JP7847195B2 - Method for predicting the quality of ready-mixed concrete, quality prediction program, and ready-mixed concrete quality prediction device. - Google Patents

Description

This disclosure relates to a method for predicting the quality of ready-mixed concrete, a quality prediction program, and a quality prediction device for ready-mixed concrete.

Patent Document 1 discloses a method for predicting the quality of ready-mixed concrete using a predictive model.

Japanese Patent Publication No. 2021-124304

This disclosure provides a method, program, and apparatus for predicting the quality of ready-mixed concrete, which can improve the accuracy of quality prediction while avoiding the complexity of the work involved in quality prediction when using a predictive model.

[1] A method for predicting the quality of ready-mixed concrete, comprising: an acquisition step of acquiring input information including first information relating to at least one of the following: the power load value of a mixer used to manufacture ready-mixed concrete by mixing concrete materials or to agitating ready-mixed concrete; vibrations caused by the dropping or flow of ready-mixed concrete; and an image of the ready-mixed concrete; and second information relating to the target quality when the ready-mixed concrete is used; a prediction model pre-constructed by machine learning to output quality information indicating the quality of the ready-mixed concrete in response to the input information; and a prediction step of predicting the quality of the ready-mixed concrete based on the input information acquired in the acquisition step.

[2] A method for predicting the quality of ready-mixed concrete as described in [1] above, wherein the quality of ready-mixed concrete includes one or more of the following qualities: slump, slump flow, and air content.

[3] The ready-mix concrete quality prediction method described in [2] above, wherein the quality of ready-mix concrete includes two or more qualities from slump, slump flow, and air content, and the prediction model is constructed to output the two or more qualities in response to the input information.

[4] The method for predicting the quality of ready-mixed concrete according to any one of [1] to [3] above, wherein the prediction model is constructed by machine learning using a neural network that does not perform operations to normalize the input information or operations to remove some connections in the fully connected layer.

[5] The method for predicting the quality of ready-mixed concrete according to any one of [1] to [4] above, wherein the second information includes information indicating the target quality of the ready-mixed concrete at the time of use, as well as information indicating the target quality of the ready-mixed concrete at the time of shipment.

[6] The method for predicting the quality of ready-mixed concrete according to any one of [1] to [5] above, wherein the prediction model is constructed by machine learning using a neural network, and in the machine learning using the neural network when constructing the prediction model, one selected from the group consisting of Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov is used as the weight update formula.

[7] The method for predicting the quality of ready-mixed concrete according to any one of [1] to [6] above, wherein the prediction model is constructed using machine learning with a neural network, and in the machine learning using the neural network used to construct the prediction model, a function is used as the loss function such that the loss value L(a) is always less than or equal to the absolute value of the error a.

[8] A quality prediction program that causes a computer to execute one of the ready-mix concrete quality prediction methods described in any one of [1] to [7] above.

[9] A ready-mix concrete quality prediction device comprising: an input information acquisition unit that acquires input information including first information relating to at least one of the following: the power load value of a mixer used to produce ready-mix concrete by mixing concrete materials or to agitating ready-mix concrete; vibrations caused by the dropping or flow of ready-mix concrete; and an image of ready-mix concrete; and second information relating to the target quality when using ready-mix concrete; a prediction model pre-built by machine learning to output quality information indicating the quality of ready-mix concrete in response to the input information; and a prediction calculation unit that predicts the quality of ready-mix concrete based on the input information acquired by the input information acquisition unit.

This disclosure provides a method, program, and apparatus for predicting the quality of ready-mixed concrete, which can improve the accuracy of quality prediction while avoiding the complexity of the work involved in quality prediction when using a predictive model.

Figure 1 is a schematic diagram showing an example of a ready-mix concrete manufacturing system. Figure 2 is a block diagram showing an example of the functional configuration of a control device. Figure 3 is an example of data related to power load values. Figure 4 is a block diagram showing an example of the hardware configuration of a control device. Figure 5(a) shows an example of the processing flow in the learning phase. Figure 5(b) shows an example of the processing flow in the evaluation phase. Figures 6(a) and 6(b) are schematic diagrams illustrating an example of the computation process by a prediction model. Figure 6(c) is a schematic diagram illustrating an example of the computation process of a neural network. Figure 7 is a graph showing an example of a loss function. Figure 8 is a graph showing an example of the results of comparing predicted values with correct values.

The following describes one embodiment with reference to the drawings. In this description, identical elements or elements with the same function are denoted by the same reference numerals, and redundant explanations are omitted. Figure 1 schematically shows a ready-mix concrete manufacturing system equipped with a quality prediction device according to one embodiment.

[Ready-mix concrete manufacturing system]
First, we will explain the overview of the ready-mix concrete manufacturing system. Manufacturing system 1, shown in Figure 1, is a system for manufacturing ready-mix concrete. Manufacturing system 1 produces ready-mix concrete by mixing concrete materials.

The concrete materials used in manufacturing system 1 include cement, admixtures, coarse aggregate, fine aggregate, water, and other admixtures. Examples of coarse aggregate include gravel, crushed stone, slag coarse aggregate, lightweight coarse aggregate, recycled coarse aggregate, recovered aggregate, or coarse aggregates mixed therewith. Gravel can be mountain gravel, land gravel, river gravel, or sea gravel. Slag coarse aggregate can be blast furnace slag aggregate, ferronickel slag aggregate, electric furnace oxidized slag aggregate, or coal gasification slag aggregate. Lightweight coarse aggregate can be natural lightweight aggregate, by-product lightweight aggregate, or artificial lightweight aggregate. Coarse aggregate may also include crushed rock or limestone.

Examples of fine aggregates include sand, crushed sand, slag aggregate, lightweight aggregate, recycled aggregate, recovered aggregate, or mixtures thereof. Sand can be mountain sand, land sand, river sand, or sea sand. Slag aggregates include blast furnace slag aggregate, ferronickel slag aggregate, copper slag aggregate, electric furnace oxidized slag aggregate, or coal gasification slag aggregate. Lightweight aggregates include natural lightweight aggregate, by-product lightweight aggregate, or artificial lightweight aggregate.

Crushed stone and crushed sand can be made from various rock types, including igneous rocks, sedimentary rocks, metamorphic rocks, silica, limestone, domaloyte, or peridotite. Igneous rocks include granite, diorite, gabbro, porphyry, diorite, rhyolite, andesite, basalt, or serpentinite. Sedimentary rocks include conglomerate, sandstone, shale, slate, or tuff. Metamorphic rocks include gneiss or schist.

The manufacturing system 1 loads the manufactured ready-mix concrete onto a transport vehicle 200. After loading, the transport vehicle 200 transports the ready-mix concrete to the site where it will be used (e.g., a construction site). Examples of transport vehicles 200 include agitator trucks (mixer trucks) or dump trucks. The manufacturing system 1 may also manufacture ready-mix concrete from concrete materials to meet the target quality (required quality) set for each site. For example, the operator of the manufacturing system 1 determines the concrete material mix to meet the target quality set for each site and inputs operation instructions to the manufacturing system 1.

The target quality set for each site differs from the quality of the ready-mixed concrete manufactured in Manufacturing System 1 and before shipment (at the time of shipment). For example, in Manufacturing System 1, quality control (inspection, etc.) is performed on the ready-mixed concrete manufactured in Manufacturing System 1 and before shipment to ensure that the target quality set for each site is met. In this disclosure, the target quality set for each site is referred to as the "target quality at the time of use of the ready-mixed concrete." "At the time of use of the ready-mixed concrete" refers to the time when the ready-mixed concrete is received at the site. The target quality at the time of use of the ready-mixed concrete may be set as follows, for example, (1) to (3) below.

(1) Japanese Industrial Standards (National Standards)
For example, the target quality when using ready-mixed concrete includes target values for at least one of the following qualities: slump, slump flow, and air content. The on-site target values for slump, slump flow, and air content may be specified by the purchaser or ordering party in accordance with "Table 1 - Types and Classifications of Ready-Mixed Concrete" in "JIS A 5308:2019 (Ready-Mixed Concrete)" or "JIS A 5308:2024 (Ready-Mixed Concrete)".
(2) Building Standards Act (Law)
In addition to the target quality specified in the above-mentioned JIS standards, target quality may also be set for designated building materials subject to certification by the Minister of Land, Infrastructure, Transport and Tourism under Article 37, Paragraph 1, Item 2 of the Building Standards Act. Ready-mixed concrete that does not conform to the JIS standards can be used in buildings if it has been certified by the Minister of Land, Infrastructure, Transport and Tourism. This allows ready-mixed concrete with target quality not specified in the above-mentioned JIS standards (for example, slump of 23 cm, slump flow of 65 cm, etc.) to be used in buildings (foundations, main structures, parts that are important for safety, fire prevention, or sanitation, etc.).
(3) Standards set by academic societies If not specified in the JIS as described above, target quality for ready-mixed concrete may be set based on technical standards set by various academic societies, such as the "Standard Specifications for Concrete" established by the Japan Society of Civil Engineers and the "Standard Specifications for Building Construction and Commentary JASS 5 Reinforced Concrete Construction" established by the Architectural Institute of Japan.
Furthermore, the "JIS A 5308:2019 (Ready-Mixed Concrete)" or "JIS A 5308:2024 (Ready-Mixed Concrete)" described in (1) correspond to the "Japanese Industrial Standards designated by the Minister of Land, Infrastructure, Transport and Tourism" as stipulated in Article 37, Paragraph 1, Item 1 of the Building Standards Act. In that case, the second piece of information concerning the target quality when using ready-mixed concrete may be the target quality of the designated building material as stipulated in Article 37, Paragraph 1, Item 1 of the Building Standards Act. Alternatively, it may be the target quality of the designated building material subject to certification by the Minister of Land, Infrastructure, Transport and Tourism as stipulated in Article 37, Paragraph 1, Item 2 of the same Act, as described in (2). In addition, it may be the target quality set by standards established by academic societies such as the "Standard Specifications for Concrete" or "Standard Specifications for Building Construction and Commentary JASS 5 Reinforced Concrete Construction" described in (3).

To control the quality of ready-mix concrete before shipment, a target quality at the time of shipment may be determined based on the target quality at the time of use. The target quality at the time of shipment may also be set according to internal standards established by each factory. In setting the target quality at the time of shipment, at least one of the following pieces of information may be considered: the condition (quality) of the materials used during manufacturing, the season (temperature), the type of concrete, the target quality at the time of use, and the transportation time. For example, the target quality at the time of shipment might be set by adding a value defined in the internal standards to the target quality at the time of use.

The manufacturing system 1 includes, for example, a manufacturing apparatus 100 and a control device 10. The manufacturing apparatus 100 is a device that manufactures ready-mixed concrete based on operation instructions from the control device 10. The manufacturing apparatus 100 mixes concrete materials to produce ready-mixed concrete. The manufacturing apparatus 100 includes, for example, a material storage area 101, a transport device 104, a storage jar 111, a measuring jar 112, a collection hopper 113, a mixer 114, and a loading hopper 115.

The material storage area 101 is a place for storing concrete materials. The material storage area 101 includes multiple silos 102. The multiple silos 102 are containers for storing at least a portion of the concrete materials, separated by material type. The multiple silos 102 include, for example, a silo 102 for storing coarse aggregate, a silo 102 for storing fine aggregate, and a silo 102 for storing cement.

The transport device 104 is a device that transports concrete materials stored in multiple silos 102 to storage jars 111. The transport device 104 includes, for example, a belt conveyor for transporting concrete materials. The transport device 104 may transport concrete materials of different types at different timings. In one example, based on operation instructions from the control device 10, a specific material from among the various concrete materials is transferred to the transport device 104 and transported to the storage jars 111.

The storage jar 111 temporarily stores various concrete materials. Various concrete materials are transported (conveyed) to the storage jar 111 from the material storage area 101 by the transport device 104. The storage jar 111 is configured to store each type of concrete material individually. Hereafter, "concrete materials" may be simply referred to as "materials." The various materials stored in the storage jar 111 are supplied to the measuring jar 112 as needed.

The measuring bottle 112 is located below the storage bottle 111. The measuring bottle 112 operates based on operation instructions from the control device 10, individually measuring various materials. When the measuring bottle 112 detects the target amount of material instructed by the control device 10, it supplies that material to the collection hopper 113. When water is supplied to the measuring bottle 112, an admixture may be mixed into the water. The collection hopper 113 is located below the measuring bottle 112. The collection hopper 113 collects the various materials discharged from the measuring bottle 112 and supplies the collected materials to the mixer 114. Note that the manufacturing apparatus 100 does not necessarily have a collection hopper 113; the various materials may be supplied directly from the measuring bottle 112 to the mixer 114.

The mixer 114 is located below the collection hopper 113. The mixer 114 is a device for mixing concrete materials. The mixer 114 produces ready-mixed concrete by mixing aggregate, cement, water, and admixtures. In other words, the mixer 114 produces ready-mixed concrete by mixing concrete materials. The ready-mixed concrete is discharged from the bottom of the mixer 114 into the loading hopper 115. The mixer 114 may be a tilting mixer, a horizontal single-axis mixer, a horizontal double-axis mixer, or a pan-type mixer. The mixer 114 includes, for example, two stirring members 114a and a mixer drive unit 114b.

The stirring members 114a are components that stir the various materials supplied to the mixer 114. Two stirring members 114a are arranged side-by-side inside the main body (container) of the mixer 114 and are rotatably mounted. Each of the two stirring members 114a includes a rotation axis extending in one horizontal direction. The mixer drive unit 114b rotates the rotation axes of each of the two stirring members 114a based on operation instructions from the control device 10. The mixer drive unit 114b includes a drive source, such as a motor, that provides driving force to the stirring members 114a. An opening is provided at the bottom of the main body of the mixer 114 for discharging the manufactured ready-mixed concrete into the loading hopper 115.

The loading hopper 115 is located below the mixer 114 and temporarily stores the ready-mix concrete. The loading hopper 115 then supplies the temporarily stored ready-mix concrete to the transport vehicle 200.

The manufacturing apparatus 100 described above is just one example of a ready-mix concrete manufacturing apparatus. A ready-mix concrete manufacturing apparatus can be configured in any way as long as it can mix concrete materials using a mixer to produce ready-mix concrete.

In this disclosure, a unit of ready-mixed concrete produced in a single mixing cycle in the mixer 114 and loaded onto the transport vehicle 200 is defined as "one batch." The process performed by the manufacturing system 1 to produce one batch of ready-mixed concrete is defined as "batch processing." In one example, 1 to 3 batches of ready-mixed concrete are loaded onto one transport vehicle 200. For example, if 2 batches of ready-mixed concrete are loaded onto one transport vehicle 200, two batch processes, each following the same manufacturing conditions, are performed at different times (sequentially).

<Control device (quality prediction device)>
The control device 10 is a device that controls the manufacturing apparatus 100. The control device 10 is composed of one or more computers. If the control device 10 is composed of multiple computers, these computers are connected to each other so as to be able to communicate with each other. The control device 10 controls the manufacturing apparatus 100 according to the set operating conditions. At least a part of the operating conditions may be determined by instructions from an operator such as a worker.

The control device 10 may be connected to an input device 12 and a monitor 14. The input device 12 is a device that inputs information indicating instructions from a worker or other user to the control device 10. The input device 12 can be any device capable of inputting desired information, and may be a keyboard (keypad), operation panel, or mouse. The monitor 14 is a device for displaying information from the control device 10 to a worker or other user. The monitor 14 can be any device capable of graphic display, and may be a liquid crystal display. The input device 12 and the monitor 14 may be integrated, such as a touch panel. The control device 10, input device 12, and monitor 14 may be integrated, such as a tablet computer (tablet terminal).

In addition to controlling the manufacturing apparatus 100, the control device 10 may also have a function to predict the quality of the ready-mixed concrete produced by the manufacturing apparatus 100. In this case, the control device 10 constitutes a quality prediction device (ready-mixed concrete quality prediction device) that predicts the quality of the ready-mixed concrete. In the following description, the quality of the ready-mixed concrete to be predicted by the control device 10 is the quality after it has been produced by the manufacturing apparatus 100 and before it is shipped to the site (i.e., the quality of the ready-mixed concrete at the time of shipment).

The quality of the ready-mixed concrete predicted by the control device 10 may include one or more of the following qualities: slump, slump flow, and air content. The quality of the ready-mixed concrete predicted by the control device 10 may include two or more of the following qualities: slump, slump flow, and air content. The quality of the ready-mixed concrete predicted by the control device 10 may be one of the following qualities: slump, slump flow, and air content, or two or more of the following qualities: slump, slump flow, and air content.

The control device 10 is configured to perform at least an acquisition step and a prediction step. The acquisition step involves acquiring input information, including information related to the power load value of the mixer 114 and information related to the target quality of the ready-mixed concrete (the target of prediction) when it is used. The prediction step involves predicting the quality of the ready-mixed concrete based on a prediction model pre-built using machine learning to output quality information indicating the quality of the ready-mixed concrete in response to the input information, and the input information acquired in the acquisition step.

Figure 2 shows an example of the functional components (hereinafter referred to as "functional blocks") of the control device 10. The control device 10 includes, for example, a motion control unit 22, a motion information acquisition unit 24, a target quality information acquisition unit 26, a model construction unit 30, a model holding unit 32, a prediction calculation unit 28, and a display output unit 34 as functional blocks. The processes executed by these functional blocks correspond to the processes executed by the control device 10.

The operation control unit 22 controls the manufacturing apparatus 100 to produce ready-mixed concrete according to predetermined operating conditions. At least a portion of these operating conditions may be determined by an operator, such as a worker, each time ready-mixed concrete is produced. The operation control unit 22 may also control the mixer drive unit 114b of the mixer 114 so that its rotational speed follows the target rotational speed specified in the operating conditions. When controlling the mixer drive unit 114b, the operation control unit 22 may also adjust the power (e.g., current value) supplied to the mixer drive unit 114b. The power load tends to be higher when the produced ready-mixed concrete is hard, and lower when the produced ready-mixed concrete is soft.

The operation information acquisition unit 24 acquires information related to the operation during the production of ready-mixed concrete. The operation information acquisition unit 24 acquires information related to the power load value of the mixer 114 (hereinafter referred to as "first information"). The power load value of the mixer 114 may be a value indicating the power (kW) supplied to the mixer 114 itself, or a value indicating the current value (A) supplied to the mixer 114. Alternatively, the power load value of the mixer 114 may be replaced with a value indicating the load oil pressure (MPa). The information related to the power load value of the mixer 114 may be continuous time-series data during the operation of the mixer 114 during the processing of one batch, or statistical data obtained from that time-series data.

Figure 3 schematically shows time-series data related to the power load value of mixer 114 during processing for one batch. The time-series data related to the power load value is obtained, for example, by repeatedly measuring the power (kW) supplied to mixer 114 at a predetermined sampling period. The operation information acquisition unit 24 may calculate (acquire) at least one of the following statistical values selected from the group consisting of the variation range (difference between maximum and minimum values), decrease (difference between maximum and ultimate values), total value within an arbitrarily set time, mean, standard deviation, coefficient of variation, median, first quartile, third quartile, kurtosis, and skewness. In addition to at least one of the above selected statistical values, the operation information acquisition unit 24 may also calculate at least one of the minimum and maximum values. The statistical data acquired by the operation information acquisition unit 24 may include statistical values obtained from calculations using two or more power load values included in the time-series data.

In the graph shown in Figure 3, "P1" represents the initial value in the time-series data. The initial value P1 is the power load value at the start of mixing in the time-series data. "P2" represents the minimum value in the time-series data. The minimum value P2 is the minimum power load value after the time when the initial value P1 was obtained. Note that the power load value at the start of mixing (initial value P1) may also be the minimum. "P3" represents the maximum value in the time-series data. The maximum value P3 is the maximum power load value after the time when the initial value P1 was obtained. "P4" represents the ultimate value in the time-series data. The ultimate value P4 is the power load value at the point when the operation control unit 22 determines that the conditions for ending mixing by the mixer 114 have been met. For example, the operation control unit 22 determines that the above conditions are met when a predetermined time has elapsed since the start of driving the stirring member 114a, and stops driving the stirring member 114a.

The target quality information acquisition unit 26 acquires information related to the target quality of the ready-mixed concrete when it is used (hereinafter referred to as "second information"). If the quality predicted by the control device 10 includes slump, the target quality information acquisition unit 26 may acquire information indicating the target value of the slump when the ready-mixed concrete is used as second information. If the quality predicted by the control device 10 includes slump flow, the target quality information acquisition unit 26 may acquire information indicating the target value of the slump flow when the ready-mixed concrete is used as second information. If the quality predicted by the control device 10 includes air content, the target quality information acquisition unit 26 may acquire information indicating the target value of the air content when the ready-mixed concrete is used as second information. The target quality information acquisition unit 26 may further acquire information indicating the target quality of the ready-mixed concrete at the time of shipment as second information. That is, the second information may include information indicating the target quality of the ready-mixed concrete at the time of shipment in addition to information indicating the target quality of the ready-mixed concrete at the time of use.

The target quality information acquisition unit 26 acquires the second information based, for example, on input from an operator such as a worker via the input device 12. In one example, the target quality information acquisition unit 26 acquires the second information along with the operating conditions related to the production of the ready-mixed concrete to be manufactured. The target quality information acquisition unit 26 may also acquire the second information based on input from an operator such as a worker via the input device 12 at the timing when the control device 10 performs quality prediction. The operation information acquisition unit 24 and the target quality information acquisition unit 26 function as input information acquisition units that acquire input information including the first and second information.

The model construction unit 30 constructs a model for predicting the quality of ready-mixed concrete (hereinafter referred to as "prediction model M"). Prediction model M is a model that outputs quality information (quality values) indicating the quality of ready-mixed concrete in response to input information including the first and second information described above. The model construction unit 30 constructs prediction model M using machine learning based on the input information and the correct quality values associated with that input information. Prediction model M may be constructed to output one or more predicted values as quality information, including predicted slump values, predicted slump flow values, and predicted air content values. Alternatively, prediction model M may be constructed to output a predicted value of the ratio of slump flow to slump (slump flow/slump) as quality information, either in lieu of or in addition to the above one or more predicted values.

Machine learning is a method in which a machine (computer) autonomously discovers laws or rules by iteratively learning based on given information. A predictive model M can be constructed using algorithms and data structures. For example, a predictive model M can be implemented using a neural network, which is an information processing model that mimics the structure of the human brain. The specific machine learning algorithm used to construct the predictive model M is not particularly limited. A neural network, for example, has an input layer, one or more hidden layers, and an output layer. Including one or more hidden layers allows for the construction of more complex predictive models M, improving prediction accuracy.

The model building unit 30 may autonomously construct a predictive model M for predicting the quality of ready-mixed concrete by performing machine learning using the data provided as input to machine learning and the ground truth data (ground truth values such as slump) from the output of machine learning. The input to machine learning consists of various datasets of input information, including first and second information. The output of machine learning is data (numerical values) indicating the quality of ready-mixed concrete. The model building unit 30 iteratively learns a model that outputs predicted values such as slump using multiple combinations of the input information datasets and ground truth values such as slump.

The stage of autonomously constructing the prediction model M corresponds to the learning phase. This learning phase may be performed before the production phase in which ready-mixed concrete is manufactured, or it may be performed in the early stages of the production phase. The model holding unit 32 holds the prediction model M constructed by the model construction unit 30. The learned prediction model M may be transferable between computers. Therefore, the prediction model M constructed in the control device 10 may be used in a manufacturing system different from the manufacturing system 1, and the model holding unit 32 may hold the prediction model M constructed in the other manufacturing system.

The prediction calculation unit 28 predicts the quality of the ready-mixed concrete to be predicted in the evaluation phase, based on the input information acquired by the operation information acquisition unit 24 and the target quality information acquisition unit 26 (input information acquisition unit), and the prediction model M. The prediction calculation unit 28 inputs the acquired input information into the prediction model M and obtains the predicted value output from the prediction model M. The input information acquired in the evaluation phase (input information for evaluation) is information where the quality of the ready-mixed concrete is unknown.

The display output unit 34 outputs information (quality value) indicating the predicted quality calculated by the prediction calculation unit 28 to the monitor 14. This displays the predicted quality value on the monitor 14, allowing operators such as workers to understand the predicted quality value of the ready-mixed concrete being predicted (the manufactured ready-mixed concrete).

As shown in Figure 4, the control device 10 includes a circuit 50. The circuit 50 comprises a processor 51, a memory 52, a storage device 53, and an input/output port 54. The storage device 53 is composed of one or more non-volatile memory devices, such as flash memory or a hard disk. The storage device 53 stores at least a quality prediction program that causes the computer to execute the acquisition and prediction processes. The storage device 53 also stores quality prediction programs for configuring each functional block of the control device 10.

Memory 52 is composed of one or more volatile memory devices, such as random access memory. Memory 52 temporarily stores the quality prediction program loaded from storage 53. The processor 51 is composed of one or more computing devices, such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The processor 51 executes the quality prediction program loaded into memory 52, thereby configuring each functional block of the control device 10. The calculation results from the processor 51 are temporarily stored in memory 52. The input/output port 54 performs information input and output between the input device 12, monitor 14, and mixer 114, etc., in response to requests from the processor 51.

The timer 55 measures elapsed time, for example, by counting a reference pulse of a fixed period. Note that the circuit 50 is not necessarily limited to having each function configured by a program. For example, the circuit 50 may have at least some functions configured by a dedicated logic circuit or an ASIC (Application Specific Integrated Circuit) that integrates such circuits. The quality prediction program may be provided fixedly recorded on a tangible recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory. Alternatively, the quality prediction program may be provided via a communication network as a data signal superimposed on a carrier wave.

[Method of manufacturing ready-mixed concrete]
Next, an example of a ready-mix concrete manufacturing method performed in manufacturing system 1 will be described. The ready-mix concrete manufacturing method includes a manufacturing process and a quality prediction process. The manufacturing process is the process of manufacturing ready-mix concrete. The quality prediction process is the process of predicting the quality of the ready-mix concrete manufactured in the manufacturing process. The quality prediction process may be performed for a period that overlaps with at least a portion of the period during which the manufacturing process is repeatedly executed.

The manufacturing process includes, for example, a conveying process, a weighing process, a loading process, a mixing process, a discharge process, and a loading process. In the conveying process, various concrete materials are conveyed to the storage jar 111 by the conveying device 104, and each material is supplied to the storage jar 111 individually. In the weighing process, each material is supplied individually from the storage jar 111 to the weighing jar 112, where it is weighed. In the weighing process, when the measured amount of each material reaches a predetermined set amount, that material is discharged into the collection hopper 113. In the loading process, after all types of materials have been collected in the collection hopper 113, the materials in the collection hopper 113 are loaded (supplied) into the mixer 114.

In the mixing process, multiple types of concrete materials are mixed in the mixer 114. During the mixing process, the control device 10 may control the mixer drive unit 114b according to predetermined operating conditions. During the mixing process, the power supplied from the control device 10 to the mixer drive unit 114b may be adjusted so that the rotational speed of the mixer drive unit 114b follows a target rotational speed.

In the discharge process, after the mixing of the concrete materials is completed in the mixer 114, the ready-mixed concrete is discharged from the mixer 114 to the loading hopper 115. In the loading process, the ready-mixed concrete discharged into the loading hopper 115 is loaded onto the transport vehicle 200.

The quality prediction process (quality prediction method) includes a model building process in the learning phase and a quality evaluation process in the evaluation phase. In the quality prediction process, the model building process is performed before the quality evaluation process. Examples of the model building process and the quality evaluation process are described below.

(Model building process)
Figure 5(a) is a flowchart showing an example of a series of processes performed in the model building process. This model building process is performed before the manufacturing process is executed in the manufacturing apparatus 100, or in the initial stages after the manufacturing process has started. In this model building process, for example, the ready-mixed concrete actually produced in the manufacturing apparatus 100 is used as ready-mixed concrete for learning.

In the model building process, step S11 is executed first. In step S11, for example, an operator such as a worker prepares training data for machine learning. The training data consists of multiple datasets. Each of the multiple datasets contains input information (training input information) including the first and second information obtained when training ready-mix concrete was manufactured, and the correct values of quality information (e.g., slump) associated with that input information.

The correct values for quality information may be obtained by actually measuring the quality of the ready-mixed concrete used for training. For example, after the ready-mixed concrete for training is loaded onto the transport vehicle 200, some of it is extracted by workers. The slump, slump flow, and air content of the extracted concrete are then measured by the workers, and at least some of these measured values are used as the correct values in the training data.

Next, step S12 is executed. In step S12, for example, the model building unit 30 of the control device 10 constructs a predictive model M by performing machine learning using the training data (multiple datasets) prepared in step S11. The model building unit 30 may construct the predictive model M using machine learning with a neural network.

Figures 6(a) and 6(b) schematically show the prediction model M constructed by the model building unit 30. The prediction model M shown in Figure 6(a) is a model that outputs a single quality prediction value in response to input information. The prediction model M shown in Figure 6(b) is a model that outputs two or more quality prediction values in response to input information. In this disclosure, machine learning performed so that the prediction model M outputs a single quality prediction value (only one type of quality prediction value) is referred to as "single-task learning." Furthermore, machine learning performed so that the prediction model M outputs two or more types of quality prediction values is referred to as "multi-task learning."

The model building unit 30 may construct a predictive model M by performing single-task learning. In this case, the model building unit 30 may construct a predictive model M for each type of quality. For example, the model building unit 30 may construct two or more models from among a predictive model M that outputs a predicted value for slump, a predictive model M that outputs a predicted value for slump flow, and a predictive model M that outputs a predicted value for air volume. The model building unit 30 may also construct a predictive model M by performing multi-task learning. For example, the model building unit 30 may construct a predictive model M (one predictive model M) that outputs predicted values for two or more types of quality from among slump, slump flow, and air volume. In addition to one or more types of quality from among slump, slump flow, and air volume, the model building unit 30 may also construct a predictive model M (one predictive model M) that outputs the ratio of slump flow to slump.

Below, an example of a prediction model M will be explained using simplified mathematical formulas for ease of understanding. The prediction model M constructed by the model building unit 30 can be simply expressed as shown in the following equations (1) and (2).

In equation (2), Y represents the quality output value. In a predictive model M constructed by single-task learning, Y is the output value of a single quality. In a predictive model M constructed by multi-task learning, Y is the output value of one of two or more quality types, and equations (1) and (2) are calculated for each quality type. N is an integer greater than or equal to 2, representing the number of input data points. x represents various input values included in the input data. The input data corresponds to the above input information, and the input data includes at least values related to the power load value included in the first information (e.g., 2 or more statistics), and values indicating the target quality included in the second information.

wi is the weight (coefficient), and b is the bias term (coefficient). f(U) represents the activation function. The activation function can be a linear function (identity function), or a nonlinear function such as a polynomial, absolute value, step function, sigmoid function, hardsigmoid function, logsigmoid function, softmax function, logsoftmax function, softmin function, softplus function, softsine function, tanh function, tanhShrink function, hardtanh function, tanhexp function, ReLU function, ReLU6 function, Leaky-ReLU function, PReLU function, ELU function, SELU function, CELU function, Switch function, Mish function, ACON function, etc.

The model building unit 30 may repeatedly evaluate the error and loss value between Y (predicted value) obtained by equation (2) and the correct quality value using training data, and determine the weight wi and bias term b in equation (1) so as to minimize the error. The model building unit 30 may use any type of loss function as the function for evaluating the error between Y (output value from an intermediate model during the construction of the prediction model M) obtained by equation (2) and the correct quality value. For example, the model building unit 30 can use one selected from the group consisting of Huber loss function (HuberLoss), Mean Absolute Error (MAE), and ε-insensitive loss function.

The Huber loss function can be expressed, for example, as shown in equation (3). The ε-tolerance loss function can be expressed, for example, as shown in equation (4). L(a) is the loss value, and a is the error between the predicted value and the correct value. δ and ε are arbitrarily set parameters, and can be set, for example, to 1.0.


The role of the loss function is to input the predicted value and the correct value into the loss function and output a loss value L(a) from the error 'a' between the predicted value and the correct value. Then, the weight wi is calculated using the loss value L(a).

Figure 7 illustrates various loss functions. Specifically, Figure 7 shows the relationship between error a and loss value L(a) for the Huber loss function (δ = 1.0), mean absolute error, and ε-tolerant loss function (ε = 0.5), as well as the relationship between error a and loss value L(a) for mean squared error (MSE). In the graph of Figure 7, the horizontal axis represents error a, and the vertical axis represents loss value L(a). In mean squared error, when error a exceeds 1, the loss value L(a) becomes larger than the absolute value of error a, whereas for the other three loss functions, the relationship that loss value L(a) is less than or equal to the absolute value of error a holds regardless of the value of error a. From the viewpoint of improving prediction accuracy, it is preferable to use a loss function in which the loss value L(a) is always less than or equal to the absolute value of error a. Mean absolute error or ε-tolerant loss function may be used as a function in which the loss value L(a) is always less than or equal to the absolute value of error a. Alternatively, by setting the δ of the Huber loss function to 1.0 or less, it can be used as a function where the loss value L(a) is always less than or equal to the absolute value of the error a.

The model building unit 30 may repeatedly update the weights wi using the gradient method so that the error and loss value evaluated by the loss function are minimized. The model building unit 30 may use any type of update formula (weight update formula) when updating the weights wi. For example, the model building unit 30 may use one selected from the group consisting of Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov as the weight update formula. The weight update formula is also called an optimization algorithm or optimization method.

The model building unit 30 may construct a prediction model M using machine learning with a neural network that does not perform operations to normalize the input information or operations to remove some connections in the fully connected layer. The operation to normalize the input information refers to an operation (Batch Normalization) that normalizes (standardizes) each value included in the input information that becomes the input to the prediction model M so that the mean is 0 and the standard deviation is 1.

The operation of deleting some connections in a fully connected layer refers to the process of performing machine learning using a neural network while deactivating an arbitrary percentage of nodes (dropout). Figure 6(c) schematically shows the calculation process of the prediction model M when some connections in the fully connected layer are deleted (when dropout is performed). The model construction unit 30 may also construct the prediction model M using machine learning with a neural network that performs at least one of the operations of normalizing the input information and deleting some connections in the fully connected layer.

Returning to Figure 5(a), step S13 is executed after step S12. In step S13, for example, the model storage unit 32 stores the prediction model M constructed in step S12. This completes the model construction process.

(Quality evaluation process)
Figure 5(b) is a flowchart showing an example of a series of processes performed in the quality evaluation process. This quality evaluation process is performed, for example, during a period that overlaps with at least a portion of the period during which the above-mentioned manufacturing process is performed by the manufacturing apparatus 100.

In the quality evaluation process, the control device 10 first executes step S21. In step S21, for example, the control device 10 waits until the evaluation timing arrives, which is the time to evaluate the quality of the ready-mixed concrete to be evaluated. The ready-mixed concrete to be evaluated is also the ready-mixed concrete produced by the manufacturing equipment 100. The evaluation timing may be predetermined for a specific time period within the day, or it may be predetermined for the execution of a particular batch process within the day. The evaluation timing may also be the timing at which an operator, such as a worker, instructs the evaluation to be performed.

Next, the control device 10 executes step S22. In step S22, for example, the operation information acquisition unit 24 and the target quality information acquisition unit 26 acquire input information including the first information (information related to power load value) and the second information (information related to target quality) when the ready-mixed concrete to be evaluated was manufactured. The input information acquired in step S22 is evaluation input information where the quality information (e.g., slump) is unknown.

Next, the control device 10 executes step S23. In step S23, for example, the prediction calculation unit 28 predicts the quality information of the ready-mixed concrete to be evaluated based on the input information acquired in step S22 and the prediction model M held in the model holding unit 32. In one example, the prediction calculation unit 28 inputs the input information acquired in step S22 into the prediction model M and obtains predicted values for one or more quality parameters such as slump, slump flow, and air content, which are output from the prediction model M.

Next, the control device 10 executes step S24. In step S24, for example, the display output unit 34 displays the predicted quality value obtained in step S23 on the monitor 14. This allows operators, such as workers, to confirm the predicted quality value.

The quality evaluation process is now complete. The control device 10 may execute the series of processes from steps S21 to S24 each time a batch of ready-mixed concrete is produced (for each batch process). Alternatively, the control device 10 may execute the series of processes from steps S21 to S24 each time multiple batches of ready-mixed concrete are produced (for each batch process).

[Variations]
The series of processes shown in Figures 5(a) and 5(b) are examples and can be modified as appropriate. In the above series of processes, one step and the next step may be executed in parallel, and some steps may be executed in a different order than the example above. In place of at least some of the steps in the above series of processes, or in addition to the above series of processes, steps with content different from the example above may be executed.

The first piece of information described above may include at least one of the initial, minimum, maximum, and final values in the time-series data related to the power load value. The input information acquired by the control device 10 (input information acquisition unit) and used as input to the prediction model M may include information related to mixing by the mixer 114 (information other than the power load value). Examples of information related to mixing include the amount of concrete mixed per batch, the mixing time, the time at which the power load value reached its maximum value, and the time from when the power load value reached its maximum value until the fresh concrete was discharged from the mixer 114. The input information used as input to the prediction model M may also include information indicating the specified strength, the type of cement, the type of admixture, the type of fine aggregate, the type of coarse aggregate, and the amount of additives added.

The first piece of information included in the input information for the prediction model M may be information related to vibrations caused by the dropping or flow of ready-mixed concrete, instead of the power load value of the mixer 114. The vibration information may also be information related to vibrations caused by the dropping or flow of ready-mixed concrete from the mixer 114. As shown in Figure 1, the manufacturing system 1 may be equipped with a vibration sensor 18.

The vibration sensor 18 is a sensor that detects vibrations (magnitude of vibrations) caused by the dropping or flow of ready-mixed concrete produced by the mixer 114. Vibrations caused by dropping or flowing include vibrations caused by both the dropping and flowing of the ready-mixed concrete. The vibration sensor 18 may be capable of detecting the magnitude of vibrations in two or more directions caused by the dropping or flowing of the ready-mixed concrete. For example, the vibration sensor 18 is attached to the loading hopper 115 to detect the magnitude of vibrations in the loading hopper 115. The vibration sensor 18 may be installed on any side wall of the loading hopper 115. Ready-mixed concrete discharged from the mixer 114 may fall onto the side wall on which the vibration sensor 18 is installed.

The vibration sensor 18 can be of any type as long as it is capable of detecting the magnitude of vibration of the loading hopper 115. For example, the vibration sensor 18 is a sensor that detects the acceleration of the loading hopper 115. The vibration sensor 18 may detect acceleration in two mutually orthogonal directions along the side wall of the loading hopper 115, and acceleration in a direction perpendicular to the side wall of the loading hopper 115. The vibration sensor 18 may detect acceleration in each direction at a predetermined sampling period. The vibration sensor 18 outputs information indicating the detection results to the control device 10.

The vibration-related information among the input information for the prediction model M may be time-series data of detection values obtained by the vibration sensor 18 during the period including the time when the ready-mix concrete was discharged into the loading hopper 115 (hereinafter referred to as the "discharge period"). Alternatively, the vibration-related information may be statistical quantities obtained from the time-series data obtained during the discharge period, either in addition to or instead of the time-series data.

The first piece of information included in the input information for the prediction model M may be information relating to an image of the ready-mixed concrete, instead of the power load value of the mixer 114. The information relating to the image of the ready-mixed concrete is information relating to an image obtained by imaging the ready-mixed concrete being manufactured in the manufacturing apparatus 100. As shown in Figure 1, the manufacturing system 1 may be equipped with an image sensor 16.

The image sensor 16 is a sensor (camera) that images the ready-mixed concrete during manufacturing. In this disclosure, imaging the ready-mixed concrete during manufacturing in the manufacturing apparatus 100 includes imaging the ready-mixed concrete after it has been manufactured in the mixer 114, as well as imaging the ready-mixed concrete during manufacturing in the mixer 114. Imaging the ready-mixed concrete after it has been manufactured in the mixer 114 includes imaging the ready-mixed concrete in the loading hopper 115 and imaging the ready-mixed concrete being loaded from the loading hopper 115 into the transport vehicle 200. The image data obtained by imaging with the image sensor 16 may be either still image data or video data. The image sensor 16 may perform imaging based on operation instructions from the control device 10. The image sensor 16 outputs the image data obtained by imaging to the control device 10.

The image sensor 16 may be positioned to capture images of the ready-mixed concrete being produced in the mixer 114. The image sensor 16 may be positioned to capture images of the ready-mixed concrete contained in the space within the loading hopper 115. The image sensor 16 may be positioned in a location less susceptible to image interference from splashing ready-mixed concrete in the mixer 114 or loading hopper 115. The image sensor 16 may be positioned to capture images of the ready-mixed concrete being discharged from the loading hopper 115. The manufacturing system 1 may include two or more image sensors 16 located in two or more different locations.

In addition to predicting the quality of the ready-mixed concrete by the control device 10, the manufacturing apparatus 100 may periodically (for example, several times a day) measure the quality of the ready-mixed concrete. In this case, the prediction model M may be updated based on the measured value of the ready-mixed concrete quality and the input information obtained at the time the measured value was acquired (for example, information related to the power load value, or an image of the ready-mixed concrete). In the prediction process described above, the quality of the ready-mixed concrete may be predicted using the updated prediction model M. Even when the updated prediction model M is used, the process of predicting the quality of the ready-mixed concrete based on the prediction model M and the input information acquired in the acquisition process described above remains unchanged.

The first information included in the input information acquired by the input information acquisition unit of the control device 10 and used as input to the prediction model M may be information relating to two or more of the following: vibrations caused by the dropping or flow of ready-mixed concrete from the mixer 114, images of the ready-mixed concrete, and the power load value of the mixer 114. As described above, the first information included in the input information acquired by the input information acquisition unit of the control device 10 and used as input to the prediction model M may be information relating to at least one of the following: vibrations caused by the dropping or flow of ready-mixed concrete from the mixer 114, images of the ready-mixed concrete, and the power load value of the mixer 114.

The manufacturing system 1 may also include, separately from the control device 10, a quality prediction device (a ready-mix concrete quality prediction device) as a functional block, comprising at least an operation information acquisition unit 24, a target quality information acquisition unit 26, a prediction calculation unit 28, and a model holding unit 32. The computer constituting this quality prediction device may be connected to the control device 10 in a communicative manner. Hereinafter, the quality prediction device comprising at least the operation information acquisition unit 24, the target quality information acquisition unit 26, the prediction calculation unit 28, and the model holding unit 32 will simply be referred to as the "quality prediction device."

In the above example, the control device 10, which functions as a quality prediction device, predicts the quality of ready-mixed concrete after it has been manufactured by the manufacturing device 100 and before it is shipped to the site. The timing of quality prediction by the quality prediction device is not limited to this example. The quality prediction device may also predict the quality of ready-mixed concrete while it is being transported to the site where it will be used (including upon arrival at the site). For example, the transport vehicle 200 is equipped with a mixer 204 for agitating the ready-mixed concrete (see Figure 1), and the input information acquisition unit of the quality prediction device acquires information related to the power load value of the mixer 204 as at least part of the first information. Note that the mixer 204 in a transport vehicle 200 such as an agitator vehicle is also called a drum. As described above, at least part of the first information may be information related to the power load value of the mixer (114, 204) that manufactures ready-mixed concrete by mixing concrete materials, or agitates the ready-mixed concrete after manufacture.

In the above example, the ready-mix concrete is manufactured at a location other than the construction site (manufacturing equipment 100). The location where the ready-mix concrete is manufactured is not limited to this example. Concrete materials (for example, concrete materials excluding water) may be transported to the construction site by a transport vehicle, and the ready-mix concrete may be manufactured at the site. In this case, water may be added to the concrete materials transported by the transport vehicle using a mixer, and the materials may be mixed to produce the ready-mix concrete. The quality prediction device may predict the quality of the ready-mix concrete at the construction site when it is manufactured there. The input information acquisition unit of the quality prediction device may acquire information related to the power load value of the mixer installed on the transport vehicle when the ready-mix concrete is manufactured at the site, as at least part of the first information.

In one of the various examples described above, at least some of the matters described in the other examples may be combined.

[Verification of prediction results using predictive models]
Next, we will explain the results of verifying the prediction of quality information using a predictive model M that takes input information including the first and second pieces of information described above as input, using a dataset with known correct values.

(Impact of target quality)
The impact of second-order information, which indicates the target quality when ready-mixed concrete is used, was investigated. For this investigation, 2765 training datasets were prepared to build a predictive model using machine learning, and 789 evaluation datasets were prepared for evaluation, with known ground truth values. In the training and evaluation datasets, various input values included in the input information are associated with the ground truth values of the quality information. As the first-order information in the training and evaluation datasets, statistical quantities obtained from time-series data related to power load values were used. A predictive model M was constructed using the 2765 training datasets, taking input information including the first and second-order information as input. The predictive model M was constructed using multi-task learning to output slump, slump flow, and air content.

To examine the impact of secondary information on prediction accuracy, a comparison model was constructed using the same 2765 training datasets, but with the secondary information removed from the input data. The impact of secondary information on prediction accuracy was evaluated for slump, slump flow, and air volume. Using 789 evaluation datasets, the prediction results of both the prediction model M and the comparison model were compared with the ground truth values in the evaluation dataset. A ground truth was defined as a prediction result falling within a range obtained by adding a predetermined tolerance to the ground truth values in the evaluation dataset, and the accuracy rate was used as the evaluation metric. Figure 8 shows a graph illustrating the results of the examination of the impact of secondary information.

Regarding the slump verification results in Figure 8, when the tolerance is "±0.5 cm," datasets where the model's prediction result (predicted value) falls within the ±0.5 cm range of the correct value are considered correct, and the percentage of datasets judged as correct is shown as the accuracy rate (%). The accuracy rate (%) is calculated similarly for tolerances of "±1.0 cm," "±1.5 cm," "±2.0 cm," and "±2.5 cm." "Yes" indicates the evaluation result using prediction model M, which includes the second information (target quality) as input information, while "No" indicates the evaluation result using the comparison model, which does not include the second information as input information. It can be seen that the accuracy rate (%) when predicting slump using prediction model M is higher than when predicting slump using the comparison model.

Regarding the slump flow verification results in Figure 8, similar to the slump verification, the accuracy rate (%) was calculated for each tolerance of "±2.5 cm," "±3.0 cm," "±5.0 cm," and "±7.5 cm." It can be seen that, for slump flow as well, the accuracy rate (%) when predicting slump flow using prediction model M is higher than when predicting slump flow using the comparison model.

Regarding the verification results for air volume in Figure 8, similar to the slump verification, the accuracy rate (%) was calculated for each tolerance of "±0.3%", "±0.5%", "±1.0%", and "±1.5%". It can be seen that the accuracy rate (%) when predicting air volume using prediction model M is higher than when predicting air volume using the comparison model.

(Effects of Batch Normalization and Dropout)
This study investigated the effects of batch normalization (normalizing input data) and dropout (partially deleting data in the fully bonded layer) when constructing a model to predict the quality information of ready-mixed concrete. Predictive models M were constructed while varying the presence or absence of these operations and the deletion rate. The accuracy of acceptable discrimination (%) was then calculated for slump, slump flow, and air content. The validation used the aforementioned 2765 training datasets and the aforementioned 789 evaluation datasets. The accuracy of acceptable discrimination (%) for slump is shown in Table 1, for slump flow in Table 2, and for air content in Table 3.

In Tables 1-3, "BN" represents Batch Normalization (normalization operation), "DO" represents Dropout (an operation that removes some connections), and "P" represents the missing value in Dropout. "BN, DO not used" indicates that the predictive model M was constructed by training a neural network without Batch Normalization or Dropout. Columns with only "BN" indicate that the predictive model M was constructed by training a neural network that includes Batch Normalization but does not include Dropout. Columns with only "DO" indicate that the predictive model M was constructed by training a neural network that includes Dropout but does not include Batch Normalization. Columns with "BN + DO" indicate that the predictive model M was constructed by training a neural network that includes both Batch Normalization and Dropout.

The results shown in Tables 1 and 2 show that, regarding slump and slump flow, there is no significant difference in accuracy (%) between the use of Batch Normalization and Dropout. The results shown in Table 3 show that, regarding air volume, the accuracy (%) tends to be higher when Batch Normalization and Dropout are not used compared to when at least one of them is used.

(Influence of the type of weight update formula)
This study investigated the effect of different weight update formulas used in building models to predict the quality information of ready-mixed concrete. Predictive models M were constructed while varying the weight update formulas, and the accuracy rate (%) of acceptable air content was calculated. The validation used the aforementioned 2765 training datasets and the aforementioned 789 evaluation datasets. The validation results are shown in Tables 4 and 5 below.

In Table 5, "(×0.01)" indicates that each value in the input information of the training dataset was multiplied by 0.01 so that it fell within the range of -1 to 1, and then machine learning was performed. From the results shown in Tables 4 and 5, it can be seen that using one of the weight update formulas selected from the group consisting of Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov results in a higher accuracy (%) compared to using AdamW, Lars, Adadelta, and Sgd. Although the data is omitted here, when calculating the accuracy (%) for slump and slump flow for each update formula shown in Tables 4 and 5, the accuracy (%) was lower when using Lars and Sgd compared to when using the other update formulas.

(Effect of the type of loss function)
The effect of the type of loss function used when constructing a model to predict the quality information of ready-mixed concrete was investigated. Predictive models M were constructed while varying the type of loss function, and the accuracy (%) of the acceptable deviation was calculated for slump, slump flow, and air content. The Huber loss function used was the function shown in equation (3) above, with the value of δ set to 1.0. The ε-acceptable loss function used was the function shown in equation (4) above, with the value of ε set to 1.0. The verification used the 2765 training datasets and the 789 evaluation datasets mentioned above. The verification results are shown in Tables 6, 7, and 8 below.

The results shown in Tables 6, 7, and 8 indicate that using one loss function selected from the group consisting of the Huber loss function, mean absolute error, and ε-tolerance loss function results in a higher accuracy (%) compared to using the Mean Squared Error (MSE). The difference between the Mean Squared Error and the other loss functions is that the Mean Squared Error results in a larger loss value when the error is large. The results shown in Tables 6, 7, and 8 confirm that selecting a loss function where the loss value is always less than or equal to the absolute value of the error is preferable from the perspective of improving prediction accuracy.

(Influence of learning methods)
This study investigated the impact of learning methods when constructing models to predict the quality information of ready-mixed concrete. The accuracy (%) was compared between constructing individual prediction models M (one for each of the three quality characteristics) using single-task learning, and constructing a single prediction model M that outputs the predicted values for all three quality characteristics together using multi-task learning. The validation used the aforementioned 2765 training datasets and the aforementioned 789 evaluation datasets. The validation results are shown in Tables 9, 10, and 11 below.

In the multi-task learning column of Table 9, the accuracy (%) is calculated using the output value related to slump from the prediction model M. In the multi-task learning column of Table 10, the accuracy (%) is calculated using the output value related to slump flow from the prediction model M. In the multi-task learning column of Table 11, the accuracy (%) is calculated using the output value related to air volume from the prediction model M. From the results shown in Tables 9, 10, and 11, it can be seen that there is no significant difference in accuracy (%) whether single-task learning or multi-task learning is used.

[Summary of this disclosure]
The ready-mixed concrete quality prediction method described above includes an acquisition step of acquiring input information including first information relating to at least one of the following: the power load value of a mixer (114, 204) that manufactures ready-mixed concrete by mixing concrete materials or agitates ready-mixed concrete, vibrations caused by the dropping or flow of ready-mixed concrete, and an image of the ready-mixed concrete, and second information relating to the target quality when the ready-mixed concrete is used; a prediction model (M) pre-built by machine learning to output quality information indicating the quality of the ready-mixed concrete in response to the input information; and a prediction step of predicting the quality of the ready-mixed concrete based on the input information acquired in the acquisition step.

When manufacturing ready-mixed concrete (for example, when manufacturing takes place in a factory equipped with a manufacturing system (1)), a target quality for the ready-mixed concrete to be used (on-site) is specified, and the concrete is manufactured to meet that target quality. By using such target quality as one of the input pieces for a predictive model (M) constructed using machine learning, the accuracy of quality prediction has been improved, as described above. Acquiring target quality as information is simple. Therefore, the above method allows for improved quality prediction accuracy while avoiding the complexity of the work associated with quality prediction when using a predictive model (M).

In the ready-mix concrete quality prediction method described above, the quality of the ready-mix concrete may include one or more of the following qualities: slump, slump flow, and air content. In this case, it becomes easy to control whether the slump, slump flow, or air content of the manufactured ready-mix concrete can meet the target quality for use.

In the ready-mix concrete quality prediction method described above, the ready-mix concrete quality may include two or more of the following qualities: slump, slump flow, and air content. The prediction model (M) may be constructed to output two or more qualities depending on the input information. As mentioned above, it was found that there is no significant difference in prediction accuracy between single-task learning, which constructs a prediction model (M) that outputs only one quality, and multi-task learning, which constructs a prediction model (M) that outputs two or more qualities. By constructing the prediction model (M) using multi-task learning, as described above, it is possible to simplify the work required to construct the model while maintaining prediction accuracy.

In the ready-mix concrete quality prediction method described above, the prediction model (M) may be constructed using machine learning with a neural network that does not perform operations to normalize input information or operations to delete some connections in the fully connected layer. Generally, in machine learning using neural networks, it is considered preferable to perform operations to normalize input information and operations to delete some connections in the fully connected layer from the viewpoint of stabilizing learning and suppressing overfitting. However, as mentioned above, there is no significant difference in prediction accuracy whether or not these operations are performed, and for certain quality (air content), prediction accuracy tends to be higher when these operations are not performed. Therefore, by constructing the prediction model (M) as described above, it is possible to simplify the work required to construct the model while maintaining prediction accuracy.

In the ready-mix concrete quality prediction method described above, the second piece of information may include not only information indicating the target quality at the time of use of the ready-mix concrete, but also information indicating the target quality at the time of shipment. In factories where the manufacturing system (1) is installed, the target quality at shipment is set to meet the target quality at the time of use of the ready-mix concrete (i.e., the required quality at the time of acceptance). Therefore, the target quality at shipment correlates with the target quality at the time of use of the ready-mix concrete. In the above method, by further including the target quality at shipment in the input information of the prediction model (M), the accuracy of quality prediction can be further improved.

In the ready-mix concrete quality prediction method described above, the prediction model (M) may be constructed using machine learning with a neural network. In the machine learning using a neural network used to construct the prediction model (M), one of the following formulas may be used as the weight update formula: Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov. As mentioned above, it has been found that using these update formulas results in higher prediction accuracy compared to using other update formulas. Therefore, the above method can further improve prediction accuracy.

In the ready-mix concrete quality prediction method described above, the prediction model (M) may be constructed using machine learning with a neural network. In the machine learning using a neural network used to construct the prediction model (M), a function may be used as the loss function such that the loss value L(a) is always less than or equal to the absolute value of the error a. As mentioned above, it has been found that using such a loss function results in higher prediction accuracy compared to using other loss functions. Therefore, the above method can further improve prediction accuracy.

The quality prediction program described above is a program that causes a computer to execute the above quality prediction method. Because this quality prediction program can execute the above quality prediction method, it is possible to improve the accuracy of quality predictions while avoiding the complexity of the work associated with quality prediction when using a prediction model (M).

The quality prediction device (10) described above includes: an input information acquisition unit (24, 26) that acquires input information including first information relating to at least one of the following: the power load value of the mixer (114, 204) that manufactures or agitates ready-mixed concrete by mixing concrete materials, vibrations caused by the dropping or flow of ready-mixed concrete, and an image of the ready-mixed concrete; and second information relating to the target quality when the ready-mixed concrete is used; a prediction model (M) pre-built by machine learning to output quality information indicating the quality of the ready-mixed concrete in response to the input information; and a prediction calculation unit (28) that predicts the quality of the ready-mixed concrete based on the input information acquired by the input information acquisition unit (24, 26). In this quality prediction device (10), similar to the quality prediction method described above, when predicting quality using the prediction model (M), it is possible to improve the accuracy of quality prediction while avoiding the complexity of the work associated with quality prediction.

1…Manufacturing system, 10…Control device, 24…Operation information acquisition unit, 26…Target quality information acquisition unit, 28…Predictive calculation unit, 30…Model construction unit, M…Predictive model, 100…Manufacturing equipment, 114…Mixer, 114a…Agitation member, 114b…Mixer drive unit, 204…Mixer.

Claims (8)

An acquisition process that acquires input information including first information relating to the power load value of a mixer used to manufacture ready-mixed concrete by mixing concrete materials or to agitate ready-mixed concrete, and second information relating to the target quality when using ready-mixed concrete .
A prediction model pre-built by machine learning to output quality information indicating the quality of ready-mixed concrete in response to the input information, and a prediction step that predicts the quality of ready-mixed concrete based on the input information acquired in the acquisition step.
A construction step of constructing the prediction model using machine learning with a neural network,
Includes,
The quality of the ready-mixed concrete predicted in the aforementioned prediction process is either the quality of the ready-mixed concrete at the time of production or the quality of the ready-mixed concrete during transportation.
A method for predicting the quality of ready-mixed concrete, wherein in the machine learning using a neural network when constructing the prediction model in the construction process, one selected from the group consisting of Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov is used as the weight update formula. An acquisition process that acquires input information including first information relating to the power load value of a mixer used to manufacture ready-mixed concrete by mixing concrete materials or to agitate ready-mixed concrete, and second information relating to the target quality when using ready-mixed concrete .
A prediction model pre-built by machine learning to output quality information indicating the quality of ready-mixed concrete in response to the input information, and a prediction step that predicts the quality of ready-mixed concrete based on the input information acquired in the acquisition step.
A construction step of constructing the prediction model using machine learning with a neural network,
Includes,
The quality of the ready-mixed concrete predicted in the aforementioned prediction process is either the quality of the ready-mixed concrete at the time of production or the quality of the ready-mixed concrete during transportation.
In the machine learning using a neural network when constructing the prediction model in the construction process, a function is used as the loss function such that the loss value L(a) is always less than or equal to the absolute value of the error a, in a method for predicting the quality of ready-mixed concrete. The quality of ready-mixed concrete includes one or more of the following qualities: slump, slump flow, and air content.
A method for predicting the quality of ready-mixed concrete according to claim 1 or 2. The quality of ready-mixed concrete includes two or more of the following qualities: slump, slump flow, and air content.
The prediction model is constructed to output two or more types of quality in response to the input information.
The method for predicting the quality of ready-mixed concrete according to claim 3. In the aforementioned construction process, the prediction model is constructed by machine learning using a neural network that does not perform both the operation of normalizing the input information and the operation of deleting some of the connections in the fully connected layer.
A method for predicting the quality of ready-mixed concrete according to claim 1 or 2. A quality prediction program that causes a computer to execute the ready-mixed concrete quality prediction method described in claim 1 or 2. An input information acquisition unit acquires input information including first information relating to the power load value of a mixer that manufactures ready-mixed concrete by mixing concrete materials or stirs ready-mixed concrete, and second information relating to the target quality when using ready-mixed concrete .
A prediction model pre-built by machine learning to output quality information indicating the quality of ready-mixed concrete in response to the input information, and a prediction calculation unit that predicts the quality of ready-mixed concrete based on the input information acquired by the input information acquisition unit,
A model building unit that constructs the prediction model using machine learning with a neural network,
Equipped with,
The quality of the ready-mixed concrete predicted by the prediction calculation unit is either the quality of the ready-mixed concrete at the time of manufacture or the quality of the ready-mixed concrete during transportation.
The aforementioned model construction unit is a ready-mix concrete quality prediction device that, in machine learning using a neural network when constructing the prediction model, utilizes one selected from the group consisting of Adam, AdamBelief, Adamax, AdaBound, Adagrad, AMSGRAD, AMSBound, RMSprop, SgdW, Momentum, and Nesterov as the weight update formula. An input information acquisition unit acquires input information including first information relating to the power load value of a mixer that manufactures ready-mixed concrete by mixing concrete materials or stirs ready-mixed concrete, and second information relating to the target quality when using ready-mixed concrete .
A prediction model pre-built by machine learning to output quality information indicating the quality of ready-mixed concrete in response to the input information, and a prediction calculation unit that predicts the quality of ready-mixed concrete based on the input information acquired by the input information acquisition unit,
A model building unit that constructs the prediction model using machine learning with a neural network,
Equipped with,
The quality of the ready-mixed concrete predicted by the prediction calculation unit is either the quality of the ready-mixed concrete at the time of manufacture or the quality of the ready-mixed concrete during transportation.
The aforementioned model construction unit is a ready-mix concrete quality prediction device that, in machine learning using a neural network when constructing the prediction model, utilizes a function as the loss function such that the loss value L(a) is always less than or equal to the absolute value of the error a.