Fresh concrete evaluation system and fresh concrete evaluation method

Abstract

To provide a fresh concrete evaluation system 3 and a fresh concrete evaluation method that enable accurate and easy evaluation on whether or not fresh concrete has the properties suitable for placement.SOLUTION: A fresh concrete evaluation system 3 includes: concrete identification means constructed by a first mechanical learning model 35 which identifies and outputs a slump test body 5 of image data 38; viscosity evaluation means constructed by a second mechanical learning model 36 which evaluates and outputs viscosity of fresh concrete; surface roughness evaluation means constructed by a third mechanical learning model 37 which evaluates and outputs surface roughness of fresh concrete; determination means for determining quality of fresh concrete properties, on the basis of output information of the second mechanical learning model 36 and the third mechanical learning model 37; and output means for outputting transmission information in which the quality of fresh concrete properties is associated with the image data 38.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to a fresh concrete evaluation system and a fresh concrete evaluation method for evaluating whether fresh concrete has properties suitable for pouring, for example. [Background technology]

[0002] Fresh concrete shipped from the factory is inspected for quality control by taking samples at the site where it is unloaded from the agitator truck. For example, one of the inspections at the time of unloading is a fresh concrete slump test conducted in accordance with JISA1101 "Concrete Slump Test Method" of the Japanese Industrial Standards, and a slump flow test conducted in accordance with JISA1150 "Concrete Slump Flow Test Method".

[0003] However, even if two slump test specimens have the same slump value, the fluidity of the cement paste and the state of material separation may differ, so it is known that the slump value and slump flow value alone cannot adequately evaluate whether the fresh concrete has properties suitable for pouring.

[0004] For this reason, it is conceivable that workers would check the fluidity of the cement paste and the state of material separation to determine whether the fresh concrete has properties suitable for pouring. However, in this case, not only would the judgment of the properties of the fresh concrete depend on the experience of the worker, but there would also be problems with increasing the amount of work required by the worker.

[0005] Therefore, various techniques have been proposed for evaluating the properties of fresh concrete at the unloading site by using imaging information obtained by imaging a slump test specimen in a slump test or a slump flow test.

[0006] For example, Patent Document 1 discloses a system for evaluating material segregation of fresh concrete based on image information obtained by imaging a slump test specimen from directly above in a slump test or a slump flow test.

[0007] In this patent document, a slump test specimen is extracted from the image information and binarized, and then material segregation of fresh concrete is evaluated based on the variation in the number of low-brightness pixels.

[0008] However, since Patent Document 1 only evaluates the material segregation of fresh concrete, it cannot fully evaluate whether the fresh concrete has properties suitable for pouring, and there is room for improvement. [Prior art documents] [Patent documents]

[0009] [Patent Document 1] JP 2020-201231 A Summary of the Invention [Problem to be solved by the invention]

[0010] In view of the above-mentioned problems, the present invention aims to provide a fresh concrete evaluation system and a fresh concrete evaluation method that can accurately and easily evaluate whether fresh concrete has properties suitable for pouring. [Means for solving the problem]

[0011] The present invention is a fresh concrete evaluation system that evaluates the properties of fresh concrete based on imaging information obtained by imaging a slump test specimen in a slump test and / or a slump flow test, and is characterized in that it includes a concrete identification means constructed with a first machine learning model that takes the imaging information as input information and identifies and outputs the slump test specimen in the imaging information, a viscosity evaluation means constructed with a second machine learning model that takes the output information of the first machine learning model as input information and evaluates and outputs the viscosity of the fresh concrete, a surface roughness evaluation means constructed with a third machine learning model that takes the output information of the first machine learning model as input information and evaluates and outputs the surface roughness of the fresh concrete, a judgment means for judging the quality of the properties of the fresh concrete based on the output information of the second machine learning model and the output information of the third machine learning model, and an output means for outputting output information that associates at least the quality of the properties of the fresh concrete with the imaging information.

[0012] The present invention also provides a fresh concrete evaluation method for evaluating the properties of fresh concrete based on imaging information obtained by imaging a slump test specimen in a slump test and / or a slump flow test, and is characterized in that the method includes a concrete identification process in which a concrete identification means constructed with a first machine learning model takes the imaging information as input information and identifies and outputs the slump test specimen from the imaging information, a viscosity evaluation process in which a viscosity evaluation means constructed with a second machine learning model takes the output information of the first machine learning model as input information and evaluates and outputs the viscosity of the fresh concrete, a surface roughness evaluation process in which a surface roughness evaluation means constructed with a third machine learning model takes the output information of the first machine learning model as input information and evaluates and outputs the surface roughness of the fresh concrete, a determination process in which a determination means determines whether the properties of the fresh concrete are good or bad based on the output information of the second machine learning model and the output information of the third machine learning model, and an output process in which an output means outputs output information that associates at least the good or bad properties of the fresh concrete with the imaging information.

[0013] The above-mentioned image information refers to still images or moving images captured during a slump test or a slump flow test. Evaluating the viscosity of fresh concrete and evaluating the surface roughness of fresh concrete refer to evaluating the degree of viscosity and the degree of surface roughness by quantifying them, or evaluating them in categories according to the degree of viscosity and the degree of surface roughness.

[0014] According to this invention, the viscosity of fresh concrete and the surface roughness of fresh concrete are evaluated based on the slump test specimen of the image information identified by the concrete identification means, so that the properties of fresh concrete can be evaluated in more detail than when only material segregation is evaluated.

[0015] In this case, imaging information from slump tests and slump flow tests is used as input information to be input into the first machine learning model of the concrete identification means, so the fresh concrete evaluation system does not need to prepare and image samples other than slump test specimens at the unloading site. As a result, the fresh concrete evaluation system can reduce the burden on workers at unloading sites.

[0016] Furthermore, since the viscosity evaluation means for evaluating viscosity is constructed using a second machine learning model, and the surface roughness evaluation means for evaluating surface roughness is constructed using a third machine learning model, the fresh concrete evaluation system can evaluate the viscosity and surface roughness of fresh concrete more efficiently and with stable accuracy compared to when an operator evaluates based on their experience.

[0017] Furthermore, because the fresh concrete evaluation system is equipped with a judgment means for judging whether the properties of fresh concrete are good or bad, it saves workers the trouble of judging whether the properties of fresh concrete are good or bad, and can reliably judge whether the properties of fresh concrete are good or bad.

[0018] Therefore, the fresh concrete evaluation system and the fresh concrete evaluation method using the same can accurately and easily evaluate whether the fresh concrete has properties suitable for pouring.

[0019] Therefore, the fresh concrete evaluation system and fresh concrete evaluation method can enable 100% evaluation of the properties of fresh concrete at the site where the fresh concrete is unloaded, similar to slump tests and slump flow tests.

[0020] As an aspect of the present invention, the first machine learning model of the concrete identification means may be configured to directly identify the slump test specimen from the imaging information including the foreground and background. According to this configuration, the slump test specimen can be more efficiently identified from the image data than when the slump test specimen is identified by deleting the foreground and background of the image data.

[0021] As another aspect of the present invention, the second machine learning model of the viscosity evaluation means is configured to output a viscosity evaluation point indicating the degree of viscosity of the fresh concrete based on the slump test specimen of the imaging information, and the third machine learning model of the surface roughness evaluation means is configured to output a surface roughness evaluation point indicating the degree of surface roughness of the fresh concrete based on the slump test specimen of the imaging information, and the judging means is configured to judge whether the viscosity of the fresh concrete is good or bad based on the viscosity evaluation point, and to judge whether the surface roughness of the fresh concrete is good or bad based on the surface roughness evaluation point.

[0022] According to this configuration, the properties of fresh concrete can be evaluated in more detail, so that the quality of the properties of fresh concrete can be determined with greater accuracy. Furthermore, by displaying output information, to which, for example, a viscosity evaluation score and a surface roughness evaluation score are assigned, on a display means, the fresh concrete evaluation system can objectively convey the properties of the fresh concrete to the worker.

[0023] As another aspect of the present invention, the output means may be configured to output the output information including the viscosity evaluation point associated with a viscosity classification that classifies the degree of viscosity of the fresh concrete, and the surface roughness evaluation point associated with a surface roughness classification that classifies the degree of surface roughness of the fresh concrete.

[0024] The viscosity classifications refer to, for example, low viscosity, appropriate viscosity, and high viscosity. The surface roughness classification means, for example, rough surface roughness, appropriate surface roughness, smooth surface roughness, etc. In addition, smooth surface roughness can also be said to be fine surface roughness.

[0025] According to this configuration, by transmitting output information to, for example, a production factory located away from the unloading site, it is possible to more objectively communicate to a production manager involved in the production of fresh concrete the properties of fresh concrete that are difficult to convey using only imaging information, viscosity evaluation points, and surface roughness evaluation points.

[0026] Furthermore, since the imaging information of the output information is associated with a viscosity classification and a surface roughness classification, the fresh concrete evaluation system can use the output information to train workers who have little experience in judging the properties of fresh concrete.

[0027] In another aspect of the present invention, the average value of evaluation points obtained by a plurality of engineers to evaluate the slump test specimen in the imaging information according to the degree of viscosity is defined as a viscosity average value, and the second machine learning model of the viscosity evaluation means is a learning model that uses the imaging information to which the viscosity average value has been assigned as teacher data, and the average value of evaluation points obtained by a plurality of engineers to evaluate the slump test specimen in the imaging information according to the degree of surface roughness is defined as a surface roughness average value, and the third machine learning model of the surface roughness evaluation means is a learning model that uses the imaging information to which the surface roughness average value has been assigned as teacher data.

[0028] According to this configuration, highly accurate second machine learning models and third machine learning models can be constructed using training data that takes into account the experience of multiple engineers. As a result, the fresh concrete evaluation system can bring the degree of viscosity evaluated by the second machine learning model closer to the degree of viscosity determined by an engineer, compared to when, for example, information indicating the characteristics due to differences in viscosity is used as training data.

[0029] Similarly, the fresh concrete evaluation system can bring the degree of surface roughness assessed by the third machine learning model closer to the degree of surface roughness determined by an engineer, compared to when, for example, information indicating the characteristics due to differences in surface roughness is used as training data. Therefore, the fresh concrete evaluation system can evaluate the properties of fresh concrete with even greater accuracy.

[0030] In another aspect of the present invention, the evaluation point given by the engineer according to the degree of viscosity may be one of the evaluation points associated with each viscosity classification indicating a classification of the degree of viscosity, and the evaluation point given by the engineer according to the degree of surface roughness may be one of the evaluation points associated with each surface roughness classification indicating a classification of the degree of surface roughness.

[0031] The viscosity classifications refer to, for example, low viscosity, appropriate viscosity, and high viscosity. The surface roughness classification means, for example, rough surface roughness, appropriate surface roughness, smooth surface roughness, etc. In addition, smooth surface roughness can also be said to be fine surface roughness. According to this configuration, it is possible to improve the accuracy of the teacher data used for the second machine learning model and the teacher data used for the third machine learning model.

[0032] Specifically, when an engineer evaluates the slump test specimen from the image information to generate training data, the engineer only needs to determine which viscosity category the viscosity of the slump test specimen falls into and which surface roughness category the surface roughness of the slump test specimen falls into.

[0033] Therefore, compared to a case where an engineer directly assigns an evaluation score to a slump test specimen based on image information, the fresh concrete evaluation system can reduce variation in the evaluation scores given by engineers to the slump test specimen based on image information.

[0034] As a result, the fresh concrete evaluation system can construct a second machine learning model and a third machine learning model using training data with stable accuracy, thereby enabling the viscosity and surface roughness of fresh concrete to be evaluated with stable accuracy.

[0035] In another embodiment of the present invention, the viscosity classification may have four levels: a suitable state having appropriate fluidity, a state having a lower fluidity than the suitable state, a state having a higher fluidity than the suitable state, and a state having an even higher fluidity than the high fluidity state; and the surface roughness classification may have three levels: a suitable state in which the coarse aggregate is not noticeable, a state in which the coarse aggregate is clearly visible, and an intermediate state between the suitable state and the state in which the coarse aggregate is clearly visible.

[0036] According to this configuration, compared to when the viscosity classification and surface roughness classification are each only at two levels, good or bad, the teacher data used for the second machine learning model and the teacher data used for the third machine learning model can be teacher data suitable for evaluating the properties of fresh concrete in detail.

[0037] As a further aspect of the present invention, there may be provided a notification means for notifying a poor property of the fresh concrete when the determination means determines that the fresh concrete is defective. The above-mentioned notification means refers to notification means provided in a terminal owned by the worker, or notification means provided in another terminal connected via a communication line.

[0038] According to this configuration, for example, by notifying a terminal held by the worker, the worker performing the slump test can be promptly notified of poor properties of the fresh concrete.

[0039] Alternatively, by reporting to a terminal at the production plant connected via a communication line, the fresh concrete evaluation system can inform the production manager at the production plant of poor properties of the fresh concrete and easily encourage inspection of the production process, etc. As a result, the fresh concrete evaluation system can prevent fresh concrete that does not satisfy the desired properties from being used for pouring. Effect of the Invention

[0040] According to the present invention, it is possible to provide a fresh concrete evaluation system and a fresh concrete evaluation method that can accurately and easily evaluate whether fresh concrete has properties suitable for pouring. [Brief description of the drawings]

[0041] [Figure 1] Schematic diagram showing an outline of a concrete construction site. [Diagram 2] FIG. 1 is a diagram showing the configuration of a fresh concrete evaluation system. [Diagram 3] FIG. 1 is a block diagram showing the internal configuration of a fresh concrete evaluation system. [Figure 4] FIG. 2 is an explanatory diagram for explaining an outline of image data. [Diagram 5] FIG. 1 is an explanatory diagram illustrating an outline of a first machine learning model. [Figure 6] FIG. 1 is an explanatory diagram illustrating an outline of a second machine learning model. [Figure 7] FIG. 13 is an explanatory diagram outlining a third machine learning model. [Figure 8] FIG. 2 is an explanatory diagram illustrating the properties of concrete. [Figure 9] FIG. 2 is a sequence diagram showing the processing operation of the fresh concrete evaluation system. [Figure 10] 4 is a flowchart showing the processing operation of a property evaluation process. [Figure 11] FIG. 13 is an explanatory diagram for explaining an outline of a judgment result screen. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0042] An embodiment of the present invention will now be described with reference to the drawings. In this embodiment, a fresh concrete evaluation system 3 for evaluating whether fresh concrete has properties suitable for pouring will be described with reference to Figs. 1 to 8.

[0043] FIG. 1 shows a schematic diagram of a concrete construction site, FIG. 2 shows a configuration diagram of the fresh concrete evaluation system 3, and FIG.

[0044] Furthermore, Figure 4 shows an explanatory diagram outlining the image data 38, Figure 5 shows an explanatory diagram outlining the first machine learning model 35, and Figure 6 shows an explanatory diagram outlining the second machine learning model 36. In addition, FIG. 7 shows an explanatory diagram outlining the third machine learning model 37, and FIG. 8 shows an explanatory diagram explaining the properties of concrete.

[0045] First, as shown in FIG. 1, fresh concrete (hereinafter referred to as concrete) is loaded onto an agitator truck 2 from a production factory 1 and transported to an unloading site while being agitated.

[0046] Then, when the agitator truck 2 arrives at the unloading site, the worker M unloads the concrete from the agitator truck 2 and checks whether the unloaded concrete is of a quality suitable for pouring.

[0047] Specifically, Worker M conducts a slump test based on the Japanese Industrial Standard JISA1101 "Method of slump test for concrete" and a slump flow test based on the Japanese Industrial Standard JISA1150 "Method of slump flow test for concrete" to check the quality of the unloaded concrete.

[0048] Furthermore, following the above-mentioned slump test and slump flow test, the worker M uses the fresh concrete evaluation system 3 to evaluate and determine whether the properties of the concrete are suitable for pouring.

[0049] As shown in Figures 1 and 2, such a fresh concrete evaluation system 3 includes a production management terminal 10 located in a production plant 1 that produces concrete, a worker terminal 20 located at the site where the concrete is unloaded, and a server 30 connected to the production management terminal 10 and the worker terminal 20 via a communication line 4.

[0050] As shown in FIG. 2, the production management terminal 10 is a terminal located in the production management office 1a of the production plant 1, and is used by a production manager who manages the concrete production process, concrete production volume, and concrete quality.

[0051] In addition, the worker terminal 20 is a mobile phone type terminal placed at an unloading site away from the production factory 1, as shown in Figure 2, and is used by a worker M involved in the slump test and / or slump flow test.

[0052] The server 30 is placed at a location away from the production factory 1 and the unloading site, and is communicatively connected to the production management terminal 10 and the worker terminal 20 via a communication line 4. This server 30 has a function of sending and receiving various information with the production management terminal 10, a function of sending and receiving various information with the worker terminal 20, and a function of analyzing image data 38 acquired from the worker terminal 20 to evaluate the properties of the concrete.

[0053] In detail, as shown in FIG. 3, the production management terminal 10 comprises an operation reception unit 11 that receives various operations from the production manager, a display unit 12 that displays various information, a memory unit 13 that stores various information, a line connection unit 14 that connects to the communication line 4, and a control unit 15 that controls the operation of these units.

[0054] Specifically, as shown in FIG. 2, the operation reception unit 11 is composed of, for example, a keyboard 11a and a mouse 11b, and has a function of receiving input operations by the production manager and a function of outputting information indicating the received input contents to the control unit 15. The display unit 12 is configured with, for example, a liquid crystal display, as shown in FIG. 2, and has a function of displaying various information in response to a control signal from the control unit 15.

[0055] The storage unit 13 is composed of a hard disk or a non-volatile memory, and has a function of writing and storing various information, and a function of reading out various information. Various programs related to concrete production management, programs related to sending and receiving various information to and from the server 30, and the like are stored in the storage unit 13.

[0056] The line connection unit 14 is constituted by, for example, a wired LAN board, and has a function of connecting to the communication line 4 and a function of transmitting and receiving various information via the communication line 4. The control unit 15 is composed of hardware such as a CPU and a memory, and software such as a control program.

[0057] This control unit 15 has a processing function related to the exchange of various information with the server 30, a processing function related to the exchange of various signals with the operation reception unit 11, the display unit 12, the memory unit 13 and the line connection unit 14, and a function to control the operation of each unit connected via a specified bus.

[0058] As shown in FIG. 3, the worker terminal 20 includes an operation display unit 21 that displays various information and accepts various operations from the worker M, a terminal memory unit 22 that stores various information, a camera 23 that captures images of the outside, a line connection unit 24 that connects to the communication line 4, and a terminal control unit 25 that controls the operations of these units.

[0059] Specifically, the operation display unit 21 is composed of, for example, a touch panel display, and has the function of accepting various operations from the user and the function of displaying various information necessary for imaging operations of the slump test specimen 5 (see Figure 1) and for analyzing image data 38 (see Figure 4) obtained by imaging the slump test specimen 5.

[0060] The terminal storage unit 22 is composed of a hard disk or a non-volatile memory, and has a function of writing and storing various information and a function of reading out various information. The terminal storage unit 22 stores a processing program related to the evaluation of the properties of concrete, etc.

[0061] The camera 23 has a function of capturing an image of a subject based on a control signal from the terminal control unit 25, and a function of outputting an image signal capturing the image of the subject to the terminal control unit 25. The line connection unit 24 is formed of, for example, a wireless communication module, and has a function of connecting to the communication line 4 and a function of transmitting and receiving various information via the communication line 4.

[0062] The terminal control unit 25 is composed of hardware such as a CPU and a memory, and software such as a control program. This terminal control unit 25 has processing functions related to the exchange of various information with the server 30, processing functions related to the exchange of various signals with the operation display unit 21, the terminal memory unit 22, the camera 23 and the line connection unit 24, and a function to control the operation of each unit connected via a specified bus.

[0063] As shown in FIG. 3, the server 30 includes a line connection unit 31 that connects to the communication line 4, a server storage unit 32 that stores various information, and a server control unit 33 that controls the operations of these units. Specifically, the line connection unit 31 is configured by, for example, a wired LAN board, and has a function of connecting to the communication line 4 and a function of transmitting and receiving various information via the communication line 4.

[0064] The server storage unit 32 is composed of a hard disk or a non-volatile memory, and has the function of writing and storing various information, and the function of reading out various information. As shown in Figure 3, this server memory unit 32 stores project data 34 set and registered for each construction site, a property evaluation program (not shown) that evaluates and determines the properties of concrete, and a first machine learning model 35 that analyzes image data 38 of the project data 34 (described later) to identify slump test specimens 5.

[0065] Furthermore, the server memory unit 32 stores a second machine learning model 36 that analyzes and evaluates the degree of viscosity (hereinafter referred to as the viscosity degree) of the slump test specimen 5, and a third machine learning model 37 that analyzes and evaluates the degree of surface roughness (hereinafter referred to as the surface roughness degree) of the slump test specimen 5.

[0066] The server control unit 33 is composed of hardware such as a CPU and a memory, and software such as a control program. This server control unit 33 has processing functions related to the exchange of various information with the production management terminal 10 and the worker terminal 20, processing functions related to the exchange of various signals with the line connection unit 31 and the server memory unit 32, and a function to control the operation of each unit connected via a specified bus.

[0067] Here, the above-mentioned project data 34, first machine learning model 35, second machine learning model 36, and third machine learning model 37 will be described in detail. First, the project data 34 is a data group in which various information indicating the construction location, the scheduled start date and time of pouring, the scheduled end date and time of pouring, and the completion date and time of pouring, as well as image data 38, a viscosity evaluation image 39, and a surface roughness evaluation image 40 are associated with information indicating the project name.

[0068] 4, image data 38 of this project data 34 is an image capturing the state of a slump test and / or a slump flow test, in which at least a slump test specimen 5 is captured as a subject, and a slump plate 6 and a slump scale 7 are captured as subjects other than the slump test specimen 5. This image data 38 includes information indicating the date and time of capture, information indicating the name of the test specimen, and the like.

[0069] The viscosity evaluation image 39 of the project data 34 is image data output by a second machine learning model 36 described later. This viscosity evaluation image 39 is provided with a viscosity evaluation point according to the degree of viscosity described later, and a viscosity label indicating a viscosity category corresponding to the viscosity evaluation point.

[0070] As will be described in detail later, the viscosity classification is a classification of the degree of viscosity of the slump test specimen 5 into four categories: "cement paste running ahead," "tending to separate," "adequate viscosity," and "excessive viscosity."

[0071] Moreover, the surface roughness evaluation image 40 of the project data 34 is image data output by a third machine learning model 37 described later. This surface roughness evaluation image 40 is provided with a surface roughness evaluation score according to the degree of surface roughness described later, and a surface roughness label indicating a surface roughness category corresponding to the surface roughness evaluation score. As will be described in detail later, the surface roughness classification is a classification of the degree of surface roughness of the slump test specimen 5 into three categories: "rough", "smooth" and "intermediate".

[0072] Moreover, the first machine learning model 35 is an image recognition algorithm using a neural network consisting of, for example, an input layer 35a to which input information is input, an intermediate layer 35b, and an output layer 35c to which output information is output, as shown in FIG. 5.

[0073] The first machine learning model 35 is constructed to analyze image data 38 to identify each object in the image data 38 and to assign a name label to each identified object indicating what it is.

[0074] Specifically, as shown in FIG. 5, the first machine learning model 35 is constructed so that image data 38 is used as input information, and an identification image 38A that detects each object in the image data 38, such as the slump plate 6, the slump scale 7, and the slump test specimen 5, is obtained as output information for the input information.

[0075] Such a first machine learning model 35 is constructed by repeatedly learning a plurality of learning images 51, which are image data different from the image data 38 that is the input information, as training data T1 for the input information. At this time, the first machine learning model 35 learns by repeatedly adjusting the weighting and the decision threshold based on the training data T1 so that the output information for the input information becomes the classified image 38A.

[0076] As shown in FIG. 5, in the learning image 51 that is the teacher data T1, at least the shapes of a slump test piece 51a, a slump plate 51b, and a slump measurement scale 51c are identified in advance. Furthermore, the learning image 51 is assigned a name label (not shown) indicating a slump plate 51b, a name label (not shown) indicating a slump measurement scale 51c, and a name label (not shown) indicating a slump test specimen 51a to the identified shapes.

[0077] In addition, as shown in Figure 6, the second machine learning model 36 is an image recognition algorithm using a neural network consisting of an input layer 36a to which, for example, a recognition image 38A output by the first machine learning model 35 is input, an intermediate layer 36b, and an output layer 36c to which a viscosity evaluation image 39 is output.

[0078] As shown in Figure 6, this second machine learning model 36 is constructed to analyze the degree of material separation from the horizontal spread and shape deformation of the slump test specimen 5 in the identification image 38A, and to assign a viscosity evaluation point according to the degree of viscosity of the concrete and a viscosity label indicating the viscosity category corresponding to the viscosity evaluation point, based on the degree of material separation.

[0079] Specifically, the second machine learning model 36 is constructed to take the identification image 38A as input information, and to obtain, as output information in response to the input information, a viscosity evaluation image 39 in which an image of a region including the slump test specimen 5 is cut out from the identification image 38A, and a viscosity evaluation point according to the degree of viscosity and a viscosity label indicating the viscosity classification are assigned to the image.

[0080] As an example, the second machine learning model 36 is constructed to calculate a viscosity evaluation score ranging from 0 to 15 points depending on the degree of viscosity. Furthermore, the second machine learning model 36 is constructed so that when the viscosity evaluation score is between 0 and 7 points, the viscosity classification is “slightly separated”, when the viscosity evaluation score is between 7 and 12 points, the viscosity classification is “appropriate viscosity”, and when the viscosity evaluation score is between 12 and 15 points, the viscosity classification is “excessive viscosity”.

[0081] In addition, the viscosity classification of "slightly separated" indicates a state in which the cement paste components have high fluidity and the coarse aggregate and cement paste have separated, "appropriate viscosity" indicates a state in which the cement paste components have appropriate fluidity and viscosity, and "excessive viscosity" indicates a state in which the fluidity of the cement paste components is extremely low, the base of the slump test specimen 5 does not spread, and the slump test specimen 5 has deformed into a rounded shape.

[0082] Such a second machine learning model 36 is constructed by repeatedly learning a plurality of learning images 52, which are image data different from the image data 38, as training data T2 for the input information, as shown in FIG. 6.

[0083] At this time, the second machine learning model 36 learns by repeatedly adjusting the weighting and judgment threshold based on the training data T2 so that the output information for the input information becomes a viscosity evaluation image 39 to which a viscosity evaluation point and a viscosity classification have been assigned.

[0084] In addition, the training images 52, which are the teacher data T2, are assigned an average viscosity value obtained by averaging the viscosity evaluation scores of the training images 52 by multiple engineers who are able to judge the degree of viscosity of concrete, and a viscosity label indicating the viscosity classification. The viscosity evaluation score of this learning image 52 is determined by an engineer such as an operator based on Table 1, which is a viscosity degree judgment table.

[0085] [Table 1] More specifically, the viscosity degree judgment table is a table that associates the viscosity degree classified into four levels, "cement paste running ahead," "tending to separate," "adequate viscosity," and "excessive viscosity," as shown in Table 1, with an evaluation score, with "10" being the best score.

[0086] Among these four levels of viscosity classification, "cement paste running" is a state in which the fluidity of the cement paste components is high and the coarse aggregate and cement paste are extremely separated, as shown in Figure 8. This "cement paste running" is associated with an evaluation score of "0".

[0087] In addition, as shown in Fig. 8, "separated" is a state in which the fluidity of the cement paste components is lower than that of "cement paste run-off" and the bottom of the slump test specimen 5 spreads out. This "separated" is assigned an evaluation score of "5".

[0088] Also, as shown in Figure 8, "adequate viscosity" is an intermediate state between the "slightly separated" state and the "excessive viscosity" described below, and is a state having appropriate fluidity and viscosity. This "adequate viscosity" is associated with the best possible evaluation score of "10".

[0089] Moreover, as shown in Fig. 8, "excessive viscosity" is a state in which the fluidity of the cement paste components is extremely low, the bottom of the slump test specimen 5 does not spread, and the slump test specimen 5 is deformed into a rounded shape. This "excessive viscosity" is associated with an evaluation score of "15."

[0090] Using such a viscosity degree judgment table, an engineer determines which viscosity category in the viscosity degree judgment table of Table 1 the viscosity degree of the slump test specimen 52a of the learning image 52 falls into, and determines the evaluation score for the corresponding viscosity category as the viscosity evaluation score of the learning image 52.

[0091] Moreover, as shown in FIG. 7, the third machine learning model 37 is an image recognition algorithm using a neural network consisting of an input layer 37a to which, for example, a recognition image 38A output by the first machine learning model 35 is input, an intermediate layer 37b, and an output layer 37c to which a surface roughness evaluation image 40 is output.

[0092] As shown in Figure 7, this third machine learning model 37 is constructed to analyze the surface roughness from the surface texture of the slump test specimen 5 in the identification image 38A, and to assign a surface roughness evaluation point according to the surface roughness degree and a surface roughness label indicating the surface roughness category corresponding to the surface roughness evaluation point, assuming the degree of surface roughness to be the surface roughness degree of concrete.

[0093] Specifically, the third machine learning model 37 is constructed to take the identification image 38A as input information, and to obtain, as output information in response to the input information, a surface roughness evaluation image 40 in which an image of a region including the slump test specimen 5 is cut out from the identification image 38A, and a surface roughness evaluation point according to the degree of surface roughness and a surface roughness label indicating the surface roughness classification are assigned to the image.

[0094] As an example, the third machine learning model 37 is constructed to calculate a surface roughness evaluation score ranging from 0 to 10 points depending on the degree of surface roughness. Furthermore, the third machine learning model 37 is constructed to classify the surface roughness as “rough” when the surface roughness evaluation score is between 0 and 3 points, to classify the surface roughness as “medium” when the surface roughness evaluation score is between 3 and 6 points, and to classify the surface roughness as “smooth” when the surface roughness evaluation score is between 6 and 10 points.

[0095] In terms of surface roughness classification, "rough" indicates a state in which the mortar covering the coarse aggregate is thin and the shape of the coarse aggregate is clear, "smooth" indicates a state in which the coarse aggregate is moderately covered with mortar and is not noticeable, and "intermediate" indicates a state intermediate between "rough" and "smooth."

[0096] Such a third machine learning model 37 is constructed by repeatedly learning a plurality of learning images 53, which are image data different from the image data 38, as training data T3 for the input information, as shown in FIG. 7. At this time, the third machine learning model 37 learns by repeatedly adjusting the weighting and judgment thresholds based on the training data T3 so that the output information for the input information becomes a surface roughness evaluation image 40 to which a surface roughness evaluation score and a surface roughness classification have been assigned.

[0097] In addition, the training image 53, which is the teacher data T3, is assigned an average surface roughness value obtained by averaging the surface roughness evaluation scores of the training image 53 by multiple engineers who are able to judge the degree of surface roughness of concrete, and a surface roughness label indicating the surface roughness classification. The surface roughness evaluation score of this learning image 53 is determined by an engineer such as an operator based on the surface roughness degree judgment table shown in Table 2.

[0098] [Table 2] More specifically, the surface roughness rating table is a table that associates surface roughness classification into three levels, "rough," "medium," and "smooth," with evaluation scores, with "10" being the best, as shown in Table 2.

[0099] Of these three levels of surface roughness classification, "rough" is a state in which the mortar covering the coarse aggregate is thin and the shape of the coarse aggregate is clearly visible, as shown in Figure 8. This "rough" is associated with an evaluation score of "0."

[0100] Also, as shown in Figure 8, "Intermediate" is an intermediate state between "Rough" and "Smooth" (described later), where the mortar is thicker than "Rough" and the shape of the coarse aggregate is not clear. This "Intermediate" is associated with an evaluation score of "5."

[0101] Furthermore, "smooth" refers to a state in which the coarse aggregate is adequately covered with mortar and is not noticeable, as shown in Figure 8. This "smooth" is associated with the best possible evaluation score of "10."

[0102] Using such a surface roughness degree judgment table, an engineer determines which surface roughness category in the surface roughness degree judgment table in Table 2 the surface roughness degree of the slump test specimen 53a in the learning image 53 falls into, and determines the evaluation score for the corresponding surface roughness category as the surface roughness evaluation score for the learning image 53.

[0103] Next, the processing operation in the fresh concrete evaluation system 3 having the above-mentioned configuration will be described with reference to Figs. 9 shows a sequence diagram of the processing operation of the fresh concrete evaluation system 3, FIG. 10 shows a flowchart of the processing operation of the property evaluation process, and FIG. 11 shows an explanatory diagram outlining the judgment result screen 60.

[0104] First, when worker M operates the worker terminal 20 to execute a processing program related to concrete property evaluation, the terminal control unit 25 of the worker terminal 20 displays a menu screen (not shown) on the operation display unit 21 to guide users through various operations.

[0105] This menu screen displays, for example, guidance messages prompting the input of the project name and test specimen name, a camera button for starting to capture images of the slump test specimen 5 after the slump test and / or slump flow test, and a registration button for registering image data 38 captured of the slump test specimen 5 to the server 30.

[0106] When the menu screen is displayed on the operation display unit 21 of the operator terminal 20, the operator M inputs the project name and the test specimen name according to the guidance on the menu screen, and then presses the camera button.

[0107] When the terminal control unit 25 detects that the worker M has pressed the camera button, it starts the image capturing process (step S101) as shown in FIG. 9, and displays on the operation display unit 21 a screen (not shown) prompting the user to capture an image of the slump test specimen 5 after the slump test and / or the slump flow test.

[0108] At this time, the worker M follows the guidance displayed on the operation display unit 21 and holds the worker terminal 20 so as to capture an image of the slump test specimen 5 on the slump slab 6 in a direction other than directly above, preferably in a direction that clearly shows the extent to which the slump test specimen 5 spreads and the degree to which it collapses in the height direction (see FIG. 1). After that, the worker M presses the shutter button displayed on the operation display unit 21.

[0109] When detecting that the worker M has pressed the shutter button, the camera 23 captures an image of the subject based on a control signal from the terminal control unit 25, and then outputs an image capture signal to the terminal control unit 25. The terminal control unit 25 acquires the imaging signal from the camera 23 , generates image data 38 that associates the imaging signal with the test piece name and the imaging date and time, and stores the image data 38 in the terminal storage unit 22 .

[0110] Thereafter, when detecting that the worker M has pressed the registration button, the terminal control unit 25 starts an image transmission process for transmitting the image data 38 to the server 30, as shown in FIG. 9 (step S102). At this time, the terminal control unit 25 transmits the project name to the server 30 in association with the image data 38 .

[0111] On the other hand, when the server 30 receives the image data 38 from the worker terminal 20, as shown in FIG. 9, the server 30 starts a property evaluation process for analyzing the image data 38 and evaluating the properties of the concrete (step S103).

[0112] Specifically, when the property evaluation process is started, the server control unit 33 of the server 30 associates the image data 38 acquired from the worker terminal 20 with the project data 34 corresponding to the project name and stores it, as shown in FIG. 10 (step S111).

[0113] Thereafter, the server control unit 33 starts an image identification process to analyze the image data 38 and identify each object in the image data 38, as shown in FIG. 10 (step S112).

[0114] Specifically, the server control unit 33 inputs the image data 38 as input information to the input layer 35a of the first machine learning model 35 read from the server storage unit 32. At this time, the first machine learning model 35 passes input information from the input layer 35a to the intermediate layer 35b, and from the intermediate layer 35b to the output layer 35c, while comparing the input information with pre-learned weights and thresholds to identify each object in the image data 38.

[0115] Thereafter, the first machine learning model 35 outputs, as output information, information indicating each object and a classification image 38A in which name labels indicating the names of each object are added to the image data 38 (see FIG. 5).

[0116] For example, the first machine learning model 35 outputs, as output information, information indicating the slump test specimen 5 in the image data 38 and an identification image 38A in which the name label "Slump test specimen" is added to the image data 38.

[0117] Upon completion of the image identification process, the server control unit 33 analyzes the slump test specimen 5 in the identification image 38A, as shown in FIG. 10, and starts a viscosity evaluation process to evaluate the degree of viscosity of the concrete from the degree of horizontal spread and deformation of the shape of the slump test specimen 5 (step S113).

[0118] Specifically, the server control unit 33 inputs the identification image 38A as input information to the input layer 36a of the second machine learning model 36 read from the server storage unit 32. At this time, the second machine learning model 36 passes input information from the input layer 36a to the intermediate layer 36b, and from the intermediate layer 36b to the output layer 36c, while performing comparison judgments with pre-learned weights and thresholds, etc., to assign a viscosity evaluation point indicating the degree of viscosity of the concrete.

[0119] Thereafter, the second machine learning model 36 outputs, as output information, a viscosity evaluation image 39 to which the viscosity evaluation points and the viscosity classifications corresponding to the viscosity evaluation points have been assigned (see FIG. 6). Upon completing the viscosity evaluation process, the server control unit 33 starts a surface roughness evaluation process to analyze the surface texture of the slump test specimen 5 in the identification image 38A as shown in FIG. 10, and evaluate the degree of surface roughness of the concrete (step S114). Specifically, the server control unit 33 inputs the identification image 38A as input information to the input layer 37a of the third machine learning model 37 read from the server storage unit 32.

[0120] At this time, the third machine learning model 37 passes the input information from the input layer 37a to the intermediate layer 37b, and from the intermediate layer 37b to the output layer 37c, while performing comparison judgments with pre-learned weights and thresholds, etc., to assign a surface roughness evaluation point indicating the degree of surface roughness of the concrete. Thereafter, the third machine learning model 37 outputs, as output information, a surface roughness evaluation image 40 to which the surface roughness evaluation points and the surface roughness classes corresponding to the surface roughness evaluation points have been assigned (see FIG. 7).

[0121] When the surface roughness evaluation process is completed, the server control unit 33 determines whether or not the viscosity evaluation points of the viscosity evaluation image 39 satisfy the viscosity determination condition (step S115), as shown in Fig. 10. Note that, here, as an example, the viscosity determination condition is a viscosity evaluation point exceeding 7 points and not exceeding 12 points.

[0122] If the viscosity evaluation point does not satisfy the viscosity determination condition (step S115: No), the server control unit 33 generates a viscosity notification flag for notifying that the degree of viscosity is poor (step S116), and then proceeds to the process of step S117.

[0123] On the other hand, if the viscosity evaluation point satisfies the viscosity judgment condition (step S115: Yes), the server control unit 33 judges whether the surface roughness evaluation point of the surface roughness evaluation image 40 satisfies the surface roughness judgment condition (step S117). Note that, here, as an example, the surface roughness judgment condition is a surface roughness evaluation point exceeding 6 points and not exceeding 10 points.

[0124] If the surface roughness evaluation point does not satisfy the surface roughness judgment condition (step S117: No), the server control unit 33 generates a surface roughness notification flag to notify that the surface roughness level is poor (step S118), and then proceeds to step S119.

[0125] On the other hand, if the surface roughness evaluation points satisfy the surface roughness judgment conditions (step S117: Yes), the server control unit 33 generates transmission information to be transmitted to the worker terminal 20 (step S119).

[0126] Specifically, the server control unit 33 generates transmission information that associates image data 38 with the image capture date and time, an identification image 38A with a name label, a viscosity evaluation image 39 with a viscosity evaluation point and a viscosity classification, and a surface roughness evaluation image 40 with a surface roughness evaluation point and a surface roughness classification.

[0127] Furthermore, if the viscosity notification flag and / or the surface roughness notification flag have been generated, the server control unit 33 adds the viscosity notification flag and / or the surface roughness notification flag to the transmission information.

[0128] Thereafter, the server control unit 33 stores the transmitted information in association with the project data 34 corresponding to the project name, and then determines whether or not a viscosity notification flag and / or a surface roughness notification flag have been added to the transmitted information (step S120).

[0129] If the viscosity notification flag and / or surface roughness notification flag is not assigned to the transmitted information (step S120: Yes), the server control unit 33 transmits the transmitted information to the worker terminal 20 via the communication line 4 (step S121), and then terminates the property evaluation process.

[0130] On the other hand, if the viscosity notification flag and / or surface roughness notification flag are assigned to the transmitted information (step S120: No), the server control unit 33 transmits the transmitted information to both the worker terminal 20 and the production management terminal 10 via the communication line 4 (step S122), and then terminates the property evaluation process.

[0131] Returning to step S103 in FIG. 9, the terminal control unit 25 of the worker terminal 20 that has acquired the transmission information from the server 30 generates a judgment result screen 60 based on the transmission information acquired from the server 30, and starts a display process to display the judgment result screen 60 on the operation display unit 21 (step S104).

[0132] As shown in FIG. 11, the judgment result screen 60 displays the image data 38, the name of the test specimen, and the image capture date and time at the top of the screen, as well as a pass / fail column 61 indicating whether the properties of the concrete are good or bad. Furthermore, the judgment result screen 60 displays an item called "Viscosity of slump test specimen," along with a viscosity evaluation image 39, a viscosity classification bar 62 indicating the viscosity classification, and a viscosity evaluation point in the center of the screen.

[0133] In addition, the judgment result screen 60 displays an item called "Surface roughness of slump test specimen," along with a surface roughness evaluation image 40, a surface roughness classification bar 63 indicating the surface roughness classification, and a surface roughness evaluation score at the bottom of the screen.

[0134] When the viscosity notification flag and / or the surface roughness notification flag are not set in the transmitted information, "Good" is displayed in the pass / fail column 61 of the judgment result screen 60. On the other hand, when the viscosity notification flag and / or the surface roughness notification flag are set in the transmitted information, "Poor viscosity" and / or "Poor surface roughness" are displayed in the pass / fail column 61 of the judgment result screen 60.

[0135] Returning to step S103 in FIG. 9, the control unit 15 of the production management terminal 10, which has received the transmission information from the server 30, starts a notification process to notify of poor concrete properties based on the transmission information received from the server 30 (step S105).

[0136] For example, the control unit 15 of the production management terminal 10 notifies the production manager of poor concrete properties by displaying a screen similar to the above-mentioned judgment result screen 60 on the display unit 12 and by issuing an audio alert.

[0137] In this way, the fresh concrete evaluation system 3 of this embodiment evaluates whether the properties of fresh concrete are suitable for pouring, and if the properties are poor, it notifies the worker M and the production manager, thereby supporting quality control of fresh concrete.

[0138] As described above, the fresh concrete evaluation system 3 of the present embodiment is a system for evaluating the properties of fresh concrete based on image data 38 obtained by capturing an image of the slump test specimen 5 in a slump test and / or a slump flow test.

[0139] This fresh concrete evaluation system 3 includes a concrete identification means constructed with a first machine learning model 35 that receives image data 38 as input information, identifies the slump test specimen 5 in the image data 38, and outputs the identified data.

[0140] Furthermore, the fresh concrete evaluation system 3 is equipped with a viscosity evaluation means constructed of a second machine learning model 36 that uses the output information (identification image 38A) of the first machine learning model 35 as input information, evaluates and outputs the viscosity of fresh concrete.

[0141] In addition, the fresh concrete evaluation system 3 is equipped with a surface roughness evaluation means constructed by a third machine learning model 37 that uses the output information (identification image 38A) of the first machine learning model 35 as input information, evaluates and outputs the surface roughness of the fresh concrete.

[0142] Furthermore, the fresh concrete evaluation system 3 is equipped with a judgment means (server control unit 33) that judges the quality of the properties of the fresh concrete based on the output information (viscosity evaluation image 39) of the second machine learning model 36 and the output information (surface roughness evaluation image 40) of the third machine learning model 37.

[0143] The fresh concrete evaluation system 3 is equipped with an output means (server control unit 33) that outputs transmission information in which the quality of the properties of at least the fresh concrete is associated with image data .

[0144] In addition, the fresh concrete evaluation method in this embodiment includes a concrete identification process in which a concrete identification means constructed by the first machine learning model 35 takes image data 38 as input information, identifies and outputs the slump test specimen 5 in the image data 38, and a viscosity evaluation process in which a viscosity evaluation means constructed by the second machine learning model 36 takes output information (identification image 38A) of the first machine learning model 35 as input information, evaluates and outputs the viscosity of the fresh concrete.

[0145] Furthermore, the fresh concrete evaluation method includes a surface roughness evaluation process in which a surface roughness evaluation means constructed by the third machine learning model 37 uses the output information (identification image 38A) of the first machine learning model 35 as input information and evaluates and outputs the surface roughness of the fresh concrete, and a judgment process in which a judgment means (server control unit 33) judges whether the properties of the fresh concrete are good or bad based on the output information (viscosity evaluation image 39) of the second machine learning model 36 and the output information (surface roughness evaluation image 40) of the third machine learning model 37.

[0146] The fresh concrete evaluation method includes an output step in which the output means (server control unit 33) outputs transmission information in which the quality of the properties of the fresh concrete is associated with image data 38.

[0147] According to this configuration, the viscosity of fresh concrete and the surface roughness of fresh concrete are evaluated based on the slump test specimen 5 of the image data 38 identified by the concrete identification means, so that the properties of fresh concrete can be evaluated in more detail than when only material separation is evaluated.

[0148] In this case, image data 38 from the slump test and slump flow test are used as input information to be input to the first machine learning model 35 of the concrete identification means, so the fresh concrete evaluation system 3 does not need to prepare and image samples other than the slump test specimen 5 at the unloading site. Thereby, the fresh concrete evaluation system 3 can reduce the burden on the worker M at the unloading site.

[0149] Furthermore, since a viscosity evaluation means for evaluating viscosity is constructed by the second machine learning model 36, and a surface roughness evaluation means for evaluating surface roughness is constructed by the third machine learning model 37, the fresh concrete evaluation system 3 can evaluate the viscosity and surface roughness of fresh concrete more efficiently and with stable accuracy compared to when worker M evaluates based on experience.

[0150] Furthermore, since the fresh concrete evaluation system 3 is equipped with a judgment means (server control unit 33) for judging whether the properties of fresh concrete are good or bad, it saves the worker M the trouble of judging whether the properties of fresh concrete are good or bad, and can reliably judge whether the properties of fresh concrete are good or bad.

[0151] Therefore, the fresh concrete evaluation system 3 and the fresh concrete evaluation method using the same can accurately and easily evaluate whether the fresh concrete has properties suitable for pouring.

[0152] Therefore, the fresh concrete evaluation system 3 and the fresh concrete evaluation method can enable 100% evaluation of the properties of fresh concrete at the site where the fresh concrete is unloaded, similar to slump tests and slump flow tests.

[0153] In addition, since the first machine learning model 35 of the concrete identification means directly identifies the slump test specimen 5 from the image data 38 including the foreground and background, the fresh concrete evaluation system 3 can more efficiently identify the slump test specimen 5 in the image data 38 compared to identifying the slump test specimen 5 by deleting the foreground and background of the image data 38.

[0154] Further, the second machine learning model 36 of the viscosity evaluation means is configured to output a viscosity evaluation point indicating the degree of viscosity of the fresh concrete based on the slump test specimen 5 of the image data 38.

[0155] Furthermore, the third machine learning model 37 of the surface roughness evaluation means is configured to output a surface roughness evaluation score indicating the degree of surface roughness of fresh concrete based on the slump test specimen 5 of the image data 38.

[0156] The determination means (server control unit 33) is configured to determine whether the viscosity of the fresh concrete is good or bad based on the viscosity evaluation score, and to determine whether the surface roughness of the fresh concrete is good or bad based on the surface roughness evaluation score.

[0157] According to this configuration, the properties of fresh concrete can be evaluated in more detail, so that the quality of the properties of fresh concrete can be determined with greater accuracy. Furthermore, by displaying the transmitted information with the viscosity evaluation score and surface roughness evaluation score on the worker terminal 20, the fresh concrete evaluation system 3 can objectively convey the properties of the fresh concrete to the worker M.

[0158] In addition, the output means (server control unit 33) is configured to output transmission information including a viscosity evaluation point associated with a viscosity classification that classifies the degree of viscosity of fresh concrete, and a surface roughness evaluation point associated with a surface roughness classification that classifies the degree of surface roughness of fresh concrete.

[0159] According to this configuration, for example, to a production manager involved in the production of fresh concrete at a production factory 1 away from the unloading site, the properties of fresh concrete that are difficult to convey using only image data 38, viscosity evaluation points, and surface roughness evaluation points can be communicated more objectively.

[0160] Furthermore, since the viscosity classification and surface roughness classification are associated with the image data 38 of the transmitted information, the fresh concrete evaluation system 3 can use the output information to train workers M who have little experience in judging the properties of fresh concrete.

[0161] In addition, the average value of the evaluation points obtained by multiple engineers in evaluating the slump test specimen 5 of the image data 38 according to the degree of viscosity is taken as the viscosity average value, and the second machine learning model 36 of the viscosity evaluation means is a learning model that uses the image data 38 to which the viscosity average value has been assigned as training data T2.

[0162] Furthermore, the average value of the evaluation points obtained by multiple engineers in evaluating the slump test specimen 5 in the image data 38 according to the degree of surface roughness is set as the average surface roughness value, and the third machine learning model 37 of the surface roughness evaluation means is a learning model that uses the image data 38 to which the average surface roughness value has been assigned as training data T3.

[0163] According to this configuration, it is possible to construct highly accurate second machine learning model 36 and third machine learning model 37 using teacher data T2 and T3 that take into account the experience of multiple engineers. As a result, the fresh concrete evaluation system 3 can bring the degree of viscosity evaluated by the second machine learning model 36 closer to the degree of viscosity determined by an engineer, compared to when, for example, information indicating characteristics due to differences in viscosity is used as training data.

[0164] Similarly, the fresh concrete evaluation system 3 can bring the degree of surface roughness evaluated by the third machine learning model 37 closer to the degree of surface roughness determined by an engineer, compared to when, for example, information indicating characteristics due to differences in surface roughness is used as training data. Therefore, the fresh concrete evaluation system 3 can evaluate the properties of fresh concrete with even greater accuracy.

[0165] The evaluation score given by the engineer according to the degree of viscosity is one of the evaluation scores associated with each viscosity class indicating a class of the degree of viscosity, and the evaluation score given by the engineer according to the degree of surface roughness is one of the evaluation scores associated with each surface roughness class indicating a class of the degree of surface roughness.

[0166] According to this configuration, the accuracy of the teacher data T2 used for the second machine learning model 36 and the teacher data T3 used for the third machine learning model 37 can be improved.

[0167] Specifically, when an engineer evaluates the slump test specimen 5 of the image data 38 to generate the teacher data T2 and T3, the engineer only needs to determine which viscosity category the viscosity of the slump test specimen 5 falls into and which surface roughness category the surface roughness of the slump test specimen 5 falls into.

[0168] Therefore, compared to a case where an engineer directly assigns an evaluation score to the slump test specimen 5 in the image data 38, the fresh concrete evaluation system 3 can reduce variation in the evaluation scores given by engineers to the slump test specimen 5 in the image data 38 from one engineer to another.

[0169] As a result, the fresh concrete evaluation system 3 can construct a second machine learning model 36 and a third machine learning model 37 using teacher data T2 and T3 with stable accuracy, thereby enabling the viscosity and surface roughness of fresh concrete to be evaluated with stable accuracy.

[0170] In addition, there are four levels of viscosity classification: "appropriate viscosity", which is an appropriate state with appropriate fluidity; "excessive viscosity", which is a state with lower fluidity than "appropriate viscosity", "tending to separate", which is a state with higher fluidity than "appropriate viscosity", and "cement paste running ahead", which is a state with higher fluidity than "tending to separate". On the other hand, the surface roughness is classified into three levels: "smooth", which is a suitable state in which the coarse aggregate is not noticeable; "rough", which means that the coarse aggregate is clearly visible; and "intermediate", which is an intermediate state between "smooth" and "rough".

[0171] According to this configuration, compared to when the viscosity classification and surface roughness classification each have two levels of good or bad, the teacher data T2 used in the second machine learning model 36 and the teacher data T3 used in the third machine learning model 37 can be made teacher data suitable for detailed evaluation of the properties of fresh concrete.

[0172] The fresh concrete evaluation system 3 also includes notification means (the production management terminal 10 and the worker terminal 20) that notifies the user of poor properties of the fresh concrete when the determination means (the server control unit 33) determines that the fresh concrete is defective. According to this configuration, by notifying the worker M via the worker terminal 20 held by the worker M, it is possible to promptly notify the worker M performing the slump test of poor properties of the fresh concrete.

[0173] Furthermore, by reporting to the production management terminal 10 connected via the communication line 4, the fresh concrete evaluation system 3 can inform the production manager at the production plant 1 of poor properties of the fresh concrete and easily encourage inspection of the production process, etc. Thereby, the fresh concrete evaluation system 3 can prevent fresh concrete that does not satisfy the desired properties from being used for pouring.

[0174] In the configuration of the present invention and the correspondence with the above-mentioned embodiment, The imaging information of the present invention corresponds to the image data 38 of the embodiment. Similarly, The output information of the first machine learning model corresponds to the classified image 38A, The output information of the second machine learning model corresponds to the viscosity evaluation image 39, The output information of the third machine learning model corresponds to a surface roughness evaluation image 40, The determination means and the output means correspond to the server control unit 33. The output information corresponds to the transmitted information; The notification means corresponds to the production management terminal 10 and the worker terminal 20, The concrete identification process corresponds to step S112. The viscosity evaluation step corresponds to step S113. The surface roughness evaluation step corresponds to step S114. The determination step corresponds to steps S115 and S117. The output step corresponds to step S119. The present invention is not limited to the configurations of the above-described embodiments, and many other embodiments can be obtained.

[0175] For example, in the above-described embodiment, the production management terminal 10 used by the production manager was used as the terminal to be placed in the production plant 1, but this is not limited to this, and any terminal placed in the production plant 1 may be a terminal used by a quality manager who manages the quality of concrete, for example. In addition, the terminal used by worker M is a mobile phone type worker terminal 20, but this is not limited to this, and the terminal may be a tablet type terminal or the like as long as it is a terminal that worker M can carry and use.

[0176] In addition, the terminal control unit 25 of the worker terminal 20 stores image data 38 of the slump test specimen 5 in the terminal memory unit 22 in step S101, and then transmits the image data 38 to the server 30 by operation of the worker M in step S102, but this is not limited to this. For example, in step S101 of FIG. 9, after image data 38 is generated by capturing an image of the slump test specimen 5, the image data 38 associated with the project name may be automatically transmitted to the server 30.

[0177] Further, the menu screen is displayed on the operation display unit 21 by the worker M operating the worker terminal 20 and executing the processing program stored in the terminal storage unit 22, but this is not limiting. For example, the configuration may be such that worker M operates worker terminal 20 to launch a web browser on worker terminal 20 and access server 30 via the web browser, thereby displaying a menu screen on the web browser.

[0178] At this time, the server 30 transmits and receives various information to and from the worker terminal 20, and then performs various processes related to imaging the slump test specimen 5 and evaluating the concrete properties, and then transmits the transmission information to the worker terminal 20 to display the judgment result screen 60. In this case, the image data 38 of the slump test specimen 5 may be stored in both the worker terminal 20 and the server 30, or may be stored only in the server 30.

[0179] Further, the image data 38 obtained by capturing an image of the slump test specimen 5 in the slump test and / or slump flow test has been used in the description, but the present invention is not limited to this and may be a moving image obtained by capturing an image of the slump test specimen 5 .

[0180] In addition, the image data 38 is taken in which the slump plate 6 and the slump scale 7 are taken as subjects other than the slump test specimen 5, but this is not limited to this. For example, in the case of overseas slump tests conducted based on standards equivalent to the Japanese Industrial Standards, the image data 38 may be taken in which a convex is taken as the subject instead of the slump scale 7.

[0181] Moreover, the slump test specimen 5 was photographed using the mobile phone-type worker terminal 20, but the present invention is not limited to this, and the slump test specimen 5 may be photographed using, for example, a digital camera. In this case, the image data 38 stored in the digital camera may be imported into a portable personal computer to evaluate the properties of the concrete.

[0182] In addition, the first machine learning model 35, the second machine learning model 36, and the third machine learning model 37 use neural networks, but are not limited to this and may be any appropriate image recognition algorithm that can identify the slump test specimen 5, the viscosity and surface roughness of the slump test specimen 5 in the image data 38.

[0183] In addition, the first machine learning model 35 outputs the identification image 38A, the second machine learning model 36 outputs the viscosity evaluation image 39, and the third machine learning model 37 outputs the surface roughness evaluation image 40, but this is not limited to this.

[0184] For example, one machine learning model may be configured to output an identification image 38A, a viscosity evaluation image 39, and a surface roughness evaluation image 40, or one machine learning model may be configured to output an identification image 38A, and another machine learning model may be configured to output a viscosity evaluation image 39 and a surface roughness evaluation image 40.

[0185] Furthermore, the training image 51, which is the training data T1 of the first machine learning model 35, the training image 52, which is the training data T2 of the second machine learning model 36, and the training image 53, which is the training data T3 of the third machine learning model 37, may be the same image data or different image data from each other, as long as they are different from the image data 38.

[0186] In addition, the viscosity evaluation image 39 output by the second machine learning model 36 is configured by adding a viscosity evaluation point and a viscosity label to an image obtained by cutting out the area including the slump test specimen 5 from the identification image 38A, but this is not limited to this, and the viscosity evaluation image 39 may be the identification image 38A to which a viscosity evaluation point and a viscosity label have been added.

[0187] Similarly, the surface roughness evaluation image 40 output by the third machine learning model 37 is configured by adding a surface roughness evaluation point and a surface roughness label to an image extracted from the identification image 38A of a region including the slump test specimen 5, but this is not limited to this, and the identification image 38A to which the surface roughness evaluation point and the surface roughness label have been added may also be used as the surface roughness evaluation image 40.

[0188] Furthermore, the viscosity evaluation score, the surface roughness evaluation score, the evaluation score in the viscosity degree judgment table, and the evaluation score in the surface roughness degree judgment table are not limited to those in the above-described embodiment, and may be any appropriate evaluation score. In addition, the viscosity classification, the surface roughness classification, the viscosity classification in the viscosity degree judgment table, and the surface roughness classification in the surface roughness degree judgment table are not limited to the above-mentioned embodiment, and may be an appropriate classification according to the viscosity degree, and a surface roughness classification according to the surface roughness degree. In this case, the viscosity evaluation point corresponding to the viscosity classification and the surface roughness evaluation point corresponding to the surface roughness classification may be an appropriate evaluation point.

[0189] In addition, in step S115 of FIG. 10, it is determined whether the viscosity evaluation point satisfies the viscosity judgment condition, and in step S117 it is determined whether the surface roughness evaluation point satisfies the surface roughness judgment condition, but this is not limiting. For example, in step S115, it may be determined whether the viscosity classification is a desired viscosity classification, and in step S117, it may be determined whether the surface roughness classification is a desired surface roughness classification.

[0190] In step S115 of FIG. 10, the viscosity determination condition is set to a viscosity evaluation score of more than 7 points and not more than 12 points, but is not limited to this and may be any suitable range of viscosity evaluation points. In step S117 of FIG. 10, the surface roughness judgment condition is set to a surface roughness evaluation score of more than 6 points and not more than 10 points, but is not limited to this and may be a surface roughness evaluation score in an appropriate range.

[0191] In addition, the first machine learning model 35, the second machine learning model 36, and the third machine learning model 37 are stored in the server 30, and the image data 38 acquired from the worker terminal 20 is analyzed, but this is not limited to the above. For example, the first machine learning model 35, the second machine learning model 36, and the third machine learning model 37 may be stored in the worker terminal 20, and the image data 38 may be analyzed in the worker terminal 20. [Explanation of symbols]

[0192] 3. Fresh concrete evaluation system 5…Slump test specimen 10...Production control terminal 20...Worker terminal 33...Server control unit 35…First machine learning model 36…Second machine learning model 37…Third machine learning model 38...Image data 38A…Identification image 39...Viscosity evaluation image 40...Surface roughness evaluation image T2: Teacher data T3: Teacher data

Claims

1. A fresh concrete evaluation system for evaluating properties of fresh concrete based on imaging information obtained by imaging a slump test specimen in a slump test or / and a slump flow test, A concrete identification means constructed by a first machine learning model that uses the imaging information as input information, identifies the slump test specimen of the imaging information, and outputs the identified slump test specimen; A viscosity evaluation means constructed by a second machine learning model that uses output information of the first machine learning model as input information, evaluates and outputs the viscosity of the fresh concrete; A surface roughness evaluation means constructed by a third machine learning model that uses the output information of the first machine learning model as input information, evaluates and outputs the surface roughness of the fresh concrete; A determination means for determining whether the properties of the fresh concrete are good or bad based on output information of the second machine learning model and output information of the third machine learning model; and an output means for outputting output information relating at least the quality of the property of the fresh concrete to the imaging information. Fresh concrete evaluation system.

2. The first machine learning model of the concrete identification means is The slump test specimen is directly identified from the imaging information including the foreground and background. The fresh concrete evaluation system according to claim 1.

3. The second machine learning model of the viscosity evaluation means is A configuration in which a viscosity evaluation point indicating a degree of viscosity of the fresh concrete is output based on the slump test piece of the imaging information, The third machine learning model of the surface roughness evaluation means is A surface roughness evaluation score indicating a degree of surface roughness of the fresh concrete is output based on the slump test piece of the imaging information, The determination means is The viscosity of the fresh concrete is judged based on the viscosity evaluation score, and the surface roughness of the fresh concrete is judged based on the surface roughness evaluation score. The fresh concrete evaluation system according to claim 1.

4. The output means includes: The output information includes the viscosity evaluation point associated with a viscosity class that classifies the degree of viscosity of the fresh concrete, and the surface roughness evaluation point associated with a surface roughness class that classifies the degree of surface roughness of the fresh concrete. The fresh concrete evaluation system according to claim 3.

5. The average of the evaluation scores obtained by a plurality of engineers in accordance with the degree of viscosity of the slump test specimen of the image information is defined as an average viscosity value. The second machine learning model of the viscosity evaluation means is a learning model using the imaging information to which the viscosity average value is assigned as training data, The average value of the evaluation scores obtained by a plurality of engineers in accordance with the degree of surface roughness of the slump test specimen of the image information is defined as an average surface roughness value; The third machine learning model of the surface roughness evaluation means is A learning model using the imaging information to which the average surface roughness value is assigned as training data. The fresh concrete evaluation system according to claim 1.

6. The evaluation points given by the engineer according to the degree of viscosity were as follows: is one of the evaluation points associated with each viscosity category indicating a category of the degree of viscosity, The evaluation points given by the engineer according to the degree of surface roughness are as follows: is one of the evaluation points associated with each surface roughness class indicating the class of the degree of surface roughness. The fresh concrete evaluation system according to claim 5.

7. The viscosity category is There are four levels: a suitable state having moderate fluidity, a state having lower fluidity than the suitable state, a state having higher fluidity than the suitable state, and a state having even higher fluidity than the high fluidity state. The surface roughness classification is There are three levels: a desirable state where the coarse aggregate is not noticeable, a state where the coarse aggregate is clearly visible, and an intermediate state between the desirable state and the state where the coarse aggregate is clearly visible. The fresh concrete evaluation system according to claim 6.

8. The apparatus is provided with a notification means for notifying a poor property of the fresh concrete when the determination means determines that the fresh concrete is defective. The fresh concrete evaluation system according to claim 1.

9. A fresh concrete evaluation method for evaluating properties of fresh concrete based on imaging information obtained by imaging a slump test specimen in a slump test or / and a slump flow test, comprising: A concrete identification process in which a concrete identification means constructed by a first machine learning model uses the imaging information as input information, identifies the slump test specimen in the imaging information, and outputs the identified concrete identification information; A viscosity evaluation step in which a viscosity evaluation means constructed by a second machine learning model uses output information of the first machine learning model as input information, evaluates and outputs the viscosity of the fresh concrete; A surface roughness evaluation step in which a surface roughness evaluation means constructed by a third machine learning model uses the output information of the first machine learning model as input information, evaluates the surface roughness of the fresh concrete, and outputs the result; A determination step in which a determination means determines whether the properties of the fresh concrete are good or bad based on output information of the second machine learning model and output information of the third machine learning model; and an output step in which an output means outputs output information relating at least the quality of the properties of the fresh concrete to the imaging information. Fresh concrete evaluation methods.

Images