Arithmetic device and model generation method

Abstract

To provide a calculating device easily obtaining a quality index of a ground.SOLUTION: A calculating device 1 for obtaining a quality index of compaction of an embankment by a vibration wheel 31 comprises: a first input portion for inputting behavior data of the vibration wheel; and a processing portion calculating to obtain a quality index. The processing portion obtains a quality index from a quality determination model by inputting the behavior data into the quality determination model to which machine learning is performed to output a quality index to an input value including a behavior data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a computing device and a model generation method. [Background technology]

[0002] Japanese Patent Publication No. 2003-193416 (hereinafter referred to as Patent Document 1) discloses a method for obtaining a quality index of the ground by determining the "disturbance rate" from the frequency spectrum of the change in vibration acceleration of a vibrating roller due to compaction, and then determining the rigidity of the ground from the disturbance rate. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2003-193416 Summary of the Invention

[0004] However, in the method of Patent Document 1, in order to calculate the ground stiffness from changes in the frequency spectrum of the vibration acceleration of the vibrating roller, it is necessary to calculate the "disturbance rate" using the numerical simulation model used to derive the ground stiffness evaluation formula. In other words, this method requires two-stage calculations to calculate the ground stiffness, which is cumbersome. Therefore, we provide a calculation device that can easily obtain a ground quality index and a model generation method that can be suitably used to obtain a ground quality index.

[0005] According to one embodiment, the calculation device is a calculation device that calculates a quality index of compaction of an embankment using a vibrating wheel, and includes a first input unit that inputs behavioral data of the vibrating wheel, and a processing unit that performs calculations to obtain the quality index, and the processing unit is configured to obtain the quality index from the quality determination model by inputting the behavioral data into the quality determination model that has been machine-trained to output a quality index for an input value including the behavior data.

[0006] According to one embodiment, the model generation method comprises performing machine learning using behavior data of a vibrating wheel obtained by a simulation model in which the movement of a vibrating wheel that vibrates for compacting an embankment is equivalently represented by a vibration model, and a quality index obtained from the embankment, as training data, to generate a quality assessment model in which the behavior data is input and the quality index is output.

[0007] Further details will be described in the following embodiments. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a schematic diagram illustrating an example of the configuration of a computing system according to an embodiment. [Figure 2] FIG. 2 is a schematic block diagram illustrating an example of the configuration of a calculation device according to the embodiment. [Figure 3] FIG. 3 is a schematic diagram illustrating an example of a simulation model used for machine learning of a quality determination model used in a computing device. [Figure 4] FIG. 4 shows the results of a comparison made by the inventors between the acceleration change obtained from the simulation model and the acceleration change of the vibrating wheel measured when compaction was actually performed with a vibrating roller. [Figure 5] FIG. 5 shows the results of a comparison made by the inventors between the acceleration change obtained from the simulation model and the acceleration change of the vibrating wheel measured when compaction was actually performed with a vibrating roller. [Figure 6] FIG. 6 is a diagram showing an outline of a method for generating a quality determination model. [Figure 7] FIG. 7 is a diagram showing an example of the flow of management of embankment compaction by a vibrating roller in a computing system. [Figure 8] FIG. 8 is a flowchart showing an example of the processing flow in the arithmetic device. [Figure 9] FIG. 9 is a diagram showing an outline of a method for re-learning a quality determination model. [Figure 10]FIG. 10 is a diagram showing the spring coefficient of the ground obtained by the calculation device. [Figure 11] FIG. 11 shows the measured values of the ground stiffness. [Figure 12] FIG. 12 is a diagram showing the results of comparing the ground stiffness converted from the spring coefficient in FIG. 10 with the measured ground stiffness in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0009] <1. Overview of the arithmetic device and model generation method>

[0010] (1) The computing device according to the embodiment is a computing device for calculating a quality index of the compaction of an embankment using a vibrating wheel, and includes a first input unit for inputting behavioral data of the vibrating wheel, and a processing unit for performing calculations to obtain the quality index. The processing unit is configured to input the behavioral data into a quality determination model that has been machine-trained to output a quality index for an input value including the behavior data, thereby obtaining the quality index from the quality determination model.

[0011] Compaction of embankments using vibrating rollers refers to the process of compacting the ground by applying pressure to the ground by vibrating the vibrating rollers. The quality index of embankment compaction is an index value used in quality control of embankment compaction that indicates the condition of the ground after compaction.

[0012] By using this computing device, it is possible to easily obtain a quality index using behavior data without conducting tests to obtain the quality index at the construction site where the embankment is compacted using a vibratory roller.

[0013] (2) Preferably, the quality assessment model is machine-learned to output a quality index using as-construction parameters obtained during compaction in addition to the behavior data as input values, and further includes a second input unit for inputting the as-construction parameters, and the processing unit is configured to obtain the quality index from the quality assessment model by inputting the behavior data and the as-construction parameters to the quality assessment model. This enables a more accurate quality index to be obtained from the quality assessment model.

[0014] (3) Preferably, the construction parameters include at least one of soil conditions, soil parameters, and construction condition parameters. By inputting the behavior data and at least one of the soil conditions, soil parameters, and construction condition parameters into the quality assessment model, a more accurate quality index can be obtained from the quality assessment model.

[0015] (4) Preferably, the computing device has a communication unit, the communication unit includes a first input unit, and is configured to communicate with another device and receive behavior data from the other device. This allows a device other than the device that obtains the behavior data to function as the computing device. Therefore, it is possible to obtain the quality index remotely.

[0016] (5) Preferably, the communication unit is configured to transmit the quality indicator to another device, whereby the quality indicator can be obtained by the other device.

[0017] (6) Preferably, the processing unit is configured to re-train the quality determination model using the behavior data and at least the measured quality index, thereby improving the accuracy of the quality determination model.

[0018] (7) Preferably, the retraining of the quality determination model includes retraining the quality determination model using the behavior data, the quality index, and construction parameters obtained during compaction, thereby improving the accuracy of the quality determination model.

[0019] (8) Preferably, the quality index includes a parameter related to ground stiffness, whereby the quality index related to ground stiffness can be obtained by a simple calculation.

[0020] (9) Preferably, the processing unit is further configured to store the quality index in the memory in association with the position information associated with the behavior data, thereby obtaining the quality index corresponding to the position where the behavior data was obtained from the memory.

[0021] (10) A model generation method according to an embodiment includes: performing machine learning using, as training data, behavior data of a vibrating wheel obtained by a simulation model that equivalently expresses the motion of a vibrating wheel that vibrates for compacting an embankment using a vibration model, and quality indices measured on the embankment after construction, to generate a quality assessment model that uses the behavior data as an input value and outputs a quality indices. By using the model generated in this way, it is possible to easily obtain a quality indices using the behavior data without conducting tests to obtain the quality indices at the construction site where the vibrating wheel is used to compact the embankment.

[0022] 2. Examples of arithmetic devices and model generation methods

[0023] 1 is a schematic diagram showing an example of the configuration of a computing system 100 according to this embodiment. The computing system 100 includes a computing device 1, and performs computations for managing the compaction of an embankment by a vibrating roller 3. The vibrating roller 3 has vibrating wheels 31, and applies pressure to the ground on which the embankment has been constructed by vibrating the vibrating wheels 31, thereby compacting (compacting) the ground G. The computing system 100 performs computations for quality control of the ground G after compaction by the vibrating roller 3.

[0024] Performing calculations for quality control includes determining a quality index of the compaction of the embankment. The quality index is a parameter that indicates the state of the ground G after compaction, and is used for quality control of the compaction of the embankment. One example of the quality index is the rigidity of the ground G. In the following description, the quality index is the subgrade reaction coefficient of the ground G, which is a rigidity parameter. Another example of the quality index may be an index that indicates the density or strength of the ground G.

[0025] 1, the vibrating roller 3 has a frame 32 connected to the front of a body 30, and a vibrating wheel 31 connected to this frame 32. For more details, see the enlarged view of part A in FIG. 1, where the vibrating wheel 31 is connected to the frame 32 via vibration-isolating rubber 33. An acceleration sensor 34 that measures the vibration acceleration of the vibrating wheel 31 is attached to the bearing of the vibrating wheel 31.

[0026] The calculation device 1 obtains information for quality control from the vibrating roller 3. As an example, the vibrating roller 3 is equipped with a tablet 5 having a communication function. The calculation device 1 obtains the information for quality control from the vibrating roller 3 by communicating with the tablet 5 via a communication network 9 such as the Internet. The tablet 5 may be a device separate from the vibrating roller 3, or may be a component incorporated into the vibrating roller 3 as an integrated part.

[0027] As another example, the arithmetic device 1 may be included in a tablet 5. That is, the arithmetic operations performed by the arithmetic device 1 may be realized by the tablet 5.

[0028] The information for quality control includes behavior data of the vibrating wheel 31. The behavior data of the vibrating wheel 31 is data representing the vibration of the vibrating wheel 31 used to estimate the quality index of the ground G, and may be, for example, a measurement value of vibration acceleration obtained by the acceleration sensor 34. As another example, the behavior data may be numerical data representing the measurement value of acceleration at each measurement time, a time change in the measurement value of acceleration, a spectral waveform obtained by performing a calculation such as a Fast Fourier Transform on the acceleration, speed data or displacement data of the vibrating wheel 31 calculated based on the acceleration data of the vibrating wheel 31, or a combination of at least two of them. The behavior data may also be numerical data or image data representing a waveform or the like.

[0029] Preferably, the information for quality control includes position information of the vibrating roller 3. That is, the behavior data of the vibrating wheels 31 may be associated with the position information of the vibrating roller 3. As an example, the position information may be obtained by a receiver 35 mounted on the vibrating roller 3 communicating with a GNSS (Global Navigation Satellite System) satellite SA.

[0030] The vibration acceleration measured by the acceleration sensor 34 may be transmitted from the acceleration sensor 34 to the tablet 5 via wireless communication, or may be input to the tablet 5 by a user operation. Position information of the vibration roller 3 may also be transmitted from the receiver 35 to the tablet 5 via wireless communication, or may be input to the tablet 5 by a user operation. In this way, information for quality control is transmitted from the tablet 5 to the calculation device 1.

[0031] 2 is a schematic block diagram showing an example of the configuration of the arithmetic device 1. Referring to FIG. 1, the arithmetic device 1 is configured as a computer having a processor 11 and a memory 12. The processor 11 is, for example, a CPU. The memory 12 includes a flash memory, an EEPROM, a ROM, a RAM, etc. Alternatively, the memory 12 may be a primary storage device or a secondary storage device.

[0032] The memory 12 stores a computer program 121 that is executed by the processor 11. The processor 11 executes the computer program 121 to perform calculations for quality control.

[0033] The memory 12 includes a behavior data storage unit 122. The behavior data storage unit 122 is an area for storing behavior data of the vibrating wheel 31 received from the tablet 5.

[0034] The memory 12 has a construction parameter storage unit 123. The construction parameter storage unit 123 is a storage area for storing construction parameters. The construction parameters are parameters obtained when the ground G is compacted by the vibrating roller 3, and are obtained by testing and measuring the ground G, testing and measuring samples made from the ground G, measuring necessary locations of the vibrating roller 3, etc. The construction parameters include at least one of the soil conditions of the ground G during compaction, soil parameters of the ground G during compaction, quality indicators measured from the ground G after compaction, and construction condition parameters for compaction.

[0035] The soil conditions of the ground G during compaction refer to the conditions of the embankment, and include, for example, the thickness of the embankment compacted in one application, called the rolled thickness. The soil parameters of the ground G during compaction are parameters that represent the soil quality of the ground G after compaction, are measured from the ground G after compaction, and include at least one of the soil type, density, water content, and grain size. The compaction application condition parameters are parameters that represent the conditions during compaction application by the vibrating roller 3, and include the mass of the vibrating wheel 31 and the vibration frequency of the vibrating wheel 31.

[0036] Preferably, the construction parameters are stored in association with the behavior data. The corresponding construction parameters and behavior data are behavior data of the vibrating wheel 31 measured at a certain position on the ground G, and the construction parameters obtained at that time.

[0037] Corresponding between the construction parameters and the behavior data may be achieved, for example, by assigning common identification information to the corresponding construction parameters and behavior data. As a result, the corresponding construction parameters and behavior data can be obtained from the memory 12 by extracting them using the identification information. As another example, data indicating the measurement time can be assigned to each of the construction parameters and behavior data. As a result, the corresponding construction parameters and behavior data can be obtained from the memory 12 by extracting those with similar measurement times.

[0038] The memory 12 has a management data storage unit 124. The management data storage unit 124 is an area for storing management data. The management data is data for quality control, including quality parameters obtained by processing by the processor 11. As an example, the management data may be a combination of the quality parameters and location information.

[0039] The computing device 1 has a communication unit 13. The communication unit 13 includes, for example, a wireless communication module and has a function of communicating with other devices via a communication network 9 such as the Internet. The communication unit 13 inputs data received from the tablet 5 into the memory 12.

[0040] The communication unit 13 is an example of a first input unit that inputs behavior data of the vibrating wheel 31 from the tablet 5 to the processor 11. The input behavior data is passed to the memory 12 and stored in the behavior data storage unit 122. The communication unit 13 is also an example of a second input unit that inputs construction parameters from the tablet 5 to the processor 11. The input construction parameters are passed to the memory 12 and stored in the construction parameter storage unit 123. The processor 11 reads the data from the memory 12, and the data is input to the processor 11. The communication unit 13 may also be an example of a quality index output unit that transmits the obtained quality index to another device.

[0041] The arithmetic device 1 may have an input device 15. The input device 15 is a keyboard, a media reader, or the like, and has a function of accepting data input. The input device 15 may be another example of the first input unit or another example of the second input unit.

[0042] The computing device 1 may have a display 14. The display 14 can display information based on the obtained quality parameters under the control of the processor 11. The display 14 may be another example of an output unit for the quality parameters.

[0043] The calculations for quality control of the ground G executed by the processor 11 include a determination process 111. The determination process 111 includes obtaining a quality index, and as a specific example, includes obtaining the stiffness of the ground G using a quality determination model 112.

[0044] The quality determination model 112 is a model that has been machine-learned to output a quality index for input values including behavior data. As a specific example, the quality determination model 112 has been machine-learned to output the stiffness of the ground G by inputting behavior data. The machine learning is, for example, deep learning. The machine learning may also be another learning method. The determination process 111 includes inputting behavior data to the quality determination model 112 to obtain the stiffness of the ground G from the quality determination model 112.

[0045] The quality assessment model 112 is a model generated by machine learning using behavior data of the vibrating wheels obtained by the simulation model and quality indicators measured on the embankment after construction as training data. As an example, the simulation model is a model that equivalently expresses the movement of the vibrating wheels that vibrate for compacting the embankment using a vibration model.

[0046] Fig. 3 is a schematic diagram showing an example of a simulation model 7 used for machine learning of the quality judgment model 112 used in the computing device 1. The simulation model 7 in Fig. 3 reproduces the acceleration response waveform of the vibrating roller using an interaction model of the soil-compaction machine system proposed by Fujiyama Tetsuo and Tateyama Kazuyoshi in "Method for Evaluating the Stiffness of Compacted Ground Using the Acceleration Response of a Vibrating Roller" (Fujiyama and Tateyama Papers of the Japan Society of Civil Engineers, No. 652 / III-51, 115-123, 2000).

[0047] 3, the simulation model 7 equivalently represents the frame portion 71 of the vibratory roller, the connection portion (vibration-isolating rubber) portion 72 between the frame and the vibrating wheel, the vibrating wheel portion 73, and the ground portion 74. Preferably, the simulation model 7 used in the quality determination model 112 for obtaining a quality index is a model that equivalently represents the frame 32 portion 71, the vibration-isolating rubber 33 portion 72, the vibrating wheel 31 portion 73, and the ground G portion 74 of the vibratory roller 3.

[0048] Specifically, simulation model 7 is formed by simplifying the section from the vibrating ring to the ground and the section from the vibrating ring to the frame with a Voigt model using the mass m1 of frame 32, the mass m2 of vibrating ring 31, the spring coefficient kr of vibration-isolating rubber 33, the viscous damping coefficient cr of vibration-isolating rubber 33, the stiffness ks of ground G, and the viscous damping coefficient cs of ground G, and then connecting these in series. In simulation model 7, the equation of motion for frame 32 is expressed by equation (1). The equation of motion for vibrating ring 31 is expressed by equation (2).

[0049] Regarding simulation model 7, the inventors compared the acceleration change obtained by simulation model 7 with the acceleration change of the vibrating wheel 31 measured when compaction was actually performed with the vibrating roller 3. The inventors compared the acceleration change obtained by simulation model 7 with the acceleration change of the vibrating wheel 31 measured at the number of compaction cycles corresponding to the simulation using simulation model 7. The number of compaction cycles corresponding to the simulation refers to the number of compaction cycles at which the measured subgrade reaction coefficient of the ground G becomes a value corresponding to the stiffness ks of the ground G that is applied to simulation model 7.

[0050] Figure 4 shows the change in acceleration when the stiffness ks of the ground G given to the simulation model 7 is 40.5MN / m, and the coefficient of subgrade reaction of the ground G is 81.0MN / m, which corresponds to the stiffness ks of 40.5MN / m. 3 Fig. 5 shows the comparison results between the acceleration change when the stiffness ks of the ground G given to the simulation model 7 is 91.6 MN / m and the acceleration change when the modulus of subgrade reaction of the ground G is 183.1 MN / m, which corresponds to a stiffness ks of 1.6 MN / m. 3 4 and 5, waveform (A) represents the acceleration change obtained by simulation model 7, and waveform (B) represents the spectral waveform obtained by subjecting that acceleration change to a fast Fourier transform. Waveform (C) represents the acceleration change measured from vibrating wheel 31, and waveform (D) represents the spectral waveform obtained by subjecting that acceleration change to a fast Fourier transform.

[0051] Comparing waveforms (A) and (C) in Figure 4, it can be seen that they are roughly similar acceleration waveforms. Waveforms (B) and (D) can also be said to have similar spectral waveforms. Similarly, in Figure 5, it can be said that waveforms (A) and (C) are roughly similar acceleration waveforms, and waveforms (B) and (D) are roughly similar spectral waveforms. This verified that the simulation model 7 can obtain acceleration changes that are close to the acceleration changes of the vibrating wheel 31 measured when compaction is actually performed with the vibrating roller 3.

[0052] Furthermore, comparing Figures 4 and 5, it can be seen that waveforms (A) and (C) in Figure 4 show regular sinusoidal wave behavior, while waveforms (A) and (C) in Figure 5 show waveform disturbances such as steps in the amplitude. While waveforms (B) and (D) in Figure 4 show that there is one dominant frequency, waveforms (B) and (D) in Figure 5 show that the vibrations contain multiple frequencies. This shows that the vibration of the vibrating wheel 31 changes due to the influence of the rigidity of the ground.

[0053] FIG. 6 is a diagram illustrating an outline of a method for generating the quality assessment model 112. As shown in FIG. 6, the quality assessment model 112 is generated by machine learning using the simulation model 7 as an example, with various stiffnesses ks and the acceleration waveforms obtained when these stiffnesses ks are applied to the simulation model 7, as pairs of training data T1, T2, ..., in which the acceleration waveforms are input values and the stiffnesses ks are output values. Instead of the acceleration waveforms, the input values may be spectral waveforms obtained by subjecting the acceleration waveforms to a fast Fourier transform. The quality assessment model 112 generated in this manner outputs the stiffness ks of the ground G as a quality index in the assessment process 111, when behavior data is applied as an input value.

[0054] The calculation for quality control of the ground G includes a storage process 113. The storage process 113 includes storing the obtained quality parameters in a memory as management data. As a specific example, the storage process 113 includes storing data including the obtained stiffness ks in a memory as management data.

[0055] The memory is, for example, the management data storage unit 124 of the memory 12. As a result, the quality index obtained in the calculation device 1 is stored as management data. As another example, the memory may be a memory of another device such as the tablet 5, or may be a storage medium. In this case, the storage process 113 includes passing the stiffness ks to the communication unit 13 and causing it to be transmitted to a storage destination such as the tablet 5. As a result, the quality index is stored in the required memory.

[0056] Preferably, the storage process 113 includes storing the quality index in memory in association with position information. The position information is input in association with the behavior data and is information representing the position where the behavior data was obtained. By storing the quality index in memory in association with the position information representing the position where the behavior data was obtained, a quality index for each position of the ground G is stored in memory.

[0057] The calculation for quality control of the ground G includes output processing 114. The output processing 114 includes outputting the management data stored in the management data storage unit 124. One example of the output destination is the tablet 5. In this case, the output processing 114 includes passing the management data including the quality index to the communication unit 13 and causing it to be transmitted to the tablet 5. This allows the display 51 of the tablet 5 to display based on the quality index, such as stiffness ks.

[0058] When the management data includes quality indices associated with location information, the location information can be used to display the quality indices on an output device such as a tablet 5. For example, the display of the quality indices using the location information may be a display in which the corresponding position on a map representing the ground G is colored in accordance with the stiffness ks. This makes it easier to visually recognize the quality indices on the ground G.

[0059] Fig. 7 is a diagram showing an example of the flow of embankment compaction management using a vibrating roller 3 in the computing system 100. The left half of Fig. 7 shows the operation at the construction site of compacting the embankment using a vibrating roller 3, and the right half shows the processing by the computing device 1 on the cloud, which was sent via the communication network 9.

[0060] 7, at a compaction construction site, the vibration acceleration of the vibrating wheel 31 measured by the acceleration sensor 34 is input to the tablet 5 (step S1), and behavior data D1 indicating the vibration acceleration of the vibrating wheel 31 is transmitted from the tablet 5 to the calculation device 1 (step S2). As a result, the behavior data is input to the calculation device 1.

[0061] Furthermore, as the construction parameters measured at the compaction construction site are input to the tablet 5 (step S3), data D2 indicating the construction parameters is transmitted from the tablet 5 to the arithmetic device 1 (step S4). As a result, the construction parameters are input to the arithmetic device 1.

[0062] The arithmetic device 1 executes the determination process 111 to obtain the stiffness ks of the ground G (step S5). Then, the arithmetic device 1 executes the storage process 113 to store the obtained stiffness ks in the memory 12 (step S6). The arithmetic device 1 transmits data D3 including the obtained stiffness ks to the tablet 5 (step S7).

[0063] In detail, Fig. 8 is a flowchart showing an example of the processing flow in the arithmetic device 1. Referring to Fig. 8, the processor 11 of the arithmetic device 1 reads behavior data from the memory 12, thereby inputting the behavior data to the processor 11 (step S101). The processor 11 also reads construction parameters from the memory 12, thereby inputting the construction parameters to the processor 11 (step S103).

[0064] The processor 11 inputs the behavior data into the quality determination model 112 (step S105) to obtain the stiffness ks (quality parameter) (step S107). The processor 11 stores management data including the obtained stiffness ks in memory (step S109). The processor 11 also transmits the obtained stiffness ks to the tablet 5 (step S111).

[0065] The tablet 5 receives the data D3 sent from the arithmetic device 1 and performs display based on the stiffness ks (step S8). If location information is associated with the stiffness ks received from the arithmetic device 1 in step S8, the tablet 5 displays the stiffness ks according to the location, for example, by displaying it on a map. This makes it possible to know the stiffness ks using the tablet 5 when compacting an embankment using the vibrating roller 3. In other words, the arithmetic device 1 obtains the stiffness ks using the behavior data, so that the stiffness ks can be obtained without conducting a test on the ground G.

[0066] Furthermore, the stiffness ks can be obtained in real time when compacting an embankment using the vibrating roller 3. This means that when the tablet 5 is mounted on the vibrating roller 3, compaction can be performed while checking the stiffness ks. Therefore, the obtained stiffness ks can be used to determine whether construction is necessary, for example, by performing additional compaction in areas where the stiffness ks is insufficient or by completing compaction in areas where the stiffness ks is sufficient.

[0067] Preferably, the calculation for quality control of the ground G includes a re-learning process 115. The re-learning process 115 includes re-learning the quality determination model 112 using behavior data of the vibrating wheel 31 obtained from the tablet 5 and quality indicators such as stiffness ks measured from the ground G after compaction by the vibrating roller 3. This can improve the accuracy of the quality determination model 112.

[0068] Preferably, the re-learning process 115 includes re-learning the quality determination model using behavior data of the vibrating wheel 31, quality indicators such as stiffness ks measured from the ground G after compaction by the vibrating roller 3, and also construction parameters during compaction by the vibrating roller 3. In the re-learning process 115, the processor 11 extracts the corresponding construction parameters and behavior data from the memory 12 and uses them for re-learning the quality determination model 112.

[0069] Fig. 9 is a diagram showing an outline of a method for relearning the quality determination model 112. Referring to Fig. 9, in a relearning process 115, the processor 11 relearns the quality determination model 112 using training data T3 in which behavior data and construction parameters of the vibrating wheel 31 are used as input values and quality indicators such as stiffness ks measured from the ground G are used as output values.

[0070] The relearning process 115 may be performed at a timing related to the embankment compaction management of Fig. 7 or at an independent timing. An example of an independent timing is when a user performs an operation to instruct the computing device 1 to execute the relearning process 115.

[0071] Furthermore, by re-learning using the construction parameters, the quality determination model 112 can be updated to a model that outputs a quality index by inputting behavior data and construction parameters. By using such a model, a more accurate quality index can be obtained.

[0072] The inventors verified the accuracy of the calculation device 1 by comparing the quality index obtained by the calculation device 1 with the quality index obtained from the ground G after compaction. FIG. 10 is a diagram showing the spring coefficient of the ground G obtained by the calculation device 1.

[0073] The quality judgment model 112 used for verification is a model trained by machine learning using the acceleration response waveform obtained by the simulation model 7, and 5,000 images of the acceleration waveform were used as training data. The inventors input the acceleration waveform of the vibrating wheels 31 actually measured from the vibrating roller 3 during compaction into the quality judgment model 112, and obtained the spring coefficient as its output value. Figure 10 shows the relationship between the number of compactions performed by the vibrating roller 3 when the acceleration waveform input into the quality judgment model 112 was obtained and the obtained spring coefficient.

[0074] Figure 11 shows the measured values of the subgrade reaction coefficient of ground G, which was used for comparison with the results in Figure 10. The subgrade reaction coefficient can be obtained, for example, by converting the maximum value of impact acceleration measured with a measuring device called a simple bearing capacity measuring instrument. Figure 11 shows the relationship between the number of compactions performed with the vibrating roller 3 during measurement and the subgrade reaction coefficient converted from the measurement value of the simple bearing capacity measuring instrument.

[0075] Comparing Figure 10 with Figure 11, it can be seen that in both cases the number of compactions increases up to about 10 times, after which the rate of increase converges and compaction is completed. In other words, Figure 10 shows the same trend as Figure 11.

[0076] Fig. 12 is a diagram showing the results of comparing the subgrade reaction coefficient obtained by converting the spring coefficient in Fig. 10 with the measured value of the subgrade reaction coefficient in Fig. 11, with the vertical axis showing the subgrade reaction coefficient obtained by arithmetic device 1 and the horizontal axis showing the measured value of the subgrade reaction coefficient. The closer the subgrade reaction coefficient obtained by arithmetic device 1 is to the actual measured value, the more linear the plot becomes.

[0077] The results in Figure 12 show that the correlation coefficient between the subgrade reaction coefficient obtained by the calculation device 1 and the measured value of the subgrade reaction coefficient is 0.556, confirming that there is a correlation between them. The slight variation in the plot is thought to include variation in the measurement of the subgrade reaction coefficient using the simple bearing capacity measuring device itself. Therefore, it can be said that the calculation device 1 can generally obtain quality parameters with high accuracy.

[0078] <3. Notes> The present invention is not limited to the above-described embodiment, and various modifications are possible. [Explanation of symbols]

[0079] 1: Arithmetic device 3: Vibration roller 5: Tablet 7: Simulation model 9: Communication network 11: Processor 12: Memory 13: Communications Department 14: Display 15: Input device 30: Body 31: Vibrating wheel 32: Frame 33: Anti-vibration rubber 34: Acceleration sensor 35: Receiver 51: Display 71: Frame part 72 :part 73: Vibration ring part 74: Ground part 100: Calculation system 111: Judgment process 112: Quality judgment model 113: Storage process 114: Output processing 115: Re-learning process 121: Computer Programs 122: Behavior data storage unit 123: Construction parameter memory unit 124: Management data storage unit D1: Behavioral data D2: Data D3: Data G: Ground SA:satellite T1: Training data T2: Training data T3: Training data cr: viscous damping coefficient cs: viscous damping coefficient kr: spring coefficient ks: Stiffness m1 :Mass m2: mass

Claims

1. A calculation device for calculating a quality index including a parameter related to ground stiffness of an embankment compacted by a vibrating wheel, a first input unit for inputting behavior data, which is measurement data representing the vibration of the vibrating wheel; a second input unit for inputting execution condition parameters representing the compaction execution conditions and including the mass of the vibrating wheel and the vibration frequency of the vibrating wheel; a processing unit that performs calculations to obtain a quality index including parameters related to the ground stiffness, The processing unit is configured to input the behavior data and the construction condition parameters to a quality determination model that has been machine-learned to output a quality index including the parameter related to ground stiffness in response to an input value including the behavior data and the construction condition parameters, thereby obtaining a quality index including the parameter related to ground stiffness from the quality determination model. Computing device.

2. It has a communication unit, The communication unit includes the first input unit and is configured to communicate with another device and receive the behavior data from the other device. The computing device of claim 1 .

3. The communication unit is configured to transmit a quality indicator including the parameter related to the ground stiffness to the other device. The computing device according to claim 2 .

4. The processing unit is configured to re-learn the quality determination model using the behavior data and a quality index including at least a parameter related to the measured ground stiffness. The computing device according to any one of claims 1 to 3.

5. Re-learning the quality determination model includes re-learning the quality determination model using the behavior data, the quality index including the parameter related to the ground stiffness, the construction condition parameters, and construction parameters obtained during the compaction, The construction parameters include soil conditions and soil parameters. The computing device according to claim 4.

6. The processing unit is further configured to store in the memory a quality index including the parameter related to the ground stiffness in association with position information associated with the behavior data. The computing device according to any one of claims 1 to 5.

7. and performing machine learning using behavior data of the vibrating wheels obtained by a simulation model in which the movement of the vibrating wheels that vibrate for compacting the embankment is equivalently expressed by a vibration model and quality indices obtained from the embankment as training data, thereby generating a quality judgment model in which the behavior data is used as an input value and the quality indices are used as an output value, the quality index is a parameter related to the ground stiffness of the compacted embankment, The behavior data is measurement data representing vibration of the vibrating wheel, The training data includes construction parameters including the mass and frequency of the vibrating wheel obtained during the compaction, The input values include the construction parameters. Model generation method.

Images