Machine learning model generation system and machine learning model generation method

Abstract

The goal is to generate machine learning models for estimating the deterioration of piping components without using vast amounts of actual measurement data. [Solution] The machine learning model generation system comprises a vibration analysis unit that determines a vibration transfer function corresponding to each deterioration state by performing vibration analysis using fixed parameters determined in response to hammering of the piping members and variable parameters determined for each different deterioration state of the piping members; a deterioration degree calculation unit that calculates the degree of deterioration corresponding to each vibration transfer function determined by the vibration analysis unit; a learning utilization data storage unit that stores learning utilization data that associates the vibration transfer function determined by the vibration analysis unit with the degree of deterioration calculated by the deterioration degree calculation unit for each deterioration state; and a machine learning model generation unit that generates a machine learning model that predicts the degree of deterioration of the piping members in response to the input of a vibration transfer function using the learning utilization data for each deterioration state.

Description

Technical Field

[0001] The present invention relates to a machine learning model generation system and a machine learning model generation method.

Background Art

[0002] There is known a technique for generating a learning model that estimates the deterioration state of a pipe by learning the relationship between the result of deterioration diagnosis determination of a pipe sample that has been subjected to a pipe extraction inspection in the past and various parameters related to the pipe of the pipe sample (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] For example, in the technique described in Patent Document 1, when generating a machine learning model, it is necessary to acquire an enormous amount of measured data for use in learning.

[0005] In consideration of the above problems, an object of the present invention is to enable the generation of a machine learning model for estimating the deterioration of a pipe member without using an enormous amount of measured data.

Means for Solving the Problems

[0006] One aspect of the present invention that solves the above-mentioned problems is a machine learning model generation system comprising: a vibration analysis unit that obtains a vibration transfer function corresponding to each deterioration state by performing vibration analysis using predetermined fixed parameters determined in response to striking of the piping member with a hammer and variable parameters determined for each different deterioration state of the piping member; a deterioration degree calculation unit that calculates the degree of deterioration corresponding to each vibration transfer function obtained by the vibration analysis unit; a learning utilization data storage unit that stores learning utilization data that associates the vibration transfer function obtained by the vibration analysis unit with the degree of deterioration calculated by the deterioration degree calculation unit for each deterioration state; and a machine learning model generation unit that generates a machine learning model that predicts the degree of deterioration of the piping member in response to the input of a vibration transfer function using the learning utilization data for each deterioration state.

[0007] A machine learning model generation method in a machine learning model generation system, comprising: a vibration analysis step in which a vibration analysis unit performs a vibration analysis using predetermined fixed parameters defined in response to striking of a piping member with a hammer and variable parameters defined for each different deterioration state of the piping member to determine a vibration transfer function corresponding to each deterioration state; a deterioration degree calculation step in which a deterioration degree calculation unit calculates a deterioration degree corresponding to each vibration transfer function obtained in the vibration analysis step; and a machine learning model generation step in which a machine learning model generation unit generates a machine learning model that predicts the deterioration degree of a piping member in response to an input vibration transfer function, using training data that associates the vibration transfer function obtained by the vibration analysis unit with the deterioration degree calculated by the deterioration degree calculation unit for each deterioration state. [Effects of the Invention]

[0008] According to the present invention, it is possible to generate a machine learning model for estimating the deterioration of piping components without using a large amount of actual measurement data. [Brief explanation of the drawing]

[0009] [Figure 1] This figure shows an example of the overall configuration of the pipe deterioration prediction system in this embodiment. [Figure 2] This figure shows an example of the overall configuration of the machine learning model generation system in this embodiment. [Figure 3] This figure shows an example of the functional configuration of the simulation device in this embodiment. [Figure 4] This figure shows an example of the learning data stored in the learning data storage device in this embodiment. [Figure 5] This shows an example of a processing procedure performed by the simulation device in this embodiment in relation to the simulation. [Figure 6] This figure shows a first example of the functional configuration of the learning model generation device in this embodiment. [Figure 7] This figure shows an example of the processing procedure for generating a machine learning model using the first example of a learning model generation device in this embodiment. [Figure 8] This figure shows a second example of the functional configuration of the learning model generation device in this embodiment. [Modes for carrying out the invention]

[0010] [Pipe deterioration prediction system] Figure 1 shows an example of the overall configuration of the pipe deterioration prediction system in this embodiment. The pipe deterioration prediction system of this embodiment predicts the deterioration status of target pipe components in actual pipes 10 that are actually installed in a building, for example. The pipe deterioration prediction system of this embodiment predicts the deterioration status of pipe components using a machine learning model generated by the machine learning model generation system described later.

[0011] The pipe deterioration prediction system shown in Figure 1 comprises a sensor 100 and a pipe deterioration prediction device 200. Sensor 100 is, for example, an acceleration sensor that detects acceleration (vibration acceleration) corresponding to vibration. In this embodiment, sensor 100 is provided so as to be able to detect vibrations occurring in the piping members that are the target of deterioration prediction among the piping members that make up the actual piping 10. Specifically, sensor 100 may be provided so as to be attached to the target piping member in the actual piping 10.

[0012] In this embodiment, in order to predict the deterioration status of the piping member to be predicted, the worker strikes the piping member to be predicted with a hammer 20 to generate vibrations in the piping member to be predicted. The sensor 100 detects the vibration acceleration corresponding to the vibration generated in the piping member to be predicted.

[0013] The pipe deterioration prediction device 200 is a device that predicts the deterioration status of a target pipe component based on the vibration acceleration detected by the sensor 100. The pipe deterioration prediction device 200 may be configured with hardware such as a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and storage devices such as an HDD (Hard Disk Drive) or SSD (Solid State Drive). The pipe deterioration prediction device 200 may also be equipped with a GPU (Graphics Processing Unit) as hardware. The functions of the pipe deterioration prediction device 200 shown in the figure are realized by the CPU and GPU in the pipe deterioration prediction device 200 executing a program.

[0014] The pipe deterioration prediction device 200 comprises a data interface unit 201, a user interface unit 202, a control unit 203, and a storage unit 204. The data interface unit 201 handles data exchange. In this embodiment, the data interface unit 201 receives vibration acceleration data output from the sensor 100.

[0015] The user interface unit 202 includes operators and input devices, and receives user operations. The user interface unit 202 also includes a display device, and displays an image on the display device for the user. The user interface unit 202 also includes a speaker or the like, and outputs sound for the user. The user interface unit 202 may also include indicators, lamps, etc., and notify the user by lighting the indicators and lamps.

[0016] The control unit 203 executes various controls in the piping deterioration prediction device 200. The control unit 203 includes a conversion unit 231 and a prediction unit 232.

[0017] The conversion unit 231 converts the vibration acceleration detected by the sensor 100 into a vibration transfer function by performing frequency analysis such as FFT (Fast Fourier Transform). The vibration transfer function may be based on, for example, the vibration spectrum of the vibration acceleration.

[0018] The prediction unit 232 predicts the deterioration status of the piping member to be predicted based on the vibration transfer function obtained by the conversion unit 231. The prediction unit 232 predicts the deterioration status using the machine learning model 241 stored in the storage unit 204.

[0019] [Machine Learning Model Generation System] In the present embodiment, the machine learning model 241 shown in FIG. 1 may be created by a machine learning model generation system as follows.

[0020] FIG. 2 shows an overall configuration example of the machine learning model generation system of the present embodiment. The machine learning model generation system of the present embodiment includes a simulation parameter storage device 300, a simulation device 400, a learning data storage device 500 (an example of a learning data storage unit), and a machine learning model generation device 600 (an example of a machine learning model generation unit). In the machine learning model generation system of this embodiment, the simulation parameter storage device 300, the simulation device 400, the learning data storage device 500, and the machine learning model generation device 600 may each be configured as separate devices, or at least some of them may be configured as a single device.

[0021] The simulation parameter storage device 300 stores the parameters (simulation parameters) used in the simulation performed by the simulation device 400. The simulation parameter storage device 300 stores fixed parameters and variable parameters as simulation parameters.

[0022] Fixed parameters are parameters defined according to the conditions under which a piping member is struck by a hammer. Fixed parameters include, for example, the excitation point on the piping member struck by the hammer, the excitation force applied when the piping member is struck by the hammer, the detection point where vibrations resulting from the hammering are detected on the piping member, and shape information indicating the shape of the piping member. Boundary conditions defined during the simulation may also be included as fixed parameters.

[0023] The variable parameters are parameters defined according to the different deterioration status of each piping component. The variable parameters corresponding to a single deterioration status may include, for example, the diameter of the piping component, the pipe type of the piping component, the deteriorated physical properties of the piping component, the thickness of the deteriorated layer of the piping component, the thickness of the sound layer of the piping component, and the void / crack rate corresponding to the deterioration status.

[0024] The parameter items in the above-mentioned variable parameters may be created not by measuring actual piping components, but by using information such as the specifications and characteristics of piping components, past empirical rules, and calculations. Therefore, even if there are many variable parameters, it is not necessary to take the trouble to measure vibrations by, for example, hammering piping components in various deterioration conditions, and the parameters can be created without taking much time, for example, by processing information. Furthermore, if the vibrations of piping components are measured as described above, a huge amount of data is required to obtain the data used for training, so a large-capacity storage device is necessary. In contrast, with the variable parameters of this embodiment, predetermined parameter items can be obtained for each deterioration condition by calculation, so a large-capacity storage device is not necessary.

[0025] The simulation device 400 performs a vibration analysis using the simulation parameters stored in the simulation parameter storage device 300 to determine the vibration transfer function corresponding to hammering of the piping members for each different deterioration state of the piping members. The simulation device 400 also performs a simulation to calculate the degree of deterioration of the piping members corresponding to the vibration transfer function determined for each deterioration state of the piping members.

[0026] The learning data storage device 500 stores the vibration transfer function and degree of deterioration for each deterioration state of the piping member, obtained by the simulation device 400 performing a simulation, as learning data.

[0027] The machine learning model generation device 600 generates a machine learning model 241 using the training data stored in the training data storage device 500. The machine learning model 241 generated by the machine learning model generation device 600 predicts the deterioration status according to the vibration transfer function obtained by striking the target piping member with a hammer.

[0028] The simulation performed by the simulation device 400 using the simulation parameters stored in the simulation parameter storage device 300 will be described with reference to Figure 3. Figure 3 shows the simulation parameters stored in the simulation parameter storage device 300. Figure 3 also shows an example of the functional configuration of the simulation device 400.

[0029] The simulation device 400 may be configured with hardware such as a CPU, ROM, RAM, and storage devices such as an HDD or SSD. The simulation device 400 may also be equipped with a GPU. The functions of the simulation device 400 shown in the figure are realized by the CPU and GPU in the simulation device 400 executing programs.

[0030] The simulation parameter storage device 300 in Figure 3 stores one fixed parameter and N variable parameters 1 to N as simulation parameters. Each of the N variable parameters 1 to N corresponds to a different N pattern of deterioration status of the piping member. Each of the variable parameters 1 to N may include the diameter of the piping member, the pipe type of the piping member, the deteriorated physical properties of the piping member, the thickness of the deteriorated layer of the piping member, the thickness of the sound layer of the piping member, and the void / crack rate.

[0031] The simulation device 400 includes a vibration analysis unit 401 and a degradation degree calculation unit 402. The vibration analysis unit 401 receives the fixed parameters and variable parameters 1 to N stored in the simulation parameter storage device 300 as input. The vibration analysis unit 401 uses the input fixed parameters and variable parameters 1 to N to simulate the vibration of the piping member, thereby determining the vibration transfer functions 1 to N corresponding to each deterioration state of each variable parameter 1 to N. Each of the vibration transfer functions 1 to N thus obtained exhibits characteristics unique to each deterioration state of the piping member defined in relation to each variable parameter 1 to N.

[0032] The deterioration degree calculation unit 402 calculates deterioration degrees 1 to N, which indicate the deterioration status corresponding to each vibration transfer function 1 to N obtained by the vibration analysis unit 401. Deterioration degrees 1 to N may be, for example, a single value that comprehensively quantifies the degree of deterioration of the piping member, or they may include numerical values ​​corresponding to the degree of deterioration of each of the multiple items that constitute the deterioration status.

[0033] As described above, the vibration transfer functions 1 to N and the degradation degrees 1 to N obtained by the simulation device 400 are stored in the learning data storage device 500 as learning data. Figure 4 shows an example of learning data stored in the learning data storage device 500. The learning data storage device 500 in this figure stores learning data 1 to N. Learning data 1 to N each contain information including the vibration transfer function and degree of degradation obtained by the simulation device 400 corresponding to the same degradation situation. That is, learning data 1 contains vibration transfer function 1 and degree of degradation 1. Similarly, learning data 2 to N each contain vibration transfer functions 2 to N and degrees of degradation 2 to N.

[0034] Referring to the flowchart in Figure 5, an example of the processing procedure performed by the simulation device 400 in relation to the simulation will be explained.

[0035] Step S100: Individual parameters stored in the simulation parameter storage device 300 are input to the simulation device 400. The input of individual parameters may be performed, for example, in response to operator operations, or according to a predetermined program or automated execution process. The vibration analysis unit 401 in the simulation device 400 acquires the input individual parameters.

[0036] Step S102: The vibration analysis unit 401 is initialized by assigning "1" to the variable n corresponding to the simulation order.

[0037] Step S104: The simulation device 400 receives the variable parameter n from among the variable parameters 1 to N stored in the simulation parameter storage device 300. The input of the variable parameter may be performed, for example, in response to the operator's actions, or according to a predetermined program or automatic execution process. The vibration analysis unit 401 acquires the input variable parameter n.

[0038] Step S106: The vibration analysis unit 401 performs a vibration analysis using the fixed parameters obtained in step S100 and the variable parameters n obtained in step S104 to determine the vibration transfer function n when a piping member with a deteriorated state corresponding to the variable parameters n is struck with a hammer.

[0039] Step S108: Next, the deterioration degree calculation unit 402 calculates the degree of deterioration n for the piping member corresponding to the deterioration status of the fluctuating parameter n, based on the vibration transfer function n obtained in step S106.

[0040] Step S110: The simulation device 400 stores the learning data n, which includes the vibration transfer function n obtained in step S106 and the degree of degradation n calculated in step S108, in the learning data storage device 500.

[0041] Step S112: In the simulation device 400, the vibration analysis unit 401 determines whether the current variable n is the maximum value "N" among the numbers 1 to N assigned to the fluctuation parameters.

[0042] Step S114: If it is determined in step S112 that the current variable n is less than "N", the vibration analysis unit 401 increments the variable n and proceeds to step S104. This process calculates the vibration transfer function n and the degree of degradation n using the next variable parameter n, and stores the learning data n, which includes the calculated vibration transfer function n and degree of degradation n, in the learning data storage device 500.

[0043] If it is determined in step S112 that the current variable n is greater than or equal to "N", then the simulation corresponding to each variable parameter 1 to N is completed, and the process shown in the figure is terminated.

[0044] [Example Configuration of a Learning Model Generation Device (Example 1)] Figure 6 shows a first example of the functional configuration of the machine learning model generation device 600. The machine learning model generation device 600 may be configured with hardware such as a CPU, ROM, RAM, and storage devices such as an HDD or SSD. The machine learning model generation device 600 may also be equipped with a GPU as hardware. The functions of the machine learning model generation device 600 shown in the figure are realized by the CPU and GPU provided in the simulation device 400 executing programs.

[0045] The machine learning model generation device 600 comprises a learner 601 and a reward calculation unit 602. The learner 601 receives sequential input of the vibration transfer functions 1 to N contained in the learning utilization data 1 to N stored in the learning utilization data storage device 500. The learner 601 predicts the degree of degradation corresponding to each input vibration transfer function 1 to N and sequentially outputs the predicted degradation degrees 1 to N.

[0046] The reward calculation unit 602 compares each predicted degradation level 1 to N output by the learner 601 with the degradation levels 1 to N contained in the oscillation transfer functions 1 to N stored in the learning utilization data storage device 500. Based on the comparison results, the reward calculation unit 602 sequentially calculates the rewards 1 to N corresponding to each predicted degradation level 1 to N output by the learner 601. The reward calculation unit 602 inputs the rewards 1 to N that it has calculated sequentially, and feeds them back to the learner 601. Each time a reward of 1 to N is input, the learner 601 modifies the parameters it uses to predict the degree of degradation so that a larger reward can be obtained. The learner 601 then generates a machine learning model 241 by applying the parameters obtained through this learning process.

[0047] Referring to the flowchart in Figure 7, an example of the processing procedure for generating a machine learning model using the machine learning model generation device 600 according to the first example shown in Figure 6 will be explained.

[0048] Step S200: The learner 601 is initialized by assigning "1" to the variable n corresponding to the predicted order of degradation.

[0049] Step S202: The learning device 601 receives the vibration transfer function n contained in the learning data n stored in the learning data storage device 500. The input of the vibration transfer function n may be performed, for example, in response to an operator's operation, or according to a predetermined program or automated execution process. The learning device 601 acquires the input vibration transfer function n.

[0050] Step S204: The learner 601 predicts the degree of degradation according to the vibration transfer function n obtained in step S202 and outputs the predicted degree of degradation n.

[0051] Step S206: The reward calculation unit 602 receives the degree of degradation n contained in the learning data n stored in the learning data storage device 500. The input of the degree of degradation n may be performed, for example, in response to an operator's operation, or according to a predetermined program or automated execution process. The reward calculation unit 602 obtains the input degree of degradation n.

[0052] Step S208: The reward calculation unit 602 compares the predicted degradation degree n output in step S204 with the degradation degree n obtained in step S206. For the comparison, the reward calculation unit 602 may, for example, determine the degree of approximation between the predicted degradation degree n and the actual degradation degree n. The reward calculation unit 602 calculates the reward n based on the result of comparing the predicted degradation degree n and the actual degradation degree n.

[0053] Step S210: The reward calculation unit 602 inputs the reward n calculated in step S208 to the learner 601.

[0054] Step S212: The learner 601 modifies the parameters it uses to predict the degree of degradation in response to the reward n input in step S210, for example, so that an optimized value is obtained.

[0055] Step S214: The learner 601 determines whether the current variable n is the maximum value "N" among the numbers 1 to N assigned to the training data.

[0056] Step S216: If it is determined in step S214 that the current variable n is less than "N", the learner 601 increments the variable n and proceeds to step S202. This process yields a predicted degradation degree n corresponding to the next oscillation transfer function n, and the parameters of the learner 601 are changed based on the reward n corresponding to the predicted degradation degree n.

[0057] Step S218: On the other hand, if it is determined in step S214 that the current variable n is greater than or equal to "N", the learner 601 generates a machine learning model 241 by applying the parameters obtained in the previous step S212.

[0058] [Example Configuration of a Learning Model Generation Device (Example 2)] Figure 8 shows a second example of the functional configuration of the machine learning model generation device 600. In the machine learning model generation device 600 shown in the figure, the learner 603 generates a machine learning model 241 by learning from multiple training datasets using supervised learning.

[0059] In this case, the learning unit 603 receives each of the learning data 1 to N stored in the learning data storage device 500 as a learning dataset. In the learning data used as a learning dataset, the vibration transfer function is the input data, and the degree of degradation is the label corresponding to the input data. By learning with the learning data 1 to N as the learning dataset, the learning unit 603 generates a machine learning model that predicts the degree of degradation in response to the input vibration transfer function.

[0060] Furthermore, the system may have an integrated configuration in which all or some of the simulation parameter storage device 300, simulation device 400, learning data storage device 500, and machine learning model generation device 600 shown in Figure 2 are combined. Furthermore, at least a portion of the simulation parameter storage device 300, the simulation device 400, the training data storage device 500, and the machine learning model generation device 600 may be configured as a server on a network or a cloud server. Furthermore, the functions of the simulation device 400 shown in Figure 3 may be distributed across multiple devices or servers on a network. Similarly, the functions of the machine learning model generation device 600 shown in Figures 6 and 8 may be distributed across multiple devices or servers on a network.

[0061] In the above embodiment, the machine learning model generation device 600 may generate the machine learning model 241 by using both the training data obtained by the simulation of the simulation device 400 and the training data based on actual measurements. By using the training data based on actual measurements in combination, it is expected that the prediction accuracy of the machine learning model 241 can be improved, for example. Learning data based on actual measurements may be obtained as follows: For example, a person in charge actually strikes an actual piping component with a hammer under conditions equivalent to those of the fixed parameters. The actual piping component has a predetermined state of deterioration, and an acceleration sensor is installed in the same way as the conditions indicated by the fixed parameters. The acceleration sensor detects the vibration generated in the actual piping component in response to the striking, and a vibration transfer function corresponding to the detected data is obtained through information processing. By associating the vibration transfer function obtained in this way with the degree of deterioration corresponding to the state of deterioration of the actual piping component, learning data based on actual measurements can be obtained.

[0062] Furthermore, the programs for realizing the functions of the aforementioned pipe deterioration prediction device 200, simulation device 400, machine learning model generation device 600, etc., may be recorded on a computer-readable recording medium, and the programs recorded on this recording medium may be loaded into a computer system and executed to perform the processing of the aforementioned pipe deterioration prediction device 200, simulation device 400, machine learning model generation device 600, etc. Here, "loading the programs recorded on the recording medium into a computer system and executing them" includes installing the programs into the computer system. Here, "computer system" includes hardware such as the OS and peripheral devices. Also, "computer system" may include multiple computer devices connected via a network including communication lines such as the Internet, WAN, LAN, and dedicated lines. Also, "computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, and storage devices such as hard disks built into the computer system. Thus, the recording medium storing the program may be a non-transient recording medium such as a CD-ROM. Also, the recording medium includes internal or external recording media that can be accessed from a distribution server for distributing the program. The program code stored on the distribution server's recording medium may be different from the program code in an executable format on the terminal device. In other words, as long as it can be downloaded from the distribution server and installed in an executable format on the terminal device, the format in which it is stored on the distribution server is irrelevant. Furthermore, the program may be divided into multiple parts, downloaded at different times and then combined on the terminal device, and each of the divided programs may be distributed by a different distribution server. In addition, "computer-readable recording medium" includes volatile memory (RAM) within computer systems that act as servers or clients when a program is transmitted over a network, which retains the program for a certain period of time. Moreover, the above program may be intended to implement only a part of the functions described above.Furthermore, the above-mentioned functions may be implemented in combination with programs already recorded in the computer system, such as so-called differential files (differential programs).

[0063] <Note> (1) One aspect of this embodiment is a machine learning model generation system comprising: a vibration analysis unit (401) that determines a vibration transfer function corresponding to each deterioration state by performing vibration analysis using predetermined fixed parameters determined in response to striking of the piping member with a hammer and variable parameters determined for each different deterioration state of the piping member; a deterioration degree calculation unit (402) that calculates the degree of deterioration corresponding to each vibration transfer function determined by the vibration analysis unit; a learning utilization data storage unit (e.g., learning utilization data storage device 500) that stores learning utilization data that associates the vibration transfer function determined by the vibration analysis unit with the degree of deterioration calculated by the deterioration degree calculation unit for each deterioration state; and a machine learning model generation unit (e.g., machine learning model generation device 600) that generates a machine learning model that predicts the degree of deterioration of the piping member in response to the input of a vibration transfer function using the learning utilization data for each deterioration state.

[0064] (2) One aspect of this embodiment is the machine learning model described in (1), wherein the fixed parameters may include an excitation point in the piping member struck by the hammer, an excitation force when the piping member is struck by the hammer, a detection point where vibrations caused by being struck by the hammer are detected in the piping member, and shape information indicating the shape of the piping member.

[0065] (3) One aspect of this embodiment is a machine learning model generation system described in (1) or (2), wherein the variable parameters may be all or some of the following: diameter of the piping member, type of piping member, deteriorated physical properties of the piping member, thickness of the deteriorated layer of the piping member, thickness of the sound layer of the piping member, and void crack rate.

[0066] (4) One aspect of this embodiment is a machine learning model generation system described in any one of (1) to (3), wherein the machine learning model generation unit predicts the degree of degradation corresponding to each degradation state in accordance with the input of the vibration transfer function included in the learning data, calculates a reward based on the predicted degree of degradation and the degree of degradation included in the learning data, and generates a machine learning model by changing the parameters based on the calculated reward.

[0067] (5) One aspect of this embodiment is a machine learning model generation system described in any one of (1) to (3), wherein the machine learning model generation unit may generate a machine learning model that predicts the deterioration status of a piping member in accordance with the input of a vibration transfer function by inputting a dataset of vibration transfer functions and degree of deterioration included in the learning utilization data corresponding to each deterioration status into a learner and training it.

[0068] (6) One aspect of this embodiment is a machine learning model generation system described in any one of (1) to (5), wherein the machine learning model generation unit generates a machine learning model using learning data that associates the vibration transfer function obtained by the vibration analysis unit with the degree of deterioration calculated by the degree of deterioration calculation unit for each deterioration state, and learning data obtained based on actual measurements.

[0069] (7) One aspect of this embodiment is a machine learning model generation method in a machine learning model generation system, comprising: a vibration analysis step in which a vibration analysis unit performs a vibration analysis using predetermined fixed parameters determined in response to striking of a piping member with a hammer and variable parameters determined for each different deterioration state of the piping member to determine a vibration transfer function corresponding to each deterioration state; a deterioration degree calculation step in which a deterioration degree calculation unit calculates a deterioration degree corresponding to each vibration transfer function obtained in the vibration analysis step; and a machine learning model generation step in which a machine learning model generation unit generates a machine learning model that predicts the deterioration degree of a piping member in response to an input vibration transfer function, using learning data that associates the vibration transfer function obtained by the vibration analysis unit with the deterioration degree calculated by the deterioration degree calculation unit for each deterioration state. [Explanation of symbols]

[0070] 10 Actual piping, 20 Hammer, 100 Sensor, 200 Piping deterioration prediction device, 201 Data interface unit, 202 User interface unit, 203 Control unit, 204 Storage unit, 231 Conversion unit, 232 Prediction unit, 241 Machine learning model, 300 Simulation parameter storage device, 400 Simulation device, 401 Vibration analysis unit, 402 Deterioration degree calculation unit, 500 Learning utilization data storage device, 600 Machine learning model generation device, 601, 603 Learner, 602 Reward calculation unit

Claims

1. A vibration analysis unit that determines a vibration transfer function corresponding to each deterioration state by performing vibration analysis using predetermined fixed parameters defined in response to hammering of piping members and variable parameters defined for each different deterioration state of the piping members, A degradation degree calculation unit calculates the degree of degradation corresponding to each vibration transfer function obtained by the vibration analysis unit, A learning data storage unit stores learning data that associates the vibration transfer function obtained by the vibration analysis unit with the degree of deterioration calculated by the degree of deterioration calculation unit for each of the aforementioned deterioration conditions. A machine learning model generation unit generates a machine learning model that predicts the degree of deterioration of piping members in accordance with the input of a vibration transfer function, using the learning data for each deterioration state. A machine learning model generation system equipped with the following features.

2. The fixed parameters include the excitation point in the piping member struck by the hammer, the excitation force when the piping member is struck by the hammer, the detection point where vibrations caused by being struck by the hammer are detected in the piping member, and shape information indicating the shape of the piping member. A machine learning model generation system according to claim 1.

3. The aforementioned variable parameters are all or some of the following: the diameter of the piping member, the type of pipe used, the deteriorated physical properties of the piping member, the thickness of the deteriorated layer of the piping member, the thickness of the sound layer of the piping member, and the void / crack rate. A machine learning model generation system according to claim 1 or 2.

4. The machine learning model generation unit predicts the degree of degradation corresponding to each degradation state in accordance with the input of the vibration transfer function included in the training data, calculates a reward based on the predicted degree of degradation and the degree of degradation included in the training data, and generates a machine learning model by changing the parameters based on the calculated reward. A machine learning model generation system according to claim 1 or 2.

5. The machine learning model generation unit generates a machine learning model that predicts the deterioration status of piping members in accordance with the input vibration transfer function by inputting a dataset of vibration transfer function and degree of deterioration included in the learning data corresponding to each deterioration status into the learner and training it. A machine learning model generation system according to claim 1 or 2.

6. The machine learning model generation unit generates a machine learning model using learning data that associates the vibration transfer function obtained by the vibration analysis unit with the degree of deterioration calculated by the degree of deterioration calculation unit for each deterioration state, and learning data for each predetermined deterioration state obtained based on actual measurements. A machine learning model generation system according to claim 1 or 2.

7. A method for generating machine learning models in a machine learning model generation system, The vibration analysis unit performs a vibration analysis using predetermined fixed parameters defined in response to hammering of the piping member and variable parameters defined for each different deterioration state of the piping member, thereby determining the vibration transfer function corresponding to each deterioration state. The degradation degree calculation unit includes a degradation degree calculation step which calculates the degree of degradation corresponding to each vibration transfer function obtained in the vibration analysis step, A machine learning model generation step involves a machine learning model generation unit generating a machine learning model that predicts the degree of deterioration of a piping member in accordance with the input vibration transfer function, using training data that associates the vibration transfer function obtained by the vibration analysis unit with the degree of deterioration calculated by the degree of deterioration calculation unit for each deterioration state. A method for generating machine learning models, including the following.

Images