for ages greater than the age of the test results. The control unit 201 repeatedly inputs prediction parameters into the prediction model 300b stored in the memory unit 202 to obtain predicted values ​​of compressive strength while sequentially changing the target age of prediction. For example, if the test results are for an age of 7 days, the control unit 201 predicts the compressive strength for each age while changing the target age of prediction from 8 to 28 days in 1-day increments.

[0101] In this process, the input to the prediction model 300b is fixed, with the age of the material in the test results (corresponding to the first age during learning), the compressive strength of the test results (corresponding to the compressive strength corresponding to the first age during learning), information on the sampling date, curing method, cement type, and nominal strength being fixed, and only the age of the material to be predicted (corresponding to the second age during learning) being changed. This allows for the prediction of the future strength development progression starting from the test results.

[0102] Furthermore, the control unit 201 calculates the prediction interval at each age using the quantile regression model 301b (step S404). Similar to the first embodiment, the prediction interval is calculated using the upper quantile model and the lower quantile model.

[0103] The control unit 201 displays the test results and the changes in compressive strength at multiple predicted ages as a graph on the display unit 205 (step S405). The graph displays the predicted values ​​and, if necessary, the prediction intervals. The test results may also be displayed on the graph. By displaying the test results together, it is possible to visually demonstrate that the prediction is made starting from the measured values.

[0104] [Predictions across different treatment methods] In the second embodiment, prediction across curing methods is possible. Prediction across curing methods means using test results obtained with one curing method to predict future compressive strength under different curing methods.

[0105] For example, suppose a test is conducted using standard curing at 7 days of age, and compressive strength is obtained (test results using the first curing method). When using these test results to predict the compressive strength from 8 days of age onward using on-site underwater curing (second curing method), the prediction unit 220 of the prediction device 200 inputs the age of the material from the test results, the compressive strength from the test results, the age of the material to be predicted, and on-site underwater curing as the curing method into the prediction model 300b. Based on these inputs, the prediction unit 220 predicts the compressive strength of the material at the age of the material to be predicted using on-site underwater curing. In this way, by specifying a curing method (on-site underwater curing) different from the curing method used in the test results (standard curing), predictions across curing methods can be achieved.

[0106] Such predictions are possible because the training dataset 400b includes data on various curing methods, and the prediction model 300b learns the characteristics of strength development depending on the curing method. In actual field work, curing methods may be changed depending on the progress after concrete placement and the site conditions, so the ability to predict between different curing methods is practically useful.

[0107] [Example Display] Figure 13 shows an example of displaying prediction results in the second embodiment. This example shows the display of prediction results using test results from a compression test conducted at 1 day after concrete placement. The horizontal axis represents age (days), and the vertical axis represents compressive strength (N / mm²). 2 ) where black circles (●) indicate test results and white circles (〇) indicate predicted values.

[0108] As shown in Figure 13, the change in compressive strength from day 2 onwards is predicted based on the test results at day 1 (black circles). The graph displays the already obtained test results and the prediction of future strength development together, allowing users to grasp the current situation and future outlook simultaneously. The target strength (nominal strength) is also indicated by a dashed line. By displaying the test results and predictions together in this way, users can appropriately determine the timing to proceed to the next process by considering both reliable information based on actual measurements and future predictions.

[0109] Figure 14 shows another example of displaying the prediction results in the second embodiment (with a prediction interval). Similar to Figure 13, this example also shows the prediction results using test results from a 1-day-old material, but in addition to the predicted value, the confidence interval (prediction interval) of the prediction is also displayed.

[0110] As shown in Figure 14, the predicted compressive strength from day 2 onwards (outlined circles) is displayed along with the predicted interval (shaded area), based on the test results at day 1 (black circles). The predicted interval allows for a quantitative evaluation of the variability of the prediction. For example, by checking the range where the lower limit of the predicted interval exceeds the target strength (dashed line), it is possible to determine the period during which the target strength can be maintained with a high probability. In this way, by displaying the test results, predicted values, and predicted intervals together, users can make decisions regarding process control that take into account reliable information based on actual measurement data, predictions of future strength development, and uncertainties in the predictions.

[0111] Note that while Figures 13 and 14 use test results from one day of age, this is merely an example for illustrative purposes. In actual operation, test results from any age can be used. For example, results from compression tests conducted at 3, 5, and 7 days of age can be used as test results to predict the change in compressive strength from the following day onward.

[0112] Even when predictions span across curing methods, the graph is displayed in the same format as in Figure 13 or Figure 14. For example, if the compressive strength after 8 days of age in on-site underwater curing is predicted using the test results obtained at 7 days of age obtained under standard curing, the future strength development trend in on-site underwater curing will be displayed, starting from the test results under standard curing. This allows for a visual understanding of prediction results that take into account the strength development characteristics of different curing methods.

[0113] Thus, according to the second embodiment, after concrete placement, the changes in compressive strength at multiple ages can be predicted using test results and displayed as a graph. Since the test results reflect actual field conditions, more accurate predictions are possible than in the first embodiment.

[0114] [Differentiation] Although embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and various modifications are possible. Several modifications will be described below.

[0115] (Variation 1: Continuous relearning) In the above embodiment, the prediction model 300 was trained using past concrete test data, but the prediction model 300 can be retrained when new test data is accumulated. This makes it possible to maintain a prediction model 300 that reflects the latest technological trends and changes in material properties. Specifically, the learning data acquisition unit 110 of the learning device 100 adds newly acquired concrete test data (data including age and compressive strength) to the existing learning dataset 400. Then, the model generation unit 120 retrains the prediction model 300 and the quantile regression model 301 using the updated learning dataset 400 including the added learning data, and overwrites and saves them to the storage unit 102. The timing of retraining can be, for example, by adding new test data periodically (every month, every year, etc.), or by adding measured values ​​obtained by actually performing compression tests after performing predictions in the prediction device 200 to the learning data.

[0116] (Variation 2: Using multiple models together) The prediction system 1 includes both the prediction model 300a (pre-casting model) of the first embodiment and the prediction model 300b (post-casting model) of the second embodiment, and the user may choose to use whichever is more appropriate.

[0117] Specifically, the memory unit 202 of the prediction device 200 stores both the pre-casting model 300a and the post-casting model 300b. The display unit 205 of the prediction device 200 displays a selection screen for the user to choose a prediction method. The selection screen is provided with, for example, a "Pre-casting prediction" button and a "Post-casting prediction" button.

[0118] When the user selects the "Pre-Placement Prediction" button, the control unit 201 selects the pre-placement model 300a and displays an input screen for the prediction conditions necessary for predicting concrete placement before it is poured (cement type, curing method, nominal strength, planned sampling date, etc.). Once the user enters the prediction conditions, the control unit 201 uses the pre-placement model 300a to predict the change in compressive strength and displays the results.

[0119] On the other hand, if the user selects the "Predict after placement" button, the control unit 201 selects the post-placement model 300b and displays an input screen for test results and prediction conditions (age and compressive strength of test results, cement type, curing method, nominal strength, sampling date, etc.). Once the user has entered the test results and prediction conditions, the control unit 201 uses the post-placement model 300b to predict the change in compressive strength and displays the results.

[0120] With this configuration, pre-placement model 300a is used to make preliminary predictions necessary for process planning before concrete placement, and after placement, once test results are obtained, it is possible to switch to post-placement model 300b to make more accurate predictions based on actual measurement data. This makes it possible to provide optimal predictions at each stage of concrete construction.

[0121] (Modification 3: Output of the test results report) In the second embodiment, the prediction device 200 may not only display the prediction results graphically but also have a function to output them in the form of a test report (also called a test report or test report form). A test report is a formal document that records the results of a concrete compressive strength test and is a standard deliverable that testing institutions deliver to test clients (construction companies, etc.) in the construction industry. Conventional test reports include the results of the compression test performed (age, compressive strength, etc.) as numerical values, but do not include predictions of future strength development. By including prediction results in the test report, it is possible to provide added value that is not present in conventional report forms.

[0122] The output process for the test results report will now be explained. In step S405 of the prediction process shown in Figure 12, the control unit 201 of the prediction device 200 displays the prediction results as a graph, or in parallel with the graph display, and receives an output instruction from the user. When the user instructs the output of the test results report via the input unit 204, the control unit 201 executes the process to generate the test results report.

[0123] Specifically, the control unit 201 first reads a test report template from the storage unit 202. The test report template defines an area for report entry items, such as the project name, test items, and test conditions, and a graph display area for displaying graphs. Next, the control unit 201 places the information entered by the user (such as the project name) and the test results data into the report entry items area. Furthermore, the control unit 201 places a graph in the graph display area that shows the changes in compressive strength corresponding to multiple predicted target ages predicted in step S403. This graph includes the age and compressive strength of the test results and the changes in the predicted compressive strength. In addition, the prediction interval can be included in the graph as needed.

[0124] The control unit 201 saves the generated test report as an electronic file in a predetermined output format (PDF, Word, Excel, etc.) to the storage unit 202, or transmits it to the user's terminal device via the network. Alternatively, the control unit 201 may print the test report as a paper document via a printer connected to the peripheral device I / F unit 206.

[0125] Figure 15 shows an example of the output of the test report 500. The test report 500 includes the conventional report information 510 (project name, test items, test conditions, etc.) and the measured values ​​of age and compressive strength obtained as test results, as well as a graph display area 520 that shows the predicted trend starting from the measured values. In the example shown in Figure 15, the measured values ​​at age 3 are recorded in the report information 510, and the graph display area 520 displays the predicted trend from age 4 onwards as a graph, starting from the test results (age 3). In this way, by including the graph display area 520 in the test report 500, the test requester can grasp the measured data and future forecasts in an integrated manner.

[0126] (Modification 4: Notification function) The prediction device 200 may also have a function to notify the display unit 205 (for example, on a graph) of various information based on the prediction results. For example, if the prediction device 200 determines that there is a high probability that the predicted compressive strength will not reach the target strength (nominal strength), it will display a message on the display unit 205 such as "There is a possibility that the target strength will not be reached at 28 days of age." Specifically, such a message will be displayed when the upper limit of the prediction interval (for example, the 80% prediction interval) calculated by the quantile regression model 301a falls below the target strength at a predetermined age (for example, 28 days). In addition, the prediction device 200 may identify the age at which the compressive strength is predicted to reach the target strength based on the predicted values ​​of the compressive strength at each age from 1 to 28, and notify the display unit 205 of the date corresponding to that age as the "date on which the material can be moved to the next process." Such notification functions allow the user to easily grasp the situation and make smooth decisions regarding process control.

[0127] (Modification 5: Addition of other features) In the above embodiment, age, sampling date, temperature, curing method, cement type, and nominal strength were used as explanatory variables, but other features may be added. For example, including regional information (regional classification, latitude and longitude, etc.) as explanatory variables makes it possible to make predictions that take regional characteristics into account.

[0128] (Variation 6: Use of other machine learning algorithms) In the above embodiment, the predictive model 300 was trained using LightGBM, but other machine learning algorithms may be used. For example, the predictive model 300 may be trained using random forests, support vector regression, neural networks, etc. Alternatively, ensemble learning combining multiple machine learning algorithms may be used.

[0129] (Modification 7: Range of age range for prediction) In the above embodiment, an example of making predictions within the range of wood age from 1 to 28 days was described, but the range of wood ages to be predicted is not limited to this. For example, predictions may be made for longer ages, such as 56 days (2 months) or 91 days (3 months). By adjusting the range of wood ages included in the training dataset 400 during the training of the prediction model 300, a prediction model 300 that corresponds to any wood age range can be constructed. Furthermore, the lower limit of the prediction is not limited to 1 day; predictions may start from an earlier age, such as 0.5 days (12 hours). In this way, the prediction system 1 can predict the changes in compressive strength within any wood age range, according to the user's needs and applications.

[0130] (Variation 8: Display format) In the above embodiment, an example of displaying the trend of compressive strength as a line graph was shown, but the display format is not limited to this. For example, the trend may be displayed using other display formats such as bar graphs, area graphs, or scatter plots. Furthermore, multiple display formats may be configured to be switchable. The graph format can be appropriately selected according to the user's visibility and application.

[0131] [Summary of Embodiments] As described above, the prediction system 1 of this embodiment can display the changes in compressive strength corresponding to multiple target ages using a prediction model 300 trained by machine learning based on a training dataset 400 in which age is the explanatory variable and compressive strength is the target variable. This allows users to visually grasp the changes in concrete strength development and efficiently consider process control and make quality control decisions.

[0132] In the first embodiment (pre-placement model), the change in compressive strength can be predicted before concrete placement without using test results, making it useful for planning the construction process. In the second embodiment (post-placement model), predictions are made using test results, enabling more accurate predictions and making it useful for on-site process management and quality assessment. The prediction system of this embodiment supports concrete quality control and process management in building and civil engineering works, and greatly contributes to ensuring the quality of structures and improving the efficiency of construction. [Explanation of symbols]

[0133] 1: Prediction System 1a: Prediction system 1b: Prediction system 100: Learning device 110: Training data acquisition unit 111: Training data acquisition unit 112: Model generation unit 120: Model generation unit 200: Prediction device 210: Model Memory Unit 220: Prediction Section 230:Result display section 300: Training dataset 300: Predictive Model 300a: Predictive Model 300b: Predictive Model 301a: Quantile regression model 301b: Quantile regression model 400a: Training dataset 400b: Training dataset 500: Exam Score Report 510: Items to be included on the report card 520: Graph display area

Claims

1. A prediction system for predicting the compressive strength of concrete, A generation unit that generates a predictive model using machine learning with a training dataset in which wood age is the explanatory variable and compressive strength is the target variable, A prediction unit that uses the aforementioned prediction model to predict the compressive strength at the input target age, The system includes a display unit that displays the progression of compressive strength corresponding to a plurality of predicted target ages predicted by the prediction unit. Prediction system.

2. Each training data set other than the aforementioned training dataset is: The explanatory variable includes a specific age of the wood. The objective variable includes the compressive strength corresponding to the specific age of the material. The prediction system according to claim 1, wherein the prediction unit predicts the compressive strength at the age to be predicted, taking the age to be predicted as input, before concrete is poured.

3. Each training data set other than the aforementioned training dataset is: The explanatory variables include a first age, the compressive strength corresponding to the first age, and a second age greater than the first age. The objective variable includes the compressive strength corresponding to the second age of the material, The prediction system according to claim 1, wherein the prediction unit, after concrete placement, takes the age of the material from the test results, the compressive strength from the test results, and the age of the material to be predicted as input, and predicts the compressive strength at the age of the material to be predicted.

4. The prediction system according to any one of claims 1 to 3, wherein each training data further includes information about the collection date as an explanatory variable.

5. The prediction system according to claim 4, wherein the information relating to the sampling date includes numerical information of the sampling date, and the numerical information represents the time elapsed since a predetermined reference date.

6. The prediction system according to claim 4, wherein each training data further includes information about the month to which the data collection date belongs as an explanatory variable.

7. The prediction system according to any one of claims 1 to 3, wherein each training data further includes a cultivation method as an explanatory variable.

8. The prediction system according to any one of claims 1 to 3, wherein each training data further includes cement type and specified strength as explanatory variables.

9. The prediction system according to claim 2, wherein each training data further includes temperature information related to the sampling date as an explanatory variable.

10. The prediction system according to claim 9, wherein the temperature information is temperature information for a predetermined period prior to the sampling date.

11. The prediction unit predicts the prediction interval along with the compressive strength at each of the plurality of target ages for prediction. The prediction system according to any one of claims 1 to 3, wherein the display unit displays the change in compressive strength and the prediction interval.

12. Each training data set further includes the cultivation method as an explanatory variable. The aforementioned test results were obtained using the first curing method. The prediction system according to claim 3, wherein the prediction unit takes the age of the material in the test results, the compressive strength of the material in the test results, the age of the material to be predicted, and a second curing method different from the first curing method as input, and predicts the compressive strength of the second curing method.

13. The prediction system according to claim 3, which outputs a test results table including the age and compressive strength of the test results and a graph showing the trend of compressive strength corresponding to the multiple predicted target ages.

14. A predictive device for predicting the compressive strength of concrete, A memory unit that stores a prediction model trained by machine learning to predict compressive strength using explanatory variables including age as input, A prediction unit that predicts the compressive strength at the target age using the prediction model, A prediction device comprising: a display unit that displays the changes in compressive strength corresponding to a plurality of predicted target ages predicted by the prediction unit; and a prediction device comprising:

15. The aforementioned prediction model takes the age of the material as input and outputs the compressive strength corresponding to that age. It is, The prediction device according to claim 14, wherein the prediction unit inputs the target age to be predicted into the prediction model before concrete placement and predicts the compressive strength at the target age.

16. The prediction model takes a first age, the compressive strength corresponding to the first age, and a second age greater than the first age as inputs to predict the compressive strength at the second age. The prediction device according to claim 14, wherein the prediction unit, after concrete placement, inputs the age of the material from the test results, the compressive strength from the test results, and the age to be predicted into the prediction model, and predicts the compressive strength at the age to be predicted.

17. The aforementioned storage unit is A first prediction model that takes age as input and outputs the compressive strength corresponding to that age, A second prediction model takes a first age, the compressive strength corresponding to the first age, and a second age greater than the first age as inputs and outputs the compressive strength corresponding to the second age, and stores these. The system further includes a selection unit that accepts a selection from the user for either the first or second prediction model. The prediction unit, If the first prediction model is selected, the target age of the material is input to the first prediction model to predict the compressive strength. The prediction device according to claim 14, wherein, when the second prediction model is selected, the age of the material from the test results, the compressive strength from the test results, and the age of the material to be predicted are input to the second prediction model to predict the compressive strength.

18. A program for causing a computer to function as a prediction device according to any one of claims 14 to 17.

Images