Quality prediction program, quality prediction method, and quality prediction apparatus

Abstract

To provide a quality prediction program, a quality prediction method, and a quality prediction apparatus for aggregate, configured to improve the convenience for a user.SOLUTION: A quality prediction apparatus includes: an evaluation information acquisition unit which acquires evaluation input information based on captured image data which is obtained by imaging aggregate to be evaluated; a prediction computing unit which inputs the evaluation input information to a prediction model constructed so as to output a value indicating a mass fraction for each grade scale included in a plurality of grade scales, in accordance with an input of input information, to obtain, for each scale, a predicted value output from the prediction model; and a display output unit which causes a monitor to display an inference graph having a first axis denoting nominal dimensions of screens and a second axis denoting mass fractions of grains passing through the screens.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present disclosure relates to a quality prediction program, a quality prediction method, and a quality prediction device. [Background technology]

[0002] Patent Document 1 discloses a method for estimating the particle size distribution of soil. Patent Document 2 discloses a method for estimating the quality of aggregate. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-117625 [Patent Document 2] JP 2021-135199 A Summary of the Invention [Problem to be solved by the invention]

[0004] The present disclosure provides a quality prediction program, a quality prediction method, and a quality prediction device that are useful for improving user convenience. [Means for solving the problem]

[0005] [1] A quality prediction program that causes a computer to execute the following steps: an acquisition step of acquiring evaluation input information based on image data obtained by imaging an aggregate to be evaluated; a first prediction step of inputting evaluation input information to a prediction model that has been constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for training and on ground truth values ​​of mass fractions for each of a plurality of particle size classifications of the aggregate corresponding to the input information, and acquiring predicted values ​​output from the prediction model for each of the classifications; and a result display step of displaying on a monitor an inference graph whose first axis represents the nominal size of a sieve and whose second axis represents the mass fraction of grains passing through each sieve, wherein the inference graph displayed on the monitor in the result display step includes an inference result calculated based on the predicted value for each of the classifications, and a standard range for the second axis that represents a standard range according to the type of aggregate to be evaluated.

[0006] [2] The quality prediction program described in [1] above, wherein the standard range is defined by a first boundary line based on the mass fraction of particles passing through each sieve when the overall particle size of the type of aggregate being evaluated is maximum, and a second boundary line based on the mass fraction of particles passing through each sieve when the overall particle size of the type of aggregate being evaluated is minimum.

[0007] [3] The quality prediction program of [1] or [2] above, further comprising: causing a computer to execute an identification process for identifying the type of aggregate to be evaluated from among a plurality of aggregate types based on user input indicating the type of aggregate to be evaluated; and in the result display process, the standard range is displayed on the monitor according to the type of aggregate to be evaluated identified in the identification process.

[0008] [4] The quality prediction program according to any one of [1] to [3] above, further causing the computer to execute an image display process of displaying an image obtained by photographing the aggregate to be evaluated on the monitor before the acquisition process.

[0009] [5] The quality prediction program described in [4] above, wherein the image display process includes rotating the captured image displayed on the monitor when a user input instructing to rotate the image is input, and in the acquisition process, the evaluation input information is determined based on the rotated state of the captured image in the image display process.

[0010] [6] The quality prediction program described in any one of [1] to [5] above, wherein the quality prediction program further causes the computer to execute a second prediction step of calculating a coarse aggregate ratio of the aggregate to be evaluated from the predicted values ​​for each category obtained in the first prediction step, and in the result display step, the calculation result of the coarse aggregate ratio of the aggregate to be evaluated in the second prediction step is displayed on the monitor together with the inference graph.

[0011] [7] The quality prediction program described in [6] above, wherein the second prediction process includes correcting the predicted value obtained in the first prediction process and calculating the coarse grain ratio of the aggregate to be evaluated from the corrected predicted value, and in correcting the predicted value in the second prediction process, the corrected predicted value is calculated for each category so that the sum of the corrected predicted values ​​for each category is 100%.

[0012] [8] The second prediction step includes correcting the predicted value obtained in the first prediction step and calculating the coarse grain ratio of the aggregate to be evaluated from the corrected predicted value, and in the correction of the predicted value in the second prediction step, if the predicted value output from the prediction model is smaller than 0, the corrected predicted value is corrected to be 0, and if the predicted value output from the prediction model is greater than 0, the predicted value is corrected to be the corrected predicted value as is. The quality prediction program described in [6] or [7] above.

[0013] [9] A quality prediction method comprising: an acquisition step of acquiring evaluation input information based on captured image data obtained by imaging an aggregate to be evaluated; a first prediction step of inputting the evaluation input information to a prediction model constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for training and on ground-truth values ​​of mass fractions for each of a plurality of particle size classifications of the aggregate corresponding to the input information, so as to output a value indicating a mass fraction for each classification included in the plurality of particle size classifications in response to the input of the input information, and acquiring a predicted value output from the prediction model for each of the classifications; and a result display step of displaying on a monitor an inference graph whose first axis represents the nominal size of a sieve and whose second axis represents the mass fraction of grains passing through each sieve, wherein the inference graph displayed on the monitor in the result display step includes an inference result calculated based on the predicted value for each classification, and a standard range for the second axis that represents a standard range according to the type of aggregate to be evaluated.

[0014]

[10] A quality prediction device comprising: an evaluation information acquisition unit that acquires evaluation input information based on captured image data obtained by imaging an aggregate to be evaluated; a prediction calculation unit that inputs evaluation input information to a prediction model constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for training and on ground-truth values ​​of mass fractions for each of a plurality of particle size classifications of the aggregate corresponding to the input information, and acquires predicted values ​​output from the prediction model for each classification; and a display output unit that displays on a monitor an inference graph whose first axis represents the nominal size of a sieve and whose second axis represents the mass fraction of grains passing through each sieve, wherein the inference graph displayed on the monitor by the display output unit includes an inference result calculated based on the predicted value for each classification, and a standard range for the second axis that represents a standard range according to the type of aggregate to be evaluated. Effect of the Invention

[0015] According to the present disclosure, a quality prediction program, a quality prediction method, and a quality prediction device are provided that are useful for improving user convenience. [Brief description of the drawings]

[0016] [Figure 1] FIG. 1 is a schematic diagram illustrating a ready-mix concrete manufacturing system and a quality prediction device. [Diagram 2] FIG. 2 is a block diagram illustrating an example of a hardware configuration of the quality prediction device. [Diagram 3] FIG. 3 is a block diagram illustrating an example of a functional configuration of the quality prediction device. [Figure 4] Fig. 4(a) is a schematic diagram showing an example of a predictive calculation using a predictive model, and Fig. 4(b) is a cross-sectional view showing a schematic diagram of acquiring image data for learning. [Diagram 5] FIG. 5 is a diagram showing an example of image data for learning. [Figure 6] FIG. 6 is a graph showing an example of a comparison result between the predicted value and the correct value. [Figure 7] FIG. 7 is a flow diagram showing an example of a series of processes executed by the quality prediction device. [Figure 8] 8(a) and 8(b) are schematic diagrams showing an example of a display screen on a monitor. [Figure 9] 9(a) and 9(b) are schematic diagrams showing an example of a display screen on a monitor. [Figure 10] 10(a) and 10(b) are schematic diagrams showing an example of a display screen on a monitor. [Figure 11] FIG. 11 is a schematic diagram showing an example of a display screen on a monitor. [Figure 12] FIG. 12 is a schematic diagram showing an example of a display screen on a monitor. [Figure 13] 13(a) and 13(b) are graphs showing an example of the comparison result between the predicted value and the correct value. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] Hereinafter, an embodiment will be described with reference to the drawings. In the description, the same elements or elements having the same functions are given the same reference numerals, and duplicated descriptions will be omitted. Fig. 1 shows a schematic diagram of a quality prediction device according to an embodiment, and a ready-mix concrete manufacturing system in which the quality prediction device is used.

[0018] [Ready-mix concrete manufacturing system] First, an overview of a ready-mix concrete manufacturing system will be described. The manufacturing system 100 shown in Fig. 1 is a system that manufactures ready-mix concrete. The manufacturing system 100 kneads and mixes concrete materials to manufacture (produce) ready-mix concrete.

[0019] The concrete materials used in the manufacturing system 100 include cement, admixtures, coarse aggregates, fine aggregates, water, and admixtures. Examples of the coarse aggregates include gravel, crushed stone, slag coarse aggregates, lightweight coarse aggregates, recycled coarse aggregates, reclaimed aggregates, and coarse aggregates made by mixing these. Examples of the gravel include mountain gravel, land gravel, river gravel, and sea gravel. Examples of the slag coarse aggregates include blast furnace slag aggregates, ferronickel slag aggregates, electric furnace oxidized slag aggregates, and coal gasification slag aggregates. Examples of the lightweight coarse aggregates include natural lightweight aggregates, by-product lightweight aggregates, and artificial lightweight aggregates. The coarse aggregates may include crushed rock, or limestone crushed stone.

[0020] Examples of fine aggregates include sand, crushed sand, slag fine aggregate, lightweight fine aggregate, recycled fine aggregate, recovered aggregate, and fine aggregates made by mixing these. Sand includes mountain sand, land sand, river sand, sea sand, etc. Slag fine aggregates include blast furnace slag aggregate, ferro-nickel slag aggregate, copper slag aggregate, electric furnace oxidized slag aggregate, coal gasification slag aggregate, etc. Lightweight fine aggregates include natural lightweight aggregate, by-product lightweight aggregate, artificial lightweight aggregate, etc.

[0021] Examples of rock types of crushed stone and crushed sand include igneous rocks, sedimentary rocks, metamorphic rocks, quartzite, limestone, domalite, and peridotite. Igneous rocks include granite, diorite, gabbro, porphyrite, diabase, rhyolite, andesite, basalt, and serpentine. Sedimentary rocks include conglomerate, sandstone, shale, slate, and tuff. Metamorphic rocks include gneiss and schist. In the present disclosure, coarse aggregate and fine aggregate may be collectively referred to as "aggregate". In this case, "aggregate" refers to coarse aggregate, fine aggregate, or both coarse aggregate and fine aggregate.

[0022] The manufacturing system 100 loads the manufactured ready-mix concrete onto a transport vehicle 200. After the ready-mix concrete is loaded, the transport vehicle 200 transports the ready-mix concrete to a site (e.g., a construction site) where the ready-mix concrete is to be used. Examples of the transport vehicle 200 include an agitator vehicle (mixer vehicle) or a dump truck. The manufacturing system 100 may manufacture ready-mix concrete from concrete materials so as to satisfy a target quality (required quality) set for each site. In one example, the target quality of the ready-mix concrete includes a target value for slump or slump flow. The manufacturing system 100 includes, for example, a material storage yard 101, a transport device 104, and a manufacturing device 110.

[0023] The material yard 101 is a place where concrete materials are stored. The material yard 101 includes a plurality of silos 102. The plurality of silos 102 are containers that store at least a portion of the concrete materials by material type. The plurality of silos 102 include a silo 102 that stores coarse aggregate, a silo 102 that stores fine aggregate, and a silo 102 that stores cement.

[0024] The transporting device 104 is a device that transports the concrete materials stored in the multiple silos 102 to the manufacturing device 110. The transporting device 104 includes, for example, a belt conveyor that transports the concrete materials. The transporting device 104 may transport the concrete materials by type at different times. In one example, based on an operation instruction from a control device included in the manufacturing system 100, a specific material from among the various concrete materials is transferred to the transporting device 104 and transported to the manufacturing device 110.

[0025] The manufacturing apparatus 110 is an apparatus for manufacturing ready-mixed concrete by mixing concrete materials including aggregate. The manufacturing apparatus 110 includes, for example, a storage bottle 111, a measuring bottle 112, a collecting hopper 113, a mixer 114, and a loading hopper 115.

[0026] Storage bottle 111 temporarily stores various types of concrete materials. Various types of concrete materials are transported (conveyed) to storage bottle 111 from material storage yard 101 by transport device 104. Storage bottle 111 is configured to store various types of concrete materials individually. Hereinafter, "concrete material" may be simply referred to as "material". Various materials stored in storage bottle 111 are supplied to measuring bottle 112 as necessary.

[0027] The measuring bottle 112 is disposed below the storage bottle 111. The measuring bottle 112 operates based on an operation command from the control device, and weighs each material individually. When the measuring bottle 112 detects a target amount of material instructed by the control device, it supplies the material to the collecting hopper 113. When water is supplied to the measuring bottle 112, an admixture may be mixed into the water. The collecting hopper 113 is disposed below the measuring bottle 112. The collecting hopper 113 collects each material discharged from the measuring bottle 112, and supplies the collected materials to the mixer 114. Note that the collecting hopper 113 does not have to be provided, and in this case, each material is supplied from the measuring bottle 112 to the mixer 114.

[0028] The mixer 114 is disposed below the collecting hopper 113. The mixer 114 is a device that mixes concrete materials. The mixer 114 produces ready-mixed concrete by mixing (kneading) aggregate, cement, water, admixtures, and the like. The ready-mixed concrete is discharged from the bottom of the mixer 114 into the loading hopper 115. The loading hopper 115 is disposed below the mixer 114, and temporarily stores the ready-mixed concrete. The loading hopper 115 supplies the temporarily stored ready-mixed concrete to the transport vehicle 200.

[0029] The manufacturing system 100 described above is one example of a ready-mix concrete manufacturing system, and the ready-mix concrete manufacturing system may be configured in any manner as long as it is capable of mixing concrete materials and manufacturing ready-mix concrete.

[0030] [Quality prediction device] The quality prediction device 1 is a device that predicts the coarse aggregate ratio as the quality of aggregate contained in concrete material. The quality prediction device 1 is configured with one or more computers. The computers that configure the quality prediction device 1 may be a personal computer, a tablet computer (tablet terminal), a smartphone, a wearable device, a workstation, a server computer, or a general-purpose computer. The quality prediction device 1 may be connected to a communication network via a wired or wireless connection.

[0031] The quality prediction device 1 includes, for example, a main body 10, an input device 12, a monitor 14, and a camera 16. The main body 10 is a device (computer main body) that executes the main functions of the computer that constitutes the quality prediction device 1.

[0032] The input device 12 is a device for inputting information to the main body 10. The input device 12 may be any device capable of inputting desired information, and specific examples thereof include operation interfaces such as a keypad, a mouse, and an operation controller.

[0033] The monitor 14 is a device for displaying information output from the main body 10. The monitor 14 may be any device capable of displaying graphics, a specific example of which is a liquid crystal panel. The monitor 14 and the input device 12 may be integrated as a touch panel.

[0034] The camera 16 is a device capable of capturing an image of the aggregate. The camera 16 captures an image of the aggregate to generate image data. The camera 16 is, for example, a digital camera that captures an image of the imaging range based on visible light, or a video camera. The image data generated by the camera 16 is used to predict the quality of the aggregate. When the camera 16 generates a video of the imaging target, still images included in the video may be acquired as image data.

[0035] The camera 16 may be disposed in the manufacturing system 100 so as to be able to capture an image of an aggregate at a specific location. For example, the camera 16 may be disposed so as to be able to capture an image of an aggregate being transported by the transport device 104. The camera 16 may be portable, and a user such as a worker may carry the camera 16 to an imaging position. The camera 16 may be mounted on a mobile body such as a drone, and the camera 16 may capture an image at any location in the manufacturing system 100 by a user operating the mobile body, etc.

[0036] In the quality prediction device 1, the main body 10, the input device 12, the monitor 14, and the camera 16 may be integrated into one device, such as a smartphone with a built-in (mounted) camera or a tablet computer with a built-in camera. In one example, the quality prediction device 1 is configured by installing an application for predicting the quality of aggregate on a smartphone.

[0037] As shown in FIG. 2, the main body 10 includes a circuit 50. The circuit 50 includes a processor 51, a memory 52, a storage 53, and an input / output port 54. The storage 53 is configured with one or more non-volatile memory devices such as a flash memory or a hard disk. The storage 53 stores a quality prediction program that causes a computer to execute at least an acquisition step, a first prediction step, and a result display step. The quality prediction program may cause a computer to execute a specification step, an image display step, an acquisition step, a first prediction step, a second prediction step, and a result display step. Details of these steps will be described later. The storage 53 stores a quality prediction program for configuring each functional block of the main body 10.

[0038] The memory 52 is composed of one or more volatile memory devices such as a random access memory. The memory 52 temporarily stores a quality prediction program loaded from the storage 53. The processor 51 is composed of one or more arithmetic devices such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The processor 51 configures each functional block of the main body 10 by executing the quality prediction program loaded into the memory 52. ​​The calculation results by the processor 51 are temporarily stored in the memory 52. ​​The input / output port 54 inputs and outputs information between the input device 12, the monitor 14, the camera 16, etc. in response to a request from the processor 51.

[0039] The quality prediction program may be provided in a state of being fixedly recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, the quality prediction program may be provided via a communication network as a data signal superimposed on a carrier wave. The main body 10 (quality prediction device 1) may be composed of a plurality of computers communicably connected to each other. The main body 10 (quality prediction device 1) may be a part of a control device included in the manufacturing system 100.

[0040] 3 shows an example of functional components (referred to as "functional blocks" in this disclosure) included in the main body 10. The main body 10 includes, as functional blocks, a user input acquisition unit 22, an image data acquisition unit 24, an evaluation information acquisition unit 26, a prediction model storage unit 28, a prediction calculation unit 30, a display output unit 32, and an aggregate information storage unit 34. The processes executed by these functional blocks correspond to the processes executed by the main body 10 (quality prediction device 1).

[0041] The user input acquisition unit 22 is configured to acquire user input information indicating an input from a user required for predicting the coarse grain ratio of the aggregate to be evaluated. The user input acquisition unit 22 may acquire the user input information based on a user input to an interface screen displayed on the monitor 14.

[0042] The image data acquisition unit 24 is configured to acquire image data generated by capturing an image of the aggregate to be evaluated (hereinafter, referred to as "captured image data"). The image data acquisition unit 24 acquires captured image data obtained by capturing an image by the camera 16.

[0043] The evaluation information acquiring unit 26 is configured to acquire evaluation input information based on captured image data acquired by capturing an image of the aggregate to be evaluated. The evaluation information acquiring unit 26 may acquire, for example, the captured image data as is as the evaluation input information, or may acquire image data acquired by performing a predetermined processing on the captured image data as the evaluation input information. A user input may determine whether or not to process the captured image data in order to obtain the evaluation input information, and how to process the captured image data.

[0044] The prediction model holding unit 28 is configured to hold a model for predicting the coarse grain ratio of the aggregate to be evaluated (hereinafter referred to as "prediction model M"). The prediction model M is a model constructed by machine learning based on input information based on learning image data (learning image data) obtained by imaging the learning aggregate and on the ground-truth values ​​of the mass fractions of the aggregate corresponding to the input information, so as to output a value indicating the mass fraction for each of the plurality of granularity classes in response to the input of the input information. The construction of the prediction model M will be described later.

[0045] The prediction model holding unit 28 may hold (store) a plurality of types of prediction models M. The plurality of types of prediction models M held by the prediction model holding unit 28 may be associated with a plurality of types of aggregate, respectively. Each of the plurality of types of prediction models M may be constructed by machine learning based on learning data obtained from the corresponding type of aggregate. For example, when the type of aggregate to be evaluated is "crushed stone 2005", the coarse aggregate ratio may be predicted using the prediction model M associated with the crushed stone 2005 among the plurality of types of prediction models M.

[0046] Machine learning refers to a method in which a machine (computer) autonomously finds laws or rules by repeatedly learning based on given information. The prediction model M can be constructed using an algorithm and a data structure. The prediction model M is realized, for example, using a neural network, which is an information processing model that mimics the mechanism of the human brain. There is no particular limitation on the specific algorithm of the machine learning performed when constructing the prediction model M. The neural network has an input layer, one or more intermediate layers, and an output layer. By including one or more intermediate layers, a more complex prediction model M can be constructed, and prediction accuracy can be improved.

[0047] The prediction model M may be constructed by machine learning using a convolutional neural network (CNN). The prediction model M is constructed by machine learning based on the correct values ​​of the mass fractions of the aggregate's multiple particle size divisions. The prediction model M may output only a value indicating the mass fraction of each division included in the multiple particle size divisions in accordance with the above input information. In other words, the prediction model M may output only a plurality of values ​​each indicating a mass fraction for the multiple particle size divisions.

[0048] The prediction model M is constructed to predict the mass fraction for each particle size division by performing machine learning using, for example, data provided as input for machine learning and correct answer data (correct answer value of mass fraction for each particle size division) of the machine learning output. The input for the machine learning is various data sets of input information based on learning image data obtained by imaging learning aggregate. In the various data sets of input information, at least the image data included in the input information is different from each other. The output of the machine learning is data (numerical value) indicating the mass fraction for each particle size division. When constructing the prediction model M, a model that outputs a predicted value of the mass fraction for each particle size division is iteratively trained using multiple combinations of input information and correct answer values ​​such as mass fraction.

[0049] The stage in which the prediction model M is autonomously constructed corresponds to the learning phase. The learning phase may be performed before the production phase in which ready-mix concrete is manufactured in the manufacturing system 100, or may be performed at an early stage of the production phase. The prediction model M, which is a trained model, is portable between computers. Therefore, the prediction model M constructed in another device may be stored in the prediction model holding unit 28. The stage in which the quality prediction device 1 predicts the coarse particle ratio of the aggregate to be evaluated using the prediction model M corresponds to the evaluation phase.

[0050] The prediction calculation unit 30 is configured to input evaluation input information to the prediction model M and acquire predicted values ​​output from the prediction model M for each of the particle size divisions. That is, the prediction calculation unit 30 acquires multiple predicted values, each indicating a mass fraction value, for multiple particle size divisions from the prediction model M. The prediction calculation unit 30 calculates the coarse particle ratio from the predicted values ​​(multiple predicted values) regarding the mass fraction for each particle size division.

[0051] The display output unit 32 is configured to display an inference graph on the monitor 14, the first axis of which is the nominal size of the sieve and the second axis of which is the mass fraction of the particles passing through each sieve. The inference graph displayed on the monitor 14 includes an inference result calculated based on the above-mentioned predicted value for each of the multiple particle size categories, and, on the second axis, a standard range that indicates a standard range according to the type of aggregate to be evaluated.

[0052] The aggregate information storage unit 34 stores information necessary for displaying the standard ranges included in the inference graph. The aggregate information storage unit 34 stores, for example, information indicating standard ranges associated with each of a plurality of types of aggregate. That is, the aggregate information storage unit 34 may store information indicating a corresponding standard range for each of a plurality of types of aggregate. Details of the inference graph including the inference result and the standard range will be described later.

[0053] <Prediction model M> Here, the prediction model M used to predict the coarse grain ratio will be further described with reference to Figs. 4 and 5. In the learning phase for constructing the prediction model M, aggregate of the same type as the aggregate that is used in the manufacturing system 100 and that can be evaluated is used as the learning aggregate. Fig. 4(a) shows a schematic diagram of the calculation process for predicting the coarse grain ratio using the prediction model M from the image Im, which is input information. The input of the prediction model M is the image Im, and the output of the prediction model M is the predicted value of the mass fraction of each of a plurality of grain size classes.

[0054] When constructing the prediction model M, learning data for machine learning is prepared. The learning data includes various image data (input information) obtained by imaging the learning aggregate, and answer data associated with each of the various image data. In preparing the learning images, for example, various aggregates whose mass fractions for each granularity classification are known are prepared. The above various aggregates do not mean different types of aggregate, but mean the same type of aggregates with different granularities.

[0055] In one example, various types of learning aggregates are individually imaged by camera 130 in a stacked state on measurement sheet 132 (see FIG. 4(b)). A stacked state (piled state) refers to a state in which some particles contained in the aggregate overlap other particles when viewed vertically from above. By individually imaging the various types of aggregates by camera 130, learning image data is obtained for each type of aggregate. Camera 130 may be a digital camera, and images an area including the stacked aggregates from a position a predetermined distance away from the aggregates on measurement sheet 132 (for example, a predetermined position above).

[0056] The sheet 132 may be a rubber sheet. As shown in FIG. 4(b), the sheet 132 may be a sheet formed to be curved in imitation of the belt conveyor of the conveying device 104, or may be a sheet formed in a flat shape, unlike the example shown in FIG. 4(b). The amount of aggregate placed on the sheet 132 is not particularly limited, but may be about 1 kg to 10 kg. FIG. 4(b) shows a schematic example of aggregate piled up on the sheet 132, and shows piled up states when the amount of aggregate is 2 kg, 5 kg, and 8 kg. Note that the sheet 132 may be a container other than a rubber sheet. If it is a container with a rim, it is preferable from the viewpoint that it can be used to uniformly adjust the size of the captured aggregate by correcting the size of the image data based on the distance calculation result after calculating the distance from the camera to the aggregate.

[0057] In one example, samples of various aggregates are prepared under the conditions shown in Table 1 below, and images of the prepared samples are acquired to obtain learning images. The mass fractions of each gradation range of the various aggregates prepared under the conditions shown in Table 1 are known, and the known values ​​are the correct answer data for machine learning. When the mass fractions of each gradation range are known, the coarse aggregate fraction is also known.

[0058] [Table 1]

[0059] The coarse grain ratio (FM) is calculated, for example, by the following formula (1) specified in the Standard Method for Concrete. In the measurement specified in the "Standard Specifications for Building Construction and Commentary JASS5 Reinforced Concrete Construction", 9 sieves are used excluding the 80 mm sieve in formula (1). The coarse grain ratio predicted in this disclosure may be specified by any method.

number

[0060] In Table 1, this means that aggregates of 10 levels of gradation (coarse aggregate ratio) are prepared. Category I represents a gradation category of particles smaller than 2.5 mm. Category I includes particles that cannot be retained on a 2.5 mm sieve. Category II represents a gradation category of particles 2.5 mm or larger and smaller than 5 mm. Category II includes particles that can be retained on a 2.5 mm sieve but cannot be retained on a 5 mm sieve. Category III represents a gradation category of particles 5 mm or larger and smaller than 10 mm. Category III includes particles that can be retained on a 5 mm sieve but cannot be retained on a 10 mm sieve. Category IV represents a gradation category of particles 10 mm or larger and smaller than 20 mm. Category IV includes particles that can be retained on a 10 mm sieve but cannot be retained on a 20 mm sieve.

[0061] For level 1, the coarse aggregate ratio can be calculated from the mass fractions of each grain size category (mass fractions of each of categories I to IV) as follows. For level 1 aggregate, the mass percentages of the sample that are retained on the 80 mm, 40 mm, and 20 mm sieves are 0, the mass percentages of the sample that are retained on the 10 mm sieve are 85, and the mass percentages of the sample that are retained on the 5 mm sieve are 100 (=15+85). Also, since all the grains are retained on the 5 mm sieve, the mass percentages of the sample that are retained on the 2.5 mm, 1.2 mm, 600 μm, 300 μm, and 150 μm sieves are 100. Therefore, the coarse aggregate ratio is calculated as shown in the following formula (2).

number

[0062] For levels 2 to 10, the coarse aggregate ratio is calculated in the same manner as for level 1. The conditions of the learning aggregate shown in Table 1 are merely examples, and the number of levels, the coarse aggregate ratio and mass fraction for each grading class at each level, and the method of setting the grading classes may be set arbitrarily. As with the multiple grading classes shown in Table 1, the grading classes in multiple classes including the smallest class (<150 μm) may be combined into one grading class. Grain sizes above a certain class (e.g., 20 to 40 mm) may not be included in multiple grading classes. Also, unlike the setting method in Table 1, grain sizes below a certain class (e.g., 1.2 to 2.5 μm) may not be included in multiple grading classes.

[0063] FIG. 5 shows images T1 to T10 obtained by imaging learning aggregates prepared according to the conditions of each of levels 1 to 10 in Table 1. Level n (n is an integer from 1 to 10) corresponds to image Tn. For example, image T1 is obtained by imaging learning aggregates prepared according to level 1. Learning aggregates according to each level n contain grains of multiple grain size divisions. In this way, learning aggregates may be prepared so as to contain grains of two or more grain size divisions, and input information may be prepared based on image data obtained by imaging the learning aggregates.

[0064] For each level n, multiple images Tn in which the states (more specifically, pile states) of the learning aggregate are different from each other may be prepared. In one example, a first image is taken with the learning aggregate placed on the measurement sheet 132, to obtain one image Tn. Then, the learning aggregate is temporarily removed from the sheet 132, and the same learning aggregate (same sample) is placed again on the sheet 132 so that the pile state of the multiple grains is different from that in the first image. Then, a second image is taken. By repeating the above-mentioned image taking, multiple images Tn may be prepared for each level n.

[0065] One data set is composed of one image Tn and the ground truth data of mass fraction at level n associated with the image Tn. As learning data, a plurality of data sets each consisting of a combination of an image Tn and ground truth data of mass fraction are prepared. Then, machine learning is performed using such learning data to construct a prediction model M. The prediction model M thus constructed outputs, for example, four mass fraction values ​​for four particle size classes.

[0066] In order to verify the method of predicting the coarse aggregate ratio using the prediction model M based on machine learning using a convolutional neural network, a comparison was made between the prediction results by the prediction model M and the correct values ​​using a dataset in which the correct values ​​were known. For the learning aggregate, 10 levels of aggregate with different coarse aggregate ratios were prepared according to the conditions in Table 1 above. For each level, images were taken 10 times indoors with the aggregate piled up on the measurement sheet 132 in different states, to obtain 10 images.

[0067] In addition, a total of 100 images (10 levels x 10 shots) were prepared according to the standard conditions shown in Table 2 below. As a result, 100 data sets were prepared as learning data, each of which corresponds to an image and a correct answer value.

[0068] [Table 2]

[0069] The prediction model M was constructed using 80 of the 100 datasets, and 20 datasets were used for validation. At each level, 8 data were used for model construction, and 2 datasets were used for validation. In the validation using 20 datasets, the prediction results calculated using the image and the prediction model M for each dataset were compared with the correct value of the mass fraction associated with the image. The prediction model M was constructed using a convolutional neural network with six intermediate layers consisting of a convolutional layer and a pooling layer, followed by four fully connected layers. The Huber function was used as the loss function. In each dataset, the number of images was increased by an image augmentation layer.

[0070] FIG. 6 is a graph showing the relationship between the predicted values ​​of mass fraction (%) in the sections III and IV in Table 1 and the corresponding correct values ​​of mass fraction (%). In the graph shown in FIG. 6, the horizontal axis is the correct value of mass fraction, and the vertical axis is the predicted value of mass fraction output from the prediction model M. The prediction results in section III (5 mm to 10 mm) are plotted with △ marks, and the prediction results in section IV (10 mm to 20 mm) are plotted with ◯ marks. A solid line where the correct value and the predicted value match is drawn, and the closer the plotted results are to the line, the higher the prediction accuracy. From the results shown in FIG. 6, the prediction results are plotted near the line, and it can be seen that the mass fraction can be predicted with high accuracy by the prediction model M. Since the coarse particle fraction can be obtained from the mass fraction, if the mass fraction can be predicted with high accuracy by the prediction model M, the coarse particle fraction can also be predicted with high accuracy.

[0071] [Quality prediction method] Next, a quality prediction method executed in the quality prediction device 1 will be described with reference to Figs. 7 to 12. Hereinafter, the quality prediction method will be described using an example in which the quality prediction device 1 is configured by a smartphone with a camera. This quality prediction method includes, for example, a specification step, an image display step, an acquisition step, a first prediction step, a second prediction step, and a result display step. The specification step, the image display step, the acquisition step, the first prediction step, the second prediction step, and the result display step are all steps executed in the evaluation phase. That is, these steps are executed after the prediction model M has already been constructed. In the learning phase, a plurality of prediction models M may be constructed according to a plurality of types of aggregate (a plurality of aggregate types).

[0072] The identifying step is a step of identifying the type of aggregate to be evaluated from among a plurality of aggregate types, based on a user input indicating the type of aggregate to be evaluated. The image displaying step is a step of displaying an image obtained by imaging the aggregate to be evaluated on the monitor 14 before the acquiring step. The image displaying step may include rotating the image displayed on the monitor 14 when a user input instructing image rotation is input. In the rotation of the image displayed on the monitor 14, for example, the image is rotated 90° to the right or left so that the vertical and horizontal directions on the image are switched. When the outer frame of the image is a square, the coordinates are converted in accordance with the rotation of the image.

[0073] The acquisition step is a step of acquiring evaluation input information based on evaluation image data acquired by capturing an image of the aggregate to be evaluated. In the acquisition step, the evaluation input information may be determined based on the rotation state of the captured image in the image display step. In one example, in the image display step, data indicating the current captured image (at the time of performing the acquisition step) displayed on the monitor 14 is acquired as the evaluation input information. Even if the captured image is the same, the pixel values ​​for each coordinate are different between an image that is not rotated by 90 degrees and an image that is rotated by 90 degrees, and therefore different evaluation input information is obtained.

[0074] The first prediction step is a step of inputting evaluation input information to the prediction model M and acquiring predicted values ​​output from the prediction model M for each granularity classification. In the first prediction step, a prediction model M according to the identification result of the type of aggregate to be evaluated in the identification step may be selected from among a plurality of prediction models M, and a predicted value for each granularity classification may be acquired using the selected prediction model M. The second prediction step is a step of calculating the coarse aggregate ratio of the aggregate to be evaluated from the predicted value for each granularity classification acquired in the first prediction step. In the second prediction step, for example, the coarse aggregate ratio is calculated from the predicted value for each granularity classification according to the above formula (1).

[0075] The result display step is a step of displaying on the monitor 14 an inference graph in which the first axis is the nominal size of the sieve and the second axis is the mass fraction of the particles passing through each sieve. The inference graph displayed on the monitor 14 includes an inference result calculated based on the result obtained in the first prediction step and a standard range according to the type of aggregate to be evaluated. In the result display step, the standard range may be displayed on the monitor 14 according to the type of aggregate to be evaluated identified in the identification step. That is, the standard range on the inference graph differs depending on the type of aggregate to be evaluated. In the result display step, the calculation result of the coarse grain ratio of the aggregate to be evaluated in the second prediction step may be displayed on the monitor 14 together with the inference graph.

[0076] <Processing flow> FIG. 7 is a flow diagram illustrating a series of processes executed in the quality prediction method. A user holds the quality prediction device 1 and executes step S01 at any timing when the user wants to predict the coarse aggregate ratio of the aggregate to be evaluated. In step S01, for example, the user starts an application for predicting quality by operating on a screen displayed on the monitor 14. In one example, as shown in FIG. 8(a), in a state where a home screen is displayed on the monitor 14 of the quality prediction device 1, the user clicks, taps, or presses an icon 60 representing an application for predicting quality.

[0077] When the quality prediction device 1 receives the start request of step S01, it executes step S02. In step S02, for example, the display output unit 32 of the quality prediction device 1 displays a home screen of an application that executes quality prediction on the monitor 14. An example of the home screen is shown in Fig. 8(b), and on the home screen, an execute button 62 for switching to a mode for photographing (taking an image of) an aggregate and an execute button 64 for switching to a mode for executing another function (for example, a mode for viewing past prediction results) are displayed.

[0078] Next, the user executes step S03. In step S03, for example, the user presses (e.g., taps) the execute button 62 displayed on the monitor 14 to select a mode for photographing the aggregate. When the selection of step S03 is made, the quality prediction device 1 executes step S04. In step S04, for example, the display output unit 32 switches the display content on the monitor 14 to a screen corresponding to the mode for photographing the aggregate. FIG. 9(a) shows an example of the screen after switching in step S04.

[0079] Next, the user executes step S05. In step S05, for example, the user operates monitor 14 to select a type corresponding to the aggregate to be evaluated from among a plurality of aggregate types. As also shown in FIG. 9(b), display output unit 32 may display on monitor 14 a pull-down menu 66 from which the user can select the type of aggregate to be evaluated. In pull-down menu 66, the user may select one that matches the type of aggregate to be evaluated from a plurality of aggregate types previously determined. FIG. 10(a) shows an example of a display screen after the type of aggregate to be evaluated has been selected.

[0080] When the selection of step S05 is performed, the quality prediction device 1 executes step S06. In step S06, for example, the prediction calculation unit 30 and the display output unit 32 identify the type of aggregate to be evaluated according to the selection result of step S05. After the execution of steps S05 and S06, the display output unit 32 displays a start button 68 for inputting an instruction to start the camera 16. Before the execution of steps S05 and S06, the display output unit 32 may display the start button 68 on the monitor 14 in a state in which the user cannot tap it. In FIG. 9(a) and the like, the button shown by the dashed line represents a state in which the user cannot tap it. Alternatively, the display output unit 32 may not display the start button 68 on the monitor 14 before the execution of steps S05 and S06, and may display the start button 68 on the monitor 14 after the execution of steps S05 and S06.

[0081] Next, the user executes step S07. In step S07, for example, the user instructs the camera 16 to be started, and then the camera 16 performs image capture. In one example, when the user taps the start button 68, the quality prediction device 1 starts the camera 16, and as shown in FIG. 10(b), the display output unit 32 displays a preview image 70 captured by the camera 16 and an image capture execution button 71 for instructing the camera 16 to perform image capture on the monitor 14. The user taps the image capture execution button 71 while viewing the preview image 70 on the monitor 14, whereby the camera 16 performs image capture of the aggregate to be evaluated.

[0082] After the imaging in step S07, the quality prediction device 1 executes step S08. In step S08, for example, the image data acquisition unit 24 acquires captured image data generated by imaging using the camera 16, and the display output unit 32 displays an image 72 (captured image) based on the captured image data on the monitor 14. Fig. 11 illustrates a state in which the image 72 based on the captured image data, a rotation execution button 74 for instructing rotation of the image 72, and a prediction execution button 76 for instructing execution of prediction using the prediction model M are displayed on the monitor 14.

[0083] Next, the user executes step S09. In step S09, for example, the user taps the rotation execution button 74 as necessary so as to rotate the image 72 on the monitor 14 by 90° per tap. If the user does not need to rotate the image 72 displayed on the monitor 14, he or she does not tap the rotation execution button 74. In one example, when the image 72 is captured by the camera 16 mounted on a device such as a smartphone constituting the quality prediction apparatus 1, the vertical and horizontal directions of the image 72 may be automatically converted. In this case, the user can adjust the image 72 to match the vertical direction, horizontal direction, up and down, and left and right within the field of view of the camera 16 by tapping the rotation execution button 74 one or more times.

[0084] Also, in step S09, for example, when the user determines that the image 72 displayed on the monitor 14 is acceptable as an input to the prediction model M, the user taps the prediction execution button 76. Before step S08 is executed, the display output unit 32 may display the prediction execution button 76 on the monitor 14 in a state in which the user cannot tap it. Alternatively, the display output unit 32 may not display the prediction execution button 76 on the monitor 14 before step S08 is executed, and may display the prediction execution button 76 on the monitor 14 after step S08 is executed.

[0085] When the quality prediction device 1 receives an execution instruction for prediction in step S09, it executes step S10. In step S10, for example, the evaluation information acquisition unit 26 acquires evaluation input information to be input to the prediction model M based on the rotation state of the image 72 displayed on the monitor 14 at the time when the prediction execution button 76 is tapped. The evaluation information acquisition unit 26 acquires, for example, pixel values ​​for each coordinate from the image 72 displayed at the time when the prediction execution button 76 is tapped, as at least a part of the evaluation input information.

[0086] Also, in step S10, for example, the prediction calculation unit 30 inputs the acquired evaluation input information to a prediction model M and acquires a predicted value of the mass fraction for each gradation class output from the prediction model M. The prediction calculation unit 30 acquires a predicted value of the mass fraction for each gradation class using the prediction model M associated with the type of aggregate to be evaluated identified in step S06. The prediction calculation unit 30 calculates the coarse aggregate ratio from the mass fraction for each gradation class.

[0087] Next, the quality prediction device 1 executes step S11. In step S11, for example, the display output unit 32 displays the result of the prediction calculation executed in step S10 on the monitor 14. Fig. 12 illustrates an example of the state in which the result of the prediction calculation in step S10 is displayed on the monitor 14.

[0088] The display output unit 32 displays an inference graph 82 including a first axis and a second axis on the monitor 14. The inference graph 82 displayed on the monitor 14 may be a two-dimensional graph (two-dimensional Cartesian coordinate system). The first axis represents the nominal size (mm) of the sieve, and is, for example, the horizontal axis. The nominal size of the sieve may also be referred to as the particle size of the sieve. The second axis represents the mass fraction (%) of the particles passing through each sieve, and is, for example, the vertical axis. The mass fraction of the particles passing through each sieve may also be referred to as the weight passing percentage, or the passing weight percentage.

[0089] When a point plotted at a certain position on the inference graph 82 is expressed as (Xmm, Y%), that point means that the mass fraction of particles passing through a sieve with a nominal dimension of X (mm) is Y (%). The inference graph 82 includes an inference result calculated based on the predicted value for each granularity class obtained from the prediction model M in step S10, and a standard range on the second axis (vertical axis) that indicates a standard range according to the type of aggregate to be evaluated. The inference result calculated based on the predicted value for each granularity class may be a plurality of points plotted on a graph, or may be a plurality of points and a line obtained by connecting those points with line segments.

[0090] The inference result will be described using a specific example. For example, assume that the predicted value of the mass fraction in section I (<2.5 mm) is 3.2%, the predicted value of the mass fraction in section II (2.5-5.0 mm) is 6.5%, the predicted value of the mass fraction in section III (5.0-10 mm) is 32.0%, and the predicted value of the mass fraction in section IV (10-20 mm) is 58.3%. In this case, the display output unit 32 may display the inference result by plotting four points, (2.5 mm, 3.2%), (5 mm, 9.7%), (10 mm, 41.7%), and (20 mm, 100%), on the inference graph 82.

[0091] The display output unit 32 may display an inference line C as an inference result on the monitor 14 as shown in Fig. 12. When the above-mentioned assumed result is obtained, the inference line C is drawn, for example, by the above-mentioned four points and straight lines connecting adjacent points. Note that the four points may have a size approximately equal to the width of the straight lines, and the inference line C may be drawn only by straight lines.

[0092] The standard range included in the inference graph 82 is defined by a first boundary line BL1 and a second boundary line BL2. The second boundary line BL2 is located above the first boundary line BL1 on the graph. With respect to the vertical axis (mass fraction), the range between the first boundary line BL1 and the second boundary line BL2 represents the standard range for the type of aggregate being evaluated. The first boundary line BL1 is a line based on the mass fraction of particles passing through each sieve when the overall particle size of the type of aggregate being evaluated is maximum. The second boundary line BL2 is a line based on the mass fraction of particles passing through each sieve when the overall particle size of the type of aggregate being evaluated is minimum.

[0093] The standard range will be explained using a specific example. For example, in JIS A5005:2020 (crushed stone and crushed sand for concrete), the grain size of crushed stone 2005 is defined as shown in Table 3 below. The numerical range according to the nominal size in Table 3 indicates the mass fraction (%) of the grains passing through each sieve. Hereinafter, the mass fraction of the grains passing through each sieve may be simply referred to as the "mass fraction," and the combination of mass fractions according to various nominal sizes is referred to as the distribution of mass fractions. Note that for crushed stone 2005, the mass fraction is not defined in the range where the nominal size of the sieve is 40 mm or more and 1.2 mm or less, so these ranges are omitted in Table 3 below. Note that, although it has been explained that the distribution of the mass fraction of crushed stone 2005 is shown in Table 3, it can be said that crushed stone having the mass fraction distribution shown in Table 3 is classified as crushed stone 2005. [Table 3] In Table 3, "-" means that the mass fraction (%) is not defined.

[0094] Table 4 shows the mass fraction distribution when the overall particle size is maximum within the range of the mass fraction distribution shown in Table 3. The overall particle size is maximum means that the particle size is the largest when evaluating all the particles contained in the aggregate within the range of the mass fraction distribution classified into a specific type. [Table 4]

[0095] 12 is drawn according to the distribution of mass fractions shown in Table 4. Specifically, the display output unit 32 plots five points, (2.5 mm, 0%), (5 mm, 0%), (10 mm, 20%), (20 mm, 90%), and (25 mm, 100%), and then draws the first boundary line BL1 by connecting adjacent points with straight lines.

[0096] Table 5 shows the mass fraction distribution when the overall particle size is the smallest within the range of the mass fraction distribution shown in Table 3. The overall particle size is the smallest means that the particle size is the smallest when evaluating all the particles contained in the aggregate within the range of the mass fraction distribution classified into a specific type. [Table 5]

[0097] 12 is drawn according to the distribution of mass fractions shown in Table 5. Specifically, the display output unit 32 plots four points, (2.5 mm, 5%), (5 mm, 10%), (10 mm, 55%), and (20 mm, 100%), and then draws the second boundary line BL2 by connecting adjacent points with straight lines.

[0098] If the inference result of the mass fraction by the prediction model M (e.g., inference line C) falls within the standard range included in the inference graph 82, the predicted mass fraction can be evaluated as appropriate for the type of aggregate to be evaluated. If the inference result of the mass fraction by the prediction model M (e.g., inference line C) falls outside the standard range included in the inference graph 82, the predicted mass fraction can be evaluated as inappropriate for the type of aggregate to be evaluated.

[0099] It is also possible to calculate the coarse aggregate ratio from the mass fraction using the prediction model M, and evaluate the predicted value of the coarse aggregate ratio to evaluate whether the quality of the aggregate is as desired. However, even if the predicted value of the coarse aggregate ratio is within the desired range, when observing the mass fraction (predicted value) for each nominal size of the sieve, it may be found to be outside the desired range. In response to this, by displaying the inference graph 82 described above on the monitor 14, the user can easily grasp whether the quality of the aggregate is as desired from the viewpoint of the distribution of the mass fraction as well.

[0100] In step S11, the display output unit 32 may display first result information 84a, second result information 84b, and a back button 86 on the monitor 14 in addition to the inference graph 82. The first result information 84a includes the type of aggregate to be evaluated selected by the user and the coarse aggregate ratio obtained from the predicted value of the mass fraction for each of the above-mentioned gradation classes. The second result information 84b includes the predicted value of the mass fraction for each of the above-mentioned gradation classes.

[0101] When the user taps the back button 86, for example, the process executed by the quality prediction device 1 returns to step S02, and the display screen exemplified in Fig. 8(b) is displayed on the monitor 14. The user may repeatedly predict the quality of the aggregate to be evaluated and evaluate the quality by using the quality prediction device 1 at different times.

[0102] [Variations] The process flow shown in Fig. 7 is an example and can be modified as appropriate. For example, the quality prediction device 1 may execute step S10 after executing step S07, without going through steps S08 and S09. When there is only one type of aggregate to be predicted in quality by the quality prediction device 1, steps S05 and S06 may be omitted. Even if there is only one type of aggregate, the standard range in the inference graph 82 displayed in that case is still the range corresponding to that one type.

[0103] In step S11, the display output unit 32 may display any information on the monitor 14 as long as it displays at least the inference graph 82 on the monitor 14. The display output unit 32 may display other graphs in addition to the inference graph 82. The display output unit 32 may display, as the other graphs, a graph in which the predicted values ​​of the mass fraction for each particle size class are plotted as is. In step S09, the quality prediction device 1 may not have a function of changing the angle of the image 72 via the operation of the rotation execution button 74.

[0104] The quality prediction device 1 may be a server connected to a terminal device such as a smartphone that can be operated by a user via a communication network. In this case, the terminal device may execute steps S01, S03, S05, S07, and S09, and the quality prediction device 1 as the server may execute steps S02, S04, S06, S08, S10, and S11. The display output unit 32 included in the quality prediction device 1 may display various information on a monitor included in the terminal device, which is a device other than the quality prediction device 1.

[0105] The quality prediction device 1 as the server may exchange information with the terminal device by presenting to the user a Web page or the like that can be viewed and accessed from the terminal device operated by the user. In this case, in step S01, login may be executed instead of starting an application. Alternatively, instead of a Web page, an application that cooperates with the quality prediction device 1 may be prepared in the terminal device, and the quality prediction device 1 as the server may provide information to the terminal device.

[0106] The prediction model M described above is an example, and the prediction model M may be any model as long as it can output at least a plurality of predicted values ​​related to mass fractions for a plurality of gradation classes from information based on image data of aggregates. The input data to the prediction model M may include information indicating physical quantities that may affect the mass fraction or the coarse aggregate fraction, in addition to image data obtained by imaging the aggregate.

[0107] As a result of the study by the present inventors, it was found that when the aggregate is not air-dried but wet, the prediction accuracy of the coarse aggregate ratio may decrease depending on the amount of aggregate to be photographed. In step S10, the prediction calculation unit 30 may correct the predicted value of the mass fraction for each gradation class output from the prediction model M, and calculate the coarse aggregate ratio based on the corrected predicted value (the predicted value obtained by the correction).

[0108] Specifically, the prediction calculation unit 30 may perform the following first and second corrections. In the first correction, when the predicted value output from the prediction model M for each particle size classification is smaller than 0, the prediction calculation unit 30 performs a correction so that the corrected predicted value is set to 0, and when the predicted value output from the prediction model M is larger than 0 (or the predicted value is equal to or larger than 0), the prediction calculation unit 30 performs a correction so that the predicted value is used as the corrected predicted value as is.

[0109] In the second correction, the prediction calculation unit 30 calculates a corrected predicted value for each particle size division so that the sum of the corrected predicted values ​​for each particle size division is 100%. Specifically, in the second correction, the prediction calculation unit 30 divides the mass fraction value for each particle size division by the sum of the mass fractions before the second correction, and then multiplies the result by 100. The prediction calculation unit 30 may perform the second correction after performing the first correction. That is, when performing the second correction, the prediction calculation unit 30 may perform the second correction on the predicted value obtained after the first correction. The calculation of the coarse particle ratio based on the corrected predicted value for the mass fraction for each particle size division is performed in the same manner as the calculation in the above formula (2).

[0110] The first and second corrections will be further described using specific values. Table 6 below shows an example of calculation when image data of level 3 shown in Table 1 is input to the prediction model M and corrected.

[0111] [Table 6]

[0112] In Table 6, "before correction" represents the value output from the prediction model M, "first correction" represents the value obtained by applying the first correction to the value before correction, and "second correction" represents the value obtained by applying the second correction to the value after the first correction. In the value before correction, the mass fraction value of category II is negative, and the sum of the mass fractions exceeds 100. By applying the first and second corrections, the negative values ​​become 0, and the sum of the mass fractions becomes 100. The calculated value of the coarse particle ratio before correction is 6.95, while the calculated value of the coarse particle ratio after applying the first and second corrections is 6.75. In calculating the coarse particle ratio, negative values ​​are added as they are, and the percentage of the sample remaining on each sieve is assumed to be 100 for categories I and below. If the mass fraction of each division is expressed by the division name, the coarse particle ratio can be calculated by {(Division IV) + (Division IV + Division III) + (Division IV + Division III + Division II) + (Division IV + Division III + Division II + Division I) + 100 × 3} / 100.

[0113] The graph shown in FIG. 13(a) shows the results of the verification using samples with different moisture conditions on the aggregate surface. In this verification, the "air-dried state" is a state in which the sample aggregate is left indoors for a predetermined time to dry (naturally dry). In addition, the "wet state" is a state in which water is added to the sample aggregate at a predetermined ratio (for example, 1% to 5% by mass ratio) relative to the mass of the aggregate, and mixed in a container. FIG. 13(a) shows the result when the coarse aggregate ratio is calculated without performing the first and second corrections, and FIG. 13(b) shows the result when the coarse aggregate ratio is calculated with the first and second corrections. In addition, evaluations are also performed when the amount of aggregate is 2 kg, 5 kg, and 8 kg, and prediction models M are constructed for each of the cases in which the amount of aggregate is 5 kg and 8 kg.

[0114] The average absolute difference for the coarse aggregate ratio was calculated as an index showing how far the predicted value deviates from the correct value. The average absolute difference is the arithmetic average of the absolute values ​​of the differences between the correct value and the predicted value in the 20 validation data sets. The vertical axis represents the average absolute difference, and "Tg" represents the target value. If the average absolute difference is equal to or less than the target value Tg (0.10), the accuracy of the prediction using the prediction model M can be evaluated as good. The target value Tg was set to 0.10, taking into consideration the error in the coarse aggregate ratio currently permitted in ready-mix concrete plants.

[0115] From the results shown in FIG. 13(a), it can be seen that in the air-dry state, the absolute difference average value is equal to or less than the target value Tg, and can be predicted with good accuracy by the prediction model M. On the other hand, in the wet state, when the amount of aggregate is 8 kg, the absolute difference average value exceeds the target value Tg. In contrast, as shown in FIG. 13(b), when the coarse aggregate ratio is calculated after the correction, the absolute difference average value is equal to or less than the target value Tg in both the air-dry state and the wet state. In other words, it can be seen that even in the wet state, the coarse aggregate ratio can be predicted with good accuracy by performing the first correction and the second correction and calculating the coarse aggregate ratio. Note that the prediction calculation unit 30 may perform either the first correction or the second correction and then calculate the coarse aggregate ratio from the predicted value of the mass fraction for each gradation class.

[0116] In one example among the various examples described above, at least a part of the matters described in the other examples may be combined.

[0117] [Summary of this disclosure] The quality prediction program described above is a program for making a computer execute the following steps: an acquisition step of acquiring evaluation input information based on captured image data obtained by capturing an image of an aggregate to be evaluated; a first prediction step of inputting evaluation input information to a prediction model (M) constructed to output a value indicating a mass fraction for each of a plurality of particle size classifications according to input of the input information by machine learning based on input information based on learning image data obtained by capturing an image of an aggregate for learning and a ground truth value of the mass fraction for each of a plurality of particle size classifications of the aggregate corresponding to the input information, and acquiring a predicted value output from the prediction model (M) for each of the classifications; and a result display step of displaying an inference graph (82) on a monitor (14) in which a first axis is the nominal size of a sieve and a second axis is the mass fraction of the grains passing through each sieve. The inference graph (82) displayed on the monitor (14) in the result display step includes an inference result (C) calculated based on the predicted value for each of the classifications, and a standard range for the second axis indicating a standard range according to the type of aggregate to be evaluated.

[0118] By obtaining a predicted value for the coarse aggregate ratio of the aggregate to be evaluated by some method and evaluating the predicted value, it is possible to evaluate whether the quality of the aggregate is desired (for example, within a control range that is deemed appropriate). However, even if the predicted value of the coarse aggregate ratio is within the desired range, when the mass fraction for each nominal size of the sieve is observed, it may be outside the desired range. In response to this, in the above-mentioned quality prediction program, an inference graph (82) showing the relationship between the nominal size of the sieve and the mass fraction is displayed on the monitor (14). The inference graph (82) includes an inference result obtained from the predicted value for the mass fraction and a standard range for the mass fraction according to the type of aggregate to be evaluated. Therefore, by comparing the inference result with the standard range, the user can easily understand whether the quality of the aggregate is desired from the viewpoint of the distribution of the mass fraction as well. Therefore, the above-mentioned quality prediction program is useful for improving the convenience of the user.

[0119] The standard range is defined by a first boundary line (BL1) based on the mass fraction of particles passing through each sieve when the overall particle size of the aggregate type to be evaluated is maximum, and a second boundary line (BL2) based on the mass fraction of particles passing through each sieve when the overall particle size of the aggregate type to be evaluated is minimum. In this case, the user can more easily understand whether the inference result is within the standard range or not by looking at the relationship between the first boundary line (BL1) and the second boundary line (BL2). Therefore, it is further useful for improving the convenience of the user.

[0120] The quality prediction program may further cause the computer to execute a specification step of specifying the type of aggregate to be evaluated from among a plurality of aggregate types, based on a user input indicating the type of aggregate to be evaluated. In the result display step, a standard range may be displayed on the monitor (82) according to the type of aggregate to be evaluated specified in the specification step. In a manufacturing system (100) for manufacturing ready-mix concrete, various types of aggregate may be used. In the program, the standard range is displayed according to the type of aggregate to be evaluated, so that the user can evaluate whether the mass fraction is within an appropriate range for various types of aggregate. Therefore, it is further useful for improving user convenience.

[0121] The quality prediction program may further cause the computer to execute an image display step of displaying on the monitor (14) a captured image (72) obtained by capturing an image of the aggregate to be evaluated, prior to the acquisition step. In this case, the user can obtain a prediction result from the prediction model M by understanding whether the captured image (72) input to the prediction model M is suitable for the prediction model M, or by understanding the state of the captured image (72) input to the prediction model M. In other words, the user can understand the prediction result from the prediction model M after understanding the input contents to the prediction model M. This is therefore further useful for improving user convenience.

[0122] The image display step may include rotating the captured image (72) displayed on the monitor (14) when a user input instructing to rotate the image is input. In the acquisition step, evaluation input information may be determined based on the rotated state of the captured image (72) in the image display step. There may be cases where the vertical and horizontal relationship of the captured image (72) to be input to the prediction model M is not suitable for prediction by the prediction model M. In this case, the user can adjust the rotated state of the captured image (72) to a state suitable for prediction by the prediction model M without repeatedly acquiring the captured image (72) by imaging. This is therefore further useful for improving user convenience.

[0123] The quality prediction program may further cause the computer to execute a second prediction step of calculating the coarse aggregate ratio of the aggregate to be evaluated from the predicted values ​​for each classification obtained in the first prediction step. In the result display step, the calculation result of the coarse aggregate ratio of the aggregate to be evaluated in the second prediction step may be displayed on the monitor (14) together with the inference graph (82). In this case, the user can use the inference graph (82) to evaluate whether the mass fraction for each nominal size of the sieve is appropriate, while also evaluating whether the coarse aggregate ratio is within a desired range. This is therefore even more useful in improving user convenience.

[0124] The second prediction step may include correcting the predicted value obtained in the first prediction step, and calculating the coarse aggregate ratio of the aggregate to be evaluated from the corrected predicted value. In the correction of the predicted value in the second prediction step, the corrected predicted value may be calculated for each classification so that the sum of the corrected predicted values ​​for each classification is 100%. Since the prediction model (M) is constructed so as to obtain results close to the correct value for various inputs, the sum of the mass fractions for each granularity classification may not be 100% in the calculation result by the prediction model (M). In the above program, even if a result in which the sum of the mass fractions is different from 100% is obtained, the sum is corrected to a value that matches reality. Therefore, it is useful for improving the prediction accuracy regarding the prediction of the coarse aggregate ratio.

[0125] The second prediction step may include correcting the predicted value acquired in the first prediction step, and calculating the coarse aggregate ratio of the aggregate to be evaluated from the corrected predicted value. In the correction of the predicted value in the second prediction step, if the predicted value output from the prediction model (M) is smaller than 0, the corrected predicted value may be corrected to 0, and if the predicted value output from the prediction model (M) is larger than 0, the predicted value may be corrected to be the corrected predicted value as it is. Since the prediction model (M) is constructed so as to obtain results close to the corrected value for various inputs, the calculation result by the prediction model (M) may have a negative mass fraction. In the above program, even if the mass fraction becomes negative, it is corrected to a value that matches reality. Therefore, it is useful for improving the prediction accuracy regarding the prediction of the coarse aggregate ratio.

[0126] The quality prediction method described above includes an acquisition step of acquiring evaluation input information based on captured image data obtained by capturing an image of an aggregate to be evaluated, a first prediction step of inputting evaluation input information to a prediction model (M) constructed to output a value indicating a mass fraction for each of a plurality of granularity classifications according to input of input information by machine learning based on input information based on learning image data obtained by capturing an image of an aggregate for learning and a ground truth value of the mass fraction for each of a plurality of granularity classifications of the aggregate associated with the input information, and acquiring a predicted value output from the prediction model (M) for each classification, and a result display step of displaying an inference graph (82) on a monitor (14) in which a first axis is the nominal size of a sieve and a second axis is the mass fraction of grains passing through each sieve. The inference graph (82) displayed on the monitor (14) in the result display step includes an inference result (C) calculated based on the predicted value for each classification, and a standard range for the second axis indicating a standard range according to the type of aggregate to be evaluated. This quality prediction method is useful for improving user convenience, similar to the above-mentioned quality evaluation program.

[0127] The quality prediction device (1) described above includes an evaluation information acquisition unit (26) that acquires evaluation input information based on captured image data obtained by capturing an image of an aggregate to be evaluated, a prediction calculation unit (30) that inputs evaluation input information to a prediction model (M) that is constructed to output a value indicating a mass fraction for each of a plurality of particle size classifications according to input of input information by machine learning based on input information based on learning image data obtained by capturing an image of an aggregate for learning and a ground truth value of the mass fraction for each of a plurality of particle size classifications of the aggregate associated with the input information, and acquires a predicted value output from the prediction model (M) for each classification, and a display output unit (32) that displays an inference graph (82) on a monitor (14) in which a first axis is the nominal size of a sieve and a second axis is the mass fraction of grains passing through each sieve. The inference graph (82) displayed on the monitor (14) by the display output unit (32) includes an inference result (C) calculated based on the predicted value for each classification, and a standard range indicating a standard range according to the type of aggregate to be evaluated for the second axis. This quality prediction device (1), like the above-mentioned quality evaluation program, is useful for improving user convenience. [Explanation of symbols]

[0128] 1...quality prediction device, 14...monitor, 26...evaluation information acquisition unit, 30...prediction calculation unit, 32...display output unit, M...prediction model, 82...inference graph, BL1...first boundary line, BL2...second boundary line.

Claims

1. an acquisition step of acquiring evaluation input information based on captured image data obtained by capturing an image of the aggregate to be evaluated; a first prediction step of inputting the evaluation input information into a prediction model constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for learning and on correct values ​​of mass fractions of a plurality of granularity classifications of the aggregate associated with the input information, so as to output a value indicating a mass fraction for each classification included in the plurality of granularity classifications in response to input of the input information, and acquiring a predicted value output from the prediction model for each classification; a result display step of displaying on a monitor an inference graph in which a first axis is the nominal size of the sieve and a second axis is the mass fraction of the grains passing through each sieve; A program for causing a computer to execute the following: The inference graph displayed on the monitor in the result display step includes: An inference result calculated based on the predicted value for each of the sections; A quality prediction program including a standard range representing a standard range according to the type of aggregate to be evaluated with respect to the second axis.

2. The standard range is a first boundary line based on the mass fraction of particles passing through each sieve when the overall particle size of the aggregate type being evaluated is maximum; a second boundary line based on the mass fraction of particles passing through each sieve when the overall particle size of the aggregate type being evaluated is the smallest; and The quality prediction program according to claim 1 , wherein the quality prediction program is defined by:

3. The quality prediction program further causes the computer to execute a specification step of specifying a type of the aggregate to be evaluated from among a plurality of aggregate types based on a user input indicating a type of the aggregate to be evaluated; 3. The quality prediction program according to claim 1, wherein in the result display step, the standard range is displayed on the monitor according to the type of the aggregate to be evaluated identified in the identification step.

4. 3. The quality prediction program according to claim 1, further comprising an image display step of displaying an image obtained by imaging the aggregate to be evaluated on the monitor before the acquisition step.

5. the image display step includes rotating the captured image displayed on the monitor when a user input instructing to rotate the image is input; The quality prediction program according to claim 4 , wherein in the acquiring step, the evaluation input information is determined based on a rotation state of the captured image in the image display step.

6. The quality prediction program further causes the computer to execute a second prediction step of calculating a coarse aggregate ratio of the aggregate to be evaluated from the predicted value for each of the categories obtained in the first prediction step, 3. The quality prediction program according to claim 1, wherein in the result display step, the calculation result of the coarse grain ratio of the aggregate to be evaluated in the second prediction step is displayed on the monitor together with the inference graph.

7. The second prediction step includes: correcting the predicted value acquired in the first prediction step; and calculating the coarse particle ratio of the aggregate to be evaluated from the corrected predicted value.

7. The quality prediction program according to claim 6, wherein in the correction of the predicted value in the second prediction step, the corrected predicted value is calculated for each of the sections so that a sum of the corrected predicted values ​​for each of the sections is 100%.

8. The second prediction step includes: correcting the predicted value acquired in the first prediction step; and calculating the coarse particle ratio of the aggregate to be evaluated from the corrected predicted value. In the correction of the predicted value in the second prediction step, for each of the sections, When the predicted value output from the prediction model is smaller than 0, a correction is performed so that the corrected predicted value becomes 0; 7. The quality prediction program according to claim 6, wherein, when the predicted value output from the prediction model is greater than 0, the predicted value is corrected to be the corrected predicted value as is.

9. an acquisition step of acquiring evaluation input information based on captured image data obtained by capturing an image of the aggregate to be evaluated; a first prediction step of inputting the evaluation input information into a prediction model constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for learning and on correct values ​​of mass fractions of a plurality of granularity classifications of the aggregate associated with the input information, so as to output a value indicating a mass fraction for each classification included in the plurality of granularity classifications in response to input of the input information, and acquiring a predicted value output from the prediction model for each classification; and a result display step of displaying an inference graph on a monitor, the first axis being the nominal size of the sieve and the second axis being the mass fraction of the grains passing through each sieve; The inference graph displayed on the monitor in the result display step includes: An inference result calculated based on the predicted value for each of the sections; A quality prediction method including, on the second axis, a standard range representing a standard range according to the type of aggregate to be evaluated.

10. an evaluation information acquisition unit that acquires evaluation input information based on captured image data obtained by capturing an image of the aggregate to be evaluated; a prediction calculation unit that inputs the evaluation input information to a prediction model that is constructed by machine learning based on input information based on learning image data obtained by imaging an aggregate for learning and on ground-truth values ​​of mass fractions of a plurality of granularity classifications of the aggregate associated with the input information, so as to output a value indicating a mass fraction for each classification included in the plurality of granularity classifications in response to input of the input information, and obtains a predicted value output from the prediction model for each classification; A display output unit that displays an inference graph on a monitor, the inference graph having a first axis representing the nominal size of the sieve and a second axis representing the mass fraction of the grains passing through each sieve; The inference graph displayed on the monitor by the display output unit includes: An inference result calculated based on the predicted value for each of the sections; A quality prediction device including, with respect to the second axis, a standard range that represents a standard range according to the type of aggregate to be evaluated.

Images