Method for estimating aggregate quality and method for producing concrete

A machine learning-based predictive model for aggregate quality estimation using image data addresses the challenges of laborious and inconsistent surface water content measurement, ensuring consistent concrete quality through rapid and accurate assessment.

JP7583529B2Active Publication Date: 2024-11-14TAIHEIYO CEMENT CORP
View PDF 15 Cites 0 Cited by

Patent Information

Application Number
JP2020032219
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-02-27
Publication Date
2024-11-14
Estimated Expiration
2040-02-27

AI Technical Summary

Technical Problem

Existing methods for measuring the surface water content of aggregates in concrete production are laborious, time-consuming, and difficult to implement continuously, and they fail to accurately account for variations due to aggregate size and storage time, leading to inconsistent concrete quality.

Method used

A predictive model using machine learning and image data analysis is employed to estimate aggregate quality, including surface water content and particle size, enabling accurate and rapid assessment of aggregate properties before mixing.

Benefits of technology

Enables high-accuracy, time-efficient estimation of aggregate quality, resulting in the production of concrete with stable fresh and hardened properties.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007583529000003
    Figure 0007583529000003
  • Figure 0007583529000004
    Figure 0007583529000004
  • Figure 0007583529000001
    Figure 0007583529000001
Patent Text Reader

Abstract

To estimate the quality of an aggregate in manufacturing of a fresh concrete more quickly and more precisely.SOLUTION: A method for estimating the quality of an aggregate by using a prediction model includes the steps of: creating the prediction model by a mechanical learning using a plurality of pieces of learning data as a combination of learning input data including image data and learning output data on the physical property value showing the quality of an aggregate; inputting image data taken to include the aggregate into the prediction model and outputting a physical property value showing the quality of the imaged aggregate from the prediction model; and assuming that the output physical property value is the physical property value of the imaged aggregate.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method for estimating aggregate quality and a method for producing concrete. [Background technology]

[0002] It has been known for some time that the surface water content of aggregates in the concrete manufacturing process at ready-mix concrete plants has a significant effect on the quality of concrete after mixing. In other words, if the surface water content of aggregates is not properly controlled, it will have a significant effect on the strength properties and fluidity.

[0003] For example, regarding the surface water ratio, JIS Q 1011:2019 Conformity Assessment - Certification of Conformity to Japanese Industrial Standards - Sector-specific Certification Guidelines (Ready Mixed Concrete) states that, as part of the management of the manufacturing process, the surface water ratio of reaggregate should be measured at least once per morning and at least once per afternoon, and the surface water ratio of coarse aggregate should be measured whenever necessary. However, the surface water ratio varies widely, so it is desirable to measure it continuously.

[0004] Patent Documents 1 and 2 and Non-Patent Document 1 disclose a technique for measuring the surface water ratio of aggregate with a moisture meter and varying the amount of aggregate and mixing water poured into a mixer according to the surface water ratio. Patent Document 3 discloses a technique for acquiring image data of aggregate, calculating the surface water ratio from the acquired image data, and adjusting the amount of water to be mixed during cement production. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2007-268839 A [Patent Document 2] JP 2002-301713 A [Patent Document 3] Japanese Patent Application Publication No. 10-048204 [Non-patent literature]

[0006] [Non-Patent Document 1] http: / / www.liberty-kobe.co.jp / products / cong-ii Summary of the Invention [Problem to be solved by the invention]

[0007] Surface water ratio is measured using methods such as JIS A 1111 Test method for surface water ratio of fine aggregate, but this requires time and effort and continuous measurement is not possible. There are also electrical and microwave methods, but these can only measure localized areas. Tests are also carried out for other qualities and values ​​are calculated, but this requires a great deal of time and effort and continuous measurement is difficult.

[0008] The method of measuring the surface water content of aggregate using a moisture meter described in Patent Documents 1 and 2 and Non-Patent Document 1 is time-consuming and labor-intensive. In addition, it may be difficult to accurately determine the surface water content because the surface water content of aggregate varies, and the surface water content of aggregate measured on a conveyor belt and the surface water content of aggregate stored in a storage bin change until the concrete is mixed.

[0009] In addition, in Patent Document 3, when aggregate changes from a bone dry state (a state containing no moisture) to a state containing moisture, the color of the surface gradually darkens, so the shade is measured from the image data of the aggregate to measure the surface water content of the aggregate. However, even if the shade of the aggregate is the same, the surface water content differs depending on the size (grain size) of the aggregate, so it is difficult to properly grasp the quality of the aggregate from the shade alone.

[0010] The present invention has been made in consideration of the above circumstances, and aims to estimate physical property values ​​that indicate the quality of aggregate before mixing into concrete in a short period of time and with high accuracy by analyzing image data of aggregate using a machine learning predictive model. [Means for solving the problem]

[0011] (1) In order to achieve the above-mentioned object, the method for estimating quality of aggregate of the present invention is a method for estimating quality of aggregate using a prediction model, comprising: a step of creating the prediction model by machine learning using a plurality of training data which are a combination of training input data including image data and training output data related to physical property values ​​indicative of quality of the aggregate; a step of inputting image data photographed so as to capture at least the aggregate into the prediction model; and a step of outputting physical property values ​​indicative of quality of the photographed aggregate from the prediction model, wherein the output physical property values ​​are estimated to be physical property values ​​of the photographed aggregate.

[0012] This makes it possible to estimate the quality of aggregate in a short amount of time with high accuracy, and to produce concrete with stable fresh and hardened properties.

[0013] (2) In the method for estimating quality of aggregate of the present invention, the image data is image data of the surface of the aggregate before it is mixed into concrete. By using image data of the surface of the aggregate before it is mixed into concrete, it is possible to estimate the quality of the aggregate with higher accuracy.

[0014] (3) In the method for estimating aggregate quality of the present invention, the image data is image data obtained by successively photographing the surface of the aggregate before mixing with concrete, the plurality of image data are input to the prediction model, physical property values ​​indicative of the quality of the plurality of photographed aggregates are output from the prediction model, and the quality of the aggregate is estimated using an average value calculated from the plurality of output physical property values. In this way, since the prediction model is created using image data of the surface of the aggregate before mixing with concrete, it is possible to estimate the quality of the aggregate with higher accuracy.

[0015] (4) In addition, in the method for estimating quality of aggregate of the present invention, the quality includes at least information on the surface water ratio of the aggregate. In this way, by learning the image data and the surface water ratio of the aggregate in association with each other and creating a prediction model, it is possible to more accurately grasp the surface water ratio of the aggregate before mixing the concrete, and to produce concrete with stable fresh properties and hardened properties.

[0016] (5) In the method for estimating quality of aggregate according to the present invention, the quality includes at least information related to the grading of the aggregate. This allows the surface water ratio of the aggregate to be correlated and learned in consideration of the grading to create a prediction model, making it possible to more accurately grasp the surface water ratio of the aggregate before mixing the concrete, and to produce concrete with stable fresh and hardened properties.

[0017] (6) Furthermore, a concrete manufacturing method of the present invention is a concrete manufacturing method using the aggregate quality estimation method described above in any one of (1) to (5), characterized by including the steps of: estimating quality of aggregate before mixing concrete using the quality estimation method; weighing each concrete material including aggregate, cement, water, admixtures, and admixtures based on the estimated aggregate quality; and mixing the weighed materials.

[0018] This makes it possible to estimate the quality of aggregate in a short amount of time with high accuracy, and to produce concrete with stable fresh and hardened properties. Effect of the Invention

[0019] According to the present invention, it is possible to estimate the quality of aggregate before concrete is mixed in a short time with high accuracy, and it is possible to produce concrete with stable fresh properties and hardened properties. [Brief description of the drawings]

[0020] [Figure 1] FIG. 1 is a diagram showing an overview of an aggregate quality estimation system. [Diagram 2] FIG. 1 is a diagram showing a concrete manufacturing process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0021] FIG. 1 is a diagram showing an overview of an aggregate quality estimation system according to this embodiment. This aggregate quality estimation system 1 is used to estimate the quality of aggregate. The quality of aggregate refers to the physical property values ​​of aggregate, such as the surface water rate, water content, water absorption rate, particle size distribution, coarse aggregate rate, and actual volume rate. In the ready-mix concrete manufacturing process, aggregate is first transported to a plant on a conveyor belt and stocked in a storage bin above the plant. A camera is installed on the conveyor belt or in the storage bin, and the camera photographs the aggregate on the conveyor belt transported by the conveyor belt or the aggregate stocked in the storage bin.

[0022] In the aggregate quality estimation system 1, the image acquisition unit 11 acquires image data captured by each camera. The acquired image data is input to a prediction model created in advance by machine learning in a prediction model creation unit 13, and the control unit (analysis execution unit) 15 uses the prediction model to analyze the quality of the aggregate captured by the camera. Thereafter, a physical property value indicating the analyzed aggregate quality is output from a display unit (output unit) 17. The output physical property value of the aggregate is estimated to be the physical property value of the aggregate captured by the camera. Also, an average value calculated from the multiple output data obtained may be used as the physical property value indicating the analyzed aggregate quality.

[0023] Using the aggregate physical property values ​​estimated by the aggregate quality estimation system 1, each material of the concrete is weighed and the concrete is mixed. In this way, by estimating the quality of the aggregate before mixing the concrete, it becomes possible to appropriately mix the materials. For example, by estimating the surface water ratio, which is one of the physical property values ​​that indicate the quality of the aggregate, in the aggregate quality estimation system 1, it becomes possible to manufacture concrete with stable fresh properties and hardened properties by injecting an amount of water into the mixer that is obtained by subtracting the estimated surface water ratio from the amount of water appropriate for mixing the concrete.

[0024] (Method of estimating aggregate quality) The aggregate quality estimation method according to this embodiment is a method for estimating the quality of aggregate by using a prediction model. The prediction model is created by machine learning using multiple pieces of learning data, which are combinations of image data photographed so that at least the aggregate is visible, and learning output data related to physical property values ​​that indicate the quality of the aggregate. The photographed image data is input to the prediction model, and physical property values ​​that indicate the quality of the photographed aggregate are output from the prediction model, and the output physical property values ​​are estimated to be the physical property values ​​of the photographed aggregate. Examples of physical property values ​​that indicate the quality of aggregate include surface water rate, moisture content, water absorption rate, particle size distribution, coarse particle rate, and actual volume rate.

[0025] The prediction model is created by machine learning. Examples of learning methods used in machine learning 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 may be used alone or in combination of two or more.

[0026] Among them, a neural network is preferable from the viewpoint of being able to estimate quality with higher accuracy. As the neural network, a hierarchical neural network having one or more intermediate layers between an input layer and an output layer is preferable from the viewpoint of being able to estimate quality with higher accuracy.

[0027] Examples of neural networks include convolutional neural networks (CNNs) such as 2D or 3D convolutional neural networks (2DCNN or 3DCNN), recurrent neural networks (RNNs), and long short-term memory (LSTM) neural networks (recurrent neural networks improved using LSTM).

[0028] Among them, a convolutional neural network (a neural network having a convolutional layer, a pooling layer, or the like as an intermediate layer) is more suitable because it has excellent performance in the field of image recognition. A convolutional neural network can detect features from image data and create a prediction model capable of performing classification or regression using the features.

[0029] 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 enabling estimation with higher accuracy.

[0030] Machine learning can be performed using, for example, "TensorFlow" ("TENSORFLOW" (registered trademark)), a software library developed by Google, or "IBM Watson" ("IBM WATSON" (registered trademark)), a system developed by IBM.

[0031] The prediction model is created by machine learning using multiple pieces of training data, which are a combination of training input data including image data and training output data related to the quality of aggregate. Examples of training input data include image data of aggregate before mixing into concrete. Specifically, examples include video image data captured by a fixed camera of aggregate stored on a belt conveyor or in a storage bin, and still image data captured periodically. One type of these may be used alone, or two or more types may be used in combination. In addition, the captured image may include the aggregate portion, and the surface may be flat or uneven.

[0032] The number of image data used as learning input data is preferably 100 or more, more preferably 1000 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 estimating quality with higher accuracy. The image data may be data of a moving image or data of a still image. The image may be a two-dimensional image or a three-dimensional image.

[0033] Image data is captured by a camera installed on the conveyor belt or inside the storage bin. The camera is installed so that the aggregate part is clearly visible. In the case of moving images, still images are converted and used as input data.

[0034] In the case of cutting out an image of a specific range from an image of an aggregate as image data (used as learning input data), multiple image data may be cut out from one image of the aggregate. The cut-out ranges of the multiple image data may overlap each other. For example, multiple image data may be obtained from one image by cutting out another image data from a position shifted by 1 to 10 pixels in at least one of the vertical and horizontal directions (up and down and left and right directions) based on the position from which the first image data was cut out.

[0035] Images of aggregates stored on a conveyor belt or in a storage bin may be converted into grayscale images, in which one pixel is represented by a number in 256 levels from 0 to 255, in order to make the image data more suitable for machine learning such as convolutional neural networks.

[0036] In this way, the state of the aggregate is continuously photographed during the process of transporting it on a conveyor belt, in storage bins, etc., and a learning model created in advance from these images is used to estimate the physical properties of the aggregate. Based on the estimated physical properties, the composition of the concrete ingredients is determined, weighed, and the ingredients are mixed in a mixer, making it possible to produce concrete with stable fresh and hardened properties.

[0037] (Concrete manufacturing method) A concrete manufacturing method using the above-mentioned aggregate quality estimation method will be described. FIG. 2 is a diagram showing the concrete manufacturing process. First, aggregates (river sand, mountain sand, land sand, sea sand, crushed sand, silica sand, slag fine aggregate, lightweight fine aggregate, coarse aggregate, etc.) are transported to the upper part of the plant by a belt conveyor and stored in a storage bin. Cameras are installed on the belt conveyor and in the storage bin to take images of the aggregates when they are being transported on the belt conveyor and the aggregates stored in the storage bin (step S1). Next, the captured image data is sent to an aggregate quality estimation system. The quality estimation system analyzes the quality of the aggregate based on the input image data. That is, the quality of the aggregate is analyzed from the input image data using a prediction model, and the physical property values ​​of the aggregate are output as quality estimation results (step S2). Next, taking into consideration the output physical property values ​​of the aggregate, other concrete materials such as aggregate, cement, admixtures, and admixtures are weighed and injected into a mixer (step S3). After the injection is completed, the concrete is mixed in a mixer to produce ready-mix concrete (step S4).

[0038] (Example) Table 1 shows information about aggregates, such as the type of aggregate used as a learning sample in this example. For the two types of aggregate shown in Table 1, the surface water ratios according to the JIS A 1111 test method for surface water ratio of fine aggregate, the JIS A 1125 test method for moisture content of aggregate, and the test method for surface water ratio based on moisture content were adjusted to the values ​​shown in Table 1. In this example, the surface water ratio (%) was used as the quality (physical property value) of the aggregate, but in addition to the surface water ratio, it is also possible to use the water absorption rate, particle size distribution, coarse aggregate ratio, actual volume ratio, etc.

[0039] [Table 1]

[0040] The aggregates shown in Table 1 were placed in a container measuring 50 cm length x 50 cm width x 10 cm height, with some of the aggregates having a flat surface and some with the surface roughened to create an uneven surface. These 24 combinations (2 (aggregate type) x 6 (surface moisture content) x 2 (surface condition)) of aggregates were photographed as a whole under two lighting patterns, dark and bright, with a digital camera, and 48 pieces (24 x 2) of image data were obtained.

[0041] Forty different image data were randomly extracted from each image data, each with a size of 256 x 256 pixels, and a learning model was created using 1,920 images (48 x 40) as input data and the surface water ratio as output data.

[0042] For the machine learning, "TensorFlow" ("TENSORFLOW" (registered trademark)) was used, and learning was performed using a five-layer convolutional neural network. In addition, the least squares method was used as the error function in the learning. The number of learning cycles was 500,000, and the number of image data (learning input data) input in each learning cycle was 50 (randomly selected).

[0043] Next, the learning model created above was verified. Table 2 shows information on the aggregates, such as the type of aggregate, used as the verification sample.

[0044] [Table 2]

[0045] As shown in Table 2, aggregates with adjusted surface moisture ratios were photographed with a digital camera in the same way, and 32 images (2 (aggregate type) x 4 (surface moisture ratio) x 2 (surface condition) x 2 (brightness)) were obtained as verification image data. From each of the 32 image data, 20 different image data pieces were randomly extracted, each with a size of 256 x 256 pixels, and 640 image data pieces (32 x 20) were obtained.

[0046] Verification was carried out using these 640 acquired image data as input data and each surface water ratio as output data. In the verification, the physical property values ​​output by the learning model were deemed to be correct if they could be estimated within ±0.25% of the actual measured surface water ratio values ​​shown in Table 2. As a result of the verification, the accuracy rate was 96%, demonstrating that it is possible to estimate the surface water ratio using image data of aggregates.

[0047] As described above, according to this embodiment, the quality of aggregate before mixing into concrete can be estimated in a short time and with high accuracy, and concrete with stable fresh properties and hardened properties can be produced. [Explanation of symbols]

[0048] 1. Quality Estimation System 11 Image acquisition section 13 Prediction Model Creation Department 15 Control section (analysis execution section) 17 Display section (output section)

Claims

1. A method for estimating aggregate quality using a prediction model, comprising: A prediction model creation unit creates the prediction model by machine learning using a plurality of learning data, which is a combination of learning input data including image data captured by a camera so that at least aggregate is captured, and learning output data related to the surface water ratio of the aggregate; An image acquisition unit acquires an image of an aggregate whose quality is to be estimated, and inputs the image data of the aggregate into the prediction model; A step of a control unit analyzing image data of the aggregate using the prediction model; an output unit outputting the surface water ratio of the aggregate in the analyzed image data as a physical property value of the photographed aggregate.

2. 2. The method for estimating quality of aggregate according to claim 1, wherein the image data is image data cut out from a specific area of ​​an image taken of the surface of the aggregate before it is mixed into concrete.

3. The image data is a plurality of image data cut out in a specific range from a plurality of images continuously taken of the surface of the aggregate before concrete mixing, The image acquisition unit inputs the plurality of image data into the prediction model; The control unit calculates an average value of physical property values ​​of the plurality of photographed aggregates using the prediction model, 3. The method for estimating quality of aggregate according to claim 1, wherein the output unit outputs the calculated average value.

4. The physical property values ​​further include information regarding the grain size of the aggregate; the prediction model creation unit performs machine learning using a plurality of pieces of learning data, each of which is a combination of the learning input data and learning output data related to aggregate particle sizes; 4. The method for estimating quality of aggregate according to claim 1, wherein the output unit outputs physical property values ​​further including a grain size of the aggregate in the analyzed image data.

5. A method for producing concrete using physical property values ​​estimated by the method for estimating aggregate quality according to any one of claims 1 to 4, A step of measuring each concrete material including aggregate, cement, water, admixture, and admixture based on the estimated aggregate physical property value; and kneading the measured materials.

Citation Information

Patent Citations

  • Automatic inspection apparatus for quality of aggregate

    JP1997145348A

  • Method for correcting quantity of water in fresh mixed concrete plant

    JP1998048203A

  • Method for measuring surface moisture rate in fresh concrete production plant

    JP1998048204A

  • Method and apparatus for measurement of moisture of aggregate in ready-mixed concrete plant

    JP2002301713A

  • Ready-mixed concrete plant

    JP2007268839A