JP7394255B1 - Aggregate quality prediction method and program - Google Patents

Abstract

[Problem] To stabilize the quality of fresh concrete. [Solution] The aggregate quality prediction method is based on machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate that is associated with the input information. The input information includes a preparatory step of constructing a prediction model that outputs a value indicating the coarse grain ratio of the aggregate according to the input information, and image data obtained by imaging the aggregate to be evaluated. and a prediction step of inputting the evaluation information into the prediction model to obtain a first predicted value output from the prediction model regarding the coarse grain ratio of the aggregate. include. [Selection diagram] Figure 3

Description

The present disclosure relates to a method for predicting the quality of aggregate, a method for producing ready-mixed concrete, and a system for producing ready-mixed concrete.

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.

JP 2021-117625 Publication JP 2021-135199 Publication

The present disclosure provides a method for predicting aggregate quality useful for stabilizing the quality of ready-mixed concrete, a method for producing ready-mixed concrete, and a system for producing ready-mixed concrete.

[1] Machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate associated with the input information, a preparatory step of constructing a prediction model that outputs a value indicating the coarse grain ratio of the aggregate; and an acquisition step of acquiring evaluation information corresponding to the input information, including image data obtained by imaging the aggregate to be evaluated. and a prediction step of inputting the evaluation information into the prediction model to obtain a first predicted value output from the prediction model regarding the coarse grain ratio of the aggregate.
[2] The aggregate quality prediction according to [1] above, wherein in the preparation step, the prediction model is constructed so as to output only a value indicating the coarse grain ratio of the aggregate according to the input information. Method.
[3] The preparation step includes constructing the prediction model by machine learning further based on the correct value of the mass ratio of each of the plurality of particle size classifications of aggregate, which is associated with the input information, and the preparation step includes: In the step, the prediction model is constructed to further output a value indicating a mass ratio for each classification included in the plurality of particle size classifications according to the input information; further obtaining a second predicted value output from the predictive model regarding the mass ratio of each category by inputting the second predicted value into a prediction model, and converting the second predicted value into a coarse particle percentage of aggregate; The aggregate quality prediction method according to [1] above, comprising:
[4] The preparation step updates the intermediate model generated in the process of constructing the prediction model based on a first error regarding the coarse grain ratio of the aggregate and a second error regarding the mass ratio for each division. [3], wherein in the preparation step, the intermediate model is updated using an error corrected so that the difference between the first error and the second error is reduced. Aggregate quality prediction method.
[5] The aggregate quality prediction method according to any one of [1] to [4] above, wherein the preparation step includes constructing the prediction model by machine learning using a convolutional neural network.
[6] In the acquisition step, in a state where the aggregates to be evaluated are piled up, an area including the aggregates to be evaluated is imaged from a position away from the aggregates to be evaluated, and the evaluation information is acquired. The method for predicting the quality of aggregate according to any one of [1] to [5] above, including obtaining the aggregate quality.
[7] The preparation step includes preparing the learning aggregate so as to include grains of two or more particle size classifications, and the input information is an image obtained by imaging the learning aggregate. The aggregate quality prediction method according to any one of [1] to [6] above, including data.
[8] A manufacturing process of mixing materials containing aggregate to produce ready-mixed concrete, and a quality prediction process of predicting the coarse particle ratio of at least a portion of the aggregate used in the manufacturing process, The quality prediction step is performed by machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate that is associated with the input information. A preparatory step of constructing a prediction model that outputs a value indicating the coarse grain ratio of the aggregate according to the above, and image data obtained by imaging the aggregate to be evaluated, and obtaining evaluation information corresponding to the input information. and a prediction step of inputting the evaluation information into the prediction model to obtain a first predicted value output from the prediction model regarding the coarse grain ratio of aggregate. .
[9] A manufacturing device for producing ready-mixed concrete by kneading materials including aggregate, and a quality prediction device for predicting the coarse grain ratio of at least a portion of the aggregate used by the manufacturing device, The prediction device uses machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate that is associated with the input information to predict the input information. A preparatory step of constructing a prediction model that outputs a value indicating the coarse grain ratio of the aggregate according to the process, and image data obtained by imaging the aggregate to be evaluated, and obtaining evaluation information corresponding to the input information. and a prediction step of inputting the evaluation information into the prediction model to obtain a first predicted value output from the prediction model regarding the coarse grain percentage of the aggregate. , ready-mixed concrete manufacturing system.

According to the present disclosure, a method for predicting aggregate quality useful for stabilizing the quality of ready-mixed concrete, a method for producing ready-mixed concrete, and a system for producing ready-mixed concrete are provided.

FIG. 1 is a schematic diagram showing an example of a ready-mixed concrete manufacturing system. FIG. 2 is a schematic diagram showing an example of the hardware configuration of the control device. FIG. 3(a) is a flowchart illustrating an example of processing executed in the learning phase. FIG. 3(b) is a flowchart illustrating an example of processing executed in the evaluation phase. FIG. 4(a) is a diagram schematically showing an example of a prediction calculation using a prediction model. FIG. 4(b) is a cross-sectional view schematically showing an example of how image data for learning is acquired. FIG. 5 is a diagram showing an example of image data for learning. FIGS. 6A and 6B are graphs showing an example of a comparison result between a predicted value and a correct value. FIGS. 7A and 7B are graphs showing an example of a comparison result between a predicted value and a correct value. FIGS. 8A and 8B are graphs showing examples of comparison results between predicted values and correct values. FIGS. 9A and 9B are graphs showing examples of comparison results between predicted values and correct values. FIGS. 10A and 10B are graphs showing an example of a comparison result between a predicted value and a correct value. FIGS. 11(a) and 11(b) are graphs showing examples of comparison results between predicted values and correct values. FIG. 12 is a graph showing an example of a comparison result between a predicted value and a correct value. FIG. 13 is a graph showing an example of a comparison result between a predicted value and a correct value. FIG. 14 is a graph showing an example of a comparison result between a predicted value and a correct value.

Hereinafter, one embodiment will be described with reference to the drawings. In the description, the same elements or elements having the same function are given the same reference numerals, and redundant description will be omitted.

FIG. 1 schematically shows a ready-mixed concrete manufacturing system according to an embodiment. A manufacturing system 1 shown in FIG. 1 is a system for manufacturing ready-mixed concrete. The manufacturing system 1 mixes concrete materials and manufactures (generates) ready-mixed concrete. The manufacturing system 1 has a function of predicting the quality of concrete material in addition to a function of manufacturing ready-mixed concrete.

Concrete materials used by the manufacturing system 1 include cement, admixtures, coarse aggregates, fine aggregates, water, admixtures, and the like. Coarse aggregates include, for example, gravel such as mountain gravel, land gravel, river gravel, and sea gravel; crushed stone; blast furnace slag aggregate, ferronickel slag aggregate, electric furnace oxidation slag aggregate, coal gasification slag aggregate, etc. slag coarse aggregate; lightweight coarse aggregates such as natural lightweight aggregates, by-product lightweight aggregates, and artificial lightweight aggregates; recycled coarse aggregates; recovered aggregates, and coarse aggregates that are a mixture of these. The coarse aggregate may include, for example, crushed rock or crushed limestone. Fine aggregates include, for example, sand such as mountain sand, land sand, river sand, and sea sand; crushed sand; blast furnace slag aggregate, ferronickel slag aggregate, copper slag aggregate, electric furnace oxidation slag aggregate, and coal gasified Examples include slag fine aggregate such as slag aggregate; lightweight fine aggregate such as natural lightweight aggregate, by-product lightweight aggregate, and artificial lightweight aggregate; recycled fine aggregate; recovered aggregate or fine aggregate mixed with these. It will be done. In addition, rock types for crushed stone and crushed sand include igneous rocks such as granite, diorite, gabbro, porphyrite, diabase, rhyolite, andesite, basalt, and serpentinite; conglomerate, sandstone, shale, and slate. , sedimentary rocks such as tuff; metamorphic rocks such as gneiss and crystalline schist; and other rocks such as silica, limestone, domarite, and peridotite. In the present disclosure, coarse aggregate and fine aggregate may be collectively referred to as "aggregate." In this case, "aggregate" means coarse aggregate, fine aggregate, or both coarse and fine aggregate.

The manufacturing system 1 loads the manufactured ready-mixed concrete onto a transport vehicle C. After the ready-mixed concrete is loaded, the transport vehicle C transports the ready-mixed concrete to the site where the ready-mixed concrete is used (for example, a construction site). Examples of the transport vehicle C include an agitator vehicle (mixer vehicle) and a dump truck. The manufacturing system 1 may manufacture ready-mixed concrete from concrete materials so as to satisfy target quality (required quality) set for each site. In one example, the target quality of fresh concrete includes a target value of slump or slump flow. The manufacturing system 1 includes, for example, a material storage area 2, a transport device 8, a manufacturing device 10, an imaging device 50, and a control device 60.

The material storage area 2 is a place where concrete materials are stored. The material storage area 2 includes a plurality of silos 4. The plurality of silos 4 are containers that store at least a portion of concrete materials for each type of material. The plurality of silos 4 include a silo 4 that stores coarse aggregate, a silo 4 that stores fine aggregate, and a silo 4 that contains cement.

The transport device 8 is a device that transports concrete materials stored in a plurality of silos 4 to the manufacturing device 10. Conveying device 8 includes, for example, a belt conveyor that conveys concrete material. The width of the belt conveyor may be about 30 cm to 160 cm. The transport device 8 may transport concrete materials for each type of material at different timings. In one example, based on operation instructions from the control device 60, a specific material among various concrete materials is transferred to the transportation device 8 and transported to the manufacturing device 10.

The manufacturing device 10 is a device that mixes concrete materials including aggregate to manufacture ready-mixed concrete. The manufacturing apparatus 10 operates based on operation instructions from the control device 60. The manufacturing device 10 includes, for example, a storage bottle 12, a measuring bottle 14, a collection hopper 16, a mixer 20, and a loading hopper 30.

The storage jar 12 temporarily stores various concrete materials. Various concrete materials are transported (conveyed) to the storage jar 12 from the material storage area 2 by a transport device 8 . The storage jars 12 are configured to individually store various concrete materials. Hereinafter, "concrete material" may be simply referred to as "material." Various materials stored in the storage bottle 12 are supplied to the measuring bottle 14 as needed.

The measuring bottle 14 is arranged below the storage bottle 12. The measuring bottle 14 operates based on operation instructions from the control device 60 and measures various materials individually. When the measuring bottle 14 detects the target amount of material instructed by the control device 60, it supplies the material to the collection hopper 16. An admixture may be mixed with the water as it is supplied to the metering bottle 14. The collection hopper 16 is arranged below the weighing bottle 14. The collecting hopper 16 collects various materials discharged from the measuring bottle 14 and supplies the collected various materials to the mixer 20. If the collecting hopper 16 is not provided, various materials are supplied to the mixer 20 from the measuring bottle 14.

The mixer 20 is arranged below the collection hopper 16. The mixer 20 is a device that mixes concrete materials. The mixer 20 produces ready-mixed concrete by mixing (kneading) aggregate, cement, water, admixtures, and the like. From the bottom of the mixer 20, fresh concrete is discharged into a loading hopper 30. The loading hopper 30 is arranged below the mixer 20 and temporarily stores fresh concrete. The loading hopper 30 supplies the temporarily stored ready-mixed concrete to the transport vehicle C.

The manufacturing apparatus 10 described above is an example of a ready-mixed concrete manufacturing apparatus, and the ready-mixed concrete manufacturing apparatus may be configured in any manner as long as it can mix concrete materials and manufacture ready-mixed concrete. good. For example, the manufacturing apparatus 10 may not include the collection hopper 16, and various materials measured in the measuring bottle 14 may be supplied from the measuring bottle 14 to the mixer 20.

The imaging device 50 is a device (camera) that can take an image of aggregate, which is one of concrete materials. The imaging device 50 generates image data by imaging the aggregate. The imaging device 50 is, for example, a digital camera or a video camera that images an imaging range based on visible light. The image data generated by the imaging device 50 is used in predicting the quality of the aggregate. Prediction of aggregate quality will be discussed later. Note that the imaging device 50 may generate a moving image of the imaging target, and still images included in the moving image may be acquired as image data.

The imaging device 50 is arranged so as to image a range including the aggregate to be imaged from a position away from the aggregate to be imaged. The imaging device 50 may image the range including the aggregate to be imaged from above. The imaging device 50 may image an area including aggregates in a stacked state. Stacked aggregate refers to aggregate in which some of the grains contained in the aggregate overlap with other grains when viewed from vertically above. The imaging device 50 may be arranged so as to be able to image the aggregate being transported by the transporting device 8. The imaging device 50 may image the aggregate being transported while the belt conveyor included in the transport device 8 is stopped or in motion. Note that the imaging device 50 may image the aggregate loaded on a transport truck or a transport ship when receiving the aggregate. The imaging device 50 may be placed in an imaging room provided with a floor or a rubber sheet, and may image aggregates placed on the floor or the like.

The control device 60 is a device that controls the manufacturing device 10. The control device 60 is configured by one or more computers. When the control device 60 is composed of a plurality of computers, these computers are connected to each other so as to be able to communicate with each other. The control device 60 controls the manufacturing device 10 according to set operating conditions. At least a portion of the operating conditions may be determined by instructions from an operator such as a worker.

An input/output device 62 may be connected to the control device 60 (see FIG. 2). The input/output device 62 is a device for inputting information indicating instructions from a worker or the like into the control device 60 and outputting information from the control device 60 to the worker or the like. The input/output device 62 may include a keyboard, an operation panel, or a mouse as an input device, and may include a monitor (for example, a liquid crystal display) as an output device. The input/output device 62 may be a touch panel in which an input device and an output device are integrated. The control device 60 and the input/output device 62 may be integrated.

It is known that the quality of fresh concrete produced by the production apparatus 10 is affected by the quality of the concrete material. Specifically, the particle size of the aggregate used to produce ready-mixed concrete affects the fluidity (eg, slump and slump flow) of the ready-mixed concrete. For example, for the same unit amount of water, the larger the particle size of the aggregate, the smaller the surface area of the aggregate, the smaller the amount of water retained, and the greater the fluidity of the ready-mixed concrete. Conversely, the smaller the particle size of the aggregate, the greater the water retention capacity due to the increased surface area of the aggregate, which may reduce the fluidity of fresh concrete. In this way, in order to stabilize the quality of fresh concrete, including its fluidity, it is necessary to control the particle size of aggregate. The particle size of aggregate can be expressed as a coarse grain ratio, and in order to stabilize the quality of fresh concrete, it is necessary to control whether the coarse grain ratio of aggregate is within a desired range, for example. .

In addition to controlling the manufacturing device 10, the control device 60 may be configured to predict the quality of aggregate used for manufacturing ready-mixed concrete by the manufacturing device 10. In this case, the control device 60 constitutes a quality prediction device that predicts the quality of the aggregate. The control device 60, which functions as a quality prediction device, predicts the coarse grain ratio of at least a portion of the aggregate used by the manufacturing device 10.

The control device 60 uses machine learning based on at least input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate that is associated with the input information. The apparatus is configured to execute a preparation step for constructing a prediction model that outputs a value indicating the coarse grain ratio of aggregate in response to an input. The control device 60 also includes an acquisition step of acquiring evaluation information that includes image data obtained by imaging the aggregate to be evaluated, and that corresponds to the input information, and inputting the evaluation information into a prediction model to determine whether the aggregate a prediction step of obtaining a value output from the prediction model with respect to the coarse grain ratio of the prediction model.

The control device 60 has a functional configuration (hereinafter referred to as a "functional module"), for example, an operation control section 68, a model construction section 72, a model holding section 74, a quality prediction section 76, and an output section 78. and has. The processing executed by these functional modules corresponds to the processing executed by the control device 60. The operation control unit 68 is a functional module that controls the manufacturing apparatus 10 to manufacture ready-mixed concrete according to the above operating conditions. A worker or the like may adjust (change) the operating conditions depending on the target quality of ready-mixed concrete.

The model construction unit 72 is a functional module that constructs a model (hereinafter referred to as "prediction model M") for predicting the coarse grain ratio of aggregate. The prediction model M is a model that outputs a value indicating at least the coarse grain ratio of the aggregate in response to input information including image data obtained by imaging the aggregate. The model construction unit 72 constructs the prediction model M by machine learning based on the input information and the correct value of the coarse grain ratio of the aggregate that is associated with the input information.

Machine learning refers to a method in which a machine (computer) autonomously discovers laws or rules by repeatedly learning based on provided information. Prediction model M can be constructed using algorithms and data structures. The prediction model M is realized using, for example, a neural network that is an information processing model that imitates the mechanism of human brain nerves. The specific algorithm of machine learning performed when constructing the predictive model M is not particularly limited. A 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.

The model construction unit 72 may construct the prediction model M by machine learning using a convolutional neural network (CNN). The model construction unit 72 may construct the prediction model M by machine learning based on the correct value of the coarse particle ratio of the aggregate and the correct value of the mass ratio of each of the plurality of particle size classifications of the aggregate. The prediction model M constructed by the model construction unit 72 outputs a value indicating the coarse particle ratio of the aggregate and a value indicating the mass ratio of each category included in the plurality of particle size categories, according to the above input information. It's okay. That is, the prediction model M may output one value indicating the coarse particle ratio of the aggregate and a plurality of values each indicating the mass ratio for a plurality of particle size classifications.

The model construction unit 72 predicts the coarse grain rate by performing machine learning, for example, using data given as input for machine learning and correct data (correct value of coarse grain rate, etc.) output from machine learning. You may autonomously construct a predictive model M for The machine learning inputs are various datasets of input information including images of aggregates. In various data sets of input information, at least image data included in the input information is different from each other. The output of machine learning is data (numerical values) indicating the coarse particle ratio of aggregate and the mass ratio for each particle size classification. The model construction unit 72 uses a dataset of input information and multiple combinations of correct values such as coarse grain ratio to iterate a model that outputs predicted values of aggregate coarse grain ratio and mass ratio for each particle size category. Learn in detail.

The stage of autonomously constructing the predictive model M corresponds to the learning phase. The learning phase may be performed before the production phase in which ready-mixed concrete is manufactured, or may be performed at an early stage of the production phase. The predictive model M, which is a trained model, may be portable between computers. Therefore, the prediction model M constructed in the control device 60 may be used in a manufacturing system different from the manufacturing system 1.

The model holding unit 74 is a functional module that holds (stores) the prediction model M. The quality prediction unit 76 is a functional module that predicts the coarse grain ratio of aggregate. Hereinafter, the aggregate whose coarse grain ratio is to be predicted will be referred to as the "aggregate to be evaluated." The quality prediction unit 76 acquires information (hereinafter referred to as "evaluation information") that includes image data obtained by imaging the aggregate to be evaluated and corresponds to the above input information. While the above input information was associated with the correct value of the coarse grain ratio, the evaluation information is information in which the coarse grain ratio of the aggregate is unknown. The stage in which the quality prediction unit 76 predicts the coarse grain ratio of the aggregate using the prediction model M corresponds to the evaluation phase.

The quality prediction unit 76 inputs the evaluation information into the prediction model M, and obtains a predicted value (first predicted value) output from the prediction model M regarding the coarse grain ratio of the aggregate. In addition, the quality prediction unit 76 inputs the evaluation information to the prediction model M, and further obtains a predicted value (second predicted value) output from the prediction model M regarding the mass ratio for each particle size classification. That is, the quality prediction unit 76 acquires from the prediction model M a plurality of predicted values (a plurality of second predicted values) each indicating a mass ratio value for a plurality of particle size classifications. The quality prediction unit 76 may convert the predicted value (a plurality of predicted values) regarding the mass ratio for each particle size classification into a coarse particle ratio. By predicting the coarse grain ratio using the prediction model M using two different methods, it is possible to evaluate the reliability of the predicted value of the coarse grain ratio.

The output unit 78 is a functional module that outputs at least a predicted value regarding the coarse grain ratio of the aggregate output from the prediction model M as a prediction result of the coarse grain ratio. The output unit 78 may further output a value obtained by converting the predicted value regarding the mass ratio for each particle size classification, which is output from the prediction model M, into a coarse grain ratio as a prediction result of the coarse grain ratio. The output unit 78 may output the prediction result of the coarse grain rate to an output device (for example, a monitor) of the input/output device 62 to notify a worker or the like of the prediction result.

As shown in FIG. 2, the control device 60 includes a circuit 91. Circuit 91 includes one or more processors 92, memory 93, storage 94, input/output ports 95, and timer 96. Storage 94 includes a computer-readable storage medium, such as a non-volatile semiconductor memory. The storage 94 includes a program for causing the control device 60 to control the manufacturing apparatus 10 according to a preset control procedure, and a program for controlling the coarse particle ratio of aggregate using a preset prediction procedure. It stores a program to be executed by the device 60. For example, the storage 94 stores programs for configuring each of the functional modules described above.

The memory 93 temporarily stores programs loaded from the storage medium of the storage 94 and the results of calculations by the processor 92 . The processor 92 configures each functional module of the control device 60 by cooperating with the memory 93 and executing the above programs. The input/output port 95 inputs/outputs electrical signals to/from the manufacturing apparatus 10, the imaging device 50, the input/output device 62, etc. according to instructions from the processor 92. The timer 96 measures elapsed time, for example, by counting reference pulses of a constant period. Note that the circuit 91 is not necessarily limited to having each function configured by a program. For example, at least part of the functions of the circuit 91 may be configured by a dedicated logic circuit or an ASIC (Application Specific Integrated Circuit) that integrates the logic circuit.

[Method for manufacturing ready-mixed concrete]
Next, an example of a method for manufacturing ready-mixed concrete executed in the manufacturing system 1 will be described. The method for producing ready-mixed concrete includes a step of producing ready-mixed concrete and a step of predicting the quality of aggregate. The step of predicting the quality of the aggregate may be performed during a period that overlaps at least part of the period during which the step of producing ready-mixed concrete is performed.

The process of manufacturing ready-mixed concrete (hereinafter referred to as "manufacturing process") includes, for example, a conveying process, a measuring process, a charging process, a mixing process, a discharging process, and a loading process. In the conveyance process, various concrete materials are conveyed to the storage bottles 12 by the conveyance device 8, and the various materials are individually supplied to the storage bottles 12. In the measuring process, various materials are individually supplied from the storage bottle 12 to the measuring bottle 14, and the various materials are measured in the measuring bottle 14. In the measuring process, when the measured amount of each material reaches a predetermined set amount, that material is discharged to the collection hopper 16.

In the charging step, after all types of materials are collected in the collecting hopper 16, the materials in the collecting hopper 16 are charged (supplied) to the mixer 20. In the mixing step, a plurality of types of concrete materials are mixed in the mixer 20. In the discharge step, after the mixing of concrete materials in the mixer 20 is completed, fresh concrete is discharged from the mixer 20 to the loading hopper 30. In the loading process, the ready-mixed concrete discharged into the loading hopper 30 is loaded onto the transport vehicle C.

The process of predicting the quality of aggregate (hereinafter referred to as "quality prediction process") includes a model construction process in a learning phase and a quality evaluation process in an evaluation phase. In the quality prediction process (quality prediction method), the model construction process is performed before the quality evaluation process. An example of a model construction process and an example of a quality evaluation process will be described below. Note that an example will be described in which the coarse grain ratio of the coarse aggregate is predicted as the quality of the aggregate, and the input information includes only image data of the aggregate. Further, a description will be given using a specific example in which the prediction model M outputs a predicted value of the coarse particle ratio and a predicted value of the mass ratio for each particle size classification.

(Model construction process)
FIG. 3(a) is a flowchart showing an example of a series of processes executed in the model building process. This model construction step may be performed before the manufacturing process is performed by the manufacturing apparatus 10, or at an early stage when the manufacturing process is started. In this model construction step, a part of the coarse aggregate used in the production of ready-mixed concrete is extracted in the manufacturing apparatus 10, and a part of the coarse aggregate may be used as "coarse aggregate for learning". .

In the model construction process, step S11 is first executed. In step S11, learning data for performing machine learning is prepared, for example, by a worker or the like. The data for learning includes various image data (input information) obtained by imaging the coarse aggregate for learning, and correct answer data associated with each of the various image data. FIG. 4(a) schematically shows a calculation method for predicting the coarse grain ratio using a prediction model M from an image Im that is input information.

In this disclosure, as shown in FIG. 4(a), the prediction model M outputs two types of information, the mass ratio for each particle size classification and the coarse particle ratio, which is referred to as "multitasking." Moreover, the prediction model M outputting information on either the mass ratio for each particle size classification or the coarse particle ratio is referred to as a "single task." FIG. 4A shows an example of a calculation method in which the coarse grain rate is predicted using two types of methods using a prediction model M that executes multitasking.

In preparing images for learning, for example, a worker or the like prepares various coarse aggregates whose mass ratios and coarse grain ratios for each of a plurality of particle size classifications are known. The various coarse aggregates do not mean that the types of coarse aggregates (for example, crushed stone) are different, but rather that they are coarse aggregates that have different particle sizes. Then, as shown in FIG. 4(b), for example, the various coarse aggregates for learning are individually imaged by the camera 52 while being stacked on the measurement sheet 8a. As a result, image data for learning can be obtained for each type of coarse aggregate. The camera 52 may be a digital camera, and is provided to take an image of a range including the coarse aggregate from a predetermined position (for example, a predetermined position above) away from the coarse aggregate on the measurement sheet 8a. .

The sheet 8a may be a sheet formed to imitate the belt conveyor of the conveyance device 8, and is, for example, a rubber sheet. The width of the sheet 8a may be approximately half the width of the belt conveyor of the conveying device 8. The amount of aggregate placed on the sheet 8a is not particularly limited, but may be about 1 kg to 10 kg. FIG. 4(b) schematically illustrates how the aggregate is accumulated on the sheet 8a, and shows the accumulation state when the amount of aggregate is 2 kg, 5 kg, and 8 kg.

In one example, various coarse aggregates are prepared under the conditions shown in Table 1 below, and images obtained by imaging the coarse aggregates are acquired. The mass ratio for each particle size classification and coarse particle ratio in various coarse aggregates prepared under the conditions shown in Table 1 are known, and the known values are correct data in machine learning.

The coarse grain ratio (F.M.) is determined, for example, by the following formula (1) specified in the Concrete Standard Specifications. In addition, nine sieves excluding 80 mm in formula (1) are used in the measurement specified in "Standard Specifications for Construction Works and Commentary JASS5 Reinforced Concrete Works." The coarse grain rate predicted in the present disclosure may be defined by any method.

Table 1 means that coarse aggregates with 10 levels of grain size (coarse grain ratio) are prepared. Category I represents the particle size category for particles smaller than 2.5 mm. In division I, grains that pass through a 2.5 mm sieve are separated. Classification II represents a particle size classification of grains larger than 2.5 mm and smaller than 5 mm. In category II, grains that remain on the 2.5 mm sieve but pass through the 5 mm sieve are separated. Category III represents a particle size category of particles of 5 mm or more and smaller than 10 mm. In category III, grains that remain on the 5 mm sieve but pass through the 10 mm sieve are separated. Classification IV represents a particle size classification of grains of 10 mm or more and smaller than 20 mm. Category IV separates grains that remain on the 10 mm sieve but pass through the 20 mm sieve.

Regarding level 1, the coarse particle ratio is determined from the mass ratio for each particle size classification (mass ratio of each of classifications I to IV) as follows. For level 1 coarse aggregate, the percentage of the mass of the sample that stays on the 80 mm, 40 mm, and 20 mm sieves is 0, the percentage of the mass of the sample that stays on the 10 mm sieve is 85, and the percentage of the sample that stays on the 5 mm sieve is 0, respectively. The mass percentage of the sample is 100 (=15+85). Also, since all the grains stay on the 5 mm sieve, the percentage of the mass of the sample that stays on the 2.5 mm, 1.2 mm, 600 μm, 300 μm, and 150 μm sieves is 100 each. Therefore, the coarse grain ratio is calculated as shown in equation (2) below.

For levels 2 to 10, the coarse grain ratio is obtained in the same manner as the above calculation for level 1. The conditions for the learning aggregate shown in Table 1 are just examples, and the number of levels, the coarse grain ratio at each level, the mass ratio for each particle size category, and the method of setting the particle size categories are determined by the operator. It may be arbitrarily determined by, etc. Particle sizes in multiple categories, including the smallest category (<150 μm), may be combined into one particle size category, such as the multiple particle size categories shown in Table 1. Particle sizes in a certain category (for example, 20 to 40 mm) or more may not be included in a plurality of particle size categories. Furthermore, although this is different from the setting method in Table 1, particle sizes below a certain category (for example, 1.2 to 2.5 μm) may not be included in a plurality of particle size categories.

FIG. 5 shows images T1 to T10 obtained by capturing learning aggregates prepared according to the conditions of levels 1 to 10 in Table 1. Level n (n is an integer from 1 to 10) corresponds to image Tn. For example, the image T1 is obtained by imaging the learning aggregate prepared according to standard 1. The learning aggregate according to each level n contains grains of a plurality of grain size classifications. In this way, in step S11, the learning aggregate is prepared so as to include grains of two or more particle size classifications, and input information including image data obtained by imaging the learning aggregate is prepared. It's okay.

For each level n, a plurality of images Tn in which the state (more specifically, the deposition state) of the learning aggregate differs from each other may be prepared. In one example, one image Tn is acquired by performing the first imaging with the learning aggregate placed on the measurement sheet 8a. Then, the learning aggregate is temporarily removed from the top of the sheet 8a, and the same learning aggregate is placed on the sheet 8a again so that the stacked state of many grains is different from the first imaging. will be posted. Then, the second imaging is performed. A plurality of images Tn may be prepared for each level n by repeating the above-described imaging by a worker or the like.

A worker or the like may input the prepared image data and correct answer data to the control device 60 via the input/output device 62. One data set is composed of one image Tn and the correct data of the mass ratio and coarse grain ratio at level n associated with the image Tn. A worker or the like prepares multiple data sets as learning data.

Returning to FIG. 3(a), after step S11 is executed, step S12 is executed. In step S12, for example, the model construction unit 72 of the control device 60 creates the predictive model M by performing machine learning using the image data prepared in step S11 and the correct answer data linked to the image data. To construct. The prediction model M may be constructed to output four mass ratio values for the four particle size categories and one value for the coarse particle fraction (see also FIG. 4(a)). By converting the four mass ratio values output from the prediction model M as shown in equation (2) above, a predicted value of the coarse grain ratio can be obtained.

In this way, the prediction model M may be a model that performs multitasking. When the prediction model M is a model that performs multitasking, two coarse-grain ratio prediction values are obtained using different methods. In this case, for example, the reliability of the predicted value of the coarse grain ratio can be evaluated. In one example, when the difference between two predicted values regarding the coarse grain ratio is small, it can be evaluated that the reliability of the prediction result is high.

Hereinafter, an example of the prediction model M will be explained using simplified formulas for easy understanding. The prediction model M constructed by the model construction unit 72 can be expressed simply as, for example, the following equations (3) and (4).

In equation (4), Y represents any one of y1 to y5, and the prediction model M calculates equation (4) for each of y1 to y5. y1 to y5 are output data representing the mass ratio and coarse particle ratio for each particle size classification. When y1, y2, y3, and y4 are the output values of the mass ratio of each of the four particle size categories, y5 is the output value of the coarse particle ratio. N is an integer of 2 or more and represents the number of input data. The input data includes at least the value of each pixel in the aggregate image. wi is a weight (coefficient), and b is a bias term (coefficient). f(U) represents an activation function. The activation function is a linear function (identity function), polynomial, absolute value, step function, sigmoid function, hardsigmoid function, logsigmoid function, softmax function, logsoftmax function, softmin function, softplus function, softsign function, tanh function, tanhShrink These are nonlinear functions such as hardtanh function, tanhexp function, ReLU function, ReLU6 function, Leaky-ReLU function, PReLU function, ELU function, SELU function, CELU function, Swith function, Mish function, ACON function, etc.

The model construction unit 72 constructs a prediction model M that outputs the values of y1 to y5 by determining the weight wi and bias term b in equation (4) for each of y1 to y5 using the learning data. . The model construction unit 72 determines the weight wi for each of y1 to y5 by repeating the calculations shown in the following equations (5) and (6).

In equation (5), yj (j is an integer from 1 to 5) is the output of the model specified by the weight wi during updating, and tj is the correct value of the mass ratio or coarse grain ratio. be. E and Ej are errors representing the difference between the output yj and the correct value tj. In equation (6), w' represents the updated weight w. Repeating the calculations of equations (5) and (6) means calculating the sum of squared errors between the value of output yj and the correct value tj when a certain value is input to the weight w, and using the gradient method, This means updating the value of the weight w so that the error is minimized. The number of times the weight w is updated may be determined by a worker or the like. Examples of gradient methods include steepest descent, stochastic gradient descent (SGD), mini-batch gradient descent, Adam, Nesterov acceleration (NAG), AdaGrad, RMSprop, Adadelta, Adam, Adamax, AMSGRAD, and Momentum. be able to.

The explanation of the prediction model M using equations (3) to (6) is a general explanation regarding deep neural networks. Examples of deep neural networks include convolutional neural networks, recurrent neural networks, and recurrent neural networks. In the aggregate quality prediction method according to the present embodiment, it is preferable to use, for example, a convolutional neural network. This is because when input information including image data is used, convolutional neural networks have an advantage over normal deep neural networks when updating weights. With normal deep neural networks, when a large amount of information is involved, such as image data, there are a large number of places where weights are calculated. On the other hand, with a convolutional neural network, weights can be shared using filters, which simplifies updating of weights. In this way, it is preferable to use a convolutional neural network in that it is possible to more appropriately capture the features of an image.

After step S12 is executed, step S13 is executed. In step S13, for example, the model holding unit 74 stores the prediction model M constructed in step S12. With the above steps, the model construction process is completed.

(Quality evaluation process)
FIG. 3(b) is a flowchart showing an example of a series of processes executed in the quality evaluation process. This quality evaluation process is performed, for example, during a period when the manufacturing process described above in the manufacturing apparatus 10 is being executed.

In this quality evaluation process, first, the control device 60 executes step S21. In step S21, for example, the quality prediction unit 76 waits until the measurement timing comes. The measurement timing may be predetermined at a certain time of the day, or may be the timing at which an instruction is received from an operator such as a worker. When the measurement timing comes, the control device 60 may perform control so that the coarse aggregate is loaded from the material storage area 2 onto the conveying device 8. Note that the measurement timing may be set to match the timing when the coarse aggregate is being transported to the transporting device 8.

Next, the control device 60 executes step S22. In step S22, for example, the quality prediction unit 76 causes the imaging device 50 to image the aggregate to be evaluated, and acquires image data from the imaging device 50. Thereby, the quality prediction unit 76 acquires evaluation information including image data obtained by imaging the aggregate to be evaluated. In one example, the evaluation information includes only image data obtained by imaging the aggregate to be evaluated. In step S22, for example, while the aggregates to be evaluated are piled up on the belt conveyor of the conveyor 8, the imaging device 50 captures the aggregates to be evaluated from an upper position away from the aggregates to be evaluated. Image the range.

Next, the control device 60 executes step S23. In step S23, for example, the quality prediction unit 76 determines the coarse grain percentage of the coarse aggregate to be evaluated based on the image data (evaluation information) acquired in step S22 and the prediction model M held by the model holding unit 74. Predict. In one example, the quality prediction unit 76 inputs the image data acquired in step S22 to the prediction model M, and acquires an output value related to the coarse grain ratio directly output from the prediction model M. Furthermore, the quality prediction unit 76 inputs the image data to the prediction model M, and obtains an output value related to the mass ratio for each particle size classification that is directly output from the prediction model M. Then, the quality prediction unit 76 indirectly obtains the predicted value of the coarse grain ratio by converting the output value regarding the mass ratio for each particle size classification into the coarse grain ratio.

Next, the control device 60 executes step S24. In step S24, for example, the output unit 78 outputs the two prediction results regarding the coarse grain rate obtained in step S23. In one example, the output unit 78 displays two prediction results regarding the coarse grain ratio on the monitor of the input/output device 62. This allows a worker or the like to confirm the prediction result regarding the coarse particle ratio. Workers who have confirmed the predicted results of the coarse grain ratio should take measures to maintain the quality of ready-mixed concrete (e.g., adjusting the aggregate particle size, modifying the mix, (temporary suspension of production, investigation of the cause, etc.)

With the above steps, a series of processes for predicting the coarse grain ratio are completed. After executing step S24, the control device 60 may repeat the series of processes from steps S21 to S24. In the above example, the coarse grain ratio is predicted at each measurement timing, but the conveying device 8 continuously images the aggregate being transported and performs a 100% inspection of the aggregate used for producing ready-mixed concrete. may be performed.

The above-mentioned quality prediction method can be used as a substitute for quality testing by sampling, in addition to inspecting all raw materials (materials) of ready-mixed concrete in the ready-mixed concrete production process (production line) to stabilize quality. It is. Examples of the quality test based on sampling include an acceptance inspection that is an inspection at the time of receiving raw materials, and a process inspection that is an inspection at each manufacturing process of ready-mixed concrete. Quality tests by sampling are routinely performed in order to supply ready-mixed concrete of stable quality. Conventionally, when measuring the coarse grain ratio by a quality test by sampling, for example, coarse aggregate is sieved and the coarse grain ratio is calculated from the mass ratio. If the aggregate quality prediction method according to this embodiment is applied to daily quality tests, the time for quality prediction tests can be significantly shortened, and for example, the frequency of quality prediction tests in one day can be increased. I can do it. This makes it possible to supply ready-mixed concrete of more stable quality. When performing a quality test by sampling using the aggregate quality prediction method according to the present embodiment, for example, the extracted aggregate is deposited on a rubber sheet, and image data is generated from thereon by the imaging device 50. be able to. Note that when the aggregate quality prediction method according to the present embodiment is used for acceptance inspection, it is also possible to install the imaging device 50 on a transport truck or a transport ship, and perform quality prediction during transport, for example.

[Verification of prediction method using prediction model]
Below, in order to verify the prediction method of the coarse grain rate using the prediction model M based on machine learning using a convolutional neural network, we will use a data set whose correct value is known and compare the prediction result by the prediction model M and the correct answer. We conducted a comparison verification with the value. As learning aggregates, 10 levels of aggregates with different coarse grain ratios were prepared according to the conditions in Table 1 above. For each level, the condition of the aggregate on the measurement sheet 8a was different from each other, and 10 images were obtained by taking images indoors 10 times.

In addition, a total of 100 images (10 levels x 10 imaging times) were prepared according to the standard conditions shown in Table 2 below. As a result, 100 data sets were prepared as learning data in which images and correct values were associated with each other in each data set.

The above prediction model M was constructed using 80 data sets out of 100 data sets, and 20 data sets were used for verification. Specifically, at each level, eight data sets were used for model construction and two data sets were used for validation. In the verification using 20 data sets, for each data set, two prediction results calculated using the image and the prediction model M, and the correct value such as the coarse grain rate associated with the image were calculated. compared. The prediction model M was constructed by a convolutional neural network having six intermediate layers each consisting of a convolutional layer and a pooling layer, and then four fully connected layers. In addition, the Huber function was used as the loss function. Note that in each dataset, the number of images was increased using the ImageAugmentation layer.

FIG. 6(a) is a graph showing the relationship between the predicted values of mass ratio (%) in Class III and Class IV in Table 1 and the corresponding correct values of mass ratio (%). In the graph shown in FIG. 6(a), the horizontal axis is the correct value of the mass ratio, and the vertical axis is the predicted value of the mass ratio output from the prediction model M. The predicted results for category III (5 mm to 10 mm) are plotted as △ marks, and the predicted results for category IV (10 mm to 20 mm) are plotted as ◯ marks. A solid line is drawn where the correct value and predicted value match, and the closer the plotted result is to that line, the higher the prediction accuracy. From the results shown in FIG. 6(a), the prediction results are plotted near the above line, and it can be seen that the prediction model M can accurately predict the mass ratio.

FIG. 6(b) is a graph showing the relationship between the predicted value of the coarse grain ratio output from the prediction model M and the correct value of the coarse grain ratio. In the graph shown in FIG. 6(b), the horizontal axis is the correct value of the coarse grain ratio, and the vertical axis is the predicted value of the coarse grain ratio output from the prediction model M. The prediction results are plotted with □ marks. The solid line is the line where the correct value and the predicted value match, and the dashed upper limit line is shifted vertically by +0.10 from that line, and the lower limit line is shifted vertically by -0.10 from that line. It is depicted. Prediction results included in the area between the upper limit line and the lower limit line are defined as being within the allowable range. With reference to the error in coarse grain ratio allowed in current ready-mixed concrete plants, a range of ±0.10 was set as the allowable range. All of the prediction results using the prediction model M are within the permissible range, and it can be seen that the prediction model M can accurately predict the coarse grain ratio.

Below, we examined the influence of various factors when preparing learning data on the predictions made by the prediction model M. Specifically, at least one factor among the presence or absence of lighting, image size, amount of coarse aggregate, and moisture condition on the aggregate surface was changed from the standard conditions shown in Table 2, and the learning data was Got ready. Then, for each condition, a prediction model M was constructed from the corresponding learning data, and the prediction accuracy of the prediction model M was evaluated using the corresponding verification data.

The presence or absence of illumination differs between the graphs shown in FIGS. 7(a) and 7(b). Figure 7(a) shows the comparison results when the image is taken using only the indoor light (without lighting), and when the image is taken with LED lighting applied to the object in addition to the indoor light (with lighting). ) are shown in FIG. 7(b). Furthermore, regardless of the presence or absence of lighting, the cases where the amount of coarse aggregate was 2 kg, 5 kg, and 8 kg were examined.

The average absolute difference value was calculated as an index showing how much the predicted value deviates from the correct value. The average absolute difference value is the arithmetic mean of the absolute values of the differences between the correct value and the predicted value in the 20 verification data sets. In the graphs shown in FIGS. 7A and 7B, "direct prediction" represents the verification result regarding the output value of the coarse grain ratio directly output from the prediction model M. "Indirect prediction" represents the verification result regarding the predicted value of the coarse particle ratio obtained indirectly by converting the output value of the mass ratio for each particle size classification output from the prediction model M.

The vertical axis represents the absolute difference average value, and "Tg" represents the target value. When the average absolute difference value is less than or equal to the target value Tg (0.10), it can be evaluated that the accuracy of the prediction using the prediction model M is good. As shown in Figures 7(a) and 7(b), regardless of whether the amount of coarse aggregate is 2 kg, 5 kg, or 8 kg, or whether it is directly predicted or indirectly predicted. , regardless of the presence or absence of illumination, the average absolute difference was less than or equal to the target value Tg. Therefore, it is considered that the presence or absence of illumination has little effect on the accuracy of prediction by prediction model M.

The prediction methods are different between the graphs shown in FIGS. 8(a) and 8(b). The comparison result when using direct prediction among the predictions by prediction model M is shown in FIG. 8(a), and the comparison result when using indirect prediction among the predictions by prediction model M is shown in FIG. Shown in (b). As shown in Fig. 8(a) and Fig. 8(b), no matter whether the amount of coarse aggregate is 2 kg, 5 kg, or 8 kg, and whether with or without lighting, it is possible to directly Regardless of which method, prediction or indirect prediction, was used, the average absolute difference value was less than or equal to the target value Tg. Therefore, it is considered that the selection results of direct prediction and indirect prediction have little influence on the accuracy of prediction by prediction model M.

In each of the graphs shown in FIGS. 9(a) and 9(b), verification was performed by changing the amount of coarse aggregate. The comparison results when using direct prediction and changing the amount of coarse aggregate are shown in Figure 9(a), and the comparison results when using indirect prediction and changing the amount of coarse aggregate. The results are shown in Figure 9(b). As shown in Fig. 9(a) and Fig. 9(b), regardless of the amount of coarse aggregate, whether it is direct prediction or indirect prediction, and whether there is illumination or not, The average absolute difference value was below the target value Tg. Therefore, it is considered that the amount of coarse aggregate has little effect on the accuracy of prediction by prediction model M.

In each of the graphs shown in FIGS. 10(a) and 10(b), verification is performed by changing the image size. The comparison results when using direct prediction and changing the image size are shown in Figure 10(a), and the comparison results when using indirect prediction and changing the image size are shown in Figure 10(b). ) is shown. In each graph, the horizontal axis represents the image size ratio, with the image size under the standard conditions in Table 2 being 1. The lower limit of the image size is set within a range where the aggregate can be visually recognized from the image. As shown in FIGS. 10(a) and 10(b), regardless of whether the prediction is direct or indirect, and whether the amount of aggregate is 2 kg, 5 kg, or 8 kg, the image Regardless of the size ratio, the average absolute difference was below the target value Tg. Therefore, it is considered that the image size (number of pixels or resolution) has a small influence on the accuracy of prediction by the prediction model M.

The graphs shown in FIGS. 11(a) and 11(b) each show the results of verification conducted with different moisture conditions on the aggregate surface. The "air-dry state" in this verification is a state in which the learning aggregates are left indoors for a predetermined period of time to dry (naturally dry). In addition, "wet state" is a state in which water is added to the learning aggregate in a predetermined ratio (for example, 1% to 5% by mass) to the mass of the aggregate, and the mixture is mixed in a container. It is. From the results shown in FIGS. 11(a) and 11(b), it can be seen that in an air-dried state, the average absolute difference value is equal to or less than the target value Tg, and the prediction model M can be accurately predicted. In addition, in the image of the aggregate obtained in the wet state, when there was no illumination, the overall color was uneven and the edges of each particle of the aggregate appeared blurred compared to the image in the air-dry state. In addition, images of aggregate obtained in a wet state included aggregate particles that appeared whiter due to reflection of surface water when illuminated than in an air-dry state.

[Modified example]
As shown in FIG. 11(a), when direct prediction is adopted and the amount of coarse aggregate is 5 kg and 8 kg, the average absolute difference value exceeds the target value Tg. As shown in FIG. 11(b), when indirect prediction is adopted and the amount of coarse aggregate is 8 kg, the average absolute difference value exceeds the target value Tg. Thus, it can be seen that in a wet state, the prediction accuracy may decrease depending on the amount of coarse aggregate. In order to avoid a decrease in prediction accuracy in wet conditions, the prediction model M may be constructed differently from the above example.

The predictive model M constructed by the model construction unit 72 may output only a value indicating the coarse grain ratio of the aggregate according to input information including image data obtained by imaging the learning aggregate. . That is, the prediction model M may be a model that performs a single task. In this case, in the learning data for constructing the prediction model M, only the correct value of the coarse grain ratio of the aggregate is associated with the input information including the image data.

The model construction unit 72 may construct a prediction model M shown in the following equation (7) instead of the above equation (4).

In the process of updating the weight w for generating the prediction model M, the model construction unit 72 may repeat the calculation shown in the following equation (8) instead of the above equation (5).

FIG. 12 shows the results of verification conducted under the same conditions as the verification results shown in FIG. 11(a), except that the prediction model M shown in equation (7) was used as the prediction model. ing. As shown in FIG. 12, a result is obtained in which the average absolute difference value is equal to or less than the target value Tg in both the air-dry state and the wet state. That is, it can be seen that even in a wet state, by using the prediction model M that performs a single task, the coarse particle ratio can be predicted with high accuracy.

When the prediction model M performs multitasking, the calculation method in the process of updating the weight w may be different from the example described above. One possible reason for the decrease in prediction accuracy in a wet state is the difference between the order of error related to the mass ratio and the order of error related to the coarse particle ratio in the process of updating the weight w. For example, out of the total error E in the above formula (5), the order of E1, E2, E3, and E4, which are errors related to mass ratio (%), and the order of E5, which is an error related to coarse grain ratio. There is a difference of about 10 to 100 times. In one example, the order of error regarding mass ratio is about 1.0 to 10, and the order of error regarding coarse particle ratio is about 0.001 to 0.10.

To update the weight w by repeating the calculations of equations (5) and (6) above, the error related to the coarse particle ratio of the aggregate (first error) and the error related to the mass ratio for each particle size category (second error) This corresponds to updating an intermediate model generated in the process of building a predictive model based on the error). If the order of the error related to the coarse grain ratio is smaller than the order of the error related to the mass ratio, the contribution of E5 in the above equation (5) will be small, and in the process of updating the intermediate model, weighting will be applied to reduce E5. There is a possibility that operations that update w are rarely executed.

The model construction unit 72 uses the error corrected to reduce the difference between the error related to the coarse grain ratio of the aggregate and the error related to the mass ratio for each particle size classification, and is an intermediate model used when constructing the prediction model M. may be updated. For example, the model construction unit 72 multiplies each of the two types of errors by a coefficient so that the difference between the errors is reduced. In one example, in the process of updating the weight w for generating the predictive model M (the process of updating the intermediate model), the model construction unit 72 uses the following equation (9) instead of the above equation (5). You may repeat the calculations.

In equation (9), λ1 and λ2 are coefficients. The coefficient λ1 and the coefficient λ2 may be determined by an operator or the like based on the order of the error in the coarse grain ratio and the order of the error in the mass ratio. The coefficient λ2 may be approximately 5 times to 200 times the coefficient λ1. The coefficient λ2 may be 10 times the coefficient λ1, or may be 100 times the coefficient λ1. Note that in this disclosure, the multitasking performed by the prediction model M constructed by updating the weight w by the calculation of Equation (9) is referred to as "coefficient adjustment multitasking." Furthermore, simply "multitasking" means multitasking in which error adjustment is not performed in the process of updating the weight w.

FIG. 13 shows the verification results in a wet state for each case of multitasking, coefficient adjustment multitasking, and single task. Note that the amount of coarse aggregate is 8 kg, and the image size ratio is 0.25. In multitasking and coefficient adjustment multitasking, the coarse grain rate is predicted by direct prediction of two types of prediction methods. The coefficient λ1 is set to 1.00, and the coefficient λ2 is set to 100. The results shown in FIG. 13 show that in a wet state, the prediction accuracy in coefficient adjustment multitask and single task is higher than the prediction accuracy in multitask.

FIG. 14 shows the verification results when the values of the coefficient λ1 and the coefficient λ2 were changed under the same conditions as the verification results shown in FIG. 13. Specifically, the coefficient combinations (λ1, λ2) are (1.0, 10), (0.1, 1.0), (1.0, 100), and (0.01, 1.0 ) were set and evaluated. As shown in FIG. 14, regardless of whether the ratio of the coefficient λ2 to the coefficient λ1 was 10 or 100, the average absolute difference value was less than or equal to the target value of 0.1. From the results shown in FIG. 14, it can be seen that the average absolute difference value tends to be smaller when the ratio of the coefficient λ2 to the coefficient λ1 is 100 than when the ratio is 10.

The various prediction models explained above are just examples, and the prediction model may be any model as long as it can output at least a predicted value regarding coarse aggregate from information including image data of the aggregate. good. In the learning phase, the aggregate on the transport device 8 may be imaged to prepare a learning image (learning data). In addition to the image data of the aggregate, the input information and evaluation information input to the prediction model may include information indicating a physical quantity that may affect the coarse grain ratio.

The manufacturing system 1 may include a computer that controls the manufacturing device 10 and a computer (quality prediction device) that predicts the coarse grain ratio of aggregate. A computer that predicts the coarse grain ratio of aggregate may include a model construction section 72, a model holding section 74, a quality prediction section 76, and an output section 78. In one of the various examples described above, at least some of the matters described in the other examples may be applied.

[Summary of this disclosure]
The aggregate quality prediction method described above is based on machine learning based on input information including image data obtained by imaging the aggregate, and the correct value of the coarse grain ratio of the aggregate that is associated with the input information. This includes a preparatory step for constructing a prediction model M that outputs a value indicating the coarse grain ratio of the aggregate according to the input information, and image data obtained by imaging the aggregate to be evaluated. The method includes an acquisition step of acquiring corresponding evaluation information, and a prediction step of inputting the evaluation information into the prediction model M and acquiring a first predicted value output from the prediction model M regarding the coarse grain ratio of the aggregate.

The coarseness of aggregate affects the quality of ready-mixed concrete produced by mixing concrete materials containing aggregate. For example, in the ready-mixed concrete industry, there is a situation in which labor is used for process inspections such as fresh property tests as part of quality control work. Therefore, in order to improve the efficiency of ready-mixed concrete production, it is desired to save labor in quality control. In the above quality prediction method, the coarse grain ratio of aggregate is predicted using a prediction model M constructed by machine learning, which simplifies the work for inspecting the coarse grain ratio of aggregate used in the production of ready-mixed concrete. It is possible to aim for As a result, the inspection frequency can be increased, or labor can be used for quality control work different from controlling the coarse particle ratio, which is useful for stabilizing the quality of ready-mixed concrete.

In the quality prediction method described above, in the preparation step, the prediction model M may be constructed so as to output only a value indicating the coarse grain ratio of the aggregate according to the input information. By the prediction model M performing the above-mentioned single task, as described above, even if the surface condition of the aggregate is wet, the coarse particle ratio can be predicted with high accuracy using the prediction model M. In factories that manufacture ready-mixed concrete, aggregate often contains some amount of water, and improving prediction accuracy in wet conditions is more beneficial in stabilizing the quality of ready-mixed concrete.

In the quality prediction method described above, the preparation step includes constructing a prediction model M by machine learning based on the correct mass ratio values of each of the plurality of particle size categories of aggregate, which are associated with the input information. May include. In the preparation step, the prediction model M may be constructed to further output a value indicating the mass ratio of each section included in the plurality of particle size sections, depending on the input information. The prediction step includes inputting the evaluation information into the prediction model M, further obtaining a second predicted value output from the prediction model M regarding the mass ratio of each category, and converting the second predicted value into the coarse particle percentage of the aggregate. It may also include converting into . In this case, one prediction model M can be used to predict the coarseness of the aggregate in two different ways. Therefore, it is possible to evaluate the reliability of the coarse grain ratio prediction results while avoiding an increase in work for model construction.

In the quality prediction method described above, the preparation step is generated in the process of constructing the prediction model M based on the first error regarding the coarse grain ratio of the aggregate and the second error regarding the mass ratio of each category. In the preparation step, the intermediate model is updated using the error corrected so that the difference between the first error and the second error is reduced. By the predictive model M performing the above-mentioned coefficient adjustment multitask, as described above, even if the surface condition of the aggregate is wet, the coarse particle ratio can be predicted with high accuracy using the predictive model M. In factories that manufacture ready-mixed concrete, aggregate often contains some amount of water, and improving prediction accuracy in wet conditions is more beneficial in stabilizing the quality of ready-mixed concrete.

In the quality prediction method described above, the preparation step may include constructing the prediction model M by machine learning using a convolutional neural network. In this case, more complex models can be constructed by performing machine learning using convolutional neural networks. Therefore, the predicted value of the coarse grain ratio can be brought closer to the correct value.

In the quality prediction method described above, the acquisition step is to image the range including the aggregates to be evaluated from a position away from the aggregates to be evaluated while the aggregates to be evaluated are piled up, and obtain evaluation information. It may include obtaining. In factories that manufacture ready-mixed concrete, aggregate is often piled up. For example, at the time of receiving, aggregates are transported on transport trucks or ships. Further, during manufacturing, the aggregates are transported in a piled state by a belt conveyor or the like. In the above configuration, since imaging is performed in a state in which the aggregates are piled up, it is possible to simplify the work for acquiring evaluation information. Note that the quality prediction method described above can be used as a substitute for quality control operations such as quality sieving tests, as well as during transportation as described above. For example, at the time of receiving or manufacturing, a part of the aggregate may be collected, laid out on a floor or sheet, and imaged.

In the quality prediction method described above, the preparation step may include preparing the learning aggregate so that it contains particles of two or more particle size classifications. The input information may include image data obtained by imaging the learning aggregate. In this case, the learning aggregate is in a state close to the actual aggregate used in a factory that manufactures ready-mixed concrete, and the accuracy of prediction by the prediction model M can be improved.

The method for producing ready-mixed concrete described above consists of a production process in which materials including aggregate are mixed to produce ready-mixed concrete, and a quality prediction in which the coarse particle ratio of at least a portion of the aggregate used in the production process is predicted. process. The quality prediction process uses machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate associated with the input information. A preparatory step of constructing a prediction model M that outputs a value indicating the coarse grain ratio of the aggregate according to the process, and image data obtained by imaging the aggregate to be evaluated, and obtaining evaluation information corresponding to the input information. The method includes an acquisition step and a prediction step of inputting evaluation information into the prediction model M and acquiring a first predicted value output from the prediction model M regarding the coarse grain ratio of the aggregate. This manufacturing method is useful for stabilizing the quality of ready-mixed concrete, similar to the quality prediction method described above.

The manufacturing system 1 described above includes a manufacturing device 10 that mixes materials including aggregate to manufacture ready-mixed concrete, and a control device 60 that predicts the coarse particle ratio of at least a portion of the aggregate used by the manufacturing device 10. (Quality prediction device). The control device 60 uses machine learning based on the input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate associated with the input information to adjust the input information. A preparatory step of constructing a prediction model M that outputs a value indicating the coarse grain ratio of the aggregate according to the process, and image data obtained by imaging the aggregate to be evaluated, and obtaining evaluation information corresponding to the input information. It is configured to perform an acquisition step and a prediction step of inputting the evaluation information into the prediction model M and acquiring a first predicted value output from the prediction model M regarding the coarse grain ratio of the aggregate. This manufacturing system 1 is useful for stabilizing the quality of ready-mixed concrete, similar to the quality prediction method described above.

1... Manufacturing system, 2... Material storage area, 4... Silo, 8... Transport device, 10... Manufacturing device, 50... Imaging device, 60... Control device, 72... Model construction section, 74... Model holding section, 76... Quality prediction Part, 78... Output part, M... Prediction model.

Claims (4)

an acquisition step of acquiring evaluation information including image data obtained by imaging the aggregate to be evaluated;
Machine learning based on input information including image data obtained by imaging the aggregate and the correct value of the coarse grain ratio of the aggregate associated with the input information allows A prediction step of inputting the evaluation information into a prediction model constructed to output only a value indicating the coarse grain ratio of the aggregate, and obtaining a first predicted value output from the prediction model regarding the coarse grain ratio of the aggregate. and ,
The aggregate quality prediction method , wherein the evaluation information is information corresponding to the input information . an acquisition step of acquiring evaluation information including image data obtained by imaging the aggregate to be evaluated;
Input information including image data obtained by imaging the aggregate, a correct value of the coarse grain ratio of the aggregate associated with the input information, and a plurality of particle sizes of the aggregate associated with the input information. Based on the correct value of the mass ratio of each category, machine learning based on the above input information calculates the value indicating the coarse particle ratio of the aggregate and the mass ratio of each category included in the plurality of particle size categories. a prediction step of predicting the coarse grain ratio of the aggregate using a prediction model constructed to output the indicated value;
The evaluation information is information corresponding to the input information,
The prediction step includes:
The evaluation information is input to the prediction model, and a first predicted value is output from the prediction model regarding the coarse grain ratio of the aggregate, and a second predicted value is output from the prediction model regarding the mass ratio of each category. and obtaining
Converting the second predicted value into a coarse grain ratio of aggregate. The aggregate quality prediction method according to claim 2, further comprising an output step of outputting the first predicted value and a value obtained by converting the second predicted value into a coarse grain ratio of the aggregate. A program for causing a computer to execute the aggregate quality prediction method according to any one of claims 1 to 3.

Images