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

Abstract

Provided is a method for processing information on the state of a magnetic pipe, etc. The method for processing information causes a computer to execute the following processing: acquires measurement data obtained by measuring a magnetic property value of a magnetic pipe, and estimates wall thickness information related to a wall thickness of the magnetic pipe by inputting the acquired measurement data to a learned model, such that the wall thickness information is estimated in a state in which the measurement data is input. Preferably, the magnetic property value is a measurement value measured using an inspection probe that includes a magnet that generates a magnetic field, a yoke that is disposed on the opposite side of the magnet from the magnetic pipe, and a magnetic sensor that is disposed between the yoke and the magnetic pipe, measures a magnetic flux density that passes through the yoke, the magnet, and the magnetic pipe, and is an output voltage of the magnetic sensor that is lower the greater the magnetic flux density.

Description

Technical Field

[0001] This invention relates to information processing methods, programs, information processing devices, and model generation methods. Background Technology

[0002] Known methods for non-destructively inspecting defects such as wall thinning in magnetic tubes include Remote Field Eddy Current Testing (RFECT) and Magnetic Flux Leakage (MFL).

[0003] For example, Patent Document 1 discloses a method for determining defects in magnetic components based on the magnetic flux resistance method (MFR).

[0004] Prior art literature

[0005] Patent documents

[0006] Patent Document 1: JP Patent No. 6514592 Summary of the Invention

[0007] -The problem the invention aims to solve-

[0008] However, in these inspection methods, the wall thickness of the magnetic tube is predicted by applying measured values ​​to pre-prepared model formulas. Therefore, it is necessary to manually create model formulas for predicting wall thickness and other parameters.

[0009] In one aspect, the aim is to provide an information processing method that can appropriately estimate the state of a magnetic tube.

[0010] -Methods used to solve problems-

[0011] One aspect of the information processing method involves a computer performing the following processing: acquiring measurement data obtained by measuring the magnetic properties of a magnetic tube, and estimating wall thickness information related to the wall thickness of the magnetic tube by inputting the acquired measurement data into a learned model, so that the wall thickness information is estimated given the input measurement data.

[0012] -Invention Effects-

[0013] In one respect, it is possible to appropriately estimate the state of the magnetic tube. Attached Figure Description

[0014] Figure 1 This is an explanatory diagram illustrating a structural example of a defect estimation system.

[0015] Figure 2A This is an explanatory diagram of the measuring apparatus.

[0016] Figure 2B This is an explanatory diagram of the measuring apparatus.

[0017] Figure 3 This is a block diagram representing a server's structure.

[0018] Figure 4 This is a block diagram representing a structural example of a terminal.

[0019] Figure 5 This is an illustrative diagram showing an example of the record layout of the user database, model database, and measurement database.

[0020] Figure 6 This is an explanatory diagram showing an outline of Implementation Method 1.

[0021] Figure 7 This is an explanatory diagram related to data expansion processing.

[0022] Figure 8 This is an explanatory diagram related to the compensation processing of measurement data.

[0023] Figure 9 This is an explanatory diagram related to image transformation processing.

[0024] Figure 10 This is a flowchart representing the steps involved in generating the estimation model.

[0025] Figure 11 This is a flowchart representing the steps involved in estimating wall thickness information.

[0026] Figure 12 This is an explanatory diagram showing the outline of variation example 1.

[0027] Figure 13 This is a flowchart illustrating the steps involved in generating the estimation model in Variation Example 1.

[0028] Figure 14 This is a flowchart illustrating the steps involved in estimating wall thickness information in Modified Example 1.

[0029] Figure 15 This is an explanatory diagram showing the outline of variation example 2.

[0030] Figure 16 This is a flowchart illustrating the steps involved in generating the estimation model in Variation Example 2.

[0031] Figure 17 This is a flowchart illustrating the steps involved in estimating wall thickness information in Modified Example 2.

[0032] Figure 18 This is an explanatory diagram showing the outline of variation example 3.

[0033] Figure 19 This is a graph representing the weighted learning in variation example 3.

[0034] Figure 20 This is a flowchart illustrating the steps involved in generating the estimation model in Variation Example 3.

[0035] Figure 21 This is a flowchart illustrating the steps involved in estimating wall thickness information in Modified Example 3.

[0036] Figure 22 This is an illustrative diagram used to show the effect of Hall element voltage on magnetic flux disturbances accompanying the MFR mode.

[0037] Figure 23A This is an explanatory diagram showing the outline of the dynamic time-scaling method in Variation Example 4.

[0038] Figure 23B This is an explanatory diagram showing the outline of the dynamic time-scaling method in Variation Example 4.

[0039] Figure 24 This is a graph showing the detection voltage and estimated wall thickness information in variation example 4.

[0040] Figure 25 This is an explanatory diagram related to data expansion processing in Variation Example 5.

[0041] Figure 26 This is an explanatory diagram showing an example of a screen for uploading measurement data.

[0042] Figure 27 This is an explanatory diagram of an example of a display screen showing wall thickness information.

[0043] Figure 28 This is a flowchart illustrating the steps involved in the wall thickness information estimation process according to Implementation Method 2. Detailed Implementation

[0044] The present invention will now be described in detail with reference to the accompanying drawings illustrating this embodiment.

[0045] (Implementation Method 1)

[0046] Figure 1 This is an explanatory diagram illustrating a structural example of a defect estimation system. In this embodiment, a defect estimation system that uses a machine learning model to estimate defects such as wall thickness reduction from measurement data obtained by measuring the magnetic properties of a magnetic tube will be described. The defect estimation system includes an information processing device 1, a user terminal 2, a measuring device 3, and an inspection detector 4. The information processing device 1 and the user terminal 2 are communicatively connected via a network N such as the Internet.

[0047] In this embodiment, the magnetic tube being measured is, for example, a tube made of a magnetic material such as carbon steel, ferritic stainless steel, or a two-phase stainless steel composed of a ferritic phase and an austenitic phase. However, while carbon steel is one example of a magnetic material, the components (magnetic materials) forming the magnetic tube are not limited to these.

[0048] The measuring device 3 is a device for measuring the magnetic characteristic value of a magnetic tube, and is used for inspecting the wall thickness reduction of a magnetic tube based on the flux resistance method proposed by the applicant. The measuring device 3 measures the magnetic characteristic value (magnetic flux density) at various locations within the magnetic tube by inserting the inspection detector 4 into the tube and moving it within the tube. For example, the defect measuring device disclosed in Japanese Patent No. 6579840, Japanese Patent No. 6514592, and / or Japanese Unexamined Patent Application Publication No. 2019-100850 can be used as the measuring device 3.

[0049] Figure 2A , Figure 2B This is an explanatory diagram related to measuring device 3. Figure 2A It is a cross-sectional view orthogonal to the long side direction (axial direction) of the magnetic tube, conceptually illustrating the cross-sectional view of the magnetic tube into which the inspection detector 4 is inserted. Figure 2B This is a cross-sectional view orthogonal to the short side (radial) of the magnetic tube, conceptually illustrating the measurement of magnetic flux density at a defective portion (hereinafter referred to as the "thinned portion") of the magnetic tube where wall thickness reduction has occurred. Based on Figure 2A , Figure 2B The flux resistance method will be explained simply.

[0050] The detector 4 includes a magnetic yoke 41, a magnet 42, and a Hall element 43 (magnetic sensor). The magnetic yoke 41 is a hollow cylindrical magnetic component, such as a high-permeability metal like carbon steel. Furthermore, the shape of the magnetic yoke 41 is not limited to a hollow cylindrical shape; it can also be rod-shaped, plate-shaped, cylindrical, etc.

[0051] Magnets 42 and Hall elements 43 are mounted at equal intervals along the outer peripheral surface of the yoke 41. For example, there are 8 locations of magnets 42 and Hall elements 43 on the outer peripheral surface of the yoke 41. Magnets 42 are configured such that one magnetic pole is opposite to the yoke 41 and the other magnetic pole is opposite to the magnetic tube, and are polarized in the direction opposite to the magnetic tube.

[0052] like Figure 2A , Figure 2B As indicated by the middle arrow, the yoke 41 and the magnet 42 form a magnetic circuit. A Hall element 43 is disposed on this magnetic circuit, and the output voltage is proportional to the magnetic flux density passing through the Hall element 43. Specifically, the lower the magnetic reluctance of the Hall element 43, the higher the magnetic flux density passing through it, resulting in a higher output voltage.

[0053] Specifically, such as Figure 2B As shown in (a), when the magnet 42 and the Hall element 43 are positioned outside the magnetic tube, the magnetic reluctance is high, therefore the magnetic flux density decreases, and the output voltage of the Hall element 43 decreases. On the other hand, as Figure 2B As shown in (c), when the magnet 42 and the Hall element 43 are positioned in the normal portion of the magnetic tube (the portion without wall thinning), the magnetic reluctance is low, therefore the magnetic flux density increases, and the output voltage of the Hall element 43 increases. Figure 2B As shown in (b), with the magnet 42 and Hall element 43 arranged in the thinned section, the wall thickness of the magnetic tube is thinner than in the normal section, resulting in increased magnetic reluctance and decreased magnetic flux density. Consequently, the output voltage of the Hall element 43 is lower.

[0054] Thus, the measuring device 3 uses the output voltage of the Hall element 43 to measure the magnitude of the magnetic flux density. The output voltage of the Hall element 43 varies depending on the wall thickness of the magnetic tube; the thinner the wall, the lower the output voltage.

[0055] Compared to the internal rotary ultrasonic thickness measurement system (IRIS) which uses ultrasound, the magnetic flux resistance method has the advantage of shorter inspection time per magnetic tube and the ability to inspect multiple magnetic tubes. Other inspection methods that utilize magnetism, such as the far-field method (RFECT) and the magnetic flux leakage method (MFL), are also based on magnetic flux resistance, but these methods have lower accuracy compared to the magnetic flux resistance method.

[0056] In this embodiment, when estimating the wall thickness of the magnetic tube based on the measurement data based on the above-described flux resistance method, a machine learning model is used to estimate the wall thickness.

[0057] Furthermore, in the inspection detector 4 involved in the flux resistance method, there is such... Figure 2B The diagram shows a P-type magnetoresistive detector (Japanese Patent No. 6514592) with a gap between the yoke 41 and the Hall element 43, and a PR-type magnetoresistive detector (Japanese Patent Application Publication No. 2019-100850) with no gap between the yoke 41 and the Hall element 43 and the yoke 41 and the Hall element 43 in close contact. The output voltage of both is proportional to the magnetic flux density; however, in the P-type, the higher the magnetic reluctance, the lower the output voltage, while in the PR-type, the higher the magnetic reluctance, the higher the output voltage. For simplicity, the P-type magnetoresistive method was used as an example to illustrate the measurement principle; however, the following description will focus on the case where the PR-type detector 4 is used for measurement. Of course, either the P-type or the PR-type can also be used.

[0058] Furthermore, in this embodiment, the measurement data is described as being based on the flux resistance method, but the measurement data is not limited to this. For example, the eddy current value measured by RFECT can also be used as the measurement data. In addition, the leakage flux (magnetic flux through the wall-thinning section) measured by MFL can also be used as the measurement data. That is, the measurement data can be any value representing the magnetic characteristics of the magnetic tube, and is not limited to measurement data based on the flux resistance method.

[0059] return Figure 1 Continuing the explanation. Information processing device 1 is an information processing device capable of various information processing operations, including sending and receiving information, such as a server computer or a personal computer. In this embodiment, information processing device 1 is a server computer, which will be referred to as server 1 for simplicity. Server 1, by learning from given training data and receiving measurement data of the magnetic properties of the magnetic tube, generates an estimation model 50 (see reference) that estimates wall thickness information related to the wall thickness of the magnetic tube. Figure 6 Then, server 1 obtains the measurement data of the magnetic tube of the object being measured from user terminal 2 and inputs it into estimation model 50, thereby estimating the wall thickness information of the magnetic tube. As will be described later, the wall thickness information is the wall thickness of the magnetic tube, but in addition to the wall thickness, it can also be the wall thickness reduction width, wall thickness reduction rate, wall thickness reduction range, etc.

[0060] User terminal 2 is the terminal device used by the user of this system, such as a personal computer or tablet terminal. For simplicity, user terminal 2 will be referred to as terminal 2 below. The user of this system may be, for example, an inspection company performing non-destructive testing of magnetic tubes, but this is not particularly limited. Furthermore, in Figure 1 The diagram shows only one terminal 2, but multiple users have their own terminals 2, 2, 2... connected to server 1. Terminal 2 is connected to measuring device 3 and sends the measurement data of the magnetic tube measured by measuring device 3 to server 1, obtaining the wall thickness information estimated based on estimation model 50 from server 1.

[0061] In this embodiment, the cloud server 1 estimates the wall thickness information based on the estimation model 50, but this embodiment is not limited to this. For example, the local terminal 2 may pre-install the data of the estimation model 50 from the server 1, input the measurement data obtained from the measuring device 3 into the estimation model 50, and thereby estimate the wall thickness information. Thus, the device for generating the estimation model 50 and the device for estimating the wall thickness information based on the estimation model 50 may be different.

[0062] Figure 3This is a block diagram illustrating the structure of server 1. Server 1 includes a control unit 11, a main storage unit 12, a communication unit 13, and an auxiliary storage unit 14. The control unit 11 has one or more processing units such as CPUs (Central Processing Units), MPUs (Micro-Processing Units), and GPUs (Graphics Processing Units), and performs various information processing by reading and executing the program P1 stored in the auxiliary storage unit 14. The main storage unit 12 is a temporary storage area such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), and flash memory, which temporarily stores data required for the control unit 11 to perform information processing. The communication unit 13 is a communication module used for communication-related processing, transmitting and receiving information with external devices.

[0063] Auxiliary storage unit 14 is a non-volatile storage area such as a mass storage device or hard disk, storing program P1 and other data required for processing by control unit 11. Furthermore, auxiliary storage unit 14 stores estimation model 50, user DB141, model DB142, and measurement DB143. Estimation model 50 is a machine learning model that has learned from given training data; it is a model that estimates wall thickness information related to the wall thickness of the magnetic tube when given measurement data obtained by measuring the magnetic properties of the magnetic tube. It is assumed that estimation model 50 is used as a program module constituting part of artificial intelligence software.

[0064] User DB141 is a database storing information about users of this system. Model DB142 is a database storing information about multiple estimation models 50 prepared according to the wall thickness reduction method (shape pattern of the wall thickness reduction section), the size of the magnetic tube, etc. As will be described later, in this embodiment, multiple estimation models 50 are prepared according to the wall thickness reduction method, size, etc. of the magnetic tube being estimated. Measurement DB143 is a database storing measurement data of the magnetic tube obtained from terminal 2.

[0065] Alternatively, the auxiliary storage unit 14 can be an external storage device connected to the server 1. Furthermore, the server 1 can be a multi-computer system composed of multiple computers, or it can be a virtual machine virtualized by software.

[0066] Furthermore, in this embodiment, the server 1 is not limited to the structure described above, and may include, for example, an input unit for accepting operation input, a display unit for displaying images, etc. Additionally, the server 1 may also include a reading unit for reading a non-transitory computer-readable recording medium 1a, and reading program P1 from the recording medium 1a. Furthermore, program P1 can be executed on a single computer or on multiple computers interconnected via a network N.

[0067] Figure 4 This is a block diagram illustrating the structure of terminal 2. Terminal 2 includes a control unit 21, a main storage 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 processing devices such as CPUs and MPUs, which perform various information processing by reading and executing the program P2 stored in the auxiliary storage unit 26. The main storage unit 22 is a temporary storage area such as RAM, which temporarily stores data required for the control unit 21 to perform information processing. The communication unit 23 is a communication module used for communication-related processing, transmitting and receiving information with external devices.

[0068] Display unit 24 is a display screen such as an LCD monitor, which displays images. Input unit 25 is an operation interface such as a keyboard and mouse, which receives operation input from the user. Auxiliary storage unit 26 is a non-volatile storage area such as a hard disk or mass storage device, which stores the program P2 and other data required for the control unit 21 to perform processing.

[0069] Alternatively, terminal 2 may also have a reading unit that can read non-transitory computer-readable recording medium 2a and read program P2 from recording medium 2a. Furthermore, program P2 can be executed on a single computer or on multiple computers interconnected via network N.

[0070] Figure 5 This is an explanatory diagram showing an example of the record layout of user DB141, model DB142, and measurement DB143. User DB141 includes a user ID column, a username column, a usage history column, and a usage fee column. The user ID column stores the user ID used to identify each user. The username column, usage history column, and usage fee column respectively store the username, the user's usage history of the estimation model 50 (estimation history of wall thickness information), and the usage fee of the estimation model 50 determined based on the usage history (system usage fee), corresponding to the user ID. The usage fee is explained in detail in Implementation 2.

[0071] Model DB142 includes a Model ID column, a Model Name column, and an Object column. The Model ID column stores the model ID used to identify each estimation model 50 prepared based on factors such as the wall thickness reduction method of the magnetic tube. The Model Name column and the Object column store the model name of the estimation model 50 and magnetic tube information related to the magnetic tube that is the estimation object, respectively. For example, the Object column stores the wall thickness reduction method, size, and material of the magnetic tube.

[0072] The DB143 measurement includes a data ID column, a date and time column, a user providing the measurement, an object, a measurement data column, and a second measurement data column. The data ID column stores a data ID used to identify the measurement data provided by each user. The date and time column, user providing the measurement, object, measurement data, and second measurement data column each store, in correspondence with the data ID, the date and time the measurement data was provided (date and time of acquisition), the username of the user providing the measurement, information about the magnetic tube being measured (wall thickness reduction method, etc.), the measurement data of the magnetic tube, and the second measurement data. The second measurement data is measurement data of the wall thickness of the magnetic tube obtained using a method different from that used in the flux resistance method, for example, measurement data based on the internal rotary ultrasonic thickness measurement system (IRIS). The second measurement data will be described in detail in Embodiment 2.

[0073] Figure 6 This is an explanatory diagram showing an outline of Embodiment 1. Figure 6 The diagram conceptually illustrates the estimation of the wall thickness of various parts of a magnetic tube when measurement data of the magnetic properties of the magnetic tube, measured by the measuring device 3 (inspection detector 4), are input into the estimation model 50. Based on Figure 6 This document provides an overview of the implementation method.

[0074] As described above, server 1 generates an estimation model 50 for estimating wall thickness information in the presence of measured data of input magnetic property values ​​by learning from given training data. The estimation model 50 involved in this embodiment is a neural network generated by deep learning, such as CNN (Convolutional Neural Network, ResNet, etc.).

[0075] Furthermore, the estimation model 50 can also be a neural network other than CNN. In addition, as in the variations 1 and 2 described later, the estimation model 50 can also be a model based on learning algorithms other than neural networks such as kNN (k-Nearest Neighbor Algorithm), SVM (Support Vector Machine), random forest, decision tree, etc.

[0076] The estimation model 50 involved in this embodiment takes as input measurement data based on the magnetoresistance method measured using the measuring device 3. As described above, the measurement data is the value of the magnetic flux density through the magnetic circuit formed by the magnetic yoke 41, magnet 42, and magnetotube provided by the inspection detector 4, measured by the Hall element 43, and is the output voltage of the Hall element 43 that is proportional to the magnetic flux density.

[0077] When measuring the magnetic tube, the user inserts the inspection detector 4 into the opening of the magnetic tube and moves (scans) it to the opening on the opposite side. This allows for the measurement of the magnetic characteristic values ​​at various locations along the long side of the magnetic tube. As described above, multiple Hall elements 43 are mounted on the yoke 41 of the inspection detector 4, and the measuring device 3 measures the magnetic characteristic values ​​at various locations along the cross-section orthogonal to the long side. Specifically, multiple Hall elements 43 are periodically mounted on the outer periphery of the cylindrical yoke 41, and the measuring device 3 measures the magnetic characteristic values ​​at various locations that divide the cross-section of the cylindrical magnetic tube into equal parts along the circumferential direction.

[0078] Furthermore, in this embodiment, the case where the magnetic tube is cylindrical has been described, but the cross-sectional shape of the magnetic tube is not limited to a circle; for example, it can also be quadrilateral. That is, the measuring device 3 only needs to be able to measure the magnetic characteristic value at each position on the cross-section orthogonal to the long side direction of the magnetic tube by checking the multiple Hall elements 43 provided by the detector 4. The term "each position on the cross-section" is not limited to each position where the cross-section of the cylindrical magnetic tube is equally divided along the circumferential direction.

[0079] In the following description, the Hall element 43 that functions as a magnetic sensor will be referred to as a "channel", and each Hall element 43 will be assigned a channel number, which will be recorded as "CH1", "CH2", "CH3", ... "CH8".

[0080] The magnetic tube was measured as described above, such as... Figure 6 As shown on the left, waveform data of the magnetic characteristic values ​​(output voltage) of each channel along the long side of the magnetotube are observed. Furthermore, the horizontal axis of the waveform data represents the position of the magnetotube along its long side (axial direction), 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.

[0081] Given the aforementioned measurement data, estimation model 50 outputs (estimates) the wall thickness information of the magnetic tube. Specifically, estimation model 50 estimates the wall thickness of the magnetic tube. In this embodiment, the problem handled by estimation model 50 is defined as a classification problem, and estimation model 50 classifies the wall thickness of the magnetic tube within a given numerical range (e.g., 1.0–2.3 mm) using a given length scale (e.g., 0.1 mm scale). Alternatively, estimation model 50 can also be used as a regression model.

[0082] 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 also estimate the wall thickness reduction width obtained by subtracting the wall thickness of the thinned portion from the wall thickness of the normal portion, instead of estimating the wall thickness of the magnetic tube. Furthermore, for example, the estimation model 50 may also estimate the wall thickness reduction rate obtained by dividing the wall thickness of the thinned portion by the wall thickness of the normal portion. Furthermore, for example, the estimation model 50 may also estimate the location of the thinned portion within the magnetic tube, the wall thickness reduction range (the width of the thinned portion along the long side direction and / or the circumferential direction), the wall thickness reduction method (the shape pattern of the wall thickness reduction), etc.

[0083] Furthermore, in this embodiment, the reduction in the wall thickness of the magnetic tube is used as the estimation object, but this embodiment is not limited to this, and the increase in wall thickness caused by rust or the like can also be used as the estimation object.

[0084] 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 defect that is the object of estimation is not limited to wall thinning.

[0085] For example, estimation model 50 estimates the wall thickness of the magnetic tube for each section that is divided by a fixed length along its long side. Figure 6 The right side conceptually illustrates the wall thickness estimation results. Estimation model 50 divides the magnetic tube into fixed lengths (e.g., 3 cm) and estimates the minimum wall thickness for each section. Alternatively, estimation model 50 can estimate the wall thickness continuously without dividing into fixed lengths, or it can estimate values ​​other than the minimum (e.g., the maximum or average wall thickness for each section). Furthermore, estimation model 50 can estimate the maximum wall thickness for each channel, or it can estimate the minimum wall thickness of all channels as a representative value.

[0086] Furthermore, in the above description, the magnetic property values ​​of each position (channel) on the cross-section at each location along the long side of the magnetic tube are input into 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) along the long side of the magnetic tube into the estimation model 50 to estimate the wall thickness information at that position. Additionally, 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 into the estimation model 50 to estimate the wall thickness information at that position.

[0087] Server 1 uses training data, which corresponds to the correct values ​​(labels) of the wall thickness information, and the training measurement data set, to generate estimation model 50. The training measurement data consists of measured data from one or more magnetic tubes of the measuring device 3, specifically waveform data from each channel at various positions along the long side where magnetic property values ​​are measured. Specifically, the training measurement data is obtained by inserting the inspection detector 4 into both ends of the magnetic tube while setting a radial direction (position on the cross-section) to 0° and varying the relative position of the inspection detector 4 in the circumferential direction so that CH1 is approximately 0°, 90°, 180°, and 270°. In other words, it is data from 4 × 2 = 8 measurements performed on each magnetic tube. The correct values ​​are obtained by measuring the wall thickness of the magnetic tube using a 3D shape measuring instrument, measured from the tube end at 0.2 mm intervals and with a scale of approximately 23 μm. Server 1 uses the training data, which assigns correct wall thickness values ​​to each part of the training measurement data, to generate estimation model 50.

[0088] Figure 7 This is an explanatory diagram related to data expansion processing. Server 1 could also use only the raw measurement data obtained by the inspection detector 4 as input to the estimation model 50, but in this embodiment, to improve the estimation accuracy of the wall thickness information, certain preprocessing is performed on the raw measurement data, and the preprocessed measurement data is used as input to the estimation model 50. Furthermore, server 1 could also learn only the aforementioned training data (measured data), but in this embodiment, to increase the amount of training data, the measurement data (measured data) used for training is expanded. Server 1 learns the expanded training data and generates the estimation model 50. Figure 7 In (a) to (d), the expansion of the measurement data after pretreatment is conceptually illustrated.

[0089] exist Figure 7 In (a), a conceptual illustration is provided of the preprocessed data, which includes summary statistics such as mean and standard deviation, as well as standardized and compensated magnetic property values, on the training measurement data. Figure 7 The preprocessing of the measurement data is explained.

[0090] Server 1 uses the position along the long side of the measured magnetic property value and the original measured values ​​of each channel. Figure 7 (a) "CH1", "CH2", ... "CH8"), are used as the original measurement data. Furthermore, the position along the long side is used in addition to the measured values ​​because the output voltage of the Hall element 43 at the tube end (inlet / outlet) of the magnetic tube changes drastically; therefore, the position at the tube end is also used as an explanatory variable. For the original measurement data, server 1 first standardizes the magnetic characteristic values ​​of each channel ( Figure 7 The magnetic characteristic value outside the magnetic tube (in the air), which becomes the maximum value, is consistent with the magnetic characteristic value of the normal part, which becomes the minimum value, throughout all channels (all positions along the circumferential direction). Specifically, server 1 compensates so that the output voltage of the Hall element 43 outside the magnetic tube is 1V, and the output voltage of the normal part is 0V. In addition, in the P-type, the value is minimum outside the magnetic tube and maximum in the normal part. By standardizing the magnetic characteristic value, the inherent sensitivity deviation of each Hall element 43 can be compensated, and the effect of temperature on the Hall element 43 can be compensated.

[0091] Next, in order to minimize the impact of the removal of detector 4 (and the entire thinned section of the wall surface), server 1 compensates for the measurement data of each channel. Figure 7 (the next paragraph of (a)).

[0092] Figure 8 This is an explanatory diagram related to the compensation processing of the measured data. Figure 8 The left-hand diagram shows the measurement data before compensation. Figure 8 The diagram on the right shows the compensated measurement data. Ideally, during measurement, the detector 4 should be checked to ensure it passes through the central axis of the magnetotube, but in practice... Figure 8 As shown on the left, the position of the inspection detector 4 inside the magnetic tube depends on the user's operation, and the spacing between each channel CH1, CH2, CH3... and the tube wall varies considerably. Therefore, server 1 minimizes the liftoff effect by compensating the magnetic property values ​​of each channel to data assuming the inspection detector 4 passes through the central axis of the magnetic tube.

[0093] Specifically, server 1 smooths the measurement data by taking a moving average along the long side of each channel to determine the baseline for each channel. Then, server 1 compensates for the values ​​of each channel to make the average baseline of each channel consistent. That is, server 1 adjusts the height of the waveform to make the waveform shape of each channel almost identical after smoothing. In addition, the length of the long side for taking the moving average and the number of averages are adjusted according to the length (total length) of the magnetic tube.

[0094] Furthermore, preferably, server 1 compensates only for the localized wall thickness reduction by removing the portion of the output voltage that changes gradually from the measured data, i.e., the data for the entire wall thickness reduction portion. The entire wall thickness reduction portion refers to the portion where the change in output voltage (magnetic characteristic value) of the Hall element 43 is fixed or less. For example, server 1 determines the range along the long side where the entire wall thickness reduction portion exists by calculating the average of the baselines of all channels, replaces the output voltage of this range with the value of the normal portion (i.e., 0V), and thereby removes the data for the entire wall thickness reduction portion. Thus, the voltage change of the entire wall thickness reduction portion is cut off, and only the data for the localized wall thickness reduction portion (the portion where the output voltage change is greater than a certain value) can be adjusted.

[0095] return Figure 7 Continuing the explanation, as described above, server 1 generates standardized data from the original measurement data and compensated data that removes the effects of lift-off (and the entire thinned portion of the wall thickness). Server 1 may use only the standardized data and / or the compensated data as input to the estimation model 50, but in this embodiment, all three types of data—the original measurement data, the standardized data, and the compensated data—are used as input.

[0096] Server 1 uses the data obtained by adding a given summary statistic to each of the three types of data to estimate model 50. The summary statistic is the mean, standard deviation, skewness, and / or kurtosis, but is not limited to these. Server 1 calculates the mean, standard deviation, skewness, and / or kurtosis along the long side and / or circumferential direction based on the magnetic property values ​​at each position along the long side and / or circumferential direction, and adds them to the data.

[0097] Specifically, server 1 calculates the average value, standard deviation, skewness, and kurtosis at each position (channel) along the circumferential direction of the magnetic tube and adds them to the data. Furthermore, server 1 calculates the moving average value, moving standard deviation, moving skewness, and moving kurtosis at multiple positions (e.g., 10, 20, 30, 40, and 50 points) along the long side of the magnetic tube and adds them to the data. Figure 7 In the table, "mean," "standard deviation," "skewness," and "kurtosis" correspond to the summary statistics along the circumferential direction, while "10-point moving average" and "20-point moving average" correspond to the summary statistics (moving average) along the longer side. For ease of illustration, in... Figure 7 The moving standard deviation, moving skewness, and moving kurtosis are not illustrated. Increasing these summary statistics can improve the accuracy of wall thickness estimation. In particular, according to the inventors' research, increasing skewness and kurtosis can reduce estimation errors.

[0098] In this way, server 1 performs certain preprocessing on the raw measurement data and uses the preprocessed measurement data as input to estimation model 50. The final preprocessed measurement data becomes a matrix of N rows × 275 columns (N is the number of measurement points along the long side). In fact, the same applies when estimating the wall thickness information based on the measurement data of the magnetic tube using estimation model 50; server 1 performs preprocessing on the measurement data obtained from terminal 2 and inputs the preprocessed data into estimation model 50.

[0099] During the learning process, server 1 adjusts the settings to increase the amount of learning data. Figure 7 The measurement data shown in (a) is expanded. As a first data expansion process, server 1 generates measurement data in multiple modes by shifting the positions on the cross-section of the magnetic tube from which the magnetic property values ​​are measured along the circumferential direction. Specifically, server 1 generates measurement data by shifting the channel numbers of the measurement data for each channel. Figure 7 (b) conceptually illustrates the first data expansion process. Server 1 first shifts the channel numbers by one, generating measurement data where CH1 data is set as CH2, CH2 data as CH3, ..., and CH8 data as CH1. Next, Server 1 shifts the channel numbers by two, generating measurement data where CH1 data is set as CH3, CH2 data as CH4, ..., and CH8 data as CH2. Similarly, Server 1 shifts the channel numbers, making the number of data eight times. Thus, based on the measurement data where a wall thickness reduction is detected at CH1, measurement data where a wall thickness reduction is detected at CH2, CH3, ..., can be generated, compensating for deviations in the channel (position on the cross-section of the magnetic tube) where the wall thickness reduction is detected.

[0100] Next, as a second data expansion process, server 1 generates multiple-mode measurement data by appending multiple modes of given noise to the measurement data. Figure 7 Figure (c) illustrates the second data expansion process. For example, server 1 increases the data size by 10 times by adding 10 patterns of Gaussian noise (mean 0, variance 0.02) to the measurement data. Thus, as... Figure 7 As shown on the right side of (c), it is possible to reproduce the small positional shift of the wall thickness reduction section, as if the wall thickness reduction section were located between one channel and other channels, so that the estimation model 50 can learn.

[0101] As a third data extension process, server 1 generates measurement data with a changed sampling rate when measuring magnetic property values. Figure 7Figure (d) illustrates the third data expansion process. Server 1 performs linear interpolation between two consecutive points along the long side of the training measurement data to approximate the magnetic property values ​​at the positions between the two points. For example, to generate measurement data with sampling rates of 1 to 6 times, server 1 generates data after 1-point interpolation, 2-point interpolation, ... 20-point interpolation. This allows the variation in the moving speed (scanning speed) of the inspection detector 4 to be reproduced, enabling the estimation model 50 to learn.

[0102] In addition, Figure 7 In the tables (b) to (d), for convenience, only the measured values ​​of each channel ("CH1", "CH2", ...) are shown, but of course, the above-mentioned summary statistics and other extended data are included.

[0103] Server 1 expands the number of measured data points by 8 × 10 × 6 = 480 times through the first to third data expansion processes. This increases the amount of learning data and allows the estimation model 50 to learn deviations in the learning channels, positional shifts of minor wall thinning sections, and changes in movement speed.

[0104] return Figure 6 Continuing the explanation, Server 1 uses the expanded measurement data as training data to generate estimation model 50. Server 1 can use the measurement data (magnetic property values) as input to estimation model 50, but... Figure 6 As shown, in this embodiment, the measurement data is transformed into an image, and the image is used as the input to the estimation model 50.

[0105] Figure 9 This is an explanatory diagram related to image transformation processing. Figure 9 The diagram illustrates the transformation of measurement data into an image. In this embodiment, server 1 generates an image representing the magnetic property values ​​at each location within the magnetotube using pixel values ​​of each pixel, which is used as input to estimation model 50.

[0106] Specifically, server 1 generates an image representing the magnetic property values ​​at each location using the color (e.g., RGB) of each pixel. The vertical axis (first axis) of this image corresponds to the position along the long side of the magnetic tube, and the horizontal axis corresponds to the position on a cross-section of the magnetic tube orthogonal to the long side, i.e., the channel. For example, server 1 will represent the positions along the long side of the magnetic tube in rows, and the original measurements, standardized values, compensated values, and summary statistics of each channel in a matrix form (see [reference]). Figure 7The image is divided into rows according to a given number of rows (e.g., 224 rows). Then, server 1 sequentially stores the measurement data for rows 1-224 from the left edge, rows 225-448, and so on, thereby assigning the measurement data at each location to each pixel within the image. Finally, if any pixels remain in the image, server 1 sets the R, G, and B values ​​of the remaining pixels to 0. Additionally, depending on the sampling rate during the third data expansion, there is also a possibility that most of the values ​​will be 0.

[0107] Server 1 generates hue images of each hue (R, G, B) based on the measurement data used for training, and synthesizes these hue images to generate the final input image. Specifically, Server 1 calculates the power values ​​of the magnetic property values ​​of multiple modes with different power exponents based on the measurement data, assigns the power values ​​of each mode to each hue, and thus generates hue images.

[0108] For example, server 1 calculates the 0th, 1st, and 2nd power values ​​for the magnetic properties at various locations within the magnetotube. Then, server 1 assigns these power values ​​to R, G, and B. Figure 9 The diagram illustrates the assignment of powers of 1 to R, powers of 2 to G, and powers of 0 to B. Server 1 generates hue images by storing the powers of the magnetic property values ​​at each location as the pixel values ​​for the corresponding pixels. Server 1 then synthesizes these hue images to generate the final input image.

[0109] In the following description, the image that combines the images of each hue will be referred to as a "composite image".

[0110] Furthermore, the image transformation method described above is just one example, and this embodiment is not limited to it. For example, server 1 may also generate a composite image by assigning powers of 0 to 4 to each hue as a CMYK image instead of an RGB image. Furthermore, the hue to which the powers are assigned is not limited to primary colors (RGB). Furthermore, the exponent of the powers is not limited to natural numbers, but can also be a real number other than a natural number (e.g., 1.5). Additionally, server 1 may generate a monochrome image that only stores the magnetic property values ​​of each position within the magnetic tube in each pixel, without calculating the powers. Furthermore, server 1 may represent the magnetic property values ​​using other pixel values ​​(e.g., grayscale) instead of color. Thus, server 1 is only required to generate an image in which pixel values ​​are assigned to each pixel based on the magnetic property values ​​of each position within the magnetic tube; the generation method is not particularly limited.

[0111] During learning, to prevent hue-based overlearning, server 1 generates synthetic images for all combinations of assigning each power value to each hue. That is, server 1 generates synthetic images for all combinations of assigning power values ​​to each hue, except for synthetic images where power 0 is assigned to R, power 1 to G, power 2 to B, etc., such as synthetic images where power 0 is assigned to G, power 1 to B, power 2 to R, power 0 to B, power 1 to R, power 2 to G, and so on, thus generating six modes of synthetic images. Server 1 prevents hue-based overlearning by having the estimation model 50 learn the synthetic images for each combination.

[0112] return Figure 6 Continuing the explanation, server 1 performs data augmentation on the measurement data provided for training, transforming the augmented measurement data into synthetic images. Then, server 1 uses each synthetic image and the correct values ​​of the wall thickness information corresponding to each synthetic image as training data to generate an estimation model 50.

[0113] That is, server 1 obtains an estimate of the wall thickness information by inputting the synthetic image used for training into the estimation model 50, and compares the estimate with the correct value. Server 1 updates parameters such as the weights between neurons to make the estimate approximate the correct value. Server 1 learns sequentially using pairs of synthetic images and correct values, and finally generates an estimation model 50 that optimizes the weights, etc.

[0114] In this embodiment, server 1 learns measurement data based on the magnetic tubes that are the measurement objects in the measurement data used for training, and generates multiple estimation models 50 corresponding to various magnetic tubes. Specifically, server 1 learns measurement data based on magnetic tube information related to the magnetic tubes of the measurement objects, and generates multiple estimation models 50 corresponding to the magnetic tube information.

[0115] Magnetic tube information describes the state or properties of a magnetic tube, such as the wall thickness reduction method, dimensions, and material. These are just examples of magnetic tube information; other information may also be included. The wall thickness reduction method refers to the state of the wall-thinned portion of the magnetic tube, specified by its shape (e.g., whether the cross-sectional shape is rectangular or hemispherical). Dimensions are specified by the diameter and length of the magnetic tube. The material is specified by the type of magnetic material forming the magnetic tube.

[0116] Server 1 learns from the training data based on the information of these magnetic tubes and generates estimation models 50 corresponding to the wall thickness reduction method and size of the magnetic tubes. In addition, in this embodiment, multiple estimation models 50 are generated, but regardless of the differences in the magnetic tube information, one estimation model 50 can be used to learn from the training data and estimate the wall thickness information through a single estimation model 50.

[0117] In the actual use of estimation model 50 to estimate wall thickness information, server 1 obtains the 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 estimation model 50. Specifically, server 1 obtains the measurement data and receives specified input from the user via terminal 2 regarding magnetic tube information related to the magnetic tube being measured. Server 1 selects any one of multiple estimation models 50 based on the specified magnetic tube information. Server 1 generates hue images of each hue based on the obtained measurement data, and generates a composite image combining the hue images. Server 1 estimates the wall thickness of each section that divides the magnetic tube along its long side at a fixed length by inputting the composite image into the selected estimation model 50. Server 1 replies with the estimated wall thickness information to terminal 2 and displays it.

[0118] Furthermore, in Embodiment 2, the operation of the terminal 2 (display screen, etc.) when estimating wall thickness information will be described in detail.

[0119] According to this embodiment, by using the estimation model 50, the wall thickness reduction state (wall thickness information) of the magnetic tube can be appropriately estimated and displayed to the user.

[0120] Figure 10 This is a flowchart illustrating the steps involved in generating the estimation model 50. Based on Figure 10 The processing of generating the estimation model 50 through machine learning will be explained. The control unit 11 of server 1 obtains correct values ​​of wall thickness information related to the wall thickness of the magnetic tube from the measurement data group obtained by measuring the magnetic characteristic value of the magnetic tube (step S11). The measurement data are data measured at each position on a cross section orthogonal to the long side direction of the magnetic tube (each position where the cross section of the cylindrical magnetic tube is equally divided along the circumferential direction). The magnetic characteristic value is the measured value obtained using an inspection detector 4, which includes: a magnet 42 that generates a magnetic field; a yoke 41 disposed on the side opposite to the magnetic tube relative to the magnet 42; and a Hall element 43 (magnetic sensor) disposed between the yoke 41 and the magnetic tube, measuring the magnetic flux density through the yoke 41, the magnet 42, and the magnetic tube. This magnetic characteristic value is the output voltage of the Hall element 43, which is proportional to the magnetic flux density. The wall thickness information is, for example, the wall thickness of each portion of the magnetic tube divided along its long side by a fixed length.

[0121] The control unit 11 performs preprocessing on the measurement data used for training (step S12). Specifically, the control unit 11 standardizes the magnetic characteristic values ​​to ensure that the magnetic characteristic values ​​outside the magnetic tube are consistent with those of the normal section in each channel. Furthermore, the control unit 11 determines the baseline of each channel by taking a moving average of the magnetic characteristic values ​​along the long side. By ensuring that the average values ​​of the baselines of each channel are consistent, compensation is made for the magnetic characteristic values ​​when the detector 4 passes through the central axis of the magnetic tube. In addition, the control unit 11 adds summary statistics such as the average value and standard deviation to the measurement data.

[0122] The control unit 11 expands the number of measurement data (step S13). Specifically, the control unit 11 generates measurement data for multiple modes by staggering the channel numbers of the data measured in each channel. Furthermore, the control unit 11 generates measurement data for multiple modes by adding a given noise (Gaussian noise, etc.) to the measurement data for multiple modes. Additionally, the control unit 11 generates measurement data for multiple modes by performing linear interpolation between consecutive two points along the long side.

[0123] The control unit 11 sets the measurement data with expanded data number as the position of the vertical axis (first axis) of the image along the long side, and sets the horizontal axis (second axis) as the position on the cross section. It then transforms the image into an image with pixel values ​​assigned to each pixel based on the magnetic property values ​​at each position within the magnetic tube (step S14). Specifically, the control unit 11 calculates the power values ​​of the magnetic property values ​​of multiple modes with different power exponents, assigns each power value to a different hue, and generates a hue image for each hue. Then, the control unit 11 generates a composite image by combining the hue images. The control unit 11 generates composite images for all combinations when assigning each power value to each hue, using these as training data.

[0124] Based on the generated synthetic image and the correct value of the wall thickness information, the control unit 11 generates an estimation model 50 that estimates the wall thickness information when the measurement data is input (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 obtains an estimated value of the wall thickness information by inputting the synthetic image generated in step S14 into the estimation model 50, compares the obtained estimated value with the correct value, and optimizes the weights and other parameters of the estimation model 50 so that the two are approximately similar. The control unit 11 then completes the series of processes.

[0125] Figure 11 This is a flowchart representing the steps involved in estimating wall thickness information. Based on... Figure 11The process of estimating wall thickness information using estimation model 50 will be described. The control unit 11 of server 1 receives a selection input from terminal 2 for the estimation model 50 used to estimate wall thickness information (step S31). Specifically, the control unit 11 receives input specifying the wall thickness reduction method, size, and material of the magnetic tube. The control unit 11 selects the estimation model 50 corresponding to the specified magnetic tube information.

[0126] The control unit 11 acquires measurement data of the magnetic tube from the terminal 2 (step S32). The control unit 11 performs preprocessing on the acquired measurement data (step S33). The control unit 11 transforms the preprocessed measurement data into an image (step S34). Specifically, the control unit 11 assigns the powers of the magnetic property values ​​with different exponents to each hue, generating a composite image that combines the images of each hue.

[0127] The control unit 11 estimates the wall thickness information of the magnetic tube by inputting the image transformed in step S34 into the estimation model 50 (step S35). For example, the control unit 11 estimates the wall thickness of each section that divides the magnetic tube along its long side by a certain length. The control unit 11 sends the estimated wall thickness information to the terminal 2 (step S36), ending the series of processes.

[0128] Based on the above, according to Embodiment 1, the state of the magnetic tube can be appropriately estimated.

[0129] (Variation Example 1)

[0130] In Embodiment 1, the method of transforming the measurement data of the magnetic tube into an image and inputting it into the estimation model 50 is described. In Modification 1, the method of directly using the original measurement data as input to the estimation model 50 is described. Furthermore, the same reference numerals are used for content repeated in Embodiment 1, and descriptions are omitted.

[0131] Figure 12 This is an explanatory diagram showing the outline of Modification Example 1. In Figure 12 The diagram illustrates a scenario where wall thickness information is estimated by extracting data corresponding to the peak portion of the wall thickness reduction from the measurement data of the magnetic tube and inputting this peak portion data into the estimation model 50. Based on Figure 12 This section provides a summary of the variation.

[0132] In this variation, as estimation model 50, server 1 constructs a model that can directly process the raw measurement data (magnetic property values ​​at various locations within the magnetotube) rather than images. For example, server 1 constructs the estimation model 50 using the model involved in kNN (k-nearest neighbor method).

[0133] In addition, in this variation, the estimation model 50 is described as the model involved in kNN, but the estimation model 50 can be any model that can process the original measurement data, such as SVM, random forest, linear classifier, etc.

[0134] Server 1 generates estimation model 50 based on the same training data as in Implementation Method 1. Furthermore, it is preferable that server 1 performs principal component analysis or similar techniques on the training measurement data to reduce dimensionality. By reducing the dimensionality of the measurement data input to estimation model 50, the processing load on server 1 can be reduced. Additionally, users can arbitrarily choose whether to perform dimensionality reduction (principal component analysis).

[0135] Here, server 1 can directly input the magnetic property values ​​at all positions along the long side into estimation model 50. However, in this embodiment, only a portion of the measured data is extracted as input to estimation model 50. Specifically, server 1 extracts the data corresponding to the peak portion of the wall thickness reduction section as input to estimation model 50.

[0136] Figure 12 The diagram illustrates the extraction of peak data from raw measurement data. Server 1 extracts the data of the output voltage variation, i.e., the peak portion, based on the waveform of the output voltage of Hall element 43, which corresponds to the measurement data.

[0137] There is no particular limitation on the method for extracting the peak data. For example, server 1 determines the baseline (flat waveform) by taking the moving average of the output voltage, and determines the peak portion based on this baseline. Specifically, server 1 defines the portion above a given threshold that differs from the baseline as the peak portion. Additionally, as... Figure 12 As shown, when the Hall element 43 is close to the wall thickness reduction section, the output voltage rises to a positive value and then decreases to a negative value. However, in this modified example, the output voltage is positive, and only the part of the waveform that bulges upward is determined as the peak value.

[0138] Server 1 extracts data from the determined peak portion of the overall measurement data of the magnetic tube and uses it as input to estimation model 50. This reduces the amount of data processed by estimation model 50 and allows for appropriate estimation of the wall thickness information of the thinned portion.

[0139] Alternatively, the data extraction of the peak portion in this variant example can be applied to Implementation 1, where only the data of the peak portion is transformed into an input image, which is then used as the input to the estimation model 50 involved in Implementation 1.

[0140] Server 1 extracts peak data from the measurement data used for training, and generates an estimation model 50 based on the extracted peak data and the correct values ​​of wall thickness information at positions corresponding to those peaks. That is, Server 1 maps the peak data, which are assigned the correct values ​​of wall thickness information (representing the correct labels for the correct wall thickness), to a feature space. Furthermore, Server 1 constructs a dataset for estimating wall thickness information based on the peak data by setting an arbitrary value k involved in the k-nearest neighbor method.

[0141] In estimating wall thickness information, server 1 extracts peak data from the measurement data obtained from terminal 2 and maps it to the feature space. Furthermore, server 1 identifies k data points (peak data points with correct labels) located near the mapped position, and estimates the wall thickness information of the magnetic tube being estimated based on the correct values ​​of the wall thickness information assigned to these k data points.

[0142] Figure 13 This is a flowchart illustrating the steps of generating the estimation model 50 involved in Modification Example 1. After expanding the data concatenation of the measurement data (step S13), the server 1 performs the following processing. The control unit 11 of the server 1 extracts data corresponding to the peak portion of the wall thickness reduction section from the measurement data of the magnetic property values ​​at each position along the long side direction of the magnetic tube (step S201). Specifically, the control unit 11 determines a baseline based on the measurement data of the entire magnetic tube and extracts data of the peak portion whose difference from the baseline is above a given threshold.

[0143] Based on the peak data extracted in step S201 and the correct values ​​of the wall thickness information, control unit 11 generates estimation model 50 (step S202). Specifically, control unit 11 generates a model involved in kNN. Control unit 11 maps the peak data with the correct values ​​(correct labels) of the wall thickness information to the feature space, sets an arbitrary value k, thereby generating a dataset for estimating the wall thickness information based on the peak data, as estimation model 50. Control unit 11 then completes a series of processes.

[0144] Figure 14This is a flowchart illustrating the steps of the wall thickness information estimation process involved in 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 corresponding to the peak portion of the wall thickness reduction section from the measurement data of magnetic property values ​​at each position along the long side direction of the magnetic tube (step S221). Specifically, the control unit 11 determines a baseline based on the measurement data of the entire magnetic tube and extracts data of the peak portion whose difference from the baseline is above a threshold. The control unit 11 estimates the wall thickness information of the wall thickness reduction section by inputting the extracted peak portion data into the estimation model 50 (step S222). Specifically, the control unit 11 maps the peak portion data to the feature space and estimates the wall thickness information of the wall thickness reduction section of the magnetic tube to be estimated based on the correct values ​​of the wall thickness information of the k nearest data points. The control unit 11 then transfers the processing to step S36.

[0145] Based on Modification 1, it is also possible to estimate the wall thickness information directly from the original measurement data without visualizing the measurement data. In particular, in this modification, by extracting the data corresponding to the peak portion of the wall thickness reduction section, the wall thickness information of the wall thickness reduction section can be appropriately estimated.

[0146] (Variation Example 2)

[0147] In this variation, the method of generating the estimated model 50 using ensemble learning is explained.

[0148] Figure 15 This is an explanatory diagram showing the outline of variation example 2. In Figure 15 The illustration shows a scenario where data from multiple data intervals, slightly offset by a given length along the long side of the magnetic tube, is further extracted based on the peak data described in Variation 1, and the data from each interval is input into the estimation model 50 to estimate the wall thickness information. Figure 15 This section provides a summary of the variation.

[0149] In this variant, server 1 uses an ensemble learning approach to generate the estimated model 50. Specifically, server 1 uses gradient boosting to generate the estimated model 50 from the models involved in the decision tree (e.g., LightGBM).

[0150] Gradient boosting is a type of ensemble learning that generates a final model by sequentially generating multiple weak recognizers (models). Server 1 uses training data to generate a weak recognizer (decision tree), and generates the next weak recognizer based on the residual between the estimated value and the correct value of the generated weak recognizer. Server 1 generates weak recognizers sequentially by referring to the gradient of the loss function defined by the error between the estimated value and the correct value, taking into account the learning results of the previous weak recognizers, to generate the final recognizer, i.e., the estimated model 50.

[0151] Furthermore, while gradient boosting is used as an ensemble learning method in this variation, other ensemble learning methods, such as bagging that generates multiple weak recognizers in parallel, can also be used. Additionally, ensemble learning can be applied to the estimation model 50 involved in Variation 1, implementing ensemble learning when generating models such as kNN and SVM.

[0152] Server 1 may also use the data of the peak portion (or the measurement data of the entire magnetic tube) described in Variation 1 as input to the estimation model 50. However, in this variation, the peak portion is divided into data intervals of a given length, and the data of each data interval is input into the estimation model 50.

[0153] Figure 15 The diagram illustrates a scenario where multiple data intervals are extracted from the peak data. Each data interval is a range of measurement points along the long side of the magnetic tube, constituting part of the peak portion. For example, server 1 might use 10 consecutive measurement points as a single data interval. However, the number of measurement points constituting a data interval is not limited to 10.

[0154] Server 1 sequentially shifts the data intervals to be extracted along their longer sides, extracting data from each interval starting from the peak portion. For example, Server 1 extracts 10 data points from the starting point of the peak portion where the difference from the baseline is above a threshold, as the initial data interval. Server 1 shifts the starting points of the data intervals to be extracted by a given number (e.g., one) each time, extracting data from each interval. Thus, Server 1 extracts data from multiple intervals from a single peak portion.

[0155] When estimating wall thickness information for a peak portion, server 1 estimates the wall thickness information by inputting data from multiple extracted data intervals into estimation model 50. This allows for a more appropriate estimation of the wall thickness of the thinned portion.

[0156] Figure 16This is a flowchart illustrating the steps of generating the estimation model 50 involved in Modification 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 corresponding to the peak portion of the wall thickness reduction section from the measurement data of the entire length of the magnetic tube measured along the long side direction (step S301). Furthermore, the control unit 11 extracts data from multiple data intervals from the data of the peak portion (step S302). Specifically, the control unit 11 sequentially offsets the data intervals of a given length that are to be extracted from the starting point of the peak portion along the long side direction, and extracts data from each data interval.

[0157] 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 at the peak portion (wall thickness reduction section) (step S303). Specifically, the control unit 11 generates a decision tree using a gradient boosting method. The control unit 11 generates a weak recognizer (decision tree) using training data (data from multiple data intervals and the correct values ​​of the wall thickness information), and generates subsequent weak recognizers based on the residuals between the estimated and correct values ​​of the wall thickness information generated by the weak recognizer. The control unit 11 sequentially generates weak recognizers according to the gradient of the loss function defined by the residuals between the estimated and correct values, generating the final recognizer, i.e., the estimation model 50. The control unit 11 then concludes the series of processes.

[0158] Figure 17 This is a flowchart illustrating the steps of the wall thickness estimation process involved in Modification Example 2. After preprocessing the measured data (step S33), server 1 performs the following processing. The control unit 11 of server 1 extracts data corresponding to the peak portion of the wall thickness reduction section from the measured magnetic property values ​​at each position along the long side direction (step S321). The control unit 11 extracts data from multiple data intervals from the peak portion data (step S322). The control unit 11 estimates the wall thickness information of the wall thickness reduction section by inputting the extracted data from the multiple data intervals into the estimation model 50 (step S323). The control unit 11 then transfers the processing to step S36.

[0159] Based on Variation Example 2, the ensemble learning involved in gradient boosting can also be used to generate the estimation model 50. Furthermore, by dividing the peak portion data into multiple data intervals and inputting them into the estimation model 50, the wall thickness information of the peak portion (thinning section) can be estimated more appropriately.

[0160] (Variation Example 3)

[0161] An example using an ensemble learning method different from Variation 2 is illustrated. As a general example of ensemble learning, it is known to input the same dataset into multiple models of different learning institutions, and to use the average of the outputs from these multiple models, or the majority decision output from these multiple models, as the learning result.

[0162] In variation 3, unlike this learning, the input data is divided into multiple groups, and each group is input into a different corresponding model. The output of each corresponding model is then input into other models, and the output of the other model is used as the learning result.

[0163] Figure 18 This is an explanatory diagram showing the outline of variation example 3. For example... Figure 18 As shown, the estimation model in Variation 3 includes: a thin-walled specialization model 50a (Bag Tree-A) learned primarily based on measurement data and wall thickness information that the residual wall thickness of the magnetic tube is thinner than a given value; a thick-walled specialization model 50b (Bag Tree-B) learned primarily based on measurement data and wall thickness information that the residual wall thickness of the magnetic tube is thicker than a given value; and a final model 50c (Bag Tree-C) with the outputs of the thin-walled specialization model 50a and the thick-walled specialization model 50b as inputs.

[0164] For example, the thin-walled specialization model 50a is a model that takes 90% of the data with a wall thickness of 0.1 to 0.8 mm and 10% of the data with a wall thickness of more than 0.8 mm as objects, while the thick-walled specialization model 50b is a model that takes the remaining data in the whole set of data that was not used in the thin-walled specialization model 50a as objects. Figure 19 This example illustrates the weighted relationship between two models, 50a and 50b.

[0165] Furthermore, the wall thickness of the reference for separating the thin-walled specialized model 50a and the thick-walled specialized model 50b is set to 0.8 mm, but this reference wall thickness value is only one example, and other values ​​may also be used. In addition, as an object of the thin-walled specialized model 50a, 90% of the data with a wall thickness below the given value and 10% of the data with a wall thickness exceeding the given value are used, but this 90%:10% ratio is only one example, and other ratios may also be used.

[0166] Figure 20This is a flowchart illustrating the steps of generating the estimation models (thin-wall specialized model 50a, thick-wall specialized model 50b, and final model 50c) in Modified Example 3. The control unit 11 of server 1 acquires first training data (step S401), which includes measurement data of the magnetic tube, primarily at the thin-walled region, as described above, and correct values ​​of wall thickness information corresponding to the measurement data. Based on the acquired first training data (measurement data and correct values ​​of wall thickness information), the control unit 11 generates a thin-wall specialized model 50a that estimates and outputs the first wall thickness information when the measurement data is input (step S402).

[0167] The control unit 11 acquires the second training data (step S403), which includes measurement data of the magnetic tube, mainly at thick-walled areas, as described above, and correct values ​​of wall thickness information corresponding to the measurement data. 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 the second wall thickness information when the measurement data is input (step S404).

[0168] The control unit 11 acquires the third training data (step S405), which includes the first wall thickness information output from the thin-walled specialization model 50a and the second wall thickness information output from the thick-walled specialization model 50b, as well as the correct values ​​for establishing the corresponding wall thickness information. Based on the acquired third training data (the first and second wall thickness information and the correct values ​​for the wall thickness information), the control unit 11 generates a final model 50c that estimates and outputs the wall thickness information when the first wall thickness information and the second wall thickness information are input (step S406).

[0169] Figure 21 This is a flowchart illustrating the steps of the wall thickness information estimation process in Modified Example 3. The control unit 11 of server 1 receives selection input (step S411) from terminal 2 for the estimation model (thin-wall specialized model 50a, thick-wall specialized model 50b, and final model 50c) used to estimate the wall thickness information. Specifically, the control unit 11 receives specified inputs such as the boundary value of the wall thickness that distinguishes the thin-wall model from the thick-wall model, and the proportion of thick walls included in the thin-wall model. The control unit 11 selects the estimation model (thin-wall specialized model 50a, thick-wall specialized model 50b, and final model 50c) corresponding to the specified input.

[0170] The control unit 11 obtains the measurement data of the magnetic tube from the terminal 2 (step S412). The control unit 11 inputs the obtained measurement data into the thin-walled specialized model 50a and the thick-walled specialized model 50b (step S413). Next, the control unit 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, thereby estimating the wall thickness information of the magnetic tube (step S414). The control unit 11 sends the estimated wall thickness information to the terminal 2 (step S36), ending the series of processes.

[0171] Alternatively, in variation 3, the final model 50c can be omitted, and the average of the output from the thin-walled specialized model 50a and the output from the thick-walled specialized model 50b can be calculated, and the calculated average value can be used as the estimated wall thickness information of the magnetic tube.

[0172] According to Variation 3, by applying different appropriate models to the thin-walled side and the thick-walled side, more accurate wall thickness information can be estimated. Generally, there are more cases where the amount of data is less on the thin-walled side, but even in such cases, it is possible to improve the estimation accuracy of wall thickness information on the thin-walled side where the amount of data is less.

[0173] (Variation Example 4)

[0174] Variation 4 is an example related to the method of using dynamic time scaling to compensate for the measurement data used in the learning of the estimation model in order to reduce the influence of Hall element voltage on magnetic flux disturbances accompanying the MFR mode.

[0175] First, the effect of the Hall element voltage on the magnetic flux disturbance accompanying the above-mentioned MFR method will be explained. Figure 22 This is an explanatory diagram used to illustrate the effect. For example... Figure 22 The cross-sectional view of the magnetic tube is shown, assuming a deep local wall thinning section A is formed in the magnetic tube. When scanning such a magnetic tube using the MFR (Mean Fractional Flow) inspection detector 4, the MFR waveform that should ideally be detected by the Hall element 43, based on the change in the wall thinning shape, is waveform B. However, the actual detected MFR waveform, as shown in waveform C, sometimes reveals areas near the wall thinning section that are independent of the wall thickness (especially the areas surrounded by ○). This is because, as shown in the cross-sectional view of the magnetic tube, the magnetic field lines deform around the wall thinning section, creating blank areas of magnetic field lines (the area surrounded by the dashed line D).

[0176] To eliminate the aforementioned problems, dynamic time scaling is used to compensate for the measurement data used in the learning of the estimation model. Figure 23A , Figure 23B This is an explanatory diagram showing the outline of the dynamic time-scaling method in Variation Example 4. Figure 23A , Figure 23BIn the diagram, 'a' represents the time-series waveform of the measured data (detection voltage). Specifically, it represents the time-series waveform based on the 3-peak method (in the case of 8 channels, the sum of the voltage values ​​of the channel with the highest detection voltage and the voltage values ​​of the two channels on either side of that channel, divided by 8). 'b' represents the time-series waveform of the wall thickness value, which is the target variable being measured.

[0177] In variation example 4, such as Figure 23B As shown, the time series points of the three peak values ​​(waveform a) are correlated with the time series points of the measured wall thickness value (waveform b) to minimize the sum of the Euclidean distances between the three peak values ​​(measured data) and the measured wall thickness value (wall thickness information). For example, the time series point i of waveform b changes to the corresponding time series point i+2, which is only shifted by 2 points from waveform a.

[0178] Figure 24 This graph represents the result of estimating wall thickness information by inputting the measurement data into an estimation model that has been learned using the compensated measurement data as described above. Figure 24 In the diagram, waveform A represents the actual measured data (MFR waveform), and wall thickness reduction shape B represents the estimated wall thickness information (wall thickness reduction shape). Additionally, in... Figure 24 In the middle, the actual MFR waveform without using the dynamic time scaling method (equivalent to Figure 22 Waveform C) is illustrated with dashed lines overlapping waveform A. Furthermore, the actual wall thickness is reduced to a shape (equivalent to...). Figure 22 The wall thickness reduction shape shown is superimposed on the estimated wall thickness reduction shape B by dashed lines.

[0179] In variation 4, the dynamic time scaling method is used to compensate for the measurement data used in the learning of the estimation model. Therefore, the influence of the Hall element voltage accompanied by magnetic flux disturbance in the MFR method described above can be reduced, and the wall thickness reduction shape of the magnetic tube can be estimated with high accuracy.

[0180] (Variation Example 5)

[0181] Variation 5 is an example of using principal component analysis to increase the number of data points. Figure 25 This is an explanatory diagram related to the data expansion processing in Variation Example 5.

[0182] Server 1 can process the raw measurement data obtained by the inspection detector 4 with 8 channels (CH1~CH8), and the first set of data obtained by preprocessing the raw measurement data, including arithmetic processing. Figure 25 In the example shown, the moving average, moving standard deviation, moving skewness, moving kurtosis, and moving difference at multiple locations (10–50 points) along the long side of the magnetic tube are used as inputs to estimate model 50 (see [reference]). Figure 25(a)), but in variation 5, in order to improve the estimation accuracy of wall thickness information, in addition to the original measurement data and the first set of data, the second set of data obtained by performing principal component analysis on the original measurement data and the first set of data will also be used as the input of the estimation model 50 (refer to...). Figure 25 (b)

[0183] In addition, Figure 25 In the example shown, the first and second sets of data are moving averages, moving standard deviations, moving skewness, moving kurtosis, and moving differences, but are not limited to these. Furthermore, the second set of data is set to the same data structure as the first set, but the two sets can also have different data structures.

[0184] In variation 5, in addition to the principal component analysis results, the number of data points also increases (in... Figure 25 In the example shown, the number of variables increased from 169 to 338, thus improving the estimation accuracy of wall thickness information.

[0185] (Implementation Method 2)

[0186] In Embodiment 1, a defect estimation system for estimating defects such as wall thinning of magnetic tubes using estimation model 50 was described. In this embodiment, an implementation method in which a user actually uses this system to inspect defects in magnetic tubes is described.

[0187] Figure 26 This is an explanatory diagram showing an example of a screen for uploading measurement data. In Figure 26 The image shows the screen when terminal 2 sends (uploads) measurement data to server 1. The upload screen includes a model selection bar 171, an upload button 172, and a "Can I provide a selection?" bar 173.

[0188] Model selection field 171 is an input field for selecting the estimation model 50 used in estimating wall thickness information. It is an input field for accepting specified inputs of magnetic tube information related to the magnetic tube being estimated (measured). For example, model selection field 171 includes a wall thickness reduction method specification field 1711, a size specification field 1712, and a material specification field 1713. Terminal 2 accepts specified inputs of the magnetic tube's wall thickness reduction method, size, and material via each specified field.

[0189] The upload button 172 is used to send (upload) measurement data to server 1. Upon receiving input to the upload button 172, terminal 2 sends the measurement data to server 1.

[0190] The "Can Provide?" selection field 173 is an input field used to select whether measurement data can be provided (used) to this system. Terminal 2 accepts selection input related to whether or not measurement data can be provided, based on the operation input to the "Can Provide?" selection field 173. If the selection input indicating the intention to provide measurement data is accepted, server 1 stores the measurement data obtained from terminal 2 in measurement DB 143. As described later, the measurement data provided by the user is used for the relearning (updating) of estimation model 50.

[0191] Alternatively, regardless of whether an option is provided, all measurement data obtained from each terminal 2 can be used for the relearning of the estimation model 50.

[0192] Figure 27 This is an explanatory diagram of an example of a display screen showing wall thickness information. In Figure 27 The image shows a display screen for the wall thickness information estimated based on estimation model 50. This screen includes a magnetic tube selection bar 181, a wall thickness chart 182, a cross-sectional image 184, a download button 185, and a second measurement data upload button 186.

[0193] The magnetic tube selection bar 181 is an input field used to select a magnetic tube for displaying wall thickness information. Terminal 2 displays the wall thickness information of the magnetic tube selected in the magnetic tube selection bar 181.

[0194] Wall thickness chart 182 is a chart showing the wall thickness of each part of the magnetic tube, estimated based on estimation model 50. The horizontal axis of wall thickness chart 182 represents the position of the magnetic tube along its long side, and the vertical axis represents the wall thickness. For example, terminal 2 displays the average wall thickness of each channel in wall thickness chart 182. Alternatively, terminal 2 can also display the wall thickness of each channel in wall thickness chart 182.

[0195] Cross-sectional image 184 is an image simulating a cross-section orthogonal to the long side direction of the magnetic tube, and is an image reproducing the wall thickness at various locations on the cross-section based on wall thickness information. As described above, server 1 estimates the wall thickness of each channel (at each location on the cross-section) at various locations along the long side direction of the magnetic tube. Server 1 detects wall thickness reduction sections where the wall thickness is thinner than the normal section by comparing the estimated wall thickness of each channel with the wall thickness of the normal section (the baseline of wall thickness chart 182). Then, server 1 calculates the difference in wall thickness between the normal section and the wall thickness reduction section, i.e., the wall thickness reduction width. For example, terminal 2 displays a cross-section of the magnetic tube with CH1 set to the 0 point direction, and displays the cross-section of the magnetic tube with the detected wall thickness reduction section (at each location on the cross-section). Figure 27 The position corresponding to CH1 is reproduced and displayed, and the wall thickness reduction part (recess) corresponding to the calculated wall thickness reduction width is displayed.

[0196] For example, terminal 2 displays line 183 on wall thickness chart 182. Terminal 2 accepts a specified input specifying the position of the magnetic tube along the long side of the cross-sectional image 184, by accepting an operation to move 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.

[0197] In addition, terminal 2 displays the number of wall-thinning sections detected based on wall thickness information, the maximum value of the wall-thinning width, the average value, etc.

[0198] in addition, Figure 27 The wall thickness information display shown is an example, and this embodiment is not limited to it. For example, terminal 2 may display only the wall thickness at each location within the magnetic tube, as shown in wall thickness chart 182. Furthermore, for example, when terminal 2 displays the wall thickness at each location within the magnetic tube in wall thickness chart 182, it may change the display method (e.g., display color) of the wall thickness at each location based on the reliability (probability value representing the accuracy of the estimation result) of the wall thickness at the corresponding location based on estimation model 50. Furthermore, for example, if terminal 2 detects multiple wall thickness reduction sections based on the wall thickness information, it may also display multiple cross-sectional images 184, 184, 184... corresponding to each wall thickness reduction section at a glance. Thus, the display method of the wall thickness information is not limited to... Figure 27 As shown in the diagram.

[0199] In addition, in this embodiment, the terminal 2 displays the wall thickness information on a dedicated screen, but the terminal 2 may also simply download a file containing the estimation results of the wall thickness information (e.g., a CSV file) without displaying the wall thickness information.

[0200] The download button 185 is used to download a file containing the estimated wall thickness information. Upon receiving input to the download button 185, terminal 2 retrieves the file containing the estimated wall thickness information from server 1.

[0201] The second measurement data upload button 186 is used to upload measurement data other than the aforementioned measurement data (measurement data based on the flux resistance method), i.e., the second measurement data, to server 1. Upon receiving input to the second measurement data upload button 186, terminal 2 sends measurement data based on the water immersion rotating ultrasonic thickness measurement method (IRIS) as the second measurement data to server 1. The second measurement data will be described later.

[0202] As described above, server 1 obtains measurement data of the magnetic tube from each user's terminal 2 and estimates the wall thickness information. Then, server 1 sends the wall thickness information to the terminal 2, which is the source of the measurement data, and displays it. In this embodiment, server 1 performs relearning based on the measurement data obtained from each terminal 2 and updates the estimation model 50.

[0203] Specifically, when server 1 accepts the option to provide measurement data via the "Can provide?" selection field 173, it stores the measurement data obtained from terminal 2 in measurement DB 143. Server 1 uses the measurement data stored in measurement DB 143 as training data for relearning. Server 1 updates the estimation model 50 based on the measurement data stored in measurement DB 143 and the correct values ​​of the wall thickness information corresponding to that measurement data.

[0204] The correct value of the wall thickness information during relearning can be obtained by the user via, for example, through... Figure 27 While manual settings for the screen are possible, in this embodiment, the second measurement data provided by the user is used as the correct value. The second measurement data is obtained by measuring the wall thickness at various locations inside the magnetic tube, such as the measurement data involved in IRIS.

[0205] IRIS is a method for inspecting the wall thickness of pipes (magnetic pipes) using ultrasound. A detector equipped with an ultrasonic probe is inserted into a water-filled pipe. The ultrasonic probe senses the reflected waves of the ultrasonic beam generated by the detector, thereby determining the wall thickness. IRIS itself is well-known, so detailed description is omitted in this embodiment. Compared to inspection methods that use magnetic properties, such as magnetoresistive methods, RFECT, and MFL, IRIS is slower but has the advantage of higher measurement accuracy.

[0206] In this embodiment, upon receiving the measurement data related to IRIS from the user, server 1 uses this data as the correct value for the wall thickness information. That is, server 1 uses the pair of measurement data based on the flux resistance method and the second measurement data based on IRIS as training data for relearning.

[0207] Server 1 compares the estimated wall thickness based on the measurement data provided by the user with the correct wall thickness shown in the second measurement data. Furthermore, Server 1 updates parameters such as the weights between neurons to make the two values ​​approximately equal. Server 1 relearns from the measurement data of each magnetic tube provided by the user and updates the estimation model 50.

[0208] Server 1 can allow an estimation model 50 to learn from the measurement data of all users, but preferably it learns from the measurement data obtained from each user's terminal 2 separately, updating the estimation model 50 for each user. That is, server 1 provides the estimation model 50 with the measurement data provided by each user, constructing an estimation model 50 for each user. Thus, the estimation model 50 can be adjusted individually according to the tendency of the magnetic tube measured (inspected) by each user.

[0209] The above demonstrates that the estimation accuracy of estimation model 50 can be improved through the application of this system.

[0210] Server 1 charges each user of the system a usage fee for the estimation model 50. The usage fee could be a fixed amount (subscription) for each fixed period, but in this embodiment, it is determined based on the computational workload of the estimation process based on the wall thickness information from the estimation model 50. The computational workload used as a basis for determining the usage fee could be, for example, the number of estimations of the wall thickness information (the number of magnetic tubes), but could 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 wall thickness information, it determines the usage fee to be charged to the user based on the computational workload at the time of estimation. Server 1 stores the determined usage fee in the user DB141 and ultimately requests the usage fee from the user.

[0211] In this embodiment, server 1 adjusts the usage fee based on whether measurement data is provided (used). Specifically, server 1 deducts the usage fee when measurement data (magnetoresistance-based measurement data) that forms the basis for wall thickness information estimation is provided. Furthermore, server 1 further deducts the usage fee when second measurement data (IRIS-based measurement data) corresponding to the first measurement data is provided. This allows for efficient collection of training data for relearning.

[0212] Figure 28 This is a flowchart illustrating the steps involved in the wall thickness information estimation process according to Implementation Method 2. Based on Figure 28 The processing content of server 1 and terminal 2 involved in this embodiment will be described.

[0213] The control unit 21 of terminal 2 accepts a selection input for the estimation model 50 used to estimate wall thickness information (step S501). Specifically, the control unit 21 accepts specified inputs related to magnetic tube information, such as the wall thickness reduction method, size, and material of the magnetic tube, which is the object of estimation. Furthermore, the control unit 21 accepts a selection input regarding whether measurement data related to updating the estimation model 50 can be used (provided) (step S502). The control unit 11 sends the measurement data of the magnetic tube, along with the selections from steps S501 and S502, to server 1 (step S403). Upon obtaining measurement data from terminal 2, the control unit 11 of server 1 executes the processing steps S33 to S35, sending the wall thickness information estimated based on the estimation model 50 to terminal 2 (step S36).

[0214] Upon obtaining the wall thickness information from server 1, the control unit 21 of terminal 2 displays the estimated wall thickness information on display unit 24 (step S504). Specifically, the control unit 21 displays the wall thickness at each position (part) along the long side direction of the magnetic tube using a wall thickness chart 182, and displays the wall thickness at each position (channel) on a cross-section at any position along the long side direction using a cross-sectional image 184, etc.

[0215] Based on user input, the control unit 21 sends the second measurement data, obtained by measuring the wall thickness of the magnetic tube using a method other than the flux resistance method, to the server 1 (step S505). The second measurement data is, for example, the measurement data involved in IRIS, which is data obtained by inserting a detector equipped with an ultrasonic probe into a water-filled magnetic tube, and the ultrasonic probe sensing the reflected wave of the ultrasonic beam generated by the detector.

[0216] After executing step S36, the control unit 11 of server 1 determines the usage fee (charge) to be charged to the user (step S506). Specifically, the control unit 11 determines the usage fee based on the computational load (number of calculations, calculation time, amount of measurement data, etc.) of the estimation processing based on the wall thickness information of estimation model 50. Furthermore, the control unit 11 subtracts the usage fee based on the availability (from use to acquisition) of the measurement data and the second measurement data.

[0217] When the user accepts the selection input indicating that measurement data can be provided, the control unit 11 stores the measurement data sent in step S403 in the measurement DB143 (step S507). Furthermore, when the control unit 11 sends the second measurement data in step S405, it stores the second measurement data in the measurement DB143 in a corresponding manner with the measurement data.

[0218] The control unit 11 updates the estimation model 50 based on the measurement data stored in the measurement DB143 and the second measurement data (step S508). That is, the control unit 11 uses the measurement data measured by the user as input data for training, uses the measured wall thickness value from the second measurement data as the correct value for wall thickness information, and updates parameters such as the weights between neurons. For example, the control unit 11 learns from the measurement data obtained from each user's terminal 2 and updates the estimation model 50 for each user. The control unit 11 then completes this series of processes.

[0219] As described above, according to Embodiment 2, this system can be appropriately implemented, and the estimation model 50 can be optimized through the application of this system.

[0220] The embodiments disclosed herein should be considered illustrative in all respects and not restrictive. The scope of the invention is not as described above, but is set forth in the claims and is intended to include all modifications equivalent to and within the scope of the claims.

[0221] -Explanation of Figure Markers-

[0222] 1. Server (Information Processing Device)

[0223] 11 Control Department

[0224] 12 main storage units

[0225] 13Ministry of Communications

[0226] 14 Auxiliary Storage Section

[0227] P1 program

[0228] 142 Model DB

[0229] 143 DB measurement

[0230] 2. Terminal (User Terminal)

[0231] 21 Control Department

[0232] 22 Main Storage Section

[0233] 23 Ministry of Communications

[0234] 24 Display Unit

[0235] 25 Input Section

[0236] 26 Auxiliary Storage Unit

[0237] P2 program

[0238] 3. Measuring apparatus

[0239] 4. Check the detector

[0240] 41 Magnetic Yoke

[0241] 42 magnets

[0242] 3 Hall effect elements (magnetic sensors)

[0243] 50 estimation model

[0244] 50a thin-walled specialized model

[0245] 50b Thick-walled Specialized Model

[0246] 50c final model.

Claims

1. An information processing method in which a computer performs the following processing: The measurement data obtained by measuring the magnetic properties of the magnetic tube are used to acquire the measurement data. The wall thickness information related to the wall thickness of the magnetic tube is estimated by inputting the obtained measurement data into the learned model, so that the wall thickness information can be estimated given the input measurement data. The model includes: The first model is mainly based on the measured data and the information that the residual wall thickness of the magnetic tube is thinner than the given value for learning. The second model is mainly based on the measured data and the wall thickness information that the residual wall thickness of the magnetic tube is thicker than the given value for learning. And a third model, which is learned to output wall thickness information given wall thickness information output from the first model and wall thickness information output from the second model as input. The wall thickness information is estimated by inputting the obtained 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.

2. The information processing method according to claim 1, wherein, The magnetic characteristic value is a measured value obtained using an inspection detector, which includes: a magnet that generates a magnetic field; a magnetic yoke disposed on the opposite side of the magnetic tube relative to the magnet; and a magnetic sensor disposed between the magnetic yoke and the magnetic tube, which measures the magnetic flux density passing through the magnetic yoke, the magnet, and the magnetic tube, and the magnetic characteristic value is the output voltage of the magnetic sensor that is proportional to the magnetic flux density.

3. The information processing method according to claim 2, wherein, The measurement data is obtained by using an inspection detector with magnets and magnetic sensors periodically mounted on the outer periphery of the magnetic yoke, measuring the magnetic characteristic values ​​at various positions along the long side of the cylindrical magnetic tube, dividing the cross-section of the magnetic tube circumferentially. The magnetic property values ​​are standardized so that the magnetic property values ​​outside the magnetic tube are consistent at all positions along the circumferential direction, and the magnetic property values ​​of the normal portion of the magnetic tube without wall thickness reduction are consistent at all positions along the circumferential direction. The wall thickness information is estimated by inputting the measured data, which has been standardized with the magnetic property values, into the model.

4. The information processing method according to claim 2 or 3, wherein, The measurement data is obtained by using an inspection detector with magnets and magnetic sensors periodically mounted on the outer periphery of the magnetic yoke, and measuring the magnetic property values ​​at various locations where the cross-section of the cylindrical magnetic tube is equally divided along the circumferential direction. The magnetic property values ​​at each location are compensated to the values ​​when the inspection detector passes through the central axis of the magnetic tube. The wall thickness information at each location is estimated by inputting the measured data, which has been compensated for the magnetic property values, into the model.

5. The information processing method according to claim 2 or 3, wherein, The measurement data is obtained by using an inspection detector with magnets and magnetic sensors periodically mounted on the outer periphery of the magnetic yoke, measuring the magnetic characteristic values ​​at various positions along the long side of the cylindrical magnetic tube, dividing the cross-section of the magnetic tube circumferentially. Based on the magnetic property values ​​at each position along the long side or circumferential direction, calculate the average value, standard deviation, skewness, or kurtosis of the magnetic property values ​​along the long side or circumferential direction. The wall thickness information is estimated by inputting the measured data, with the mean, standard deviation, skewness, or kurtosis added, into the model.

6. The information processing method according to any one of claims 1 to 3, wherein, The measurement data is obtained by measuring the magnetic property values ​​at various positions along the long side of the magnetic tube. By inputting the measured data into the model, the wall thickness information of each section of the magnetic tube divided into fixed lengths along the long side direction is estimated.

7. The information processing method according to claim 6, wherein, The baseline of the measurement data is determined by taking a moving average of the magnetic property values ​​along the long side. The difference between the baseline and the magnetic property value is extracted from the measurement data, and the data representing the peak value above a given threshold is extracted. The wall thickness information of the peak portion is estimated by inputting the extracted peak portion data into the model.

8. The information processing method according to claim 7, wherein, Extract data from the measured data, ensuring that data intervals of a given length are staggered along the long side. The wall thickness information of the peak portion is estimated by inputting the extracted data from each data interval into the model.

9. The information processing method according to any one of claims 1 to 3, wherein, The measurement data is obtained by measuring the magnetic property values ​​at various positions along the long side of the magnetic tube on a cross section orthogonal to the long side. The measured data is transformed into an image where the first axis of the image is taken as the position along the long side, the second axis is taken as the position on the cross section, and the pixel value of each pixel is assigned according to the magnetic property value of each position of the magnetic tube. The wall thickness information is estimated by inputting the image into the model.

10. The information processing method according to claim 9, wherein, Based on the measured data, multiple hue images are generated, assigning the power values ​​of the magnetic property values ​​of multiple modes with different power exponents to different hues. Generate a composite image obtained by combining the multiple hue images. The wall thickness information is estimated by inputting the generated synthetic image into the model.

11. The information processing method according to any one of claims 1 to 3, wherein, Obtain magnetic tube information related to the magnetic tube of the test object from the measurement data. After learning from multiple models that have different training data based on the magneto tube information, select the model corresponding to the obtained magneto tube information. The wall thickness information is estimated by inputting the measured data into the selected model.

12. The information processing method according to any one of claims 1 to 3, wherein, The measured data is compensated to minimize the sum of the measured data and the Euclidean distances to the wall thickness information corresponding to the measured data. The model is learned based on the compensated measurement data and wall thickness information.

13. The information processing method according to any one of claims 1 to 3, wherein, The measurement data includes sensor data obtained from multiple sensors arranged in the circumferential direction of the magnetic tube, a first set of data obtained by preprocessing the sensor data including computational processing, and a second set of data obtained by performing principal component analysis on the sensor data and the first set of data. The wall thickness information is estimated by inputting the sensor data, the first set of data, and the second set of data into the model.

14. According to the information processing method of claim 1, the computer performs the following processing: The measurement data is obtained from user terminals that are communicatively connected via a network. The estimated wall thickness information is sent to the user terminal, which is the source of the measurement data.

15. The information processing method according to claim 14, wherein, The magnetic characteristic value is a measured value obtained using an inspection detector, which includes: a magnet that generates a magnetic field; a magnetic yoke disposed on the opposite side of the magnetic tube relative to the magnet; and a magnetic sensor disposed between the magnetic yoke and the magnetic tube, which measures the magnetic flux density passing through the magnetic yoke, the magnet, and the magnetic tube, wherein the magnetic characteristic value is the output voltage of the magnetic sensor, which decreases as the magnetic flux density increases.

16. The information processing method according to claim 14 or 15, wherein, Accepts specified inputs related to the magnetic tube of the object being measured, including magnetic tube information. After learning from multiple models that have different training data based on the magneto tube information, select the model corresponding to the obtained magneto tube information. The wall thickness information is estimated by inputting the measured data into the selected model.

17. The information processing method according to claim 14 or 15, wherein, The measurement data obtained from each user terminal is stored in the storage unit. Obtain the correct value of the wall thickness information corresponding to the measured data. The model is updated based on the measurement data and the correct value stored in the storage unit.

18. The information processing method according to claim 17, wherein, For each user who is the source of the measurement data, the model is updated using the measurement data obtained from that user's user terminal.

19. The information processing method according to claim 17, wherein, Based on the estimated wall thickness information, the user is charged a usage fee for the model, according to the computational cost of estimating the wall thickness information using the model.

20. The information processing method according to claim 19, wherein, When the measurement data is obtained from the user terminal, the user selects whether the measurement data is usable in connection with accepting the model update. If the input indicating that the measurement data can be used is received, the measurement data is stored in the storage unit. Based on whether the measured data can be used, the usage fee for the model charged to the user shall be deducted.

21. The information processing method according to claim 19 or 20, wherein, A second set of measurement data representing the correct value is obtained from the user terminal. This second set of measurement data is obtained by measuring the wall thickness of the magnetic tube using a different measurement method than the first set of measurement data. The model is updated based on the measured data and the second measured data. Based on whether or not the second set of measurement data is obtained, the usage fee for the model charged to the user shall be subtracted.

22. The information processing method according to claim 21, wherein, The second measurement data is based on the measurement data of the water immersion rotating ultrasonic thickness measurement method.

23. A program product that causes a computer to perform the following processing: The measurement data obtained by measuring the magnetic properties of the magnetic tube are used to acquire the measurement data. The wall thickness information related to the wall thickness of the magnetic tube is estimated by inputting the obtained measurement data into the learned model, so that the wall thickness information can be estimated given the input measurement data. The model includes: The first model is mainly based on the measured data and the information that the residual wall thickness of the magnetic tube is thinner than the given value for learning. The second model is mainly based on the measured data and the wall thickness information that the residual wall thickness of the magnetic tube is thicker than the given value for learning. And a third model, which is learned to output wall thickness information given wall thickness information output from the first model and wall thickness information output from the second model as input. The wall thickness information is estimated by inputting the obtained 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.

24. The program product according to claim 23, causing the computer to perform the following processing: The measurement data is sent to the information processing device that estimates the wall thickness information. The estimated wall thickness information is obtained from the information processing device. The obtained wall thickness information is displayed on the display unit.

25. The program product according to claim 24, wherein, The second measurement data, representing the correct value of the wall thickness information, is sent to the information processing device. This second measurement data is obtained by measuring the wall thickness of the magnetic tube using a different measurement method than the first measurement data. The wall thickness information, estimated using the model updated based on the measured data and the second measured data, is obtained from the information processing device.

26. The program product according to claim 24 or 25, wherein, The measurement data is obtained by measuring the magnetic property values ​​at various positions along the long side of the magnetic tube on a cross section orthogonal to the long side. Based on the wall thickness information, which represents the wall thickness at each location on the cross-section along the long side, a graph representing the wall thickness at each location along the long side is displayed. On the graph, input specifying a position along the long side is accepted. The image displays a cross-sectional image of the magnetic tube, showing the wall thickness at various locations on the cross-section at the specified position.

27. An information processing device comprising: The acquisition section acquires measurement data obtained by measuring the magnetic properties of the magnetic tube; and The estimation unit estimates wall thickness information related to the wall thickness of the magnetic tube by inputting the acquired measurement data into the learned model, so that the wall thickness information can be estimated given the input measurement data. The model includes: The first model is mainly based on the measured data and the information that the residual wall thickness of the magnetic tube is thinner than the given value for learning. The second model is mainly based on the measured data and the wall thickness information that the residual wall thickness of the magnetic tube is thicker than the given value for learning. And a third model, which is learned to output wall thickness information given wall thickness information output from the first model and wall thickness information output from the second model as input. The estimation unit estimates the wall thickness information by inputting the obtained 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.

28. The information processing apparatus according to claim 27, comprising: The acquisition unit acquires the measurement data from each user terminal that is communicatively connected via a network. The information processing device further includes a transmission unit that transmits the estimated wall thickness information to the user terminal, which is the source of the measurement data.

29. A model generation method, wherein a computer performs the following processing: By obtaining the correct values ​​of wall thickness information related to the wall thickness of the magnetic tube and the measurement data obtained from measuring the magnetic properties of the magnetic tube, corresponding training data is established. Based on the training data, a fully trained model is generated that estimates the wall thickness information given the input measurement data, wherein... The first training data was obtained, which mainly included measurement data and information on the wall thickness of the magnetic tube, indicating that the remaining wall thickness was thinner than a given value. Based on the first training data obtained, a first model is generated that outputs the first wall thickness information when the measurement data is input. The second training data was obtained, which mainly included measurement data and information on the wall thickness of the magnetic tube, specifically the difference between the remaining wall thickness and the given value. Based on the obtained second training data, a second model is generated that outputs second wall thickness information when the measured data is input. Obtain the third training data, which includes the first wall thickness information output from the first model, and the second wall thickness information and wall thickness information output from the second model. Based on the obtained third training data, a third model is generated that outputs wall thickness information when the first and second wall thickness information are input.

30. The model generation method according to claim 29, wherein, The measured data are obtained by measuring the magnetic property values ​​at various locations where the cross-section of the cylindrical magnetic tube is equally divided along the circumferential direction. By offsetting the positions on the cross-section of the magnetic tube for which the magnetic property values ​​have been measured along the circumferential direction, measurement data for multiple patterns are generated based on the training data. The learned model is generated using the measurement data from the multiple modes.

31. The model generation method according to claim 29 or 30, wherein, By adding a given noise of multiple patterns, measurement data for multiple patterns is generated based on the training data. The learned model is generated using the measurement data from the multiple modes.

32. The model generation method according to claim 29 or 30, wherein, The measurement data is obtained by measuring the magnetic property values ​​at various positions along the long side of the magnetic tube. By performing linear interpolation of the magnetic property values ​​between two consecutive points along the long side, measurement data for multiple patterns are generated based on the training data. The learned model is generated using the measurement data from the multiple modes.

33. The model generation method according to claim 29 or 30, wherein, Obtain the measurement data and the corresponding wall thickness information. The measured data is compensated to minimize the sum of the Euclidean distances between the obtained measured data and the wall thickness information. The model is learned based on the compensated measurement data and wall thickness information.

Images