Aggregate surface water rate estimating apparatus, model generation program, aggregate surface water rate estimating program, and learned model generation method
A machine learning-based system estimates aggregate surface water content using image and density data, simplifying the estimation process and improving efficiency and accuracy for concrete production.
Patent Information
- Application Number
- JP2023071066
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-04-24
- Publication Date
- 2024-11-06
- Estimated Expiration
- 2043-04-24
AI Technical Summary
Existing methods for determining aggregate surface water content, such as those described in Patent Document 1, require heating samples to measure moisture drying rates, making the process complicated and inefficient.
A system utilizing machine learning to estimate aggregate surface water percentage by inputting data, including photographed images and density data, to a trained model generated through learning data comprising images, density, and surface water percentage of aggregates, allowing for efficient estimation during conveyor transport.
Enables accurate and efficient estimation of aggregate surface water content immediately before shipping, ensuring proper mixing ratios for concrete production, thereby stabilizing concrete quality without requiring skilled operators.
Smart Images

Figure 2024156527000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an aggregate surface water ratio estimating device, a model generation program, an aggregate surface water ratio estimating program, and a method for generating a trained model. [Background technology]
[0002] It is known that the mixing ratio (particularly the ratio of water to cement) of the cement, water, sand, gravel, crushed stone, and other materials used in concrete affects the quality of the concrete. Moisture adheres to the surface of aggregates such as sand, gravel, and crushed stone, and the amount of moisture adhered to the surface (surface water ratio) varies depending on the storage environment of the aggregate. For this reason, when mixing concrete, it is necessary to know the surface water ratio of the aggregate used in mixing.
[0003] As a technique for grasping the surface moisture content of aggregate, for example, Patent Document 1 discloses a surface moisture content measuring method. This surface moisture content measuring method is a method in which fine aggregate having a predetermined surface moisture content is heated and dried, and the change in mass is measured, thereby performing a calibration to measure the relationship between the drying time and the drying moisture content and the relationship between the drying time and the moisture drying rate, and the measured value of the dry moisture content of the sample aggregate at the point where the moisture drying rate of the sample fine aggregate heated under the same conditions as in the calibration coincides with the moisture drying rate calculated by the calibration is regarded as an estimate of the surface moisture content of the sample fine aggregate. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] JP 2013-104710 A Summary of the Invention [Problem to be solved by the invention]
[0005] In the method disclosed in Patent Document 1, it is necessary to heat the sample aggregate to obtain the moisture drying rate, and the work for estimating the surface water ratio of the aggregate (sample fine aggregate) is complicated.
[0006] The present invention has been made in consideration of the above-mentioned problems, and has an object to provide an aggregate surface water ratio estimation device, a model generation program, an aggregate surface water ratio estimation program, and a method for generating a trained model, which can easily estimate the surface water ratio of aggregate. [Means for solving the problem]
[0007] A characteristic configuration of the aggregate surface water ratio estimation device of the present invention is that it comprises a surface water ratio estimation unit that inputs input data to a trained model generated by machine learning using training data, and obtains a value indicated by output data output from the trained model as an estimate of the surface water ratio of the aggregate, wherein the training data includes training image data indicating a photographed image of the training aggregate, which is the aggregate used for training, training density data indicating the density of the training aggregate, and training surface water ratio data indicating the surface water ratio of the training aggregate, and the input data includes input image data indicating a photographed image of the estimation target aggregate, which is the aggregate to be estimated.
[0008] According to this characteristic configuration, by inputting input data including input image data into the trained model, an estimate of the surface water ratio of the aggregate to be estimated can be obtained. Therefore, the surface water ratio of the aggregate can be easily estimated.
[0009] As another configuration, the input data may further include input density data indicating a density of the estimation target aggregate.
[0010] According to this characteristic configuration, since the input data further includes input density data, the surface water ratio of the aggregate can be estimated more accurately.
[0011] As another configuration, the estimation target aggregate may be the aggregate transported by a belt conveyor.
[0012] According to this characteristic configuration, the surface water ratio of the aggregate can be estimated while the aggregate is being transported by the belt conveyor, thereby improving the efficiency of the task of estimating the surface water ratio of the aggregate.
[0013] Alternatively, the density may represent the density of the aggregate in a surface dry, water-saturated state.
[0014] According to this characteristic configuration, the surface water ratio of the aggregate can be estimated more accurately.
[0015] As another configuration, the input image data may be data of a captured image of the estimation target aggregate immediately before it is packed in preparation for shipment.
[0016] According to this characteristic configuration, it is possible to estimate the surface water ratio of the aggregate, which is likely to change depending on the storage environment, etc., immediately before shipping, which is closer to the stage of concrete mixing work. As a result, it is possible to obtain concrete in which the cement, water, and aggregate are mixed in the appropriate ratio.
[0017] A characteristic configuration of the model generation program of the present invention is that it causes a computer to function as a model generation unit that generates a trained model by machine learning using training data, and the training data includes training image data that indicates an image of training aggregate, which is an aggregate used for training purposes, training density data that indicates the density of the training aggregate, and training surface water ratio data that indicates the surface water ratio of the training aggregate.
[0018] According to this characteristic configuration, the computer can function as a model generation unit that generates a trained model using learning data, so that a trained model for estimating the surface water ratio of aggregate can be easily generated, and as a result, the surface water ratio of aggregate can be easily estimated.
[0019] A characteristic configuration of the aggregate surface water ratio estimation program of the present invention is that it causes a computer to function as a surface water ratio estimation unit that inputs input data to a trained model generated by machine learning using training data, and obtains a value indicated by output data output from the trained model as an estimate of the surface water ratio of the aggregate, the training data including training image data indicating a photographed image of the training aggregate, which is the aggregate used for training, training density data indicating the density of the training aggregate, and training surface water ratio data indicating the surface water ratio of the training aggregate, and the input data includes input image data indicating a photographed image of the estimation target aggregate, which is the aggregate to be estimated.
[0020] According to this characteristic configuration, the computer can function as a surface water ratio estimation unit that obtains the value indicated by the output data output from the trained model as an estimate of the surface water ratio of the aggregate, thereby making it possible to easily estimate the surface water ratio of the aggregate.
[0021] As another configuration, the input data may further include input density data indicating a density of the estimation target aggregate.
[0022] According to this characteristic configuration, since the input data further includes input density data, the surface water ratio of the aggregate can be estimated more accurately.
[0023] A characteristic configuration of the trained model generating method of the present invention lies in that training data including training image data showing a photographed image of a training aggregate, which is an aggregate, training density data showing the density of the training aggregate, and training surface water ratio data showing the surface water ratio of the training aggregate are obtained, and a trained model is generated using the training data, and when input image data showing a photographed image of an estimation target aggregate, which is the aggregate to be estimated, the trained model outputs an estimate of the surface water ratio of the estimation target aggregate.
[0024] According to this characteristic configuration, a trained model can be generated that outputs an estimated value of the surface water ratio of the aggregate to be estimated by inputting input image data, thereby making it possible to easily estimate the surface water ratio of the aggregate.
[0025] In another configuration, the trained model may output an estimate of the surface water ratio of the aggregate to be estimated when input density data indicating the density of the aggregate to be estimated is input in addition to the input image data.
[0026] According to this characteristic configuration, input density data is input to the trained model in addition to input image data, thereby enabling more accurate estimation of the surface water ratio of the aggregate. [Brief description of the drawings]
[0027] [Figure 1] 1 is a diagram showing the configuration of an aggregate surface water ratio estimation system according to an embodiment. FIG. [Diagram 2] 1 is a block diagram showing a configuration of a learning device according to an embodiment. [Diagram 3] 1 is a block diagram showing a configuration of an estimation device according to an embodiment. [Figure 4] FIG. 2 is a diagram showing attributes of aggregates used in the examples and comparative examples. [Diagram 5] FIG. 13 is a diagram showing estimation results according to an example and a comparative example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0028] Hereinafter, an aggregate surface water ratio estimating device, a model generating program, an aggregate surface water ratio estimating program, and a trained model generating method according to the embodiments of the present invention will be described with reference to the drawings. However, the present invention is not limited to the following embodiments, and various modifications are possible without departing from the spirit of the present invention.
[0029] [Aggregate surface water ratio estimation system] FIG. 1 is a diagram showing the configuration of an aggregate surface water ratio estimation system 100. The aggregate surface water ratio estimation system 100 estimates the surface water ratio of aggregate 10 by utilizing AI (Artificial Intelligence) technology. The aggregate 10 includes at least one of fine aggregate and coarse aggregate. An example of fine aggregate is sand, and an example of coarse aggregate is gravel, crushed stone, etc. The surface water ratio of aggregate 10 is the ratio of the amount of water adhering to the surface of aggregate 10 divided by the mass of aggregate 10 in a surface-dry water-saturated state.
[0030] 1, the aggregate surface water ratio estimation system 100 includes an imaging device 1 that captures an image of an aggregate 10, a learning device 2 used in the learning phase, and an estimation device 3 (an example of an aggregate surface water ratio estimation device) used in the operation phase. The imaging device 1, the learning device 2, and the estimation device 3 are connected to each other via a network or the like so as to be able to communicate with each other.
[0031] [Imaging device] The imaging device 1 is a camera capable of photographing the aggregate 10, and outputs image data of the photographed image of the aggregate 10. The imaging device 1 has a lens, an imaging element that converts light entering through the lens into an electrical signal and outputs it, and an output unit that outputs image data based on the electrical signal output from the imaging element, etc. The imaging element is a CCD (Charge Coupled Device) image sensor, a CMOS image sensor, etc. In this embodiment, the imaging device 1 is provided at a position where it can photograph the aggregate 10 transported by the belt conveyor 20, and photographs the aggregate 10 transported by the belt conveyor 20.
[0032] [Learning device] Next, the configuration of the learning device 2 will be described with reference to Fig. 2. Fig. 2 is a block diagram showing the configuration of the learning device 2. The learning device 2 is configured with an information processing device such as a computer. The learning device 2 generates a trained model M2 by machine learning using training data.
[0033] As shown in FIG. 2, the learning device 2 has a first communication unit 21, a first operation unit 22, a first display unit 23, a first storage unit 24, and a first control unit 25.
[0034] The first communication unit 21 transmits and receives data. The first communication unit 21 is, for example, a network interface card (NIC), and is a communication device capable of communication conforming to a communication standard (for example, Ethernet) between the imaging device 1 and the estimation device 3. The first communication unit 21 is communicably connected to the imaging device 1 and the estimation device 3 via a network. The first communication unit 21 transmits and receives data of the trained model M2, learning data described below, and the like.
[0035] The first operation unit 22 accepts operations for the study device 2. The first operation unit 22 is composed of a touch sensor, a keyboard, and the like.
[0036] The first display unit 23 displays various information including characters, images, etc. The first display unit 23 is a liquid crystal display, an organic EL display, or the like.
[0037] The first storage unit 24 stores various data. The first storage unit 24 is an auxiliary storage device, and is configured with an HDD (Hard Disk Drive), an SSD (Solid State Drive), an SSDH (Solid State Hybrid Drive), etc. The first storage unit 24 stores a learning device control program PG1 (an example of a model generation program) and data indicating the learning model M1, learning data, etc.
[0038] The first control unit 25 controls the operation of the learning device 2. The first control unit 25 includes a processor such as a Central Processing Unit (CPU) and a Graphics Processing Unit (GPU), and a main storage device such as a Read Only Memory (ROM) and a Random Access Memory (RAM).
[0039] The first control unit 25 executes the learning device control program PG1 to function as a model generation unit 251. The model generation unit 251 acquires learning data, and generates a trained model M2 by performing machine learning (deep learning) on the learning model M1 using the learning data. That is, the first control unit 25 executes a method for generating the trained model M2.
[0040] The learning data is data for learning the learning model M1, and is data related to the aggregate 10 used for learning (hereinafter, referred to as "training aggregate 10S"; see FIG. 1). More specifically, the learning data includes a plurality of learning data sets. The learning data set is composed of learning image data showing a photographed image of the training aggregate 10S, learning density data showing the density of the training aggregate 10S, and learning surface water ratio data showing the surface water ratio of the training aggregate 10S. The surface water ratio of the training aggregate 10S corresponds to the correct answer data, and the image of the training aggregate 10S, the density of the training aggregate 10S, and the surface water ratio of the training aggregate 10S are associated in advance by a skilled technician, an expert, or the like. The density and surface water ratio of the training aggregate 10S are each normalized and used as the training data.
[0041] The density of the aggregate 10 (the learning aggregate 10S) is the density of the aggregate 10 in a surface-dried, water-saturated state, and is measured in advance in accordance with the method defined in the Japanese Industrial Standards (JIS A 1109).
[0042] In this embodiment, the learning image data included in the learning dataset is data of an image captured by the imaging device 1. More specifically, the learning image data is data representing a still image (hereinafter referred to as a "first still image") extracted from a video, and the video is a video of aggregates 10 transported by the belt conveyor 20 captured by the imaging device 1.
[0043] In this embodiment, the first still image is a still image cut out from the video after a predetermined time (e.g., 10 seconds) has elapsed since the start of imaging the training aggregate 10S. This stabilizes the brightness of the first still image (image represented by the training image data), and allows the generation of a trained model M2 with high estimation accuracy.
[0044] In addition, the first still image is subjected to a removal process to remove images with similar image characteristics, and a trimming process to cut out the periphery of the image so that it does not include images other than the learning aggregate 10S (for example, an image of the conveyor belt 20).
[0045] The training image data includes an augmented image that has been subjected to at least one of a brightness change process for changing the brightness and a left-right inversion process for inverting the left-right of the image with respect to the first still image. This makes it possible to increase the number of training data sets (images) used as training data, and generate a trained model M2 with high estimation accuracy.
[0046] The learning model M1 is configured by setting parameters in a network configuration (architecture) including an input layer to which input data (learning data set) is input, an intermediate layer to extract features from the input data (learning image data of the learning data set), and an output layer to output an estimation result estimated from the features as output data. In this embodiment, the trained model M2 is a model of a convolutional neural network such as EfficientNet, and is capable of learning (the association between the learning image data, learning density data, and learning surface water ratio data of the training aggregate 10S) by a multimodal method.
[0047] The model generation unit 251 inputs the learning data set into the learning model M1, and updates the parameters until a termination condition is met so as to minimize a loss function between the surface water ratio (learning surface water ratio data) associated with the learning image data and the estimation result indicated by the output data from the learning model M1. The parameter update is terminated when a termination condition is met, such as the number of parameter updates being equal to or greater than a specified value, or the loss function being equal to or less than a specified value. As a result, a trained model M2 is generated. The termination condition is set in advance by a system administrator, designer, or the like. In this embodiment, the model generation unit 251 performs machine learning by mini-batch learning, but the model generation unit 251 may also perform machine learning by online learning, batch learning, or the like.
[0048] The generated trained model M2 is stored in the first storage unit 24 by the model generation unit 251. In this embodiment, the trained model M2 is provided (transmitted) by the model generation unit 251 to the estimation device 3 via the first communication unit 21. Note that the trained model M2 may be provided to the estimation device 3 via a recording medium such as a Universal Serial Bus (USB) memory, a Compact Disc Read-Only Memory (CD-ROM), or a Digital Versatile Disc (DVD).
[0049] [Estimation device] Next, the configuration of the estimation device 3 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the estimation device 3. The estimation device 3 is composed of an information processing device such as a computer, and estimates the surface water ratio of an aggregate 10 (hereinafter referred to as "estimated aggregate 10T"; see Fig. 1) that is an estimation target of the surface water ratio, using a trained model M2.
[0050] As shown in FIG. 3, the estimation device 3 has a second communication unit 31, a second operation unit 32, a second display unit 33, a second storage unit , and a second control unit .
[0051] The second communication unit 31 transmits and receives data. The second communication unit 31 is, for example, a network interface card (NIC), and is a communication device capable of communication conforming to a communication standard (for example, Ethernet) between the imaging device 1 and the learning device 2. The second communication unit 31 is communicably connected to the imaging device 1 and the learning device 2 via a network. The second communication unit 31 transmits and receives data of the trained model M2, estimation data (an example of input data) described below, and the like.
[0052] The second operation unit 32 accepts operations for the estimation device 3. The second operation unit 32 is composed of a touch sensor, a keyboard, and the like.
[0053] The second display unit 33 displays various information including characters, images, etc. The second display unit 33 is a liquid crystal display, an organic EL display, or the like.
[0054] The second storage unit 34 stores various data. The second storage unit 34 is an auxiliary storage device, and is configured with an HDD, an SSD, an SSDH, etc. The second storage unit 34 stores an estimation device control program PG2 (an example of an aggregate surface water ratio estimation program), data of the trained model M2, estimation data, etc.
[0055] The second control unit 35 controls the operation of the estimation device 3. The second control unit 35 includes a processor such as a CPU or a GPU, and a main storage device such as a ROM or a RAM.
[0056] The second control unit 35 executes the estimation device control program PG2 to function as a surface water ratio estimation unit 351. The surface water ratio estimation unit 351 inputs estimation data to the trained model M2 and obtains an estimate (data indicating the estimate) of the surface water ratio of the estimation target aggregate 10T.
[0057] The estimation data is data related to the estimation target aggregate 10T, which is the aggregate 10 that is the estimation target. More specifically, the estimation data includes an estimation data set configured with estimated image data (an example of input image data) showing a captured image of the estimation target aggregate 10T. In other words, the estimation data is only estimated image data showing a captured image of the estimation target aggregate 10T, and does not include estimated density data showing the density of the estimation target aggregate 10T.
[0058] The estimated image data included in the estimation dataset is data of an image captured by the imaging device 1. In this embodiment, the estimated image data is data representing a still image (hereinafter referred to as a "second still image") cut out from a video, and the video is a video captured of the aggregate 10 transported by the belt conveyor 20. In other words, the estimation target aggregate 10T is the aggregate 10 transported by the belt conveyor 20. Also, in this embodiment, the estimation target aggregate 10T is the estimation target aggregate 10T immediately before it is packed in preparation for shipment, and the captured image data is data of an image captured of the estimation target aggregate 10T immediately before it is packed in preparation for shipment.
[0059] In this embodiment, the second still image, like the first still image, is a still image extracted from a moving image after a predetermined time (10 seconds in this embodiment) has elapsed since the start of imaging of the aggregate 10. This stabilizes the brightness of the still image, and allows the surface water ratio of the estimation target aggregate 10T to be estimated with high accuracy.
[0060] The second still image may be subjected to a removal process for removing images with similar image features, a trimming process for cutting out the periphery of the image so that images other than the estimation target aggregate 10T are not included, or the like.
[0061] The estimated image data is transmitted from the imaging device 1 to the estimation device 3 when imaging by the imaging device 1 is started.
[0062] When the surface water ratio estimation unit 351 acquires the estimated image data, it inputs the estimated image data to the trained model M2 as an estimation data set.
[0063] When the estimation data set is input, the trained model M2 outputs data (hereinafter referred to as "output data") corresponding to an estimated value of the surface water ratio of the estimation target aggregate 10T. The surface water ratio estimation unit 351 acquires the value indicated by the output data output from the trained model M2 as an estimated value of the surface water ratio of the estimation target aggregate 10T. The surface water ratio estimation unit 351 causes the second display unit 33 to display the acquired estimated value.
[0064] The estimated value of the surface water ratio is attached to a packaging material (e.g., a flexible container bag) in which the estimation target aggregate 10T is packed. More specifically, the estimated value of the surface water ratio is attached to the packaging material by a means such as attaching a tag on which the estimated value is written to the packaging material. Note that instead of writing the estimated value of the surface water ratio on the tag, a bar code or a two-dimensional code representing the estimated value of the surface water ratio may be affixed to the tag.
[0065] Thereafter, the estimation target aggregate 10T, which has been packaged with the estimated surface water ratio, is shipped to the concrete mixing site. This allows the worker mixing the concrete to easily grasp the estimated surface water ratio of the estimation target aggregate 10T. As a result, concrete in which the cement, water, and aggregate 10 are mixed in an appropriate ratio can be obtained, and the quality of the concrete can be stabilized.
[0066] [Example] Hereinafter, the generation and verification of a trained model will be described as an embodiment in comparison with a comparative example. Note that the present invention is not limited to the embodiment. For example, the embodiment using fine aggregate as aggregate is described below, but the aggregate is not limited to fine aggregate. The aggregate may be coarse aggregate, or may be a mixture of fine aggregate and coarse aggregate.
[0067] In Examples 1 to 3 and Comparative Examples 1 and 2 shown in Fig. 5, trained models were generated by machine learning using a convolutional neural network model (EfficientNet). Mean square error was used as the loss function of the machine learning.
[0068] Example 1 In Example 1, data sets (a learning data set and an estimation data set) were created for each of aggregates A to F shown in Fig. 4. Fig. 4 shows the attributes (type, rock quality, density, place of origin, manufacturer, and surface water ratio) of 14 types of aggregates A to N used in Examples 1 to 3 and Comparative Examples 1 and 2.
[0069] The learning dataset was composed of data on eight types of surface water ratios (48 sets of 6 × 8) out of data on aggregates A to F (6 types) with different surface water ratios (10 types of surface water ratios from 1 to 10) (i.e., 60 sets of 6 × 10).The learning dataset includes image data of captured images of any of aggregates A to F with any of the eight types of surface water ratios, density data showing the density of the aggregate shown in the captured image, and surface water ratio data showing the surface water ratio of the aggregate shown in the captured image.
[0070] Using the constructed training dataset, the above convolutional neural network model (EfficientNet) was subjected to machine learning to generate a trained model.
[0071] The estimation dataset used to verify the accuracy of the trained model was composed of surface water ratio data other than that used for learning (learning dataset), that is, two types of data (12 sets of 6 x 2) with different surface water ratios for each of aggregates A to F. The estimation dataset contains image data of images taken of aggregates with two types of surface water ratios used for verification, but does not contain density data. In other words, the estimation dataset is only image data of images taken of aggregates. Note that in FIG. 4, the surface water ratios of the data used as the estimation dataset in Examples 1 and 2 and Comparative Examples 1 and 2 are marked with (*).
[0072] The estimation data set was input to the generated trained model, data (output data) corresponding to the estimated value of the surface water ratio of the aggregate was obtained from the trained model, and the error of the output data relative to the correct answer data (data indicating the surface water ratio) was calculated. Figure 5 shows the error of the output data from the trained model of Example 1 relative to the correct answer data.
[0073] Example 2 In Example 2, in addition to the aggregates A to F used in Example 1, datasets (a learning dataset and an estimation dataset) were created for each of the aggregates G to N shown in Fig. 4. The learning dataset was composed of data (14 x 8 = 112 sets) of aggregates A to N (14 types) with different surface water ratios (10 types with surface water ratios of 1 to 10), and a trained model was generated in the same manner as in Example 1. The learning dataset includes image data of a photograph of any of the aggregates A to N with any of the eight types of surface water ratios, density data indicating the density of the aggregate shown in the photographed image, and surface water ratio data indicating the surface water ratio of the aggregate shown in the photographed image.
[0074] As in Example 1, the estimation dataset was composed of data on surface water ratios other than that used for learning (the learning dataset), that is, two types of data (14×2=28 sets) with different surface water ratios for each of aggregates A to N. The estimation dataset contains image data of images taken of aggregates with two types of surface water ratios used for verification, but does not contain density data. In other words, the estimation dataset contains only image data of images taken of aggregates.
[0075] The estimation data set was input to the generated trained model, data (output data) corresponding to the estimated value of the surface water ratio of the aggregate was obtained from the trained model, and the error of the output data relative to the correct answer data (data indicating the surface water ratio) was calculated. Figure 5 shows the error of the output data from the trained model of Example 2 relative to the correct answer data.
[0076] Example 3 In Example 3, a trained model was generated in the same manner as in Example 2, except that cross-validation was performed (in this example, the training set was rearranged so that the data on the surface water ratio used in the estimation dataset included all combinations) to generate a trained model, data (output data) corresponding to the estimated value of the surface water ratio of the aggregate was obtained from the generated trained model, and the error of the output data relative to the correct data (data indicating the surface water ratio) was calculated. Figure 5 shows the error of the output data from the trained model of Example 3 relative to the correct data.
[0077] Comparative Example 1 Comparative Example 1 is the same as Example 1, except that the learning dataset does not include density data indicating density. That is, a trained model was generated using image data of a photographed image of an aggregate. Then, image data of a photographed image of an aggregate was input to the trained model thus generated as an estimation dataset, data (output data) corresponding to an estimated value of the surface water ratio of the aggregate was obtained, and an error of the output data with respect to the correct answer data (data indicating the surface water ratio) was calculated. FIG. 5 shows an error of the output data from the trained model according to Comparative Example 1 with respect to the correct answer data.
[0078] Comparative Example 2 Comparative Example 2 is the same as Example 2, except that the learning dataset does not include density data indicating density. That is, a trained model was generated using image data of a photographed image of an aggregate. Then, image data of a photographed image of an aggregate was input to the trained model thus generated as an estimation dataset, data (output data) corresponding to an estimated value of the surface water ratio of the aggregate was obtained, and an error of the output data with respect to the correct answer data (data indicating the surface water ratio) was calculated. FIG. 5 shows an error of the output data from the trained model according to Comparative Example 2 with respect to the correct answer data.
[0079] 5, in Example 1, although the error values for some aggregates were larger than those in Comparative Example 1, it was shown that the error average value, which is the average value of the errors for aggregates A to F, was smaller than the error average value of Comparative Example 1. It was also shown that the error standard deviation value (rounded off to the third decimal place) which is the standard deviation value (variation) of the errors for aggregates A to F was also smaller than the error standard deviation value of Comparative Example 1. Note that the error standard deviation value indicates the standard deviation of a population with aggregates A to F as samples (the same applies to Comparative Example 1 below).
[0080] Also, in Example 2, although some aggregates showed larger error values than Comparative Example 2, it was shown that the average error values of aggregates A to N were smaller than the average error value of Comparative Example 2. It was also shown that the error standard deviation values of aggregates A to N were smaller than the error standard deviation value of Comparative Example 2. Note that the error standard deviation value indicates the standard deviation of a population with aggregates A to N as samples (the same applies to Example 3 and Comparative Example 2 below).
[0081] In Example 3, although the error values for some aggregates were larger than those of Example 2, the average error values of aggregates A to N were smaller than those of Example 2. In addition, the standard deviation value of Example 3 was also smaller than those of Examples 1 and 2.
[0082] From the above, it was confirmed from the results of Examples 1 and 2 and Comparative Examples 1 and 2 that when the surface water ratio of aggregate is estimated using only the image data of the images of the aggregate as the estimation dataset, the surface water ratio of aggregate can be estimated with higher accuracy by including density data in addition to the image data of the images of the aggregate in the training dataset.
[0083] As shown in Examples 1 to 3 above, it was confirmed that the aggregate surface water ratio estimation system 100 of this embodiment can estimate the surface water ratio of aggregate using a trained model. Also, as shown in Examples 1 to 3, it was confirmed that the trained model generated using training data (training data set) including aggregate density data can estimate the surface water ratio of aggregate with higher accuracy than Comparative Examples 1 and 2 (cases in which the surface water ratio of aggregate was estimated using a trained model generated using training data not including data indicating aggregate density).
[0084] [Effects of the embodiment] In this way, according to the aggregate surface water ratio estimation system 100, the estimation data including the estimated image data is input to the trained model M2 generated using the training data including the density data (training density data) of the aggregate 10 (training aggregate 10S), thereby obtaining an estimated value of the surface water ratio of the estimation target aggregate 10T. That is, according to the aggregate surface water ratio estimation system 100 of the present embodiment, the estimation data set includes only the estimated image data and does not include the estimated density data set, so that it is not necessary to input the estimated density data set separately for estimating the surface water ratio. This makes it possible to easily estimate the surface water ratio of the aggregate. In addition, since it is only necessary to input the estimation data to the estimation device 3, even an inexperienced and unskilled person can easily obtain the estimated value of the surface water ratio of the estimation target aggregate 10T. In addition, it is possible to suppress the variation in the estimated value due to the skill level of the worker who estimates the surface water ratio of the estimation target aggregate 10T. As a result, it is possible to stabilize the quality of the concrete using the estimation target aggregate 10T.
[0085] Furthermore, by including density data in the estimation data set, the surface water ratio can be estimated more accurately.
[0086] According to this embodiment, since the estimation target aggregate 10T is the aggregate 10 transported by the belt conveyor 20, when the estimation data does not include estimated density data, the surface water ratio can be estimated by acquiring estimated image data while the estimation target aggregate 10T is being transported by the belt conveyor 20. This can improve the efficiency of the task of estimating the surface water ratio of the aggregate.
[0087] Furthermore, according to this embodiment, since the density is the density of the aggregate 10 in a surface dry but water-saturated state, the surface water ratio of the aggregate 10 can be estimated more accurately.
[0088] Furthermore, according to this embodiment, it is possible to estimate the surface water ratio of the aggregate 10 (10T), which is likely to change depending on the storage environment, etc., immediately before shipping, which is a stage closer to the concrete mixing work. As a result, it is possible to obtain concrete in which the cement, water, and aggregate 10 are mixed in an appropriate ratio.
[0089] [Another embodiment] The present invention may be configured as follows other than the above-described embodiment (common numbers and symbols as in the embodiment are used to designate components having the same functions as in the embodiment).
[0090] (1) In the above embodiment, the input data includes only estimated image data indicating an image of the estimation target aggregate 10T, but the input data may further include estimated density data (an example of input density data) indicating the density of the estimation target aggregate 10T in addition to the estimated image data. When the input data includes estimated density data, the estimated density data is, for example, input to the estimation device 3 via the second operation unit 32 before the imaging device 1 starts imaging, and is stored in the second storage unit 34.
[0091] (2) In the above embodiment, the learning device 2 and the estimation device 3 are configured as independent information processing devices. However, the learning device 2 and the estimation device 3 may be configured as a single information processing device.
[0092] (3) In the above embodiment, the learning device 2 and the estimation device 3 are described as being connected to each other so that they can communicate with each other. However, in cases where the trained model M2 is provided to the estimation device 3 via a recording medium, for example, the learning device 2 and the estimation device 3 do not need to be connected to each other so that they can communicate with each other.
[0093] (4) In the above embodiment, the case where the generated trained model M2 is stored in the second storage unit 34 of the estimation device 3 has been described. However, the location where the trained model M2 is stored is not limited to the second storage unit 34 of the estimation device 3. The trained model M2 may be stored in, for example, the first storage unit 24 of the learning device 2. Alternatively, the trained model M2 may be stored in a storage area of a server (information processing device) further included in the aggregate surface water ratio estimation system 100. In this case, the server that stores the trained model M2 is communicatively connected to the estimation device 3. The surface water ratio estimation unit 351 may input an estimation dataset to the trained model M2 stored in the first storage unit 24 or the storage area of the server via the second communication unit 31 to acquire data indicating an estimated value.
[0094] (5) In the above embodiment, the estimated value is displayed on the second display unit 33 of the estimation device 3. However, the estimated value may be displayed on the first display unit 23 of the learning device 2, or on an information terminal (tablet terminal, computer, etc.) different from the learning device 2 and the estimation device 3 that are further provided in the aggregate surface water ratio estimation system 100 and communicatively connected to the estimation device 3. In this case, data indicating the estimated value is transmitted to the learning device 2 or the information terminal via the second communication unit 31. Furthermore, the estimated value is not limited to being displayed and output on the second display unit 33, and may be output via a speaker that the estimation device 3 further has.
[0095] (6) In the above embodiment, the second still image is a still image cut out from a video of the estimation target aggregate 10T transported by the belt conveyor 20. However, the second still image is not limited to a still image cut out from a video. The second still image may be a still image of the estimation target aggregate 10T transported by the belt conveyor 20. The second still image does not have to be a still image of the estimation target aggregate 10T immediately before being packed for shipment, but may be a still image of aggregate other than that transported by the belt conveyor 20 (for example, aggregate stored in a storage container). Similarly, the first still image is not limited to a still image cut out from a video. The first still image may be a still image of the estimation target aggregate 10T transported by the belt conveyor 20, or may be a still image of aggregate other than that transported by the belt conveyor 20 (for example, aggregate stored in a storage container).
[0096] (7) In the above embodiment, the case where the density and the surface water ratio of the learning aggregate 10S are normalized has been described. However, at least one of the density and the surface water ratio of the learning aggregate 10S does not have to be normalized.
[0097] (8) In the above embodiment, the extended image may further include a composite image obtained by combining a plurality of first still images. For example, the average value of the surface water ratio of the aggregate 10 shown in the original image is used as the ground truth data (data indicating the surface water ratio) of the composite image.
[0098] (9) Furthermore, the learning device control program PG1 executed by the first control unit 25 may be stored in a non-transitory computer-readable recording medium. The learning device control program PG1 stored in the recording medium may be installed in an information processing device such as a computer to configure a device that executes the above-mentioned processing. Similarly, the estimation device control program PG2 executed by the second control unit 35 may be stored in a non-transitory computer-readable recording medium. The estimation device control program PG2 stored in the recording medium may be installed in an information processing device such as a computer to configure a device that executes the above-mentioned processing.
[0099] (10) The trained model M2 may be generated using a neural network other than a convolutional neural network (such as a recurrent neural network), or may be generated using an algorithm other than a neural network. [Industrial Applicability]
[0100] The present invention can be used in an aggregate surface water ratio estimation device, a model generation program, an aggregate surface water ratio estimation program, and a method for generating a trained model. [Explanation of symbols]
[0101] 2: Learning device (computer) 3: Estimation device (computer; aggregate surface water ratio estimation device) 10: Aggregate 10S: Learning materials 10T: Estimated target aggregate 20: Conveyor belt 251: Model generation unit 351: Surface water rate estimation part M1: Learning model M2: Trained model PG1: Learning device control program (model generation program) PG2: Estimation device control program (aggregate surface water ratio estimation program)
Claims
1. a surface water ratio estimating unit that inputs input data to a trained model generated by machine learning using training data, and obtains a value indicated by output data output from the trained model as an estimate of the surface water ratio of the aggregate; The learning data includes learning image data showing a captured image of learning aggregate, which is the aggregate used for learning, learning density data showing a density of the learning aggregate, and learning surface water ratio data showing a surface water ratio of the learning aggregate, The input data includes input image data representing a photographed image of the estimation target aggregate, which is the aggregate to be estimated.
2. The aggregate surface water ratio estimating device according to claim 1 , wherein the input data further includes input density data indicating a density of the aggregate to be estimated.
3. The aggregate surface water ratio estimating device according to claim 1 , wherein the estimation target aggregate is the aggregate transported by a belt conveyer.
4. The aggregate surface water ratio estimating device according to claim 1 or 2, wherein the density indicates a density of the aggregate in a surface dry water-saturated state.
5. 3. The apparatus for estimating a surface water ratio of aggregate according to claim 1, wherein the input image data is data of a photographed image of the estimation target aggregate immediately before the aggregate is packed for shipment.
6. On the computer, It functions as a model generation unit that generates a trained model through machine learning using training data, A model generation program, wherein the training data includes training image data showing images of training aggregate, which is an aggregate used for training purposes, training density data showing the density of the training aggregate, and training surface water ratio data showing the surface water ratio of the training aggregate.
7. On the computer, inputting input data into a trained model generated by machine learning using training data, and causing the trained model to function as a surface water ratio estimation unit that acquires a value indicated by output data output from the trained model as an estimated value of the surface water ratio of the aggregate; The learning data includes learning image data showing a captured image of learning aggregate, which is the aggregate used for learning, learning density data showing a density of the learning aggregate, and learning surface water ratio data showing a surface water ratio of the learning aggregate, The input data includes input image data representing a photographed image of the aggregate to be estimated, the aggregate being the subject of estimation.
8. The aggregate surface water ratio estimating program according to claim 7 , wherein the input data further includes input density data indicating a density of the aggregate to be estimated.
9. acquiring learning data including learning image data showing an image of a learning aggregate, which is an aggregate, learning density data showing a density of the learning aggregate, and learning surface water ratio data showing a surface water ratio of the learning aggregate; Generate a trained model using the training data; A method for generating a trained model, in which the trained model outputs an estimated value of the surface water ratio of the estimation target aggregate when input image data indicating an image of the estimation target aggregate, which is the aggregate to be estimated, is input.
10. The method for generating a trained model according to claim 9, wherein the trained model outputs an estimate of the surface water ratio of the aggregate to be estimated when input density data indicating the density of the aggregate to be estimated is input in addition to the input image data.
Citation Information
Patent Citations
Method for measuring density in saturated surface-dry condition, rate of water absorption, surface moisture ratio of aggregate and high-temperature aggregate cooling device used for the same
JP2002131208A
Method for estimating quality of aggregate and method for manufacturing concrete
JP2021135199A
Quality predication system for ready mixed concrete and quality prediction method for ready mixed concrete
JP2023031937A
Method of measuring water content on fine aggregate surface
JP2013104710A
