Information processing method, program, information processing device, and model generation method

Abstract

The present invention provides an information processing method, etc., capable of favorably estimating the state of a magnetic pipe. In the information processing method, a computer performs a process in which measurement data in which magnetic characteristic values of a magnetic pipe have been measured is acquired, and thickness information is estimated by inputting the acquired measurement data into a model which has been trained so as to estimate the thickness information relating to the thickness of the magnetic pipe when the measurement data has been input. Preferably, the magnetic characteristic values are measurement values measured by using an inspection probe provided with a magnet which generates a magnetic field, a yoke which is disposed on the opposite side of the magnetic tube with respect to the magnet, and a magnetic sensor which is disposed between the yoke and the magnetic pipe and measures the magnetic flux density passing through the yoke, magnet, and magnetic pipe. The output voltage of the magnetic sensor decreases as the magnetic flux density increases.

Description

Technical Field

[0001] The present invention relates to an information processing method, a program, an information processing apparatus, and a model generation method.

Background Art

[0002] As a method for non-destructively inspecting defects such as wall thickness reduction in a magnetic tube, remote field eddy current testing (RFECT), magnetic flux leakage method (MFL), etc. are known.

[0003] For example, Patent Document 1 discloses a method for measuring defects in a magnetic member by the magnetic flux resistance method (MFR).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, in these inspection methods, the wall thickness of the magnetic tube is predicted by applying the measured values to a previously prepared model formula. Therefore, it is necessary to manually create a model formula for predicting the wall thickness and the like.

[0006] On one side, an object is to provide an information processing method or the like that can preferably estimate the state of a magnetic tube.

Means for Solving the Problems

[0007] One aspect of the information processing method involves acquiring measurement data of the magnetic properties of a magnetic tube, and having a computer perform a process to estimate the wall thickness information by inputting the acquired measurement data into a model that has been trained to estimate wall thickness information of the magnetic tube when the measurement data is input. [Effects of the Invention]

[0008] In one respect, it allows for a favorable estimation of the state of the magnetic tube. [Brief explanation of the drawing]

[0009] [Figure 1] This is an explanatory diagram showing an example configuration of a defect estimation system. [Figure 2A] This is an explanatory diagram of the measuring device. [Figure 2B] This is an explanatory diagram of the measuring device. [Figure 3] This is a block diagram showing an example server configuration. [Figure 4] This is a block diagram showing an example of a terminal configuration. [Figure 5] This is an explanatory diagram showing an example of the record layout for the user database, model database, and measurement database. [Figure 6] This is an explanatory diagram showing an overview of Embodiment 1. [Figure 7] This is an explanatory diagram regarding data augmentation processing. [Figure 8] This is an explanatory diagram regarding the correction process for measurement data. [Figure 9] This is an explanatory diagram regarding image conversion processing. [Figure 10] This is a flowchart showing the steps involved in generating the estimation model. [Figure 11] This flowchart shows the procedure for estimating wall thickness information. [Figure 12] This is an explanatory diagram showing an overview of the first modified example. [Figure 13] This flowchart shows the procedure for generating the estimation model related to Modification Example 1. [Figure 14]It is a flowchart showing the procedure of the estimation process of the wall thickness information according to Modification Example 1. [Figure 15] It is an explanatory diagram showing the outline of Modification Example 2. [Figure 16] It is a flowchart showing the procedure of the generation process of the estimation model according to Modification Example 2. [Figure 17] It is a flowchart showing the procedure of the estimation process of the wall thickness information according to Modification Example 2. [Figure 18] It is an explanatory diagram showing the outline of Modification Example 3. [Figure 19] It is a diagram showing the learning weighting in Modification Example 3. [Figure 20] It is a flowchart showing the procedure of the generation process of the estimation model according to Modification Example 3. [Figure 21] It is a flowchart showing the procedure of the estimation process of the wall thickness information according to Modification Example 3. [Figure 22] It is an explanatory diagram for showing the influence of the hole element voltage due to the magnetic flux disturbance in the MFR method. [Figure 23A] It is an explanatory diagram showing the outline of the dynamic time stretching method in Modification Example 4. [Figure 23B] It is an explanatory diagram showing the outline of the dynamic time stretching method in Modification Example 4. [Figure 24] It is a diagram showing the detected voltage and the estimated wall thickness information in Modification Example 4. [Figure 25] It is an explanatory diagram regarding the data augmentation process in Modification Example 5. [Figure 26] It is an explanatory diagram showing an example of the upload screen of the measurement data. [Figure 27] It is an explanatory diagram showing an example of the display screen of the wall thickness information. [Figure 28] It is a flowchart showing the procedure of the estimation process of the wall thickness information according to Embodiment 2.

Modes for Carrying Out the Invention

[0010] Hereinafter, the present invention will be described in detail based on the drawings showing its embodiments. (Embodiment 1) Figure 1 is an explanatory diagram showing an example configuration of a defect estimation system. In this embodiment, a defect estimation system that estimates defects such as wall thinning from measurement data obtained by measuring the magnetic properties of a magnetic tube using a machine learning model is described. The defect estimation system includes an information processing device 1, a user terminal 2, a measuring device 3, and an inspection probe 4. The information processing device 1 and the user terminal 2 are connected via a network N such as the Internet.

[0011] In this embodiment, the magnetic tube to be measured is a tube material formed from a magnetic material such as carbon steel, ferritic stainless steel, or duplex stainless steel consisting of a ferrite phase and an austenite phase. Note that carbon steel and the like are just examples of magnetic materials, and the members (magnetic materials) forming the magnetic tube are not limited to these.

[0012] The measuring device 3 is a measuring device for measuring the magnetic properties of a magnetic tube, and is a device for performing a wall thinning inspection of a magnetic tube using the magnetic flux resistance method proposed by the applicant. The measuring device 3 measures the magnetic properties (magnetic flux density) at each position inside the magnetic tube by inserting the inspection probe 4 into the magnetic tube and moving it inside the tube. As the measuring device 3, for example, the defect measuring device disclosed in Japanese Patent Publication No. 6579840, Japanese Patent Publication No. 6514592, and / or Japanese Patent Application Publication No. 2019-100850 can be used.

[0013] Figures 2A and 2B are explanatory diagrams of the measuring device 3. Figure 2A is a cross-sectional view perpendicular to the longitudinal direction (axial direction) of the magnetic tube, conceptually illustrating the cross-section of the magnetic tube into which the inspection probe 4 is inserted. Figure 2B is a cross-sectional view perpendicular to the short direction (radial direction) of the magnetic tube, conceptually illustrating the measurement of the magnetic flux density at a defective area of ​​the magnetic tube where thinning has occurred (hereinafter referred to as the "thinned area"). Based on Figures 2A and 2B, the magnetic flux resistance method will be briefly explained.

[0014] The inspection probe 4 comprises a yoke 41, a magnet 42, and a Hall element 43 (magnetic sensor). The yoke 41 is a hollow cylindrical magnetic material, such as a high-permeability metal like carbon steel. The shape of the yoke 41 is not limited to a hollow cylinder; it may be a rod, plate, or columnar shape, for example.

[0015] The magnets 42 and Hall elements 43 are mounted at equal intervals along the outer surface of the yoke 41. For example, the magnets 42 and Hall elements 43 are provided at eight locations on the outer surface of the yoke 41. The magnets 42 are positioned such that one magnetic pole faces the yoke 41 and the other magnetic pole faces the magnetic tube, and are polarized in the direction opposite to the magnetic tube.

[0016] As shown by the arrows in Figures 2A and 2B, the yoke 41 and magnet 42 constitute a magnetic circuit. The Hall element 43 is provided on this magnetic circuit and outputs an output voltage proportional to the magnetic flux density passing through the Hall element 43. Specifically, the Hall element 43 outputs a higher output voltage as its magnetic resistance decreases and the magnetic flux density passing through the Hall element 43 increases.

[0017] Specifically, as shown in Figure 2B(a), when the magnet 42 and Hall element 43 are placed outside the magnetic tube, the magnetic resistance is high, resulting in a low magnetic flux density and a low output voltage from the Hall element 43. On the other hand, as shown in Figure 2B(c), when the magnet 42 and Hall element 43 are placed in the sound part of the magnetic tube (the part where thinning has not occurred), the magnetic resistance is low, resulting in a high magnetic flux density and a high output voltage from the Hall element 43. As shown in Figure 2B(b), when the magnet 42 and Hall element 43 are placed in the thinned part, the thickness of the magnetic tube is thinner compared to the sound part, resulting in a higher magnetic resistance and a lower magnetic flux density. Therefore, the output voltage from the Hall element 43 is low.

[0018] In this way, the measuring device 3 measures the magnitude of the magnetic flux density by the level of the output voltage of the Hall element 43. The output voltage of the Hall element 43 changes according to the thickness of the magnetic tube, becoming lower as the thickness decreases.

[0019] Magnetic flux resistance is a water immersion rotation ultrasonic thickness measurement method that uses ultrasound. ( Compared to the IRIS (Internal Rotary Inspection System), it has the advantage of shorter inspection time per magnetic tube and the ability to inspect a large number of magnetic tubes. In addition, there are other inspection methods that utilize magnetism, such as the remote field method (RFECT) and magnetic flux leakage method (MFL), but these measurement methods have lower measurement accuracy compared to the magnetic flux resistance method.

[0020] In this embodiment, when estimating the wall thickness of the magnetic tube from the measurement data using the magnetic flux resistance method described above, a machine learning model is used to estimate the wall thickness.

[0021] Furthermore, the inspection probe 4 for the magnetic flux resistance method includes a P-type (Japanese Patent No. 6514592) in which there is a gap between the yoke 41 and the Hall element 43, as shown in Figure 2B, and a PR-type (Japanese Patent Publication No. 2019-100850) in which there is no gap between the yoke 41 and the Hall element 43 and the yoke 41 and Hall element 43 are in close contact. In both cases, the output voltage is proportional to the magnetic flux density, but in the P-type, the output voltage decreases as the magnetic resistance increases, while in the PR-type, the output voltage increases as the magnetic resistance increases. Above, the measurement principle of the magnetic flux resistance method was explained using the P-type as an example for simplicity, but in the configuration described below, the case where measurement is performed using the PR-type inspection probe 4 will be explained. Of course, either the P-type or PR-type can be used.

[0022] Furthermore, although this embodiment describes the measurement data as being measured by the magnetic flux resistance method, the measurement data is not limited to this. For example, eddy current values ​​measured by RFECT may be used as measurement data. Alternatively, leakage flux (magnetic flux passing through the thinned section) measured by MFL may be used as measurement data. In other words, the measurement data can be any value that represents the magnetic properties of the magnetic tube, and is not limited to measurement data by the magnetic flux resistance method.

[0023] Let's return to Figure 1 and continue the explanation. The information processing device 1 is an information processing device capable of various information processing and information transmission and reception, such as a server computer or a personal computer. In this embodiment, the information processing device 1 is assumed to be a server computer, and for simplicity, it will be read as server 1 below. Server 1 learns predetermined training data and generates an estimation model 50 (see Figure 6) that estimates wall thickness information of a magnetic tube when measurement data of the magnetic properties of the magnetic tube is input. Server 1 then obtains measurement data of the magnetic tube to be measured from the user terminal 2 and inputs it into the estimation model 50 to estimate the wall thickness information of the magnetic tube. As will be described later, the wall thickness information is the wall thickness (thickness) of the magnetic tube, but it may also be the wall thickness, wall thickness reduction width, wall thickness reduction rate, wall thickness reduction range, etc.

[0024] User terminal 2 is a terminal device used by users of this system, such as a personal computer or tablet. For simplicity, user terminal 2 will be referred to as terminal 2 below. Users of this system are, for example, inspection companies that perform non-destructive testing of magnetic tubes, but are not limited to such companies. Although only one terminal 2 is shown in Figure 1, it is assumed that multiple users each have their own terminals 2, 2, 2... connected to server 1. Terminal 2 is connected to measuring device 3, transmits measurement data of magnetic tubes measured by measuring device 3 to server 1, and obtains estimated wall thickness information based on estimation model 50 from server 1.

[0025] In this embodiment, the server 1 on the cloud performs the estimation of wall thickness information based on the estimation model 50, but this embodiment is not limited to this. For example, a local terminal 2 may pre-install the estimation model 50 data from the server 1 and estimate wall thickness information by inputting measurement data acquired from the measuring device 3 into the estimation model 50. Thus, the device that generates the estimation model 50 and the device that performs the estimation of wall thickness information based on the estimation model 50 may be different.

[0026] Figure 3 is a block diagram showing an example configuration of Server 1. Server 1 comprises a control unit 11, a main memory unit 12, a communication unit 13, and an auxiliary memory unit 14. The control unit 11 has one or more arithmetic processing units such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), and a GPU (Graphics Processing Unit), and performs various information processing by reading and executing a program P1 stored in the auxiliary memory unit 14. The main memory unit 12 is a temporary storage area such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), or flash memory, and temporarily stores data necessary for the control unit 11 to perform information processing. The communication unit 13 is a communication module for performing communication-related processing and sends and receives information with the outside.

[0027] The auxiliary storage unit 14 is a non-volatile storage area such as a large-capacity memory or hard disk, and stores the program P1 and other data necessary for the control unit 11 to execute processing. The auxiliary storage unit 14 also stores the estimation model 50, user DB 141, model DB 142, and measurement DB 143. The estimation model 50 is a machine learning model that has been trained on predetermined training data, and when measurement data of the magnetic properties of a magnetic tube is input, it is a model that estimates thickness information regarding the wall thickness of a magnetic tube. The estimation model 50 is intended to be used as a program module that constitutes part of the artificial intelligence software.

[0028] User DB141 is a database that stores user information for this system. Model DB142 is a database that stores information for each of the multiple estimation models 50 prepared according to the thinning pattern (shape pattern of the thinned part), the dimensions of the magnetic tube, etc. As will be described later, in this embodiment, multiple estimation models 50 are prepared according to the thinning pattern, dimensions, etc. of the magnetic tube to be estimated. Measurement DB143 is a database that stores measurement data of the magnetic tube acquired from terminal 2.

[0029] The auxiliary storage unit 14 may be an external storage device connected to the server 1. Furthermore, the server 1 may be a multi-computer system consisting of multiple computers, or it may be a virtual machine virtually constructed by software.

[0030] Furthermore, in this embodiment, the server 1 is not limited to the above configuration and may include, for example, an input unit for receiving operation input, a display unit for displaying images, etc. The server 1 may also include a reading unit for reading a non-temporary computer-readable recording medium 1a and read the program P1 from the recording medium 1a. The program P1 may be executed on a single computer or on multiple computers interconnected via a network N.

[0031] Figure 4 is a block diagram showing an example configuration of terminal 2. Terminal 2 comprises a control unit 21, a main memory unit 22, a communication unit 23, a display unit 24, an input unit 25, and an auxiliary storage unit 26. The control unit 21 is one or more arithmetic processing units such as CPUs and MPUs, and performs various information processing by reading and executing the program P2 stored in the auxiliary storage unit 26. The main memory unit 22 is a temporary storage area such as RAM, and temporarily stores data necessary for the control unit 21 to perform information processing. The communication unit 23 is a communication module for performing communication-related processing, and transmits and receives information with the outside.

[0032] The display unit 24 is a display screen such as a liquid crystal display and displays images. The input unit 25 is an operation interface such as a keyboard or mouse and accepts operation input from the user. The auxiliary storage unit 26 is a non-volatile storage area such as a hard disk or large-capacity memory and stores the program P2 and other data necessary for the control unit 21 to execute processing.

[0033] Terminal 2 may also include a reading unit for reading a non-temporary computer-readable recording medium 2a, and may read program P2 from the recording medium 2a. Furthermore, program P2 may be executed on a single computer, or on multiple computers interconnected via a network N.

[0034] Figure 5 is an explanatory diagram showing an example of the record layout of User DB 141, Model DB 142, and Measurement DB 143. User DB 141 includes User ID column, User Name column, Usage History column, and Usage Fee column. The User ID column stores the User ID to identify each user. The User Name column, Usage History column, and Usage Fee column each store the User Name, the User's Usage History of Estimation Model 50 (Estimation History of Thickness Information), and the Usage Fee for Estimation Model 50 (System Usage Fee) determined according to the Usage History, respectively, in association with the User ID. The Usage Fee will be explained in detail in Embodiment 2.

[0035] Model DB142 includes columns for Model ID, Model Name, and Target. The Model ID column stores the Model ID for identifying each estimated model 50, which is prepared according to the thinning method of the magnetic tube, etc. The Model Name column and Target column store the model name of the estimated model 50 and magnetic tube information related to the magnetic tube to be estimated, respectively. For example, the Target column stores the thinning method, dimensions, and material of the magnetic tube.

[0036] The measurement DB143 includes columns for data ID, date and time, providing user, target, measurement data, and second measurement data. The data ID column stores a data ID to identify the measurement data provided by each user. The date and time column, providing user column, target, measurement data column, and second measurement data column each store, in association with the data ID, the date and time the measurement data was provided (acquisition date and time), the username of the providing user, information about the magnetic tube that was measured (thinning form, etc.), the measurement data of the magnetic tube, and the second measurement data, respectively. The second measurement data is measurement data of the wall thickness of the magnetic tube, which is measured using a different method than the measurement data using the magnetic flux resistance method. For example, it is measurement data obtained by the water immersion rotary ultrasonic thickness measurement method (IRIS; Internal Rotary Inspection System). The second measurement data will be described in detail in Embodiment 2.

[0037] Figure 6 is an explanatory diagram illustrating the overview of Embodiment 1. Figure 6 conceptually illustrates how the wall thickness of each part of the magnetic tube is estimated when measurement data of the magnetic properties of the magnetic tube measured by the measuring device 3 (inspection probe 4) is input into the estimation model 50. Based on Figure 6, the overview of this embodiment will be explained.

[0038] As described above, Server 1 generates an estimation model 50 that estimates wall thickness information when measurement data of magnetic properties is input, by learning predetermined training data. The estimation model 50 in this embodiment is a neural network generated by deep learning, such as a CNN (Convolutional Neural Network, ResNet, etc.).

[0039] Note that the estimated model 50 may be a neural network other than a CNN. Also, as shown in the modified examples 1 and 2 described later, the estimated model 50 may be a model based on a learning algorithm other than a neural network, such as kNN (k-Nearest Neighbor Algorithm), SVM (Support Vector Machine), random forest, or decision tree.

[0040] The estimated model 50 according to this embodiment takes measurement data based on the magnetic flux resistance method, measured using the measuring device 3, as input. As already mentioned, the measurement data is the value of the magnetic flux density passing through the magnetic circuit formed by the yoke 41 and magnet 42 on the inspection probe 4 and the magnetic tube, measured by the Hall element 43, and is the output voltage of the Hall element 43 which is proportional to the magnetic flux density.

[0041] When measuring a magnetic tube, the user inserts the inspection probe 4 into the opening of the magnetic tube and moves (sweeps) it to the opposite opening. This measures the magnetic properties at each position along the longitudinal direction of the magnetic tube. As described above, multiple Hall elements 43 are attached to the yoke 41 of the inspection probe 4, and the measuring device 3 measures the magnetic properties at each position along the longitudinal direction on a cross-section perpendicular to the longitudinal direction. Specifically, multiple Hall elements 43 are periodically attached to the outer circumference of the cylindrical yoke 41, and the measuring device 3 measures the magnetic properties at each position that divides the cross-section of the cylindrical magnetic tube equally along the circumferential direction.

[0042] In this embodiment, the magnetic tube is described as being cylindrical, but the cross-sectional shape of the magnetic tube is not limited to a circle; for example, it may be square or other shapes. That is, the measuring device 3 only needs to be able to measure the magnetic property values ​​at each position on the cross-section perpendicular to the longitudinal direction of the magnetic tube using the multiple Hall elements 43 provided on the inspection probe 4, and the "each position on the cross-section" is not limited to the positions that divide the cross-section of the cylindrical magnetic tube equally along the circumferential direction.

[0043] In the following explanation, the Hall elements 43 that function as magnetic sensors will be referred to as "channels" as appropriate, and each Hall element 43 will be assigned a channel number, which will be written as "CH1", "CH2", "CH3", ... "CH8".

[0044] As described above, the measurement of the magnetic tube allows for the observation of waveform data of the magnetic characteristic values ​​(output voltage) of each channel, which are continuous along the longitudinal direction of the magnetic tube, as shown on the left side of Figure 6. The horizontal axis of the waveform data represents the position in the longitudinal direction (axial direction) of the magnetic tube, and the vertical axis represents the output voltage of the Hall element 43. Server 1 uses this waveform data as input to the estimation model 50.

[0045] The estimation model 50 outputs (estimates) wall thickness information of the magnetic tube when the above measurement data is input. Specifically, the estimation model 50 estimates the wall thickness (thickness) of the magnetic tube. In this embodiment, the problem handled by the estimation model 50 is defined as a classification problem, and the estimation model 50 classifies the wall thickness of the magnetic tube in predetermined length increments (e.g., 0.1 mm increments) within a predetermined numerical range (e.g., 1.0 to 2.3 mm). Of course, the estimation model 50 may also be used as a regression model.

[0046] Furthermore, in this embodiment, the estimation model 50 estimates the wall thickness of the magnetic tube, but this embodiment is not limited to this. For example, the estimation model 50 may estimate the wall thickness of the thinned portion by subtracting the wall thickness of the thinned portion from the wall thickness of the healthy portion, rather than the wall thickness of the magnetic tube. Alternatively, the estimation model 50 may estimate the wall thickness by dividing the wall thickness of the thinned portion by the wall thickness of the healthy portion. Also, the estimation model 50 may estimate the location of the thinned portion within the magnetic tube, the range of the thinned portion (width of the thinned portion along the longitudinal and / or circumferential directions), the thinned portion form (shape pattern of the thinned portion), etc.

[0047] Furthermore, while this embodiment estimates the thinning of the magnetic tube, it is not limited to this, and an increase in wall thickness due to rust or the like may also be estimated.

[0048] Thus, the wall thickness information output from the estimation model 50 only needs to be information related to the wall thickness of the magnetic tube. The estimated wall thickness information is not limited to the wall thickness of the magnetic tube itself, and the defects targeted for estimation are not limited to wall thinning.

[0049] For example, estimation model 50 estimates the wall thickness of the magnetic tube for each section obtained by dividing the magnetic tube into fixed-length intervals along its longitudinal direction. The right side of Figure 6 conceptually illustrates the wall thickness estimation results. Estimation model 50 divides the magnetic tube into fixed-length intervals (e.g., 3 cm) and estimates the minimum wall thickness of each section. Note that estimation model 50 may also estimate the wall thickness continuously without dividing it into fixed-length intervals, and may estimate values ​​other than the minimum value (e.g., the maximum and average wall thickness of each section). Furthermore, estimation model 50 may estimate the maximum wall thickness for each channel, or it may estimate the minimum wall thickness of all channels combined as a representative value.

[0050] In the above, the magnetic property values ​​of each position (each channel) on the cross-section at each position in the longitudinal direction of the magnetic tube are input to the estimation model 50, but this embodiment is not limited to this. For example, the server 1 may input only the magnetic property value of a single position (cross-section) in the longitudinal direction of the magnetic tube to the estimation model 50 and estimate the wall thickness information of that position. Alternatively, for example, the server 1 may input only the magnetic property value of a single position (channel) on the cross-section of the magnetic tube to the estimation model 50 and estimate the wall thickness information of that position.

[0051] Server 1 generates an estimation model 50 using training data to which the correct values ​​(labels) of wall thickness information are associated with the training measurement data set. The training measurement data consists of actual measurement data of one or more magnetic tubes by the measuring device 3, and is waveform data of each channel in which magnetic characteristic values ​​were measured at each position in the longitudinal direction. Specifically, the training measurement data is data obtained by inserting the inspection probe 4 from both ends of the magnetic tube while changing the relative position of the inspection probe 4 in the circumferential direction so that CH1 is approximately 0°, 90°, 180°, and 270°, with a certain radial direction (position on the cross-section) as 0°. In other words, there is data obtained by performing 4 × 2 = 8 measurements for each magnetic tube. The correct values ​​are data obtained by measuring the wall thickness of the magnetic tube using a 3D shape measuring instrument, and the wall thickness was measured at 0.2 mm intervals from the end of the magnetic tube, in increments of approximately 23 μm. Server 1 generates an estimation model 50 using training data to which the correct values ​​of the wall thickness of each part are assigned to the training measurement data.

[0052] Figure 7 is an explanatory diagram regarding the data augmentation process. Server 1 may use only the raw measurement data measured by the inspection probe 4 as input to the estimation model 50. However, in this embodiment, in order to improve the accuracy of the estimation of wall thickness information, the raw measurement data is subjected to certain preprocessing, and the preprocessed measurement data is used as input to the estimation model 50. Furthermore, Server 1 may learn only the training data (actual measurement data) described above. However, in this embodiment, in order to increase the amount of training data, the training measurement data (actual measurement data) is augmented. Server 1 learns the augmented training data and generates the estimation model 50. Figures 7(a) to (d) conceptually illustrate how the preprocessed measurement data is augmented.

[0053] Figure 7(a) conceptually illustrates the preprocessed data, in which summary statistics such as the mean and standard deviation are added to the training measurement data, and data with normalized and corrected magnetic property values ​​are also added. Based on Figure 7, the preprocessing of measurement data will be explained.

[0054] Server 1 uses the longitudinal position where the magnetic characteristic value was measured and the raw measured values ​​of each channel ("CH1", "CH2", ... "CH8" in Figure 7(a)) as raw measurement data. The longitudinal position is used in addition to the measured value because the output voltage of the Hall element 43 changes drastically at the ends (inlet and outlet) of the magnetic tube, so the position of the tube end is also used as an explanatory variable. For the raw measurement data, Server 1 first normalizes the magnetic characteristic value of each channel so that the magnetic characteristic value outside the magnetic tube (in air), which is the maximum value, and the magnetic characteristic value at the sound part, which is the minimum value, are the same for all channels (all positions along the circumferential direction) (middle section of Figure 7(a)). Specifically, Server 1 corrects the output voltage of the Hall element 43 outside the magnetic tube to 1V and the output voltage at the sound part to 0V. In the case of P-type, the minimum value is outside the magnetic tube and the maximum value is at the sound part. By normalizing the magnetic characteristic values, it is possible to correct the sensitivity variations inherent in each Hall element 43, as well as correct the effect of temperature on the Hall element 43.

[0055] Next, Server 1 corrects the measurement data for each channel to minimize the effects of lift-off (and overall thinning) of the inspection probe 4 (lower part of Figure 7(a)).

[0056] Figure 8 is an explanatory diagram regarding the correction process for measurement data. The left side of Figure 8 shows the measurement data before correction, and the right side shows the measurement data after correction. Ideally, the inspection probe 4 should pass through the central axis of the magnetic tube during measurement, but in reality, as shown on the left side of Figure 8, the position of the inspection probe 4 inside the magnetic tube depends on the user's operation, and the distance between each channel CH1, CH2, CH3, etc. and the tube wall is often different. Therefore, the server 1 minimizes the effect of lift-off by correcting the magnetic characteristic values ​​of each channel to data that would be obtained assuming the inspection probe 4 passes through the central axis of the magnetic tube.

[0057] Specifically, Server 1 smooths the measurement data by taking a longitudinal moving average for each channel, thereby identifying the baseline for each channel. Server 1 then corrects the values ​​of each channel so that the average values ​​of the baselines of each channel match. In other words, Server 1 adjusts the waveform height so that the waveform shapes of the smoothed channels are nearly identical. The longitudinal length for taking the moving average and the number of averages are adjusted according to the length (total length) of the magnetic tube.

[0058] Furthermore, it is preferable for Server 1 to correct the measurement data to include only localized areas of thinning by removing data from the portion where the output voltage changes smoothly, i.e., the portion where the entire wall is thinned. The portion where the entire wall is thinned refers to the portion where the change in the output voltage (magnetic characteristic value) of the Hall element 43 is below a certain level. For example, Server 1 identifies the longitudinal range where the entire wall is thinned by calculating the average value of the baseline of all channels, and removes the data from the entire wall by replacing the output voltage in that range with the value of the healthy portion (i.e., 0V). This cuts out the voltage change in the entire wall, and the data can be adjusted to include only localized areas of thinning (the portion where the change in output voltage is above a certain level).

[0059] Let's return to Figure 7 and continue the explanation. As described above, Server 1 generates normalized data from the raw measurement data and corrected data from which the effects of lift-off (and the entire wall thinning area) have been removed. Server 1 may use only the normalized data and / or the corrected data as input to the estimation model 50, but in this embodiment, all three types of data—raw measurement data, normalized data, and corrected data—are used as input.

[0060] Server 1 uses data for the estimation model 50, which includes predetermined summary statistics added to each of the three types of data. The summary statistics are, but are not limited to, the mean, standard deviation, skewness, and / or kurtosis. Server 1 calculates the mean, standard deviation, skewness, and / or kurtosis along the longitudinal and / or circumferential directions from the magnetic property values ​​at each position in the longitudinal and / or circumferential directions, and adds them to the data.

[0061] Specifically, Server 1 calculates the mean value, standard deviation, skewness, and kurtosis for each position (channel) along the circumferential direction of the magnetic tube and adds them to the data. Server 1 also calculates the moving average, moving standard deviation, moving skewness, and moving kurtosis for multiple positions along the longitudinal direction of the magnetic tube (e.g., 10, 20, 30, 40, and 50 points) and adds them to the data. In the table in Figure 7, "mean value," "standard deviation," "skewness," and "kurtosis" correspond to the circumferential summary statistics, while "10-point moving average" and "20-point moving average" correspond to the longitudinal summary statistics (moving average). For illustrative purposes, the moving standard deviation, moving skewness, and moving kurtosis are not shown in Figure 7. Adding these summary statistics improves the estimation accuracy of wall thickness information. In particular, according to the inventor's research, adding skewness and kurtosis reduced the estimation error.

[0062] In this manner, Server 1 applies certain preprocessing to the raw measurement data and uses the preprocessed measurement data as input to the estimation model 50. The final preprocessed measurement data is a matrix of N rows × 275 columns (where N is the number of measurement points along the longitudinal direction). Similarly, when estimating wall thickness information from magnetic tube measurement data using the estimation model 50, Server 1 applies preprocessing to the measurement data acquired from Terminal 2 and inputs the preprocessed data into the estimation model 50.

[0063] During training, Server 1 expands the measurement data shown in Figure 7(a) to increase the amount of training data. As a first data expansion process, Server 1 generates multiple patterns of measurement data by shifting the position on the cross-section of the magnetic tube where the magnetic properties were measured along the circumferential direction. Specifically, Server 1 generates measurement data by shifting the channel number of the measurement data for each channel. Figure 7(b) conceptually illustrates the first data expansion process. First, Server 1 shifts the channel number by one, generating measurement data where the data from CH1 becomes CH2, the data from CH2 becomes CH3, ... the data from CH8 becomes CH1. Next, Server 1 shifts the channel number by two, generating measurement data where the data from CH1 becomes CH3, the data from CH2 becomes CH4, ... the data from CH8 becomes CH2. In the same manner, Server 1 shifts the channel number and increases the amount of data by eight times. This allows for the generation of measurement data for when a thinning section is detected in CH2, CH3, etc., based on the measurement data for when a thinning section is detected in CH1, thereby correcting the bias in the channels (positions on the cross-section of the magnetic tube) where thinning sections are detected.

[0064] Next, as a second data augmentation process, Server 1 generates multiple patterns of measurement data by adding multiple patterns of predetermined noise to the measurement data. Figure 7(c) illustrates the second data augmentation process. For example, Server 1 increases the number of data points tenfold by adding 10 patterns of Gaussian noise (mean 0, variance 0.02) to the measurement data. This allows for the reproduction of minute positional shifts of thinning areas, such as when a thinning area is located between one channel and another, as shown on the right side of Figure 7(c), and enables the estimation model 50 to learn from these shifts.

[0065] As a third data augmentation process, Server 1 generates measurement data with a changed sampling rate when measuring magnetic property values. Figure 7(d) illustrates the third data augmentation process. Server 1 performs linear interpolation between two consecutive points in the longitudinal direction on the training measurement data to approximate the magnetic property values ​​at the position between the two points. For example, Server 1 generates data with 1-point interpolation, 2-point interpolation, ... 20-point interpolation to generate measurement data with sampling rates from 1 to 6 times. This reproduces changes in the movement speed (sweep speed) of the inspection probe 4 and allows the estimation model 50 to learn from it.

[0066] Note that, for convenience, the tables in Figure 7(b) to (d) only show the measured values ​​for each channel ("CH1", "CH2", ...), but of course, the data should be expanded to include the summary statistics mentioned above.

[0067] Server 1 expands the number of measurement data by 8 × 10 × 6 = 480 times by performing the first to third data augmentation processes. This increases the amount of training data and simultaneously allows the estimation model 50 to learn about channel bias, positional shifts of minute thinning areas, and changes in movement speed.

[0068] Returning to Figure 6, let's continue the explanation. Server 1 generates an estimation model 50 using the expanded measurement data as training data. Server 1 could use the measurement data (magnetic property values) as input to the estimation model 50, but as shown in Figure 6, in this embodiment, the measurement data is converted into an image, and this image is used as input to the estimation model 50.

[0069] Figure 9 is an explanatory diagram regarding the image conversion process. Figure 9 illustrates how measurement data is converted into an image. In this embodiment, Server 1 generates an image in which the magnetic property values ​​of each position in the magnetic tube are represented by the pixel values ​​of each pixel, and uses this as input to the estimation model 50.

[0070] Specifically, Server 1 generates an image in which the magnetic property values ​​of each position are represented by the color (e.g., RGB) of each pixel. In this image, the vertical axis (first axis) corresponds to the position along the longitudinal direction of the magnetic tube, and the horizontal axis corresponds to the position on the cross-section of the magnetic tube perpendicular to the longitudinal direction, i.e., the channel. For example, Server 1 divides the measurement data in matrix form (see Figure 7), where each position along the longitudinal direction of the magnetic tube is represented by rows, and the raw measured value, normalized value, corrected value, and summary statistics for each channel are represented by columns, into predetermined numbers of rows (e.g., 224 rows). Server 1 then sequentially stores the measurement data from rows 1 to 224, the measurement data from rows 225 to 448, and so on, from the left edge of the image, thereby assigning the measurement data for each position to each pixel in the image. If there are any remaining pixels in the image at the end, Server 1 sets the R, G, and B values ​​of the remaining portion to 0. Depending on the sampling rate during the third data expansion, most of the values ​​may become 0.

[0071] Server 1 generates hue images for each of the R, G, and B hues from the training measurement data, and then generates the final input image by combining these hue images. Specifically, Server 1 calculates power values ​​of multiple patterns of magnetic characteristic values ​​with different exponents from the measurement data, and generates hue images for each pattern by assigning the power values ​​of each pattern to each hue.

[0072] For example, Server 1 calculates the 0th, 1st, and 2nd power values ​​for the magnetic property values ​​at each position within the magnetic tube. Server 1 then assigns each power value to R, G, and B. Figure 9 illustrates the case where the 1st power value is assigned to R, the 2nd power value to G, and the 0th power value to B. Server 1 generates each hue image by storing the power values ​​of the magnetic property values ​​at each position as the pixel values ​​of the pixels corresponding to that position. Server 1 then synthesizes each hue image to generate the final input image.

[0073] In the following explanation, the image created by combining each hue image will be referred to as a "composite image."

[0074] The image conversion method described above is merely an example, and this embodiment is not limited thereto. For example, Server 1 may generate a composite image by assigning power values ​​from 0 to 4 to each hue, with the final composite image being a CMYK image instead of an RGB image. Furthermore, the hues to which power values ​​are assigned are not limited to primary colors (RGB). Also, the exponent of the power value is not limited to a natural number, but may be a real number other than a natural number (e.g., to the power of 1.5). In addition, Server 1 may generate a monochrome image by simply storing the magnetic property values ​​of each position in the magnetic tube in each pixel without calculating power values. Furthermore, Server 1 may represent the magnetic property values ​​with other pixel values ​​(e.g., grayscale) instead of color. Thus, Server 1 only needs to be able to generate an image in which the pixel value of each pixel is assigned according to the magnetic property value of each position in the magnetic tube, and the method of generation is not particularly limited.

[0075] During training, Server 1 generates composite images for all combinations of assigning each power value to each hue in order to prevent overfitting due to hue. That is, in addition to the composite image where the 0th power is assigned to R, the 1st power to G, and the 2nd power to B, Server 1 generates six patterns of composite images, such as the composite image where the 0th power is assigned to G, the 1st power to B, and the 2nd power to R, the composite image where the 0th power is assigned to B, the 1st power to R, and the 2nd power to G, and so on. Server 1 prevents overfitting due to hue by training the estimation model 50 with the composite images of each combination.

[0076] Returning to Figure 6, let's continue the explanation. Server 1 performs data augmentation on each measurement data given for training, and converts the augmented measurement data into a composite image. Server 1 then uses each composite image and the correct values ​​of the thickness information corresponding to each composite image as training data to generate an estimation model 50.

[0077] Specifically, Server 1 inputs the synthetic images used for training into the estimation model 50 to obtain an estimated value of the wall thickness information, and compares the estimated value with the ground truth value. Server 1 updates parameters such as the weights between neurons so that the estimated value and the ground truth value approximate each other. Server 1 sequentially learns using each pair of synthetic images and ground truth values, and finally generates an estimation model 50 with optimized weights and other parameters.

[0078] In this embodiment, Server 1 learns measurement data according to the magnetic tube being measured in the training measurement data and generates a plurality of estimation models 50 corresponding to various magnetic tubes. Specifically, Server 1 learns measurement data separately according to the magnetic tube information relating to the magnetic tube being measured and generates a plurality of estimation models 50 corresponding to the magnetic tube information.

[0079] Magnetic tube information refers to information indicating the state or attributes of the magnetic tube, such as the thinning pattern, dimensions, and material of the magnetic tube. Note that these are just examples of magnetic tube information, and other information may also be included. The thinning pattern refers to the state of the thinned portion of the magnetic tube, defined by the shape pattern of the thinned portion (for example, whether the cross-sectional shape of the thinned portion is rectangular or hemispherical). Dimensions are defined by the diameter and length of the magnetic tube. Material is defined by the type of magnetic material forming the magnetic tube.

[0080] Server 1 separately learns training data according to this magnetic tube information and generates an estimation model 50 corresponding to the thinning pattern, dimensions, etc., of the magnetic tube. In this embodiment, multiple estimation models 50 are generated, but it is also possible to train one estimation model 50 with training data regardless of the differences in magnetic tube information and estimate the wall thickness information with a single estimation model 50.

[0081] When actually estimating wall thickness information using the estimation model 50, Server 1 obtains measurement data of the magnetic tube measured by the user from Terminal 2 and estimates the wall thickness information by inputting the measurement data into the estimation model 50. Specifically, Server 1 obtains the measurement data and also accepts the user's input of magnetic tube information regarding the magnetic tube to be measured via Terminal 2. Server 1 selects one of several estimation models 50 according to the specified magnetic tube information. Server 1 generates hue images for each hue from the acquired measurement data and generates a composite image by combining each hue image. Server 1 inputs the composite image into the estimation model 50 selected above to estimate the wall thickness of each part of the magnetic tube divided into fixed-length intervals along its longitudinal direction. Server 1 returns the estimated wall thickness information to Terminal 2 for display.

[0082] The operation of terminal 2 (display screen, etc.) when estimating wall thickness information will be explained in detail in Embodiment 2.

[0083] Based on the above, according to this embodiment, the thinning state (wall thickness information) of the magnetic tube can be suitably estimated and presented to the user by using the estimation model 50.

[0084] Figure 10 is a flowchart showing the procedure for generating the estimated model 50. Based on Figure 10, the process of generating the estimated model 50 by machine learning will be explained. The control unit 11 of the server 1 acquires training data to which the correct values ​​of thickness information relating to the wall thickness of the magnetic tube are associated with a group of measurement data measuring the magnetic property values ​​of the magnetic tube (step S11). The measurement data is data measured at each position on the cross section perpendicular to the longitudinal direction (each position obtained by dividing the cross section of the cylindrical magnetic tube equally along the circumferential direction) at each position along the longitudinal direction of the magnetic tube. The magnetic property value is a measured value obtained using an inspection probe 4 which comprises a magnet 42 that generates a magnetic field, a yoke 41 positioned on the opposite side of the magnetic tube from the magnet 42, and a Hall element 43 (magnetic sensor) positioned between the yoke 41 and the magnetic tube and measuring the magnetic flux density passing through the yoke 41, the magnet 42, and the magnetic tube, and is the output voltage of the Hall element 43 which is proportional to the magnetic flux density. The thickness information is, for example, the wall thickness (thickness) of each part obtained by dividing the magnetic tube into fixed length intervals along the longitudinal direction.

[0085] The control unit 11 preprocesses the training measurement data (step S12). Specifically, the control unit 11 normalizes the magnetic characteristic values ​​so that the magnetic characteristic values ​​outside the magnetic tube and the magnetic characteristic values ​​in the healthy part match for each channel. The control unit 11 also identifies the baseline for each channel by taking a moving average of the magnetic characteristic values ​​along the longitudinal direction, and the baseline for each channel vinegar By matching the average values ​​of the lines, the magnetic characteristic values ​​are corrected to those that would occur if the inspection probe 4 passed through the central axis of the magnetic tube. The control unit 11 also adds summary statistics such as the mean and standard deviation to the measurement data.

[0086] The control unit 11 expands the number of measurement data (step S13). Specifically, the control unit 11 generates multiple patterns of measurement data by shifting the channel numbers of the data measured in each channel. The control unit 11 also generates multiple patterns of measurement data by adding multiple patterns of predetermined noise (such as Gaussian noise) to the measurement data. Furthermore, the control unit 11 generates multiple patterns of measurement data by performing linear interpolation between two consecutive points in the longitudinal direction.

[0087] The control unit 11 converts the expanded measurement data into an image in which the vertical axis (first axis) of the image is aligned to the longitudinal direction, and the horizontal axis (second axis) is aligned to the cross-sectional position, assigning pixel values ​​to each pixel according to the magnetic characteristic value of each position in the magnetic tube (step S14). Specifically, the control unit 11 calculates power values ​​of multiple patterns of magnetic characteristic values ​​with different exponents, assigns each power value to a different hue, and generates hue images for each hue. The control unit 11 then generates a composite image by combining each hue image. The control unit 11 generates composite images for all combinations when assigning each power value to each hue and uses them as training data.

[0088] The control unit 11 generates an estimation model 50 that estimates wall thickness information when measurement data is input, based on the generated composite image and the ground truth value of wall thickness information (step S15). For example, the control unit 11 generates a neural network such as a CNN as the estimation model 50. The control unit 11 inputs the composite image generated in step S14 into the estimation model 50 to obtain an estimated value of wall thickness information, compares the obtained estimated value with the ground truth value, and optimizes parameters such as the weights of the estimation model 50 so that the two approximate each other. The control unit 11 then completes the series of processes.

[0089] Figure 11 is a flowchart showing the procedure for estimating wall thickness information. Based on Figure 11, the process of estimating wall thickness information using the estimation model 50 will be explained. The control unit 11 of server 1 receives input from terminal 2 to select the estimation model 50 to be used for estimating wall thickness information (step S31). Specifically, the control unit 11 receives input specifying the thinning form, dimensions, and material of the magnetic tube. The control unit 11 selects the estimation model 50 corresponding to the specified magnetic tube information.

[0090] The control unit 11 acquires measurement data from the magnetic tube from the terminal 2 (step S32). The control unit 11 preprocesses the acquired measurement data (step S33). The control unit 11 converts the preprocessed measurement data into an image (step S34). Specifically, the control unit 11 assigns power values ​​of magnetic characteristic values ​​with different exponents to each hue, and generates a composite image by combining the images of each hue.

[0091] The control unit 11 inputs the image converted in step S34 into the estimation model 50 to estimate the wall thickness information of the magnetic tube (step S35). For example, the control unit 11 estimates the wall thickness of each part of the magnetic tube divided into fixed-length sections along its longitudinal direction. The control unit 11 transmits the estimated wall thickness information to the terminal 2 (step S36) and completes the series of processes.

[0092] Based on the above, according to this embodiment 1, the state of the magnetic tube can be suitably estimated.

[0093] (Variation 1) Embodiment 1 described a configuration in which measurement data from a magnetic tube is converted into an image and input into the estimation model 50. Modification 1 describes a configuration in which the original measurement data is used directly as input to the estimation model 50. Note that components that overlap with Embodiment 1 are denoted by the same reference numerals and their descriptions are omitted.

[0094] Figure 12 is an explanatory diagram illustrating the overview of Modification 1. Figure 12 illustrates how wall thickness information is estimated by extracting peak data corresponding to the thinned-wall portion from the measurement data of the magnetic tube and inputting the peak data into the estimation model 50. Based on Figure 12, the overview of this modification will be explained.

[0095] In this modified example, Server 1 constructs a model as the estimation model 50 that can directly handle the original measurement data (magnetic property values ​​at each position inside the magnetic tube) instead of images. For example, Server 1 constructs a model related to kNN (k-nearest neighbor method) as the estimation model 50.

[0096] In this modified example, the estimation model 50 is described as a model related to kNN, but the estimation model 50 can be any model capable of handling the original measurement data, such as SVM, random forest, or linear classifier.

[0097] Server 1 generates an estimation model 50 based on training data similar to that in Embodiment 1. It is preferable for Server 1 to perform principal component analysis or similar on the training measurement data to reduce its dimensionality. Reducing the dimensionality of the measurement data input to the estimation model 50 reduces the processing load on Server 1. The user may be allowed to arbitrarily choose whether or not to perform dimensionality reduction (principal component analysis).

[0098] Here, Server 1 may input the magnetic property values ​​of all positions along the longitudinal direction directly into the estimation model 50, but in this embodiment, only a portion of the measurement data is extracted and used as input to the estimation model 50. Specifically, Server 1 extracts the data of the peak portion corresponding to the thinned area and uses it as input to the estimation model 50.

[0099] Figure 12 illustrates how peak data is extracted from the original measurement data. Server 1 extracts data from the waveform of the output voltage of the Hall element 43, which corresponds to the measurement data, specifically the portion where the output voltage changes, i.e., the peak portion.

[0100] The method for extracting peak data is not particularly limited, but for example, Server 1 identifies the baseline (flat waveform) by taking a moving average of the output voltage and identifies the peak portion according to the baseline. Specifically, Server 1 identifies the portion where the difference from the baseline is greater than or equal to a predetermined threshold as the peak portion. As shown in Figure 12, when the Hall element 43 reaches a thinning section, the output voltage rises to a positive value and then falls to a negative value, but in this modified example, only the portion where the output voltage is positive and the waveform is convex upwards is identified as the peak portion.

[0101] Server 1 extracts data from the identified peak portion from the measurement data of the entire magnetic tube and uses it as input to the estimation model 50. This allows for the appropriate estimation of wall thickness information in the thinned-wall portion while reducing the amount of data processed by the estimation model 50.

[0102] Furthermore, the peak portion data extraction method in this modified example may be applied to Embodiment 1, and only the peak portion data may be converted from the input image and used as input for the estimation model 50 according to Embodiment 1.

[0103] Server 1 extracts peak data from the training measurement data and generates an estimation model 50 based on the extracted peak data and the ground truth values ​​of the wall thickness information at the corresponding peak locations. In other words, Server 1 maps the peak data, to which the ground truth values ​​of the wall thickness information (ground truth labels indicating the correct wall thickness) have been assigned, into the feature space. Then, by setting an arbitrary value k related to the k-nearest neighbors method, Server 1 constructs a dataset for estimating wall thickness information from the peak data.

[0104] When estimating wall thickness information, Server 1 extracts peak data from the measurement data acquired from Terminal 2 and maps it to the feature space. Server 1 then identifies k data points (peak data points with correct labels) located near the mapped position and estimates the wall thickness information of the target magnetic tube from the correct values ​​of wall thickness information assigned to these k data points.

[0105] Figure 13 is a flowchart showing the procedure for generating the estimated model 50 according to the modified example 1. After expanding the number of measurement data (step S13), the server 1 performs the following processing. The control unit 11 of the server 1 extracts data of the peak portion corresponding to the thinned portion from the measurement data of the magnetic property values ​​at each position along the longitudinal direction of the magnetic tube (step S201). Specifically, the control unit 11 identifies a baseline from the measurement data of the entire magnetic tube and extracts data of the peak portion where the difference from the baseline is greater than or equal to a predetermined threshold.

[0106] The control unit 11 generates an estimation model 50 based on the peak portion data extracted in step S201 and the ground truth values ​​for wall thickness information (step S202). Specifically, the control unit 11 generates a model related to kNN. The control unit 11 maps the peak portion data to which the ground truth values ​​(ground truth labels) for wall thickness information have been assigned to the feature space and sets an arbitrary value k to generate a dataset for estimating wall thickness information from the peak portion data as the estimation model 50. The control unit 11 then completes the series of processes.

[0107] Figure 14 is a flowchart showing the procedure for estimating wall thickness information according to Modification Example 1. After preprocessing the measurement data (step S33), Server 1 performs the following processing. The control unit 11 of Server 1 extracts data of peak portions corresponding to the thinned-wall portion from the measurement data of magnetic property values ​​at each position along the longitudinal direction of the magnetic tube (step S221). Specifically, the control unit 11 identifies a baseline from the measurement data of the entire magnetic tube and extracts data of peak portions where the difference from the baseline is greater than or equal to a threshold. The control unit 11 inputs the extracted peak portion data into the estimation model 50 to estimate the wall thickness information of the thinned-wall portion (step S222). Specifically, the control unit 11 maps the peak portion data to the feature space and estimates the wall thickness information of the thinned-wall portion of the target magnetic tube from the correct values ​​of wall thickness information of k data located nearby. The control unit 11 then proceeds to step S36.

[0108] As described above, according to Modification 1, it is possible to estimate wall thickness information using the original measurement data without converting the measurement data into an image. In particular, in this modification, by extracting data from the peak portion corresponding to the thinned area, the wall thickness information of the thinned area can be suitably estimated.

[0109] (Modification 2) This modified example describes a method for generating an estimated model 50 using an ensemble learning technique.

[0110] Figure 15 is an explanatory diagram illustrating the overview of Modification 2. In Figure 15, data from multiple data intervals, each slightly shifted along the longitudinal direction of the magnetic tube, is extracted from the peak portion data described in Modification 1. The thickness information is then estimated by inputting the data from each interval into the estimation model 50. Based on Figure 15, the overview of this modification will be explained.

[0111] In this modified example, Server 1 generates the estimated model 50 using an ensemble learning method. Specifically, Server 1 uses a gradient boosting method to generate a model related to a decision tree (e.g., LightGBM) as the estimated model 50.

[0112] Gradient boosting is a type of ensemble learning that generates a final model by sequentially generating multiple weak classifiers (models). Server 1 generates one weak classifier (decision tree) using the training data, and generates the next weak classifier based on the residual between the estimated value and the ground truth value generated by the generated weak classifier. Server 1 sequentially generates weak classifiers by referring to the gradient of the loss function defined by the error between the estimated value and the ground truth value, taking into account the learning results of the previous weak classifier, and generates the final classifier, i.e., the estimated model 50.

[0113] In this modified example, gradient boosting is used as the ensemble learning method, but other ensemble learning techniques may be used, such as bagging, which generates multiple weak classifiers in parallel. Alternatively, ensemble learning may be applied to the estimation model 50 related to Modified Example 1, and ensemble learning may be performed when generating models such as kNN and SVM.

[0114] Server 1 may use the peak portion data (or the measurement data of the entire magnetic tube) as input to the estimation model 50, as described in Modification 1. However, in this modification, the peak portion is further divided into data intervals of a predetermined length, and the data from each data interval is input to the estimation model 50.

[0115] Figure 15 illustrates how data from multiple data intervals is extracted from the peak portion of the data. These data intervals are segments of measurement points along the longitudinal direction of the magnetic tube and constitute a part of the peak portion. For example, Server 1 considers 10 consecutive measurement points as one data interval. Note that the number of measurement points constituting a data interval is not limited to 10.

[0116] Server 1 sequentially shifts the data interval to be extracted along its longitudinal direction and extracts data from the peak portion of each data interval. For example, Server 1 extracts 10 data points from the starting point of the peak portion where the difference from the baseline is greater than or equal to a threshold, as the data for the first data interval. Server 1 then shifts the starting point of the data interval to be extracted by a predetermined number (e.g., one) and extracts data from each data interval. In this way, Server 1 extracts data from multiple data intervals from the data of a single peak portion.

[0117] When Server 1 estimates wall thickness information for a single peak portion, it inputs data from multiple extracted data intervals into the estimation model 50 to estimate the wall thickness information. This allows for a more favorable estimation of the wall thickness in the thinned portion.

[0118] Figure 16 is a flowchart showing the procedure for generating the estimation model 50 according to the modified example 2. After expanding the measurement data (step S13), the server 1 performs the following processing. The control unit 11 of the server 1 extracts data for the peak portion corresponding to the thinned portion from the measurement data of the entire length of the magnetic tube measured along the longitudinal direction (step S301). Furthermore, the control unit 11 extracts data for multiple data intervals from the peak portion data (step S302). Specifically, the control unit 11 sequentially shifts the data interval of a predetermined length to be extracted along the longitudinal direction from the starting point of the peak portion, and extracts data for each data interval.

[0119] The control unit 11 generates an estimation model 50 based on the data from multiple data intervals extracted in step S302 and the correct values ​​of the wall thickness information in the peak portion (thinned portion) (step S303). Specifically, the control unit 11 generates a decision tree using the gradient boosting method. The control unit 11 generates one weak classifier (decision tree) using the training data (data from multiple data intervals and the correct values ​​of the wall thickness information), and generates the next weak classifier based on the residual between the estimated value and the correct value of the wall thickness information generated by the weak classifier. The control unit 11 sequentially generates weak classifiers according to the gradient of the loss function defined by the residual between the estimated value and the correct value, and generates the final classifier, i.e., the estimation model 50. The control unit 11 then completes the series of processes.

[0120] Figure 17 is a flowchart showing the procedure for estimating wall thickness information according to Modification Example 2. After preprocessing the measurement data (step S33), Server 1 performs the following processing. The control unit 11 of Server 1 extracts data for the peak portion corresponding to the thinned area from the measurement data of magnetic property values ​​at each position along the longitudinal direction (step S321). The control unit 11 extracts data for multiple data intervals from the peak portion data (step S322). The control unit 11 inputs the extracted data for multiple data intervals into the estimation model 50 to estimate the wall thickness information of the thinned area (step S323). The control unit 11 moves the processing to step S36.

[0121] Based on the above, according to Modification 2, it is also possible to generate the estimation model 50 using ensemble learning related to gradient boosting. Furthermore, by dividing the peak portion data into data from multiple data intervals and inputting them into the estimation model 50, the wall thickness information of the peak portion (thinned portion) can be estimated more favorably.

[0122] (Variation 3) This section describes an example using a different ensemble learning method than that described in Modification 2. A common ensemble learning method involves inputting the same dataset into multiple models with different learning mechanisms, and then using the average of the outputs from these multiple models, or the majority vote output from these multiple models, as the learning result.

[0123] In variation 3, unlike the previous learning method, the input data is divided into multiple sets, each divided set is input into a different corresponding model, the output of each corresponding model is input into yet another model, and the output of that other model is taken as the learning result.

[0124] Figure 18 is an explanatory diagram showing an overview of Modification 3. As shown in Figure 18, the estimation model in Modification 3 includes a thin-wall specialized model 50a (Bag Tree-A) which is learned mainly based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thinner than a predetermined value, a thick-wall specialized model 50b (Bag Tree-B) which is learned mainly based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thicker than a predetermined value, and a final model 50c (Bag Tree-C) which takes the outputs of the thin-wall specialized model 50a and the thick-wall specialized model 50b as input.

[0125] For example, the thin-walled specialized model 50a is a model that targets 90% of the data with a wall thickness of 0.1 to 0.8 mm and 10% of the data with a wall thickness exceeding 0.8 mm, while the thick-walled specialized model 50b is a model that targets the remaining data from the total data that was not used in the thin-walled specialized model 50a. Figure 19 shows the weighting relationship for both models 50a and 50b in such an example.

[0126] The criterion for distinguishing between the thin-walled model 50a and the thick-walled model 50b is set at 0.8 mm, but this criterion is merely an example, and other values ​​may be used. Also, for the thin-walled model 50a, 90% of the data with a wall thickness below a predetermined value and 10% of the data with a wall thickness exceeding a predetermined value were selected, but this 90%:10% ratio is merely an example, and other ratios may be used.

[0127] Figure 20 is a flowchart showing the procedure for generating the estimation models (thin-walled specialized model 50a, thick-walled specialized model 50b, and final model 50c) in modified example 3. The control unit 11 of server 1 acquires first training data, which includes measurement data mainly for thin-walled tubes as described above, and the correct values ​​of wall thickness information associated with the measurement data (step S401). Based on the acquired first training data (measurement data and correct values ​​of wall thickness information), the control unit 11 generates a thin-walled specialized model 50a that estimates and outputs first wall thickness information when measurement data is input (step S402).

[0128] The control unit 11 acquires second training data from the measurement data of the magnetic tube, mainly measurement data for thick walls as described above, and the correct values ​​of wall thickness information associated with the measurement data (step S403). Based on the acquired second training data (measurement data and correct values ​​of wall thickness information), the control unit 11 generates a thick-wall specialized model 50b that estimates and outputs second wall thickness information when measurement data is input (step S404).

[0129] The control unit 11 acquires third training data, which includes the first thickness information output from the thin-wall specialized model 50a, the second thickness information output from the thick-wall specialized model 50b, and the corresponding correct values ​​for the thickness information (step S405). Based on the acquired third training data (first and second thickness information and the correct values ​​for the thickness information), the control unit 11 generates a final model 50c that estimates and outputs the thickness information when the first and second thickness information is input (step S406).

[0130] Figure 21 is a flowchart showing the procedure for estimating wall thickness information in Modification Example 3. The control unit 11 of the server 1 receives a selection input from the terminal 2 for the estimation model to be used for estimating wall thickness information (thin-wall specialized model 50a, thick-wall specialized model 50b, and final model 50c) (step S411). Specifically, the control unit 11 receives a specification input that specifies the wall thickness boundary value that separates the thin-wall model from the thick-wall model, the proportion of thick walls included in the thin-wall model, etc. The control unit 11 selects an estimation model (thin-wall specialized model 50a, thick-wall specialized model 50b, and final model 50c) corresponding to the specification input.

[0131] The control unit 11 acquires measurement data of the magnetic tube from the terminal 2 (step S412). The control unit 11 inputs the acquired measurement data into the thin-walled specialized model 50a and the thick-walled specialized model 50b (step S413). Next, the control unit 11 inputs the output from the thin-walled specialized model 50a and the output from the thick-walled specialized model 50b into the final model 50c to estimate the wall thickness information of the magnetic tube (step S414). The control unit 11 transmits the estimated wall thickness information to the terminal 2 (step S36) and ends the series of processes.

[0132] In modification 3, instead of using the final model 50c, the average of the output from the thin-wall specialized model 50a and the output from the thick-wall specialized model 50b may be calculated, and the calculated average value may be used as the estimated wall thickness information of the magnetic tube.

[0133] According to Modification 3, different appropriate models are applied to the thin-walled and thick-walled sides, making it possible to estimate more accurate wall thickness information. Generally, the amount of data is often small on the thin-walled side, but even in such cases, it is possible to improve the accuracy of estimating wall thickness information on the thin-walled side where the amount of data is small.

[0134] (Modification 4) Modification 4 is an example of a method for correcting the measurement data used to train the estimation model using dynamic time stretching in order to reduce the effect of Hall element voltage due to magnetic flux disturbances in the MFR method.

[0135] First, we will explain the effect of the Hall element voltage due to magnetic flux disturbance in the above MFR method. Figure 22 is an explanatory diagram to illustrate this effect. As shown in the cross-sectional view of the magnetic tube in Figure 22, we assume that a deep local thinning A has occurred in the magnetic tube. M When scanning with the FR-type inspection probe 4, the MFR waveform to be detected by the Hall element 43 should ideally be waveform B, depending on the change in the thinning shape. However, the MFR waveform actually detected is like waveform C, and areas independent of wall thickness (especially the areas circled) may appear near the thinning area. This is because, as shown in the cross-sectional view of the magnetic tube, the magnetic field lines are distorted around the thinning area, resulting in a blank area of ​​magnetic field lines (dashed line). D This is because a region enclosed by the symbols will be created.

[0136] To resolve the above-mentioned problems, dynamic time stretching is used to correct the measurement data used to train the estimation model. Figures 23A and 23B are explanatory diagrams showing an overview of the dynamic time stretching in modified example 4. In Figures 23A and 23B, a shows the time-series waveform of the measurement data (detected voltage), specifically the time-series waveform of the value obtained by the 3-peak method (in the case of 8 channels, the value obtained by dividing the sum of the voltage value of the channel with the maximum detected voltage and the voltage values ​​of the two channels on either side of that channel by 8). b shows the time-series waveform of the measured wall thickness value as the target variable.

[0137] In Modification 4, as shown in Figure 23B, the time-series points of the three peak values ​​(waveform a) and the time-series points of the measured thickness value (waveform b) are linked so that the sum of the Euclidean distances between the three peak values ​​(measured data) and the measured thickness value (thickness information) is minimized. For example, the time-series point i of waveform b changes to correspond to the time-series point i+2, which is two points shifted from waveform a.

[0138] Figure 24 shows the results of estimating wall thickness information by inputting measurement data into an estimation model trained using the corrected measurement data described above. In Figure 24, waveform A shows the actual measurement data (MFR waveform), and wall thinning shape B shows the estimated wall thickness information (wall thinning shape). In Figure 24, the actual MFR waveform when dynamic time stretching is not used (corresponding to waveform C in Figure 22) is superimposed on waveform A and shown as a dashed line, and the actual wall thinning shape (corresponding to the wall thinning shape shown in Figure 22) is superimposed on the estimated wall thinning shape B and shown as a dashed line.

[0139] In the fourth modified example, dynamic time stretching is used to correct the measurement data used to train the estimation model. This reduces the influence of Hall element voltage due to magnetic flux disturbances in the MFR method described above, and allows for the estimation of the thinning shape of the magnetic tube with high accuracy.

[0140] (Variation 5) Modification 5 is an example of increasing the number of data points using the principal component analysis method. Figure 25 is an explanatory diagram of the data augmentation process in Modification 5.

[0141] Server 1 may use raw measurement data measured by inspection probe 4 having 8 channels (CH1~CH8), and first group data obtained by performing preprocessing including calculations on this raw measurement data (in the example shown in Figure 25, this includes moving average, moving standard deviation, moving skewness, moving kurtosis, and moving difference for multiple positions (10~50 points) along the longitudinal direction of the magnetic tube) as input to the estimation model 50 (see Figure 25(a)). In modification 5, in order to improve the accuracy of the wall thickness information estimation, in addition to these raw measurement data and first group data, second group data obtained by performing principal component analysis on the raw measurement data and first group data is used as input to the estimation model 50 (see Figure 25(b)).

[0142] In the example shown in Figure 25, the data for Group 1 and Group 2 are defined as moving average, moving standard deviation, moving skewness, moving kurtosis, and moving difference, respectively, but are not limited to these. Also, while the data for Group 2 has the same data structure as the data for Group 1, the data structures of the two groups may be different.

[0143] In variation 5, the number of data points is increased by adding the results of principal component analysis. ( In the example shown in Figure 25, the number of variables was increased from 169 to 338, which improves the accuracy of the wall thickness estimation.

[0144] (Embodiment 2) Embodiment 1 described a defect estimation system that estimates defects such as wall thinning in magnetic tubes using estimation model 50. This embodiment describes an embodiment in which a user actually uses this system to inspect defects in magnetic tubes.

[0145] Figure 26 is an explanatory diagram showing an example of a measurement data upload screen. Figure 26 illustrates the display screen when terminal 2 sends (uploads) measurement data to server 1. The upload screen includes a model selection field 171, an upload button 172, and a data availability selection field 173.

[0146] The model selection field 171 is an input field for selecting the estimation model 50 to be used for estimating wall thickness information, and is an input field for specifying magnetic tube information related to the magnetic tube to be estimated (measured). For example, the model selection field 171 includes the wall thinning type specification field 1711, the dimension specification field 1712, and the material specification field 1713. Terminal 2 is each designation The system accepts input for the thinning method, dimensions, and material of the magnetic tube via the provided fields.

[0147] The upload button 172 is used to send (upload) measurement data to server 1. When input is received to the upload button 172, terminal 2 sends the measurement data to server 1.

[0148] The provision availability selection field 173 is an input field for selecting whether or not to provide (use) measurement data to this system. Terminal 2 accepts the selection input regarding the provision of measurement data in response to the operation input to the provision availability selection field 173. If the selection input indicating that measurement data will be provided is received, Server 1 stores the measurement data obtained from Terminal 2 in the measurement DB 143. As will be described later, the measurement data provided by the user will be used to retrain (update) the estimation model 50.

[0149] Regardless of whether or not the option to provide the data is selected, all measurement data acquired from each terminal 2 may be used to retrain the estimation model 50.

[0150] Figure 27 is an explanatory diagram showing an example of a display screen for wall thickness information. Figure 27 illustrates the display screen for wall thickness information estimated based on the estimation model 50. This screen includes a magnetic tube selection field 181, a wall thickness graph 182, a cross-sectional image 184, a download button 185, and a second measurement data upload button 186.

[0151] The magnetic tube selection field 181 is an input field for selecting a magnetic tube whose wall thickness information will be displayed. Terminal 2 displays the wall thickness information of the magnetic tube selected in the magnetic tube selection field 181.

[0152] The wall thickness graph 182 is a graph showing the wall thickness of each part of the magnetic tube, estimated based on the estimation model 50. In the wall thickness graph 182, the horizontal axis represents the position in the longitudinal direction of the magnetic tube, and the vertical axis represents the wall thickness. For example, terminal 2 displays the average value of the wall thickness of each channel in the wall thickness graph 182. Alternatively, terminal 2 may display the wall thickness of each channel in the wall thickness graph 182.

[0153] Cross-sectional image 184 is an image that simulates a cross-section perpendicular to the longitudinal direction of the magnetic tube, and reproduces the wall thickness at each position on the cross-section according to the wall thickness information. As described above, Server 1 estimates the wall thickness of each channel (each position on the cross-section) at each position along the longitudinal direction of the magnetic tube. By comparing the estimated wall thickness of each channel with the wall thickness of the healthy section (baseline of the wall thickness graph 182), Server 1 detects the thinned section, which is thinner than the healthy section. Server 1 then calculates the thinning width, which is the difference in wall thickness between the healthy section and the thinned section. For example, Terminal 2 displays the cross-section of the magnetic tube with CH1 at the 12 o'clock position, and reproduces and displays the thinned section (recess) according to the calculated thinning width at the position corresponding to the channel where the thinned section was detected (CH1 in Figure 27).

[0154] For example, terminal 2 displays line 183 on the wall thickness graph 182. Terminal 2 accepts an input to specify the longitudinal position of the magnetic tube on which to display the cross-sectional image 184 by moving line 183 along the horizontal axis. Terminal 2 displays the cross-sectional image 184 of the magnetic tube corresponding to the position of line 183.

[0155] In addition, terminal 2 displays the number of thinned areas, the maximum and average values ​​of the thinning width, etc., based on the thickness information.

[0156] Note that the display screen for wall thickness information shown in Figure 27 is just an example, and this embodiment is not limited to it. For example, terminal 2 may simply display the wall thickness at each position inside the magnetic tube, as shown in the wall thickness graph 182. Also, for example, when terminal 2 displays the wall thickness at each position inside the magnetic tube in the wall thickness graph 182, etc., it may change the display manner (e.g., display color) of the wall thickness at each position according to the reliability of the estimation result of the estimation model 50 (a probability value representing the certainty of the estimation result). Also, for example, if terminal 2 detects multiple thinning areas from the wall thickness information, it may display a list of multiple cross-sectional images 184, 184, 184... corresponding to each thinning area. Thus, the display manner of wall thickness information is as shown in Figure 27. 27 It is not limited to the embodiments shown.

[0157] Furthermore, although in this embodiment terminal 2 displays thickness information on a dedicated screen, terminal 2 does not need to display the thickness information itself; it can simply download a file (for example, a CSV file) containing the estimated thickness information.

[0158] Download button 185 is a button for downloading a file containing the estimated wall thickness information. When input is received for download button 185, terminal 2 retrieves the file containing the estimated wall thickness information from server 1.

[0159] The second measurement data upload button 186 is used to upload second measurement data, which is measurement data other than the measurement data described above (measurement data based on the magnetic flux resistance method), to server 1. When input is received to the second measurement data upload button 186, terminal 2 sends measurement data obtained by the Immersion Rotation Ultrasonic Thickness Measurement Method (IRIS) to server 1 as second measurement data. The second measurement data will be described later.

[0160] As described above, Server 1 acquires measurement data of magnetic tubes from each user's terminal 2 and estimates the wall thickness information. Server 1 then transmits the wall thickness information to the terminal 2 that acquired the measurement data and displays it. In this embodiment, Server 1 retrains based on the measurement data acquired from each terminal 2 and updates the estimation model 50.

[0161] Specifically, if Server 1 receives a selection input indicating that measurement data will be provided via the provision availability selection field 173, it stores the measurement data acquired from Terminal 2 in the measurement DB 143. Server 1 uses the measurement data stored in the measurement DB 143 as training data for retraining. Server 1 updates the estimation model 50 based on the measurement data stored in the measurement DB 143 and the correct values ​​of the wall thickness information corresponding to that measurement data.

[0162] The correct values ​​for wall thickness information during retraining may be manually set by the user, for example, via the screen shown in Figure 27. However, in this embodiment, the second measurement data provided by the user is used as the correct values. The second measurement data is data measuring the wall thickness at each position inside the magnetic tube, and is, for example, measurement data related to IRIS.

[0163] IRIS is a method for inspecting the wall thickness of a tube (magnetic tube) using ultrasound. A probe equipped with an ultrasonic transducer is inserted into a water-filled tube, and the ultrasonic transducer detects the reflected waves of the ultrasonic beam generated from the probe to measure the wall thickness. Since IRIS itself is well known, a detailed explanation is omitted in this embodiment. Compared to inspection methods that utilize magnetic properties, such as the magnetic flux resistance method described in this embodiment, RFECT, and MFL, IRIS has the advantage of being slower in inspection speed but having relatively high measurement accuracy.

[0164] In this embodiment, when Server 1 receives measurement data related to IRIS from a user, it uses that data as the correct value for thickness information. That is, Server 1 uses a pair of measurement data based on the magnetic flux resistance method and second measurement data based on IRIS as training data for retraining.

[0165] Server 1 compares the estimated wall thickness information based on the measurement data provided by the user with the wall thickness indicated by the second measurement data, i.e., the correct value. Server 1 then updates parameters such as the weights between neurons so that the two approximate each other. Server 1 retrains on the measurement data for each magnetic tube provided by the user and updates the estimation model 50.

[0166] Server 1 may retrain a single estimation model 50 using measurement data from all users, but it is preferable to separately train the estimation model 50 using measurement data acquired from each user's terminal 2 and update it for each user. That is, Server 1 separately provides the estimation model 50 with measurement data from each user and builds an estimation model 50 for each user. This allows the estimation model 50 to be individually tuned according to the characteristics of the magnetic tubes that each user measures (inspects).

[0167] Based on the above, the estimation accuracy of estimation model 50 can be improved through the operation of this system.

[0168] Server 1 charges each user of this system a usage fee (charge) for the estimation model 50. The usage fee may be a fixed amount (subscription) every certain period, for example, but in this embodiment, it is determined according to the computational amount of the thickness information estimation process based on the estimation model 50. The computational amount that serves as the basis for determining the usage fee is, for example, the number of times the thickness information is estimated (number of magnetic tubes), but it may also be the computation time required for the estimation process, the amount of measurement data, etc. When Server 1 obtains measurement data from Terminal 2 and estimates the thickness information, it determines the usage fee to be charged to the user according to the computational amount at the time of estimation. Server 1 stores the determined usage fee in User DB 141 and finally bills the user for the usage fee.

[0169] In this embodiment, Server 1 changes the usage fee depending on whether or not measurement data is provided (used). Specifically, if measurement data that formed the basis for thickness information estimation (measurement data based on the magnetic flux resistance method) is provided, Server 1 reduces the usage fee. Furthermore, if second measurement data corresponding to the said measurement data (measurement data based on IRIS) is provided, Server 1 further reduces the usage fee. This allows for efficient collection of training data for retraining.

[0170] Figure 28 is a flowchart showing the procedure for estimating wall thickness information according to Embodiment 2. Based on Figure 28, the processing contents of Server 1 and Terminal 2 according to this embodiment will be explained. The control unit 21 of terminal 2 accepts input for selecting an estimation model 50 to be used to estimate wall thickness information (step S501). Specifically, the control unit 21 accepts input for specifying magnetic tube information to be estimated, such as the thinning form, dimensions, and material of the magnetic tube. The control unit 21 also accepts input for selecting whether or not to use (provide) measurement data related to updating the estimation model 50 (step S502). The control unit 11 transmits the measurement data of the magnetic tube along with the selections made in steps S501 and S502 to the server 1 (step S403). If measurement data is obtained from terminal 2, the control unit 11 of server 1 executes the processing in steps S33 to S35 and transmits the wall thickness information estimated based on the estimation model 50 to terminal 2 (step S36).

[0171] When thickness information is obtained from server 1, the control unit 21 of terminal 2 displays the estimated thickness information on the display unit 24 (step S504). Specifically, the control unit 21 displays the thickness of each position (part) along the longitudinal direction of the magnetic tube using a thickness graph 182, etc., and also displays the thickness of each position (channel) on the cross-section at any position along the longitudinal direction using a cross-sectional image 184, etc.

[0172] The control unit 21 transmits second measurement data, obtained by measuring the wall thickness of the magnetic tube using a method other than the magnetic flux resistance method, to the server 1 in response to user input (step S505). The second measurement data is, for example, measurement data related to IRIS, and is data obtained by inserting a probe equipped with an ultrasonic transducer into a magnetic tube filled with water, and detecting the reflected wave of the ultrasonic beam generated from the probe by the ultrasonic transducer.

[0173] After executing the process in step S36, the control unit 11 of server 1 determines the usage fee (charge) to be imposed on the user (step S506). Specifically, the control unit 11 determines the usage fee according to the computational amount of the thickness information estimation process based on the estimation model 50 (number of calculations, calculation time, amount of measurement data, etc.). Furthermore, the control unit 11 deducts the usage fee depending on whether or not measurement data and second measurement data are provided (used or acquired).

[0174] If the control unit 11 receives a selection input from the user indicating that measurement data can be provided, it stores the measurement data transmitted in step S403 in the measurement DB 143 (step S507). Furthermore, if second measurement data is transmitted in step S405, the control unit 11 stores the second measurement data in the measurement DB 143 in association with the measurement data.

[0175] The control unit 11 updates the estimation model 50 based on the measurement data stored in the measurement DB 143 and the second measurement data (step S508). Specifically, the control unit 11 uses the measurement data measured by the user as training input data, uses the measured thickness in the second measurement data as the correct value for thickness information, and updates parameters such as the weights between neurons. For example, the control unit 11 individually learns the measurement data acquired from each user's terminal 2 and updates the estimation model 50 for each user. The control unit 11 then completes the series of processes.

[0176] Based on the above, according to this second embodiment, the system can be suitably implemented, and the estimation model 50 can be optimized through the operation of the system.

[0177] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the invention is indicated by the claims and not in the sense described above, and all modifications within the sense and scope equivalent to the claims are intended to be included. [Explanation of Symbols]

[0178] 1. Server (Information Processing Device) 11 Control Unit 12 Main memory 13 Communications Department 14 Auxiliary storage P1 Program 142 Model DB 143 Measurement DB 2. Terminal (User terminal) 21 Control Unit 22 Main memory 23 Communications Department 24 Display 25 Input section 26 Auxiliary storage P2 Program 3. Measuring device 4. Test probes 41 York 42 Magnets 43. Hall element (magnetic sensor) 50 Estimated Models 50a Thin-walled specialized model 50b Thick-walled specialized model 50c Final Model

Claims

1. We obtained measurement data by measuring the magnetic properties of the magnetic tube. The thickness information is estimated by inputting the acquired measurement data into a model that has been trained to estimate the thickness information of the magnetic tube when the aforementioned measurement data is input. The computer performs the process, The model includes a first model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thinner than a predetermined value, a second model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thicker than the predetermined value, and a third model learned to output thickness information when the thickness information output from the first model and the thickness information output from the second model are input. The acquired measurement data is input to the first and second models, and the output from the first and second models is input to the third model to estimate the wall thickness information. Information processing methods.

2. The aforementioned magnetic characteristic value is a measurement value obtained using an inspection probe comprising a magnet that generates a magnetic field, a yoke positioned on the opposite side of the magnetic tube from the magnet, and a magnetic sensor positioned between the yoke and the magnetic tube for measuring the magnetic flux density passing through the yoke, the magnet, and the magnetic tube, wherein the output voltage of the magnetic sensor is proportional to the magnetic flux density. The information processing method according to claim 1.

3. The measurement data is data obtained by measuring the magnetic property values ​​at each position along the longitudinal direction of the cylindrical magnetic tube, where the cross-section of the magnetic tube is equally divided along the circumferential direction, using the inspection probe on which the magnet and magnetic sensor are periodically attached to the outer circumference of the yoke. The magnetic characteristic values ​​are normalized such that the magnetic characteristic values ​​outside the magnetic tube coincide at all positions along the circumferential direction, and the magnetic characteristic values ​​in the sound parts of the magnetic tube where no thinning has occurred coincide at all positions along the circumferential direction. The thickness information is estimated by inputting the measurement data, which is obtained by normalizing the magnetic property values, into the model. The information processing method according to claim 2.

4. The measurement data is data obtained by measuring the magnetic characteristic values ​​at each position that divides the cross-section of the cylindrical magnetic tube equally along the circumferential direction, using the inspection probe on which the magnet and magnetic sensor are periodically attached to the outer circumference of the yoke. The magnetic characteristic values ​​at each of the aforementioned positions are corrected to the values ​​obtained when the inspection probe passes through the central axis of the magnetic tube. By inputting the measurement data, which has been corrected for the magnetic properties, into the model, the wall thickness information at each of the positions is estimated. The information processing method according to claim 2 or 3.

5. The measurement data is data obtained by measuring the magnetic property values ​​at each position along the longitudinal direction of the cylindrical magnetic tube, where the cross-section of the magnetic tube is equally divided along the circumferential direction, using the inspection probe on which the magnet and magnetic sensor are periodically attached to the outer circumference of the yoke. From the magnetic property values ​​at each position in the longitudinal or circumferential direction, the mean value, standard deviation, skewness, or kurtosis of the magnetic property values ​​along the longitudinal or circumferential direction is calculated. The thickness information is estimated by inputting the measurement data, which includes the mean value, standard deviation, skewness, or kurtosis, into the model. The information processing method according to any one of claims 2 to 4.

6. The measurement data is data obtained by measuring the magnetic characteristic values ​​at each position along the longitudinal direction of the magnetic tube. By inputting the aforementioned measurement data into the model, the thickness information of each portion of the magnetic tube, divided into fixed-length segments along its longitudinal direction, is estimated. The information processing method according to any one of claims 1 to 5.

7. By taking a moving average of the magnetic characteristic values ​​along the longitudinal direction, the baseline of the measurement data is identified. From the measurement data, data of the peak portion where the difference between the baseline and the magnetic characteristic value is greater than or equal to a predetermined threshold is extracted. By inputting the extracted peak portion data into the model, the thickness information of the peak portion is estimated. The information processing method according to claim 6.

8. From the aforementioned measurement data, data is extracted from each data section obtained by shifting data sections of a predetermined length along the longitudinal direction. By inputting the extracted data from each of the aforementioned data intervals into the model, the thickness information of the peak portion is estimated. The information processing method according to claim 7.

9. The measurement data is data obtained by measuring the magnetic property values ​​at each position on a cross-section perpendicular to the longitudinal direction of the magnetic tube, at each position along the longitudinal direction of the magnetic tube. The measurement data is converted into an image in which the first axis of the image is a position along the longitudinal direction and the second axis is a position on the cross-section, and the pixel values ​​of each pixel are assigned according to the magnetic characteristic values ​​of each position in the magnetic tube. The thickness information is estimated by inputting the aforementioned image into the aforementioned model. The information processing method according to any one of claims 1 to 8.

10. From the measurement data, multiple hue images are generated by assigning power values ​​of the magnetic characteristic values, each with different power exponents, to different hues. A composite image is generated by combining the aforementioned multiple hue images. The thickness information is estimated by inputting the generated composite image into the model. The information processing method according to claim 9.

11. The magnetic tube information relating to the magnetic tube to be measured in the measurement data is obtained, From among a plurality of models that have been trained with different training data according to the magnetic tube information, select the model that corresponds to the acquired magnetic tube information. The thickness information is estimated by inputting the measurement data into the selected model. The information processing method according to any one of claims 1 to 10.

12. The measurement data is corrected so that the sum of the Euclidean distances between the measurement data and the thickness information corresponding to the measurement data is minimized. The aforementioned model is learned based on corrected measurement data and wall thickness information. The information processing method according to any one of claims 1 to 11.

13. The measurement data includes sensor data obtained from a plurality of sensors provided in the circumferential direction of the magnetic tube, first group data obtained by performing preprocessing including calculation processing on the sensor data, and second group data obtained by performing principal component analysis on the sensor data and the first group data. The thickness information is estimated by inputting the sensor data, the first group data, and the second group data into the model. The information processing method according to any one of claims 1 to 12.

14. We obtained measurement data by measuring the magnetic properties of the magnetic tube. The thickness information is estimated by inputting the acquired measurement data into a model that has been trained to estimate the thickness information of the magnetic tube when the aforementioned measurement data is input. Let the computer perform the process, The model includes a first model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thinner than a predetermined value, a second model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thicker than the predetermined value, and a third model learned to output thickness information when the thickness information output from the first model and the thickness information output from the second model are input. The acquired measurement data is input to the first and second models, and the outputs from the first and second models are input to the third model, causing the computer to perform a process to estimate the wall thickness information. program.

15. An acquisition unit that acquires measurement data by measuring the magnetic properties of a magnetic tube, An estimation unit that estimates the wall thickness information by inputting the acquired measurement data into a model that has been trained to estimate wall thickness information regarding the wall thickness of the magnetic tube when the aforementioned measurement data is input. Equipped with, The model includes a first model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thinner than a predetermined value, a second model learned primarily based on measurement data and thickness information where the remaining wall thickness of the magnetic tube is thicker than the predetermined value, and a third model learned to output thickness information when the thickness information output from the first model and the thickness information output from the second model are input. The estimation unit estimates the wall thickness information by inputting the acquired measurement data into the first model and the second model, and inputting the outputs from the first model and the second model into the third model. Information processing device.

16. Training data is obtained in which the correct values ​​of wall thickness information relating to the wall thickness of the magnetic tube are associated with measurement data of the magnetic properties of the magnetic tube. Based on the aforementioned training data, a trained model is generated that estimates the wall thickness information when the aforementioned measurement data is input. The computer performs the process, First training data is acquired, which mainly includes measurement data and thickness information indicating that the remaining wall thickness of the magnetic tube is thinner than a predetermined value. Based on the acquired first training data, a first model is generated that outputs first wall thickness information when measurement data is input. Second training data is acquired, which mainly includes measurement data and thickness information indicating that the remaining wall thickness of the magnetic tube is greater than the predetermined value. Based on the acquired second training data, a second model is generated that outputs second wall thickness information when measurement data is input. The first thickness information output from the first model and the second thickness information output from the second model, along with third training data including the thickness information, are acquired. Based on the acquired third training data, a third model is generated that outputs wall thickness information when the first and second wall thickness information is input. Model generation method.

17. The measurement data is data obtained by measuring the magnetic property values ​​at each position obtained by dividing the cross-section of the cylindrical magnetic tube equally along the circumferential direction. By shifting the position on the cross-section of the magnetic tube where the magnetic characteristic value was measured along the circumferential direction, multiple patterns of measurement data are generated from the training data. The trained model is generated using the measurement data of the multiple patterns described above. The model generation method according to claim 16.

18. By adding multiple patterns of predetermined noise, multiple patterns of measurement data are generated from the training data. The trained model is generated using the measurement data of the multiple patterns described above. The model generation method according to claim 16 or 17.

19. The measurement data is data obtained by measuring the magnetic characteristic values ​​at each position along the longitudinal direction of the magnetic tube. By performing linear interpolation of magnetic property values ​​between two consecutive points along the longitudinal direction, multiple patterns of measurement data are generated from the training data. The trained model is generated using the measurement data of the multiple patterns described above. A model generation method according to any one of claims 16 to 18.

20. The measurement data and the corresponding wall thickness information are acquired. The measurement data is corrected so that the sum of the Euclidean distances between the acquired measurement data and the wall thickness information is minimized. The model is trained based on the corrected measurement data and wall thickness information. A model generation method according to any one of claims 16 to 19.

21. The measurement data is acquired from each user terminal connected to the network in a communicative manner. The estimated wall thickness information is transmitted to the user terminal, which is the source of the measurement data. The computer executes the process. The information processing method according to claim 1.

22. The aforementioned magnetic characteristic value is a measurement value obtained using an inspection probe comprising a magnet that generates a magnetic field, a yoke positioned on the opposite side of the magnetic tube from the magnet, and a magnetic sensor positioned between the yoke and the magnetic tube for measuring the magnetic flux density passing through the yoke, magnet, and magnetic tube, wherein the output voltage of the magnetic sensor decreases as the magnetic flux density increases. The information processing method according to claim 21.

23. The system accepts the input of specified magnetic tube information regarding the magnetic tube to be measured. From among a plurality of models that have been trained with different training data according to the magnetic tube information, select the model that corresponds to the acquired magnetic tube information. The thickness information is estimated by inputting the measurement data into the selected model. The information processing method according to claim 21 or 22.

24. The measurement data acquired from each user terminal is stored in the storage unit. The correct value of the wall thickness information corresponding to the measurement data is obtained, The model is updated based on the measurement data stored in the memory unit and the correct answer value. The information processing method according to any one of claims 21 to 23.

25. For each user from whom the measurement data is obtained, the model is updated using the measurement data obtained from that user's terminal. The information processing method according to claim 24.

26. When the aforementioned wall thickness information is estimated, the user will be charged a usage fee for the model based on the computational cost of the wall thickness information estimation process using the model. The information processing method according to claim 24 or 25.

27. When acquiring the measurement data from the user terminal, the user is asked to select whether or not to use the measurement data for updating the model. If an input indicating that the measurement data is available is received, the measurement data is stored in the storage unit. Depending on whether the aforementioned measurement data can be used, the usage fee for the aforementioned model imposed on the aforementioned user will be reduced. The information processing method according to claim 26.

28. From the user terminal, second measurement data representing the correct value is obtained, which is obtained by measuring the wall thickness of the magnetic tube using a different measurement method than the measurement data. The model is updated based on the aforementioned measurement data and the second measurement data. Depending on whether or not the second measurement data is acquired, the usage fee for the model imposed on the user will be reduced. The information processing method according to claim 26 or 27.

29. The second measurement data is data obtained by the water immersion rotation ultrasonic thickness measurement method. The information processing method according to claim 28.

30. The acquisition unit acquires the measurement data from each user terminal that is connected to communicate via a network, The system includes a transmission unit that transmits the estimated wall thickness information to the user terminal, which is the source of the measurement data. The information processing apparatus according to claim 15.

31. The measurement data is transmitted to an information processing device for estimating the wall thickness information, The estimated wall thickness information is obtained from the information processing device, The acquired wall thickness information is displayed on the display unit. Let the computer perform the process. The program according to claim 14.

32. A second measurement data representing the correct value of the wall thickness information, obtained by measuring the wall thickness of the magnetic tube using a different measurement method than the first measurement data, is transmitted to the information processing device. The wall thickness information estimated using the model, which is updated based on the aforementioned measurement data and the second measurement data, is obtained from the information processing device. The program according to claim 31.

33. The measurement data is data obtained by measuring the magnetic property values ​​at each position on a cross-section perpendicular to the longitudinal direction of the magnetic tube, at each position along the longitudinal direction of the magnetic tube. Based on the thickness information indicating the thickness at each position on the cross-section along the longitudinal direction, a graph showing the thickness at each position along the longitudinal direction is displayed. On the aforementioned graph, the system accepts a specified input for a position along the longitudinal direction. Display a cross-sectional image of the magnetic tube that reproduces the wall thickness at each position on the cross-section at the specified location. The program according to claim 31 or 32.

Images