JP7653331B2 - Ready-mix concrete strength prediction system - Google Patents

Description

The present invention relates to a system for predicting the strength of fresh concrete.

One of the important properties of ready-mix concrete is concrete compressive strength (hereafter simply referred to as "compressive strength"). When designing a structure, compressive strength is used as a criterion for determining the structure and the timing for removing the formwork.

As mentioned above, the compressive strength of ready-mixed concrete is an important index used as a design standard. For this reason, ready-mixed concrete is generally inspected a specified number of days after it is poured to check whether it has developed the required compressive strength. If this inspection determines that the specified standard is not met, the ready-mixed concrete that was inspected has already been poured, and so reinforcement work must be carried out on the structure that uses that ready-mixed concrete.

In addition, although testing of the compressive strength of ready-mix concrete is generally performed using quality control specimens rather than the poured concrete itself, there has been no standardization of the management methods for these specimens. Furthermore, because poured ready-mix concrete and the specimens that are actually subject to testing are stored in different environments, it is anticipated that the compressive strength at the time of quality testing may differ.

In recent years, instead of preparing test specimens separately, attempts have been made to predict the compressive strength a certain number of days after casting based on the mix data of the ready-mixed concrete when it is shipped from the manufacturing plant. For example, the following Patent Document 1 describes a method for estimating the compressive strength of ready-mixed concrete after casting based on the water-cement ratio (C/W) and slump value, etc., using a specified formula.

JP 2009-98055 A

It is known that the compressive strength of fresh concrete is linear over time if the water-cement ratio (C/W) is within a certain range, and linear approximation equations determined for each type of cement are used to predict the compressive strength of fresh concrete.

However, the way in which ready-mixed concrete is actually produced varies depending on the mixing method in the mixer during production and the environment at the pouring site, and the way the hydration reaction progresses varies depending on the conditions. For this reason, the compressive strength of ready-mixed concrete over time does not actually develop according to a linear approximation formula based on information about the ready-mixed concrete mix, making it difficult to predict accurately.

In addition to compressive strength, the strength of fresh concrete also includes flexural strength and Young's modulus, and it is expected that flexural strength, as well as standard underwater curing strength and on-site underwater curing strength, can be predicted with high accuracy.

In view of the above problems, the present invention aims to provide a ready-mix concrete strength prediction system that can predict the strength of ready-mix concrete with high accuracy before quality inspection.

The ready-mix concrete strength prediction system of the present invention comprises:
A storage unit that stores a plurality of trained models generated by machine learning based on training data;
A data input unit that accepts input of prediction input data;
A data output unit that outputs data including strength prediction information of fresh concrete derived by the plurality of trained models,
the plurality of trained models include a first trained model that is generated by machine learning based on first teacher data and that derives primary intensity prediction information, a second trained model that is generated by machine learning based on second teacher data and that derives secondary intensity prediction information, and a third trained model that is generated by machine learning based on third teacher data and that derives tertiary intensity prediction information,
The first teacher data is data in which first learning input data including mix-related information, which is information related to the mix-related composition of the ready-mixed concrete, and first learning output data including strength information of the ready-mixed concrete are associated with each other,
the second teacher data is data in which second learning input data including primary strength prediction information derived by the first trained model and production-related information that is information related to the production of ready-mixed concrete is associated with second learning output data including strength information of ready-mixed concrete,
The third teacher data is data in which third learning input data including secondary strength prediction information derived by the second trained model and quality-related information that is information related to the strength of fresh concrete obtained after pouring and before the time to be predicted, and third learning output data including strength information of fresh concrete are associated with each other,
The input data for prediction includes the mix-related information, the production-related information, and the quality-related information of the ready-mix concrete to be predicted.

The mix-related information used in the quality prediction system of the present invention is used as an element for predicting how the hydration reaction will progress in fresh concrete after casting and the level of strength that will be developed over time. Specific details of the mix-related information are listed and explained in the section "Form for carrying out the invention."

Manufacturing-related information is information that contributes to the speed at which the hydration reaction progresses in the ready-mix concrete being produced, and is used as a factor in predicting how the hydration reaction will progress in the ready-mix concrete after it is poured, and the level of strength that will be developed over time.

The quality-related information includes information such as the compressive strength, flexural strength, and Young's modulus of fresh concrete at a specified age after casting (e.g., 1 day, 3 days, 7 days, 14 days, etc.). The quality-related information also includes information on strength after on-site underwater curing and standard underwater curing.

The strength prediction system of the present invention first derives primary strength prediction information related to the strength actually exhibited in fresh concrete from the mix-related information of the fresh concrete to be predicted, using a first trained model that is trained based on the relationship between mix-related information and the actual strength of fresh concrete.

When manufacturing-related information is input to the data input section of the strength prediction system of the present invention, the second trained model corrects the primary strength prediction information output by the first trained model based on the manufacturing-related information, and derives secondary strength prediction information. As a result of this correction, the strength prediction of fresh concrete by the strength prediction system becomes a prediction that includes deviations due to information from the manufacturing-related information that affects the development of the strength of fresh concrete, such as the mixing time in the mixer.

When quality-related information is input to the data input section, the strength prediction system of the present invention has a third trained model that corrects the secondary strength prediction information output by the second trained model based on the quality-related information, and derives tertiary strength prediction information. As a result of this correction, the strength prediction of fresh concrete by the strength prediction system is corrected by intermediate information regarding the actual strength development. In other words, the strength prediction system outputs tertiary strength prediction information that is further corrected from the secondary strength prediction information based on information regarding the degree of deviation between the strength prediction based on the secondary strength prediction information and the actual strength of fresh concrete developed.

In other words, the strength prediction system of the present invention can make predictions that take into account not only the mix-related information of fresh concrete, but also manufacturing-related information and quality-related information, which are factors that affect the strength development of fresh concrete, and can predict the strength development of fresh concrete with higher accuracy. Furthermore, since the strength prediction system of the present invention is configured to predict strength development while correcting the strength prediction information derived by the previous trained model based on the strength prediction information output by each trained model and the manufacturing-related information and quality-related information, it is possible to easily analyze which information has a significant impact on the strength development of fresh concrete by analyzing the strength prediction information derived by each trained model.

The strength prediction system includes:
In the teacher data corresponding to at least one of the plurality of trained models, learning input data including environment-related information that is information related to the environment at a manufacturing site, a pouring site, or a transportation route of the ready-mixed concrete is associated with learning output data,
As the input data for prediction, environment-related information regarding the ready-mix concrete to be predicted may be input.

Environment-related information is used as a factor to identify the type of environment to which poured ready-mix concrete will be exposed by the time to be predicted.

Temperature and humidity affect the progress of the hydration reaction in fresh concrete, and are therefore factors that affect the prediction of strength. Therefore, with the above configuration, the strength prediction system of the present invention can predict the strength of fresh concrete with greater accuracy.

The strength prediction system includes:
The temperature sensor may be provided to acquire temperature information inside the fresh concrete, temperature information inside the hardened body of the fresh concrete after pouring, temperature information around the fresh concrete, or temperature information around the hardened body of the fresh concrete after pouring, and input data including the acquired temperature information to the data input unit as the quality-related information.

In this specification, "surrounding area" refers to an area that is 1 km or less away from the surface of the fresh concrete.

Since ready-mix concrete has the characteristic of generating heat as the hydration reaction progresses, its strength can be predicted by measuring the accumulated temperature. Therefore, with the above configuration, the strength prediction system of the present invention can predict the strength of ready-mix concrete with higher accuracy.

In the above strength prediction system,
The data output unit may be configured to be able to select the strength prediction information of fresh concrete to be included in the output data from among the primary strength prediction information, the secondary strength prediction information, and the tertiary strength prediction information.

The above configuration allows each user to switch the information output from the data output unit to the information they want to refer to. For example, a production site staff member or manager may want to refer to prediction information based on mix-related information to determine whether to readjust the mix ratio of each raw material, and can switch to output primary strength prediction information from the data output unit. In addition, a pouring site manager or a ready-mix concrete manufacturing plant manager may want to understand the production process and check whether there are any problems with the production process, and can switch to output all strength prediction information from the data output unit.

In the above strength prediction system,
the plurality of trained models further include a weighting trained model for determining a coefficient for weighting each piece of information included in the prediction input data according to its contribution to the prediction of strength of fresh concrete, the weighting trained model being generated by machine learning based on a plurality of teacher data obtained by associating weighting learning input data with weighting learning output data;
the weighting learning input data includes at least one of the primary intensity prediction information, the secondary intensity prediction information, and the tertiary intensity prediction information,
The weighting learning output data may include strength measurement value information that is information regarding the actually measured strength of the ready-mixed concrete from which the strength prediction information is derived.

The strength prediction system includes:
At least one of the mix-related information, the production-related information, and the quality-related information may be configured to be acquired from an information management system in a ready-mix concrete manufacturing plant.

An "information management system in a ready-mix concrete factory" is a system that is introduced in a ready-mix concrete manufacturing factory, and is a system for managing information related to the production, shipping, transportation, etc. of ready-mix concrete. In this specification, the "information management system" is not limited to a large-scale management system, but also includes, for example, a simple system in which a database of information related to ready-mix concrete is configured and the necessary information can be read out as needed.

With the above configuration, the strength prediction system of the present invention finds the more influential factors each time it repeats predicting the strength of fresh concrete, and the trained model is automatically optimized so that predictions gradually place more emphasis on those factors.

The present invention provides a ready-mix concrete strength prediction system that can predict the strength of ready-mix concrete with high accuracy before quality inspection.

FIG. 1 is a schematic overall configuration diagram of one embodiment of a quality prediction system for ready-mix concrete. FIG. 2 is a block diagram illustrating a configuration of a main server according to an embodiment of the present invention. FIG. 2 is a block diagram illustrating details of input and output of each data item performed within the main server.

The ready-mix concrete strength prediction system of the present invention will be described below with reference to the drawings. Note that each of the drawings is a schematic illustration, and the numbers on the drawings do not necessarily correspond to the actual numbers.

The ready-mix concrete strength prediction system of the present invention is a system that predicts the strength of ready-mix concrete using a trained model generated by machine learning based on multiple training data that associates training input data, including information related to the mix ratio, manufacturing, and quality of ready-mix concrete, with training output data, including strength information of ready-mix concrete, as described below.

The trained models installed in the strength prediction system of the present invention consist of a first trained model, a second trained model, and a third trained model. The information contained in the learning input data applied to each trained model is classified into first learning input data, second learning input data, and third learning input data for the trained model to which it is applied. The strength prediction information derived from each trained model is classified into primary strength prediction information, secondary strength prediction information, and tertiary strength prediction information.

In this embodiment, the first learning input data is mix-related information, the second learning input data is manufacturing-related information, and the third learning input data is quality-related information. First, details of the mix-related information, manufacturing-related information, and quality-related information, and the relationship between this information and the strength of ready-mix concrete will be described. Note that each learning input data may also include environment-related information, so the environment-related information will also be described here.

[Formulation related information]
Specific examples of the mix-related information include the water-cement ratio (W/C), cement type, admixture type, admixture amount, and unit water amount. The mix-related information applied to the first trained model used in the present invention may be one of the above mix-related information or a combination of two or more of them.

Mix-mix related information is information that contributes to the speed at which the hydration reaction progresses in the fresh concrete being produced, and is an element for predicting how the hydration reaction will progress in the fresh concrete after it is poured, and how the strength will change over time. For this reason, strength prediction information related to the predicted value of the strength of fresh concrete can be associated with mix-mix related information.

Therefore, by applying the first trained model for predicting the strength of ready-mix concrete generated by machine learning using first training input data including mix-related information and training output data including strength prediction information, it is possible to predict the strength of the ready-mix concrete to be produced.

Other examples of mix-related information that can be used include, but are not limited to, data on the raw materials of cement clinker, data on the burning conditions of cement clinker, data on the grinding conditions of cement, etc. Mix-related information may be mix-related information for cement production, as well as mix-related information for aggregate, water, and other materials that are mixed into ready-mix concrete, and may also be information such as the shipping date and shipping time of ready-mix concrete that is taken into account in determining the mix ratio of aggregate, water, and other materials in ready-mix concrete. For example, the design strength of concrete, unit water content, unit cement content, cement density, cement specific surface area, type of coarse aggregate, unit amount of coarse aggregate, type of fine aggregate, unit amount of fine aggregate, aggregate density, coarse aggregate ratio, fine aggregate ratio, particle size distribution, density of fine aggregate and coarse aggregate, water absorption rate, water content, surface water rate, maximum size, chemical composition of cement, mineral composition, particle size distribution, sieve test residue amount, color tone, mineralogical properties and crystallographic properties of each mineral contained in cement, and hemihydration rate of gypsum contained in cement, and other data related to the entire cement can also be used. In particular, the design slump value can be preferably used. Since the design slump value information is highly correlated with the unloading slump value, the unloading slump value can be predicted more accurately by using the design slump value as the mixture related information, especially when other mixture related information is scarce.

[Manufacturing related information]
Specific examples of the manufacturing related information include the mixer type, mixer model, mixing time, mixing sound, power load value, temperature of fresh concrete during mixing, temperature and humidity inside the mixer, etc. The manufacturing related information applied to the second trained model may be one of the above manufacturing related information or a combination of two or more of them.

Production-related information is information that can confirm the hardness and fluidity of fresh concrete during mixing, and contributes to the speed at which the hydration reaction progresses in the fresh concrete being produced. It is also information that can be used to predict how the hydration reaction will progress in the fresh concrete after it is poured, and how the strength will change over time.

Therefore, a quality prediction system to which a second trained model generated by machine learning using second training input data including manufacturing-related information and training output data including strength prediction information is applied can predict the strength of fresh concrete with higher accuracy.

Other production-related information that can be used include the amount of air, temperature of each material, temperature and humidity of the storage location (container), batcher data (type of heating/cooling equipment, temperature of stored water, temperature of stored aggregate, temperature of stored cement, measuring bottle type, discharge volume, maximum discharge pressure, etc.), chloride content, crack resistance, dynamic modulus of elasticity, dynamic shear modulus of elasticity, dynamic Poisson's ratio, void volume and void size distribution of the hardened body, durability, color, separation status of each material in the fresh concrete, fluidity, rheological values (plastic viscosity, yield value, etc.), chloride content in the fresh concrete, etc.

［Quality related information］
The quality-related information is information related to the strength of fresh concrete obtained after casting and before the time to be predicted. In other words, the compressive strength, bending strength, Young's modulus, etc. of fresh concrete at a predetermined age (e.g., 1 day, 3 days, 7 days, 14 days, etc.) before the 28-day age, which is the prediction time, are included. The quality-related information applied to the third trained model may use only one of the above quality-related information, or may use two or more types in combination. In addition, the quality-related information also includes temperature information of the inside of the fresh concrete, the inside of the hardened body after casting of the fresh concrete, or the surroundings. Note that the hardened body after casting of the fresh concrete means one that has hardened at least more than the state of the fresh concrete, and is not limited to one that has hardened to the extent that it reaches a common standard strength.

The quality-related information is used to determine the extent to which the actual strength of fresh concrete after a specified time has elapsed since pouring differs from the strength prediction information derived by the first trained model and the second trained model, and to correct any deviation from the prediction.

Therefore, a strength prediction system to which a third trained model generated by machine learning using training input data including quality-related information and training output data including strength prediction information is applied can predict the strength of fresh concrete with higher accuracy.

[Environment-related information]
Specifically, the environment-related information is the amount of solar radiation, wind speed, wind direction, weather, temperature, humidity, and altitude at the casting site, or the weather, humidity, altitude, and outside temperature at the manufacturing site of the ready-mixed concrete. In particular, the outside temperature at the manufacturing site is suitable as information used to predict the strength of ready-mixed concrete. Other applicable environment-related information includes the weather, humidity, altitude, and outside temperature of the route along which the agitator vehicle travels, that is, information related to the environment on the transportation route of the ready-mixed concrete. Special means are required to acquire environment-related information during transportation, such as the temperature inside the agitator vehicle and the weather of the route along which the agitator vehicle travels, but by using the outside temperature at the manufacturing site, the environment-related information can be acquired in a relatively simple manner. The environment-related information applied to each trained model may be only one of the above-mentioned environment-related information, or may be a combination of two or more types.

Environment-related information is an element for identifying the type of environment to which fresh concrete will be exposed after pouring. In other words, it is an element for understanding whether the progress of the hydration reaction of fresh concrete will be accelerated or delayed due to the influence of the outside temperature, etc.

Therefore, a quality prediction system that applies a trained model generated by machine learning using training input data including environmental information and training output data including strength prediction information can predict the strength of fresh concrete with higher accuracy.

From the above, the quality prediction system to which the first trained model, the second trained model, and the third trained model are applied can predict the strength of fresh concrete with higher accuracy than predictions made by linear approximation based on mix-related information.

Furthermore, a quality prediction system that applies a trained model generated by performing machine learning using training input data including environmental information and training output data including strength prediction information can predict the strength of fresh concrete with higher accuracy.

At least one of the above-mentioned mix-related information, production-related information, and quality-related information may be obtained from an existing information management system in a ready-mix concrete factory. Specifically, the mix-related information may be obtained from a system that manages mix-related information of specified materials that are normally used when ready-mix concrete is shipped in an existing ready-mix concrete factory. The quality-related information may be obtained from an information management system that inputs data such as compressive strength, bending strength, and Young's modulus of ready-mix concrete at a specified age (e.g., age of 1 day, 3 days, 7 days, 14 days, etc.) obtained by taking a sample and creating a test specimen at the time of shipping the ready-mix concrete, and stores and manages the data. Furthermore, production-related information such as mixing time, mixing sound, power load value, temperature of ready-mix concrete during mixing, and temperature and humidity inside the mixer may be obtained from a management system that manages the situation when the ready-mix concrete is mixed. Note that the above-mentioned information management system is one example of a method for acquiring mix-related information, production-related information, and quality-related information, and does not limit the method for acquiring each related information.

The number of training data samples used for training each trained model is preferably 100 or more, more preferably 1,000 or more, even more preferably 10,000 or more, even more preferably 50,000 or more, and particularly preferably 100,000 or more, from the viewpoint of extracting features necessary for deriving unloading slump information from the training input data and further improving prediction accuracy. Furthermore, the number of training times for each trained model is preferably 1,000 or more, more preferably 8,000 or more, and particularly preferably 10,000 or more, from the viewpoint of improving prediction accuracy, but is not particularly limited. For example, only one sample with high quality as training data may be adopted as training data.

Next, we will explain the machine learning method for creating the trained model used in the strength prediction system of the present invention. Examples of machine learning methods for creating the trained model used in the strength prediction system of the present invention include neural networks, linear regression, decision trees, support vector regression, ensemble methods, support vector machines, discriminant analysis, naive Bayes methods, nearest neighbor methods, etc. These methods may be used alone or in combination of two or more.

Among these methods, machine learning using a neural network is preferably selected from the viewpoint of being able to predict quality with higher accuracy. From the viewpoint of being able to predict quality with higher accuracy, it is preferable for the neural network to be a hierarchical neural network having one or more intermediate layers between the input layer and the output layer.

Examples of neural networks include convolutional neural networks (CNN) such as 3D Convolutional Neural Networks (3DCNN), deep neural networks (DNN), recurrent neural networks (RNN), and long short-term memory (LSTM) neural networks (recurrent neural networks improved using LSTM).

Among these, convolutional neural networks (neural networks having a convolutional layer, a pooling layer, etc. as an intermediate layer) are more suitable because of their excellent performance in the field of image recognition. Convolutional neural networks can detect features from image data and create a prediction model capable of performing classification or regression using the features. In a convolutional neural network, the number of layers consisting of a combination of convolutional layers and pooling layers is preferably two or more, more preferably three or more, from the viewpoint of being able to perform predictions with higher accuracy.

In addition, as a tool for performing machine learning, for example, "TensorFlow (registered trademark)", a software library developed by Google, or "IBM Watson (registered trademark)", a system developed by IBM, can be used.

Next, we will explain the specific configuration of the strength prediction system 1.

The configuration of a first embodiment of the ready-mix concrete strength prediction system 1 of the present invention will be described. FIG. 1 is a schematic overall configuration diagram of one embodiment of the ready-mix concrete strength prediction system 1. As shown in FIG. 1, the strength prediction system 1 is composed of a main server 10, a data server 20, an operation terminal 30, and a plurality of reference terminals 40 that are connected to the main server 10 via a network. The number of data servers 20, operation terminals 30, and reference terminals 40 connected to the main server 10 via a network is arbitrary, and each is connected to perform data communication via wire or wirelessly.

In the first embodiment, the main server 10 starts a prediction operation when an operator operates the operation terminal 30 to start the operation. Then, when prediction input data including information to be applied to the learned model is input from the data server 20 or the operation terminal 30 to the main server 10, a predetermined process is performed in the main server 10, and strength prediction information, which is information related to the predicted value of the strength of fresh concrete, is output from the main server 10. The output data including the strength prediction information output from the main server 10 is transmitted to each reference terminal 40. Then, each reference terminal 40 displays the predicted value of the strength of fresh concrete obtained from the output data output from the main server 10 on its respective display device 50.

In the first embodiment, each reference terminal 40 is a desktop PC separate from the display device 50 as shown in FIG. 1, but it may be a terminal in which the display device 50 is integrally configured, such as a notebook PC, a smartphone, or a tablet. Also, each reference terminal 40 may also function as an operation terminal 30.

Figure 2 is a block diagram showing a schematic configuration of a first embodiment of the main server 10. As shown in Figure 2, the main server 10 in the first embodiment includes a data input unit 11, a data output unit 12, and a calculation processing unit 13.

Figure 3 is a block diagram illustrating the details of the input and output of each data performed within the main server 10. As shown in Figure 3, the calculation processing unit 13 includes a first calculation processing unit 13a, a second calculation processing unit 13b, and a third calculation processing unit 13c, and each calculation processing unit (13a, 13b, 13c) includes a first memory unit 14a, a second memory unit 14b, and a third memory unit 14c.

For ease of explanation, the main server 10 shown in FIG. 3 is shown with three sets of processing units and memory units. However, the main server 10 may be configured such that each processing unit (13a, 13b, 13c) is a processing unit such as a CPU or MPU, and each memory unit (14a, 14b, 14c) is a memory such as a flash memory or hard disk, forming the processing unit 13.

The first memory unit 14a stores a first trained model M1 generated by machine learning based on the first teacher data. The second memory unit 14b stores a second trained model M2 generated by machine learning based on the second teacher data. The third memory unit 14c stores a third trained model M3 generated by machine learning based on the third teacher data.

The first teacher data is data in which first learning input data including mix-related information, which is information related to the mix-mix of ready-mix concrete, is associated with first learning output data including strength information of ready-mix concrete. The second teacher data is data in which second learning input data including primary strength prediction information derived by the first trained model and manufacturing-related information is associated with second learning output data including strength information of ready-mix concrete. The third teacher data is data in which third learning input data including secondary strength prediction information derived by the second trained model and quality-related information is associated with third learning output data including strength information of ready-mix concrete.

The strength information of fresh concrete is the strength information used in the first trained model, the second trained model, and the third trained model. In this embodiment, the strength information of fresh concrete is divided into the first strength information, the second strength information, and the third strength information, which are described below. The first strength information, the second strength information, and the third strength information are strength information that are more reliable in this order.

The first strength information is information derived from mix-related information at the time of designing the ready-mix concrete, and is information obtained before the ready-mix concrete is actually mixed.

The second strength information is information obtained during the period from when the ready-mix concrete is actually mixed until it is transported to the pouring site and arrives, and is information obtained before the ready-mix concrete hardens.

The third strength information is information obtained from test specimens of ready-mixed concrete poured at the pouring site, and is, for example, information on the compressive strength, flexural strength, Young's modulus, etc. of ready-mixed concrete at a specified age (e.g., 1 day, 3 days, 7 days, 14 days, etc.) after the test specimen is created.

The test specimen may be prepared at a ready-mix concrete manufacturing plant and the information may be obtained by measuring the strength at the manufacturing plant, or it may be obtained from a specimen prepared at the pouring site.

The data input unit 11 in the first embodiment accepts input of prediction input data d1 (in the first embodiment, ready-mix concrete mix-related information, manufacturing-related information, and quality-related information for prediction) transmitted from the data server 20 or the operation terminal 30 to be applied to the first trained model M1. Note that the ready-mix concrete mix-related information, manufacturing-related information, and quality-related information for prediction to be applied to the first trained model M1 may be included in one prediction input data d1, or each may be included in a separate prediction input data d1.

When the data input unit 11 receives input of prediction input data, it generates calculation input data (d11, d12, 13) to be input to each calculation processing unit (13a, 13b, 13c) and outputs it to each calculation processing unit (13a, 13b, 13c).

In this embodiment, the calculation input data d11 is data including composition-related information, the calculation input data d12 is data including manufacturing-related information, and the calculation input data d13 is data including quality-related information.

In this embodiment, when any of the formulation-related information, manufacturing-related information, and quality-related information is input, the data input unit 11 sequentially outputs the information as calculation input data (d11, d12, d13) to the calculation processing unit (13a, 13b, 13c) where processing is performed using the applied learning model.

When the data output unit 12 receives the calculation output data (d21, d22, d23) output from the calculation processing unit 13, it generates transmission data d2 that can be received by the reference terminals 40 based on the input calculation output data (d21, d22, d23) as shown in FIG. 2, and transmits the transmission data d2 to each reference terminal 40.

In the present embodiment, the data output unit 12 is configured to sequentially output the input calculation output data (d21, d22, d23) to the reference terminal 40, but may be configured to output only the calculation output data (d21, d22, d23) specified by the reference terminal 40 to the reference terminal 40.

When the first calculation processing unit 13a receives the calculation input data d11 output from the data input unit 11, it reads out the first trained model M1 recorded in the first storage unit 14a and applies the calculation input data d11 to the first trained model M1.

The first calculation processing unit 13a executes calculation processing using the first trained model M1, and outputs calculation output data d21, which includes primary intensity prediction information derived based on the calculation input data d11, to the data output unit 12 and the second calculation processing unit 13b.

When the second calculation processing unit 13b receives the calculation input data d12 output from the data input unit 11 and the calculation output data d21 output from the first calculation processing unit 13a, it reads out the second trained model M2 recorded in the second storage unit 14b and applies the calculation input data d12 and the calculation output data d21 to the second trained model M2.

The second calculation processing unit 13b executes calculation processing using the second learned model M2, and outputs the calculation output data d22, which includes secondary intensity prediction information derived based on the calculation input data d12 and the calculation output data d21, to the data output unit 12 and the third calculation processing unit 13c.

When the third calculation processing unit 13c receives the calculation input data d13 output from the data input unit 11 and the calculation output data d22 output from the second calculation processing unit 13b, it reads out the second trained model M3 recorded in the third memory unit 14c and applies the calculation input data d13 and the calculation output data d22 to the third trained model M3.

The third calculation processing unit 13c executes calculation processing using the third trained model M3, and outputs to the data output unit 12 the calculation output data d23 including tertiary intensity prediction information derived based on the calculation input data d13 and the calculation output data d22.

The information used as learning input data and prediction input data in the first embodiment is as follows:

Here, we explain the verification experiment to confirm the accuracy with which the strength prediction system 1, which applies the trained models (M1, M2, M3) trained based on the above information, can predict the slump value and slump flow at the time of unloading.

Example 1
In the first embodiment, the ready-mix concrete strength prediction system 1 described above predicts the standard underwater curing concrete strength of manufactured concrete after 28 days.

(Reference Example 1)
In Reference Example 1, prediction is made by a linear approximation formula based on mix-related information. The linear approximation formula used in Reference Example 1 is a formula that has been studied to be able to predict the strength of fresh concrete with a relatively high degree of accuracy based on accumulated data.

(Reference Example 2)
Reference Example 2 is similar to Example 1, except that the input data for prediction does not include quality-related information on the ready-mix concrete to be predicted.

The number of output data for each was set to 1,043.

The first trained model M1 was generated using a deep neural network (DNN), the learning library was Tensor Flow (registered trademark), and the training was performed 200,000 times.

The accuracy rate was confirmed by considering the predicted intensity value to be within the allowable error range of the actual intensity value as a correct answer. The allowable error range was confirmed in two patterns: ±5.0% and ±10.0%. The allowable error range here is a value calculated by E = (R - P) / R, where E is the allowable error range, P is the predicted intensity value, and R is the actual intensity value.

The results were as follows:

Concrete compressive strength tests are stipulated to have a success rate of 85% or more for the nominal strength specified at the time of ordering, and that the average of three test results must be equal to or greater than the specified nominal strength (JIS A 1108 standard). In order to consistently comply with this standard, it is desirable for the success rate to be at least 90% when the error range is ±10%, and 95% when the error range is ±15.0%.

The above results confirm that the strength prediction system 1 of the present invention is more accurate than predicting the standard 28-day underwater curing concrete strength using a linear approximation formula based on mix-related information. It is also confirmed that the strength prediction system 1 of the present invention can predict with higher accuracy by predicting the strength of fresh concrete by applying the standard 7-day underwater curing concrete strength.

As for prediction accuracy, it has been confirmed that the strength prediction system 1 of the present invention can predict the standard underwater curing concrete strength after 28 days with an accuracy of 95% or more when the error range is ±10.0%.

As described above, the strength prediction system 1 can predict the strength of fresh concrete with higher accuracy than predictions using conventional linear approximation formulas based on mix-related information by applying a trained model that is trained based on mix-related information, manufacturing-related information, and quality-related information.

[Another embodiment]
Another embodiment will be described below.

<1> In the strength prediction system 1 of this embodiment, environmentally related information is not applied, but any of the trained models (M1, M2, M3) may be generated by machine learning using training input data that includes environmentally related information, and the corresponding calculation input data (d11, d12, d13) may be configured to include environmentally related information.

With the above configuration, the strength prediction system 1 can predict strength taking into account information such as weather, temperature, and humidity at the pouring site, making it possible to predict the strength of ready-mix concrete with greater accuracy.

<2> The strength prediction system 1 may be equipped with a temperature sensor that is placed inside the ready-mix concrete for which strength is to be predicted, acquires temperature information inside the ready-mix concrete, and inputs data including the acquired temperature information to the data input unit 11.

With the above configuration, the strength prediction system 1 can predict the strength of fresh concrete by measuring the accumulated temperature resulting from the hydration reaction of the fresh concrete. Therefore, with the above configuration, the strength prediction system 1 can predict the strength of fresh concrete with higher accuracy.

The temperature sensor that acquires the temperature information of the ready-mixed concrete may be configured to acquire temperature information inside the hardened body of the ready-mixed concrete after pouring, the temperature information around the ready-mixed concrete for which strength is to be predicted, or the temperature information around the hardened body of the ready-mixed concrete after pouring. Furthermore, the temperature information may be acquired, for example, from a database of the Japan Meteorological Agency, such as weather information, temperature and humidity information, etc. at the pouring site.

〈3〉 The memory unit 14 of the strength prediction system 1 may store a weighting trained model generated by machine learning based on multiple teacher data that associates weighting training input data including strength prediction information for fresh concrete with weighting training output data including strength measurement value information, which is information regarding the actually measured strength of the fresh concrete for prediction. The weighting trained model is a trained model for determining a coefficient that weights each piece of information included in the prediction input data according to its contribution to the strength prediction.

With the above configuration, the strength prediction system 1 finds the more influential factors each time it repeats predicting the strength of fresh concrete, and the trained model is automatically optimized so that predictions gradually place more emphasis on those factors.

〈4〉 The configuration of the strength prediction system 1 described above is merely an example, and the present invention is not limited to the configurations shown in the figures.

1: Strength prediction system 10: Main server 11: Data input unit 12: Data output unit 13: Arithmetic processing unit 13a: First arithmetic processing unit 13b: First arithmetic processing unit 13c: First arithmetic processing unit 14: Memory unit 14a: First memory unit 14b: First memory unit 14c: First memory unit 20: Data server 30: Operation terminal 40: Reference terminal 50: Display device M1: First trained model M2: Second trained model M3: Third trained model

Claims (5)

A storage unit that stores a plurality of trained models generated by machine learning based on training data;
A data input unit that accepts input of prediction input data;
A data output unit that outputs data including strength prediction information of fresh concrete derived by the plurality of trained models,
the plurality of trained models include a first trained model that is generated by machine learning based on first teacher data and that derives primary intensity prediction information, a second trained model that is generated by machine learning based on second teacher data and that derives secondary intensity prediction information, and a third trained model that is generated by machine learning based on third teacher data and that derives tertiary intensity prediction information,
The first teacher data is data in which first learning input data including mix-related information, which is information related to the mix-related composition of the ready-mixed concrete, and first learning output data including strength information of the ready-mixed concrete are associated with each other,
the second teacher data is data in which second learning input data including primary strength prediction information derived by the first trained model and production-related information that is information related to the production of ready-mixed concrete is associated with second learning output data including strength information of ready-mixed concrete,
The third teacher data is data in which third learning input data including secondary strength prediction information derived by the second trained model and quality-related information that is information related to the strength of the fresh concrete to be predicted, which is obtained after pouring and before the time to be predicted, and which is measured before the time to be predicted , and third learning output data including strength information of the fresh concrete are associated with each other,
In the teacher data corresponding to at least one of the plurality of trained models, learning input data including environment-related information that is information related to the environment at a manufacturing site, a pouring site, or a transportation route of the ready-mixed concrete is associated with learning output data,
A ready-mix concrete strength prediction system, characterized in that the mix-related information, the manufacturing-related information, the quality-related information , and the environmental-related information of the ready-mix concrete to be predicted are input as the input data for prediction. The fresh concrete strength prediction system according to claim 1, further comprising a temperature sensor that acquires temperature information inside fresh concrete, temperature information inside the hardened body of the fresh concrete after pouring, temperature information around the fresh concrete, or temperature information around the hardened body of the fresh concrete after pouring, and inputs data including the acquired temperature information into the data input unit as the environment -related information. The system for predicting the strength of fresh concrete according to claim 1 or 2, characterized in that the data output unit is configured to be able to select the strength prediction information of fresh concrete to be included in the output data from the primary strength prediction information, the secondary strength prediction information , and the tertiary strength prediction information. the plurality of trained models further include a weighting trained model for determining a coefficient for weighting each piece of information included in the prediction input data according to its contribution to the prediction of strength of fresh concrete, the weighting trained model being generated by machine learning based on a plurality of teacher data obtained by associating weighting learning input data with weighting learning output data;
the weighting learning input data includes at least one of the primary intensity prediction information, the secondary intensity prediction information, and the tertiary intensity prediction information,
The system for predicting the strength of fresh concrete according to any one of claims 1 to 3 , characterized in that the output data for weighting learning includes strength measurement information which is information regarding the actually measured strength of the fresh concrete from which the strength prediction information is derived. The ready-mix concrete strength prediction system according to any one of claims 1 to 4 , characterized in that at least one of the mix-related information, the production-related information, and the quality-related information is obtained from an information management system in a ready-mix concrete manufacturing plant.