JP7761900B2 - AI-based concrete fluidity estimation system - Google Patents

Description

The present invention relates to a fluidity estimation system for estimating the fluidity of concrete.

Concrete is produced by mixing ordinary Portland cement, mixing water, fine aggregate, coarse aggregate, and optional admixtures in a batcher plant mixer. The produced concrete is transported by agitator truck and unloaded at the concrete pouring site before being poured. The required strength and properties of concrete vary depending on the concrete pouring site and are specified at the time of ordering. Slump is one indicator of concrete properties. Slump is an indicator of fluidity and is officially measured as follows: Concrete is filled into a 30 cm high conical container (a slump cone), which is then inverted and gently removed upward. The concrete loses its shape and its top drops. The height of this drop is the slump; a higher value indicates greater fluidity, while a lower value indicates less fluidity. Many batcher plant mixers are equipped with slump monitors, which estimate the slump based on the current driving the mixer.

Patent No. 6680936

Patent Document 1 describes an estimation system that uses machine learning to estimate slump. In this estimation system, concrete is first photographed inside a mixer while it is being mixed, and the resulting image data is input into a neural network as an explanatory variable, with the slump at that time being used as training data for learning. Image data for the concrete whose slump is to be estimated is then input into the trained neural network. The estimated slump is then output as the objective variable.

Conventional slump estimation methods using slump monitors are also excellent in that they can estimate slump with practical accuracy. Furthermore, the slump estimation system described in Patent Document 1 is also excellent in that it can estimate slump without using a slump monitor. However, slump cannot be estimated unless a slump monitor is installed in the batcher plant, and slump monitors do not necessarily estimate accurate slump. It cannot be said that the accuracy of slump estimation in the slump estimation system using the method described in Patent Document 1 necessarily reaches a level of practical accuracy, and there appears to be room for further improvement in estimation accuracy.

The present invention aims to provide a fluidity estimation system that can estimate concrete fluidity, consisting of slump or slump flow, with higher accuracy.

The present invention provides a concrete fluidity estimation system comprising a mixer for mixing concrete, a microphone, a camera, and a control device. The control device is equipped with a fluidity estimation AI. The fluidity estimation AI is composed of first to third AIs, which are supervised learning neural networks. The first AI receives as explanatory variables multiple pieces of time-series acoustic data representing the sounds of concrete mixing obtained by a microphone at multiple timings, and outputs multiple first intermediate fluidity estimates as a response variable. The second AI receives as explanatory variables multiple pieces of time-series image data of concrete captured by a camera at multiple timings, and outputs multiple second intermediate fluidity estimates as a response variable. The third AI uses the multiple first and second intermediate fluidity estimates as explanatory variables and outputs the concrete fluidity estimate as a response variable. The multiple timings are configured to include timings every second from 10 seconds before the completion of concrete mixing until the completion of mixing.

According to the present invention, the fluidity estimation AI is made up of three AIs, namely the first to third AIs, and outputs a fluidity estimation value from acoustic data of the concrete mixing sound and image data of the concrete, making it possible to estimate the fluidity of concrete with high accuracy.

1 is a front view showing a concrete fluidity estimation system according to an embodiment of the present invention. FIG. 1 is a block diagram showing a liquidity estimation AI according to the present embodiment. A diagram showing the neural network of the first AI that constitutes the liquidity estimation AI in this embodiment. A diagram showing the neural network of the second AI that constitutes the liquidity estimation AI in this embodiment. A diagram showing the neural network of the third AI that constitutes the liquidity estimation AI in this embodiment.

The present embodiment will be described.
As shown in FIG. 1 , a concrete fluidity estimation system 1 according to an embodiment of the present invention is provided in association with a mixer 4 of a batcher plant 2. Although not shown in the figure, the batcher plant 2 is also provided with a cement bin for storing and charging ordinary Portland cement, an aggregate bin for storing and charging aggregate, multiple measuring tanks for measuring the materials charged from these bins, and measuring tanks for charging mixing water and admixtures. Materials are charged from these bins into the mixer 4, which mixes them to produce concrete. A loading hopper 5 is provided below the mixer 4, and the produced concrete is loaded onto an agitator truck 6. The mixer 4 is also provided with a slump monitor 7 that measures the torque required to drive the mixer, i.e., the power consumed, to estimate the slump of the concrete.

The fluidity estimation system 1 according to this embodiment comprises a mixer 4, a microphone 10 located near the mixer 4, a camera 12 located above the mixer 4, and a controller (controller) 14. The controller 14 is equipped with a fluidity estimation AI 16 according to this embodiment, which will be described in detail below, and is configured to estimate the fluidity of concrete. While the fluidity to be estimated can be either slump or slump flow, in this embodiment, slump is estimated. This is because the fluidity estimation AI 16 is a learning AI and is provided with slump as training data. The mixing sound of concrete being mixed in the mixer 4 is collected by the microphone 10 and photographed by the camera 12. The mixing sound is sampled and sent as audio data, and the photographed image of the concrete is sent as image data to the controller 14. The fluidity estimation AI 16 estimates the fluidity of the concrete from these data. In this embodiment, a slump monitor 7 is connected to the controller 14. The fluidity estimation AI 16 learns from the slump measured by the slump monitor 7.

As shown in Figure 2, the liquidity estimation AI 16 in this embodiment is configured by combining three AIs, namely, the first to third AIs 18, 19, and 20. The first to third AIs 18, 19, and 20 are all supervised learning neural networks.

The first AI 18 receives acoustic data, which is a sample of the concrete mixing sound, as an explanatory variable and outputs a first intermediate fluidity estimate, which is an intermediate estimate of concrete fluidity, as a response variable. Figure 3 shows a schematic representation of the neural network constituting the first AI 18. It is composed of multiple neurons arranged in multiple layers and synapses connecting neurons in adjacent layers, making it a so-called perceptron-type neural network. Specifically, it is composed of an input layer 22, multiple intermediate layers 24, and an output layer 25. Acoustic data is input as an explanatory variable to the input layer 22, processed, and the first intermediate fluidity estimate is output as a response variable from the output layer 25. As shown in Figure 2, the first AI 18 also receives the measured concrete fluidity value, which enables the first AI 18 to learn. In this embodiment, the measured concrete fluidity value is the slump measured by the slump monitor 7. This measured slump is used as the measured concrete fluidity value for learning.

In this embodiment, acoustic data is sampled at a predetermined cycle from a predetermined timing before the concrete is completely mixed until the mixing is complete. For example, sampling is performed every second from 10 seconds before the mixing is complete until the mixing is complete. The first AI 18 processes this sampled acoustic data from multiple timings and outputs a first intermediate fluidity estimation value for each. In other words, the fluidity estimation AI 16 in this embodiment outputs multiple first intermediate fluidity estimation values in chronological order.

As shown in FIG. 2, the second AI 19 constituting the fluidity estimation AI 16 receives image data of concrete images as an explanatory variable and outputs a second intermediate fluidity estimate, which is an intermediate estimate of concrete fluidity, as a dependent variable. The second AI 19 is configured similarly to the first AI 18, and as shown in FIG. 4, it consists of an input layer 27, an intermediate layer 28, and an output layer 29. Image data is input to the input layer as an explanatory variable, and the output layer 29 outputs a second intermediate fluidity estimate as a dependent variable. As shown in FIG. 2, the second AI 19 also receives the slump measured by the slump monitor 7, which is the actual concrete fluidity value, as input and learns from it. In this embodiment, image data is also captured at a predetermined interval from a predetermined timing before the completion of mixing of the concrete until the completion of mixing. For example, image data is captured every second from 10 seconds before the completion of mixing until the completion of mixing. The second AI 19 processes the image data captured at these multiple times and outputs a second intermediate fluidity estimate for each. In other words, multiple second intermediate fluidity estimates are also output in chronological order.

As shown in Figure 2, the third AI 20 constituting the fluidity estimation AI 16 inputs the first and second intermediate fluidity estimates output by the first and second AIs 18 and 19 as objective variables as explanatory variables, and outputs an estimate of concrete fluidity, i.e., the fluidity estimate, as the objective variable. The third AI 20 is configured similarly to the first AI 18, and as shown in Figure 5, it consists of an input layer 31, an intermediate layer 32, and an output layer 33. Figure 5 shows multiple first and second intermediate fluidity estimates input to the input layer. These are all first and second intermediate fluidity estimates output in multiple time series from the first and second AIs 18 and 19. The third AI 20 processes the multiple first and second intermediate fluidity estimates in these time series and outputs a single fluidity estimate. As shown in Figure 2, this third AI 20 also learns by inputting the slump measured by the slump monitor 7 as the actual concrete fluidity value.

The fluidity estimation AI 16 in this embodiment is trained as follows. First, multiple lots of concrete are produced in the batcher plant 2. That is, the concrete is mixed in the mixer 4, and acoustic data and image data are obtained for each batch of concrete. The slump is then measured by the slump monitor 7. Hereinafter, this slump is referred to as the measured slump value. Learning is performed using this training data obtained for multiple lots. First, the first AI 18 is trained from the acoustic data and measured slump value, and the second AI 19 is trained from the image data and measured slump value. As the learning of the first and second AIs 18 and 19 progresses, the first and second intermediate fluidity estimation values output by the first and second AIs 18 and 19 become stable. Once this occurs, the third AI 20 is trained from the first and second intermediate fluidity estimation values output by the first and second AIs 18 and 19 and the measured slump value.

Once the fluidity estimation AI 16 has trained, it can estimate the slump of the concrete being mixed in the batcher plant 2 as follows: Acoustic data is obtained from the sound of mixing the concrete, and image data of the concrete is obtained. By inputting these into the fluidity AI 16, an estimated fluidity value is obtained. In other words, the slump can be estimated.

The fluidity estimation system 1 according to this embodiment can be modified in various ways. For example, the fluidity estimation AI 16 can be modified. For example, input data can be added to the first to third AIs 18, 19, and 20. When manufacturing concrete, a nominal slump is specified as a target value for fluidity. Although the nominal slump does not necessarily match the slump of the concrete being manufactured, such a nominal slump can be added as an explanatory variable. For example, it is possible to add the nominal slump as an explanatory variable to the first AI 18 and the third AI 20. Other data, such as air temperature and water temperature, can also be added as explanatory variables. Explanatory variables can be added as appropriate. Modifications in other respects are also possible. In the fluidity estimation system 1 according to this embodiment, the slump measured by the slump monitor 7 is used as training data for the objective variable. However, if the slump flow can be actually measured, it can be provided as training data for learning. In this case, the estimated fluidity value estimated by the fluidity estimation system 1 will naturally be the slump flow, not the slump.

An experiment was conducted to confirm that the liquidity estimation system 1 according to this embodiment can estimate slump with sufficiently high accuracy.
Experimental method:
In the batcher plant 2, 15,131 lots of concrete with different slumps were produced to obtain learning data. That is, 15,131 sets of time-series acoustic data, time-series image data, and actual slump values measured by the slump monitor 7 were obtained.
The first AI 18 was trained by providing the acoustic data of the learning data as explanatory variables and the measured slump values as training data for the objective variables.
The second AI 19 was given not only the image data for learning but also the slump designation specified when the concrete was manufactured as an explanatory variable, and was made to learn along with the actual measured slump value.
For the third AI 20, not only the first and second intermediate fluidity estimated values but also the nominal slump were given as explanatory variables, and it was made to learn them together with the actual measured slump values.
Next, 3000 lots of concrete with different slumps were produced in the batcher plant 2, and confirmation data was obtained. That is, 3000 sets of time-series acoustic data, time-series image data, and actual slump values measured by the slump monitor 7 were obtained.
The trained first AI 18 was given the acoustic data of the confirmation data to output a first intermediate fluidity estimate value, which was then compared with the actual slump measurement value. The comparison results are shown in Table 1-1.
Similarly, the second AI 19, which had already been trained, was given the image data of the confirmation data and the slump designation, and the second intermediate fluidity estimated value was output, which was then compared with the actual slump value. The comparison results are shown in Table 1-2.
Finally, the first and second intermediate fluidity estimates and the slump output by the first and second AIs 18 and 19 were given to the trained second AI 19, which then output a fluidity estimate, which was then compared with the actual slump measurement. The results of the comparison are shown in Table 1-3.

Consideration:
Among the fluidity estimation AIs 16 according to this embodiment, the first intermediate fluidity estimates estimated by the first AI 18 had a comparatively good error of ±2.5 cm or less in 79.47% of cases compared to the actual slump values. However, the error rate of ±1.5 cm or less was only 61.09%, which is not sufficiently high. Similarly, the second flu fluidity estimates estimated by the second AI 20 had a comparatively good error of ±2.5 cm or less in 88.06% of cases compared to the actual slump values. However, the error rate of ±1.5 cm or less was only 70.90%, which is not sufficiently high.
In contrast, the estimated fluidity values estimated by the third AI 20, i.e., the estimated fluidity values estimated by the fluidity estimation AI 16 according to the present embodiment, showed that the percentage of errors within ±2.5 cm in comparison with the measured slump values was 96.31%, and the percentage of errors within ±1.5 cm was 80.84%, which were very good. It was confirmed that the fluidity estimation AI 16 according to the present embodiment can estimate slump with high accuracy.

An experiment was conducted in which the fluidity estimation AI 16 according to this embodiment was made to estimate slump without giving nominal slump as an explanatory variable.
Experimental method:
An experiment in Example 2 was carried out using 15,131 sets of training data obtained in the experiment in Example 1 and 3,000 sets of verification data.
In the experiment in Experimental Example 1, the second and third AIs 19 and 20 were given the nominal slump as an explanatory variable, but in the experiment in Example 2, learning was performed without giving the nominal slump. Then, the first to third AIs 18, 19, and 20 that had already learned were made to estimate slump without giving the nominal slump. A comparison of the slump estimated by the fluidity estimation AI 16 and the actual slump value is shown in Table 2.

Consideration:
The estimated fluidity values estimated by the fluidity estimation AI 16 according to this embodiment were compared with the measured slump values with an error of ±2.5 cm or less in 94.83% of cases and an error of ±1.5 cm or less in 78.33% of cases, which were very good results. It was confirmed that the fluidity estimation AI 16 according to this embodiment can estimate slump with sufficiently high accuracy without providing nominal slump as an explanatory variable.

In this embodiment, the microphone 10 is used to obtain acoustic data, which is a sampling of concrete mixing sounds. Mixing sounds include not only the sounds emitted by the concrete materials as they collide, disperse, and mix, but also the impact and friction sounds of the concrete materials hitting the steel plates of the mixer 4. The sound also includes the sound of the motor that drives the mixer 4. The proportion of each sound naturally varies depending on where the microphone 10 is installed. That is, if the microphone 10 is installed near the motor that drives the mixer 4, the proportion of motor sound will be high, but if the microphone 10 is installed inside the mixer 4, the proportion of sound emitted by the concrete materials will be high. The fluidity estimation AI 16 of this embodiment can estimate the fluidity of concrete regardless of the proportion of these sounds contained in the acoustic data. In other words, concrete fluidity can be estimated whether the proportion of motor sound is high or almost nonexistent among the sounds contained in the acoustic data. However, it is preferable to fix the installation position of the microphone 10 and obtain acoustic data as training data, and then obtain acoustic data for estimating concrete fluidity.

1 Fluidity estimation system 2 Batcher plant 4 Mixer 5 Hopper 6 Agitator truck 7 Slump monitor 10 Microphone 12 Camera 14 Control device 16 Fluidity estimation AI
18 First AI 19 Second AI
20 The Third AI

Claims (2)

a mixer for mixing concrete;
a microphone provided near the mixer;
a camera provided near the mixer;
a control device;
The control device is equipped with a fluidity estimation AI that inputs, as explanatory variables, a plurality of pieces of time-series acoustic data obtained by sampling the concrete mixing sound detected by the microphone at a plurality of timings and a plurality of pieces of time-series image data of the concrete photographed at the plurality of timings by the camera, and outputs, as a response variable, a fluidity estimation value that is an estimate of concrete fluidity consisting of slump or slump flow,
The plurality of timings include timings every second from 10 seconds before the completion of mixing of the concrete until the completion of mixing,
The liquidity estimation AI is composed of first to third AIs which are supervised learning neural networks,
the first AI is configured to output a first intermediate liquidity estimated value, which is a target variable, from the acoustic data input as an explanatory variable, thereby obtaining a plurality of the first intermediate liquidity estimated values for a plurality of the acoustic data;
the second AI is configured to output a second intermediate fluidity estimated value, which is a target variable, from the image data input as an explanatory variable, thereby obtaining a plurality of the second intermediate fluidity estimated values for a plurality of the image data;
the third AI is configured to output the liquidity estimate values as objective variables when the first intermediate liquidity estimate values and the second intermediate liquidity estimate values are input as explanatory variables,
The first to third AIs use the plurality of time-series acoustic data and the plurality of time-series image data obtained by mixing a plurality of lots of concrete in a mixer as explanatory variables for learning, and learn by providing actual concrete fluidity measurements consisting of an actually measured slump or an actually measured slump flow as teacher data for each of the plurality of first and second intermediate fluidity estimate values and the fluidity estimate values. The concrete fluidity estimation system according to claim 1 , wherein the nominal slump or nominal slump flow designated when mixing concrete is input to any or all of the first to third AIs.