Patent Application
20240159708
The present invention relates to an information processing method, a non-transitory computer-readable storage medium, an information processing device, and a model generation method.
As a method for inspecting defects such as thickness reduction in a magnetic tube in a non-destructive manner, remote field eddy current testing (RFECT), magnetic flux leakage (MFL), and the like are known.
For example, Japanese Patent No. 6514592 discloses a method of measuring defects in a magnetic member by magnetic flux resistance (MFR).
By the way, in the inspection methods, the wall thickness and the like of the magnetic tube are predicted by applying a measurement value to a model formula prepared in advance. Therefore, it is necessary to manually create a model formula for predicting the wall thickness and the like.
An object of an aspect is to provide an information processing method capable of suitably estimating a state of a magnetic tube, and the like.
According to an aspect, there is provided an information processing method of causing a computer to execute: processing of acquiring measurement data obtained by measuring magnetic characteristic values of a magnetic tube; and processing of estimating wall thickness information by inputting the acquired measurement data to a model trained for estimating the wall thickness information relating to the wall thickness of the magnetic tube in a case where the measurement data is input.
In the aspect, it is possible to suitably estimate the state of the magnetic tube. The above and further objects and features will more fully be apparent from the following detailed description with accompanying drawings.
Hereinafter, the invention will be described on the basis of the accompanying drawings illustrating embodiments of the invention.
For example, the magnetic tube that is set as a measurement target in this embodiment is a tube material formed from a magnetic substance such as carbon steel, ferritic stainless steel, two-phase stainless steel including a ferrite phase and an austenite phase. Note that, the carbon steel and the like are examples of the magnetic substance, and a member (magnetic substance) that forms the magnetic tube is not limited thereto.
The measurement device 3 is a measurement device that measures a magnetic characteristic value of the magnetic tube, and is a device configured to perform thickness reduction inspection on the magnetic tube by using magnetic flux resistance suggested by an applicant of the invention. The measurement device 3 measures a magnetic characteristic value (magnetic flux density) at respective positions inside the magnetic tube by inserting the inspection probe 4 into the magnetic tube and by moving the inspection probe 4 through the inside of the tube. As the measurement device 3, for example, detect measurement devices disclosed in Japanese Patent No. 6579840, Japanese Patent No. 6514592, and Japanese Patent Laid-Open Publication No. 2019-100850 can be employed.
The inspection probe 4 includes a yoke 41, a magnet 42, and a hall element 43 (magnetic sensor). The yoke 41 is a hollow cylindrical magnetic member, and is a high magnetic permeability metal such as carbon steel. Note that, the shape of the yoke 41 is not limited to the hollow cylindrical shape, and may be, for example, a rod shape, a plate shape, a circular column shape, or the like.
The magnet 42 and the hall element 43 are attached along an outer peripheral surface of the yoke 41 at equal intervals. For example, the magnet 42 and the hall element 43 are provided at eight sites on the outer periphery of the yoke 41. The magnet 42 is disposed so that one magnetic pole faces the yoke 41 and the other magnetic pole faces the magnetic tube, and is polarized in a direction facing the magnetic tube.
As indicated by an arrow in
Specifically, as illustrated in (a) of
As described above, the measurement device 3 measures the magnitude of the magnetic flux density on the basis of the magnitude of the output voltage of the hall element 43. The output voltage of the hall element 43 varies in correspondence with the wall thickness of the magnetic tube, and becomes lower as the thickness reduction occurs.
The magnetic flux resistance has a characteristic in that inspection time per one magnetic tube is shorter in comparison to an internal rotary inspection system (IRIS) using ultrasonic waves, and a plurality of magnetic tubes can be inspected.
In addition, as an inspection method using magnetism as in the magnetic flux resistance, there are a remote field method (RFECT), magnetic flux leakage (MFL), and the like, but in these measurement methods, measurement accuracy is lower in comparison to the magnetic flux resistance.
In this embodiment, when estimating the wall thickness of the magnetic tube from measurement data by the magnetic flux resistance, the wall thickness is estimated by using machine learning model.
Note that, examples of the inspection probe 4 according to the magnetic flux resistance, as shown in
In addition, in this embodiment, description will be given on the assumption that measurement data is measurement data obtained by the magnetic flux resistance, but the measurement data is not limited thereto. For example, as the measurement data, an eddy current value measured by RFECT may be used. In addition, as the measurement data, a leaked magnetic flux (magnetic flux passing through the thickness reduction portion) measured by MFL may be used. That is, the measurement data may be a value representing a magnetic characteristic of the magnetic tube, and is not limited to the measurement data obtained by the magnetic flux resistance.
Returning to
The user terminal 2 is a terminal device that is used by a user of this system, and examples thereof include a personal computer, a tablet terminal, and the like. Hereinafter, the user terminal 2 will be referred to as a terminal 2 for simplification. For example, the user of this system is an inspection company that performs non-destructive inspection of the magnetic tube, but there is no particular limitation. Note that, in
Note that, in this embodiment, it is assumed that the server 1 on a cloud estimates the wall thickness information based on the estimation model 50, but this embodiment is not limited thereto. For example, a local terminal 2 may install data of the estimation model 50 from the server 1 in advance, and may input the measurement data acquired from the measurement device 3 to the estimation model 50 to estimate the wall thickness information. In this manner, a device that generates the estimation model 50 and a device that estimates the wall thickness information based on the estimation model 50 may be different from each other.
The auxiliary storage unit 14 is a non-volatile storage region such as a large-capacity memory, and a hard disk, and stores the program P1 necessary for the control unit 11 to execute processing, and other pieces of data. In addition, the auxiliary storage unit 14 stores the estimation model 50, a user DB 141, a model DB 142, and a measurement DB 143. The estimation model 50 is a machine learning model that has learned predetermined training data, and is a model that estimates wall thickness information relating to the wall thickness of the magnetic tube in a case where measurement data obtained by measuring a magnetic characteristic value of the magnetic tube is input. It is assumed that the estimation model 50 is used as a program module that constitutes a part of artificial intelligence software.
The user DB 141 is a database that stores information of a user of this system. The model DB 142 is a database that stores information of each of a plurality of the estimation models 50 prepared in correspondence with a thickness reduction aspect (a shape pattern of a thickness reduction portion), dimensions of the magnetic tube, and the like. As to be described later, in this embodiment, the plurality of estimation models 50 are prepared in correspondence with the thickness reduction aspect, the dimensions, and the like of the magnetic tube set as an estimation target. The measurement DB 143 is a database that stores measurement data of the magnetic tube which is acquired from the terminal 2.
Note that, the auxiliary storage unit 14 may be an external storage device that is connected to the server 1. In addition, the server 1 may be a multi-computer constituted by a plurality of computers, or may be a virtual machine that is virtually constructed by software.
In addition, in this embodiment, the server 1 is not limited to the above-described configuration, and may include, for example, an input unit that accepts an operation input, a display unit that displays an image, and the like. In addition, the server 1 may be provided with a reading unit that reads a non-temporary computer-readable recording medium 1a and may read out the program P1 from the recording medium 1a. In addition, the program P1 may be executed on a single computer, or may be executed on a plurality of computers connected to each other through the network N.
The display unit 24 is a display screen such as a liquid crystal monitor, and displays an image. The input unit 25 is an operation interface such as a keyboard and a mouse, and accepts an operation input from a user. The auxiliary storage unit 26 is a non-volatile storage region such as a large-capacity memory, and a hard disk, and stores the program P2 necessary for the control unit 21 to execute processing, and other pieces of data.
Note that, the terminal 2 may be provided with a reading unit that reads a non-temporary computer-readable recording medium 2a and may read out the program P2 from the recording medium 2a. In addition, the program P2 may be executed on a single computer, or may be executed on a plurality of computers connected to each other through the network N.
The user name column, the use history column, and the usage fee column respectively store, in association with the user ID, a user name, a use history (an estimation history of wall thickness information) of the estimation model 50 by a user, and a usage fee (system usage fee) of the estimation model 50 which is determined in correspondence with the use history. Details of the usage fee will be described in Embodiment 2.
The model DB 142 includes a model ID column, a model name column, and a target column. The model ID column stores a model ID for identifying each estimation model 50 prepared in correspondence with a thickness reduction aspect and the like of the magnetic tube. The model name column and the target column respectively store a model name of the estimation model 50 and magnetic tube information relating to the magnetic tube that becomes an estimation target. For example, a thickness reduction aspect, dimensions, and a material of the magnetic tube are stored in the target column.
The measurement DB 143 includes a data ID column, a date column, a provider user column, a target column, a measurement data column, and a second measurement data column. The data ID column stores a data ID for identifying measurement data provided from each user. The date column, the provider user column, the target column, the measurement data column, and the second measurement data column respectively store, in association with the data ID, a measurement data providing date (acquisition date), a user name of a providing source user, magnetic tube information (a thickness reduction aspect and the like) relating to the magnetic tube that becomes a measurement target, measurement data of the magnetic tube, and second measurement data. The second measurement data is wall thickness measurement data of the magnetic tube which is different from the measurement data obtained by the magnetic flux resistance in a measurement method, and is measurement data obtained, for example, by internal rotary inspection system (IRIS). Details of the second measurement data will be described in Embodiment 2.
As described above, the server 1, by learning predetermined training data, generates the estimation model 50 that estimates wall thickness information in a case where measurement data of a magnetic characteristic value is input. The estimation model 50 according to this embodiment is a neural network generated by deep learning, and is, for example, a convolutional neural network (CNN, ResNet, or the like).
Note that, the estimation model 50 may be a neural network other than the CNN. In addition, as in Modification Examples 1 and 2 to be described later, the estimation model 50 may be a model such as k-nearest neighbor algorithm (kNN), support vector machine (SVM), random forest, and decision tree based on a learning algorithm other than the neural network.
In the estimation model 50 according to this embodiment, measurement data that is measured by using the measurement device 3 and is based on the magnetic flux resistance is set as an input. As described above, the measurement data is a value obtained by measuring a magnetic flux density passing through a magnetic circuit formed by the yoke 41 and the magnet 42 provided in the inspection probe 4, and the magnetic tube by using the hall element 43, and is an output voltage of the hall element 43 proportional to the magnetic flux density.
When performing measurement on the magnetic tube, a user inserts the inspection probe 4 from an opening of the magnetic tube, and moves (sweeps) the inspection probe 4 to an opening on an opposite side. According to this, a magnetic characteristic value at each position of the magnetic tube along a longitudinal direction is measured. As described above, a plurality of the hall elements 43 are attached to the yoke 41 of the inspection probe 4, and the measurement device 3 measures magnetic characteristic values of respective positions on a cross-section orthogonal to the longitudinal directions at the respective positions in the longitudinal direction. Specifically, the plurality of hall elements 43 are periodically attached onto the outer periphery of the cylindrical yoke 41, and the measurement device 3 measures the magnetic characteristic values at respective positions where the cross-section of the cylindrical magnetic tube is equally divided along a peripheral direction.
Note that, in this embodiment, description is made on the assumption that the magnetic tube is a cylindrical tube, but a cross-sectional shape of the magnetic tube is not limited to a circular shape, and may be, for example, a rectangular shape or the like. That is, the measurement device 3 may be capable of measuring the magnetic characteristic values of the respective positions on the cross-section orthogonal to the longitudinal direction of the magnetic tube by using the plurality of hall elements 43 provided in the inspection probe 4, and the “respective positions on the cross-section” is not limited to the respective positions where the cross-section of the cylindrical magnetic tube is equally divided along the peripheral direction.
In the following description, each of the hall elements 43 functioning as a magnetic sensor is appropriately referred to as “channel”, and channel numbers are allocated to the hall elements 43 and are described as “CH1”, “CH2”, “CH3” . . . and “CH8”.
As described above, when performing measurement on the magnetic tube, as illustrated in the left side of
The estimation model 50 outputs (estimates) wall thickness information of the magnetic tube in a case where the measurement data is input. Specifically, the estimation model 50 estimates the wall thickness (thickness) of the magnetic tube. In this embodiment, a problem handled by the estimation model 50 is defined as a classification problem, and the estimation model 50 classifies the wall thickness of the magnetic tube per a predetermined length (for example, per 0.1 mm) in a predetermined numerical value range (for example, 1.0 to 2.3 mm). Note that, it goes without saying that the estimation model 50 may be a regression model.
In addition, in this embodiment, it is assumed that the estimation model 50 estimates the wall thickness of the magnetic tube, but this embodiment is not limited thereto. For example, instead of the wall thickness of the magnetic tube, the estimation model 50 may estimate a thickness reduction width obtained by subtracting the wall thickness of a thickness reduction portion from the wall thickness of a sound portion. In addition, for example, the estimation model 50 may estimate a thickness reduction rate obtained by dividing the wall thickness of the thickness reduction portion by the wall thickness of the sound portion. In addition, for example, the estimation model 50 may estimate a position of the thickness reduction portion in the magnetic tube, a thickness reduction range (a width of the thickness reduction portion along a longitudinal direction and/or a peripheral direction), a thickness reduction aspect (a shape pattern of thickness reduction), and the like.
In addition, in this embodiment, the thickness reduction of the magnetic tube is set as an estimation target, but this embodiment is not limited thereto, and a wall thickness increase due to rust or the like may be set as the estimation target.
As described above, the wall thickness information output from the estimation model 50 may be information relating to the wall thickness of the magnetic tube. The wall thickness information that is estimated is not limited to the wall thickness of the magnetic tube, and a defect that is set as an estimation target is not limited to thickness reduction.
For example, the estimation model 50 estimates the wall thickness of the magnetic tube for every portion obtained by dividing the magnetic tube per a certain length along the longitudinal direction. A wall thickness estimation result is conceptually illustrated on a left side in
Note that, in the above description, it is assumed that a magnetic characteristic value at each position (each channel) on a cross-section at each position of the magnetic tube in a longitudinal direction is input to the estimation model 50, but this embodiment is not limited thereto. For example, the server 1 may be configured to input only a magnetic characteristic value at a single position (cross-section) of the magnetic tube in the longitudinal direction to the estimation model 50 to estimate the wall thickness information at the position. In addition, for example, the server 1 may be configured to input only a magnetic characteristic value at a single position (channel) on the cross-section of the magnetic tube to the estimation model 50 to estimate the wall thickness information at the position.
With respect to a measurement data group for training, the server 1 generates the estimation model 50 by using training data with which a correct value (label) of the wall thickness information is associated. The measurement data for training is actual measurement data of one or a plurality of magnetic tube obtained by performing measurement with the measurement device 3, and waveform data of each channel which is obtained by measuring a magnetic characteristic value at respective position in the longitudinal direction. Specifically, the measurement data for training is data measured by inserting the inspection probe 4 from both ends of the magnetic tube while changing a relative position of the inspection probe 4 in a peripheral direction so that CH1 becomes approximately 0°, 90°, 180°, and 270° with any radial direction (position on the cross-section) set to 0°. That is, the measurement data for training is data obtained by performing eight (4×2) measurements per one magnetic tube. The correct value is data obtained by measuring the wall thickness of the magnetic tube by using a 3D shape measurement device, and is data obtained by measuring the wall thickness per approximately 23 μm at a pitch of 0.2 mm from a tube end of the magnetic tube. With respect to the measurement data for training, the server 1 generates the estimation model 50 by using training data to which a correct value of the wall thickness of each portion is applied.
(a) of
The server 1 uses a position in the longitudinal direction where the magnetic characteristic value is measured, and a raw measurement value of respective channels (“CH1”. “CH2” . . . , and “CH8” in (a) of
Next, the server 1 corrects measurement data of each channel in order to minimize an influence of lift-off of the inspection probe 4 (and an overall thickness reduction portion) (a lower stage in (a) of
Specifically, the server 1 smooths the measurement data by taking a moving average in the longitudinal direction for every channel, and specifies base lines of respective channels. In addition, the server 1 corrects values of respective channels so that average values of base lines of the respective channels match each other. In other words, the server 1 adjusts a height of a waveform so that waveform shapes of the respective smoothed channels approximately match each other. Note that, a length in the longitudinal direction and the number of times of averaging when taking the moving average are adjusted in correspondence with the length (total length) of the magnetic tube.
Further, the server 1 appropriately corrects the measurement data to data of only a local thickness reduction portion by removing data of a portion where an output voltage changes gently, that is, the overall thickness reduction portion from the measurement data. Note that, the overall thickness reduction portion represents a portion where a change in the output voltage (magnetic characteristic value) of the hall element 43 is equal to or less than a certain value. For example, the server 1 obtains an average value of the base lines of all channels to specify a range in the longitudinal direction where the overall thickness reduction portion exists, and substitutes an output voltage in the range with a value (that is, 0 V) of the sound portion to remove data of the overall thickness reduction portion. According to this, a voltage change of the overall thickness reduction portion is cut out, and thus adjustment to data of only the local thickness reduction portion (portion where a change in the output voltage is equal to or more than a certain value) can be performed.
Returning to
The server 1 uses data obtained by adding predetermined summary statistics to each of the three kinds of data for the estimation model 50. The summary statistics include an average value, a standard deviation, the degree of distortion, and/or kurtosis, but the summary statistics are not limited thereto. The server 1 calculates the average value, the standard deviation, the degree of distortion, and/or the kurtosis along the longitudinal direction and/or the peripheral direction from a magnetic characteristic value at each position in the longitudinal direction and/or the peripheral direction, and adds the calculated values to data.
That is, the server 1 calculates the average value, the standard deviation, the degree of distortion, and the kurtosis at each position (channel) along the peripheral direction of the magnetic tube, and adds the calculated values to data. In addition, the server 1 calculates a moving average, a moving standard deviation, the degree of moving distortion, and moving kurtosis at a plurality of positions (for example, 10 points, 20 points, 30 points, 40 points, and 50 points) along the longitudinal direction of the magnetic tube, and adds the calculated values to data. In a table in
As described above, the server 1 performs certain preprocessing on raw measurement data, and the preprocessed measurement data is used as an input to the estimation model 50. Final preprocessed measurement data become matrix data of N rows ×275 columns (N represents the number of measurement points along the longitudinal direction). Actually, as in a case where the wall thickness information is estimated from the measurement data of the magnetic tube by the estimation model 50, the server 1 performs preprocessing on measurement data acquired from the terminal 2, and the preprocessed data is input to the estimation model 50.
In learning, the server 1 expands the measurement data shown in (a) of
Next, as second data expansion processing, the server 1 generates a plurality of patterns of measurement data by adding a plurality of patterns of predetermined noise to the measurement data. The second data expansion processing is shown in (c) of
As third data expansion processing, the server 1 generates measurement data in which a sampling rate when measuring a magnetic characteristic value is changed. The third data expansion processing is shown in (d) of
Note that, in tables in (b) to (d) of
The server 1 expands the number of data of the measurement data by 480 (8×10×6) times by performing the first expansion processing to the third expansion processing. According to this, it is possible to increase the number of learning data, and it is possible to cause the estimation model 50 to learn a deviation of a channel, a minute positional deviation of the thickness reduction portion, a variation of a movement speed, and the like.
Returning to
Specifically, the server 1 generates an image in which the magnetic characteristic values at the respective positions are expressed as colors (for example, RGB) of the respective pixels. In the image, the vertical axis (first axis) of the image corresponds to a position of the magnetic tube in a longitudinal direction, and the horizontal axis corresponds to a position on a cross-section of the magnetic tube which is orthogonal to the longitudinal direction, that is, a channel. For example, the server 1 divides measurement data (refer to
The server 1 generates a hue image of respective hues of R, G, and B from measurement data for training, and generates a final input image by synthesizing respective hue images. Specifically, the server 1 calculates exponentiation values of a plurality of patterns of magnetic characteristic values which are different in an exponent from the measurement data, and allocates the exponentiation values of the respective patterns to respective hues to generate respective hue images.
For example, with respect to magnetic characteristic values at respective positions in the magnetic tube, the server 1 calculates a value with an exponent of 0, a value with an exponent of 1, and a square value. In addition, the server 1 allocates respective exponentiation values to R, G, and B.
In the following description, the image obtained by synthesizing the respective hue images will be referred to as “synthetic image”.
Note that, the above-described image converting method is illustrative only, and this embodiment is not limited thereto. For example, the server 1 may generate the synthetic image by allocating a value with an exponent of 0 to a value with an exponent of 4 to respective hues by setting a final synthetic image as a CMYK image instead of the RGB image. In addition, hues to which the exponentiation values are allocated are not limited to original colors (RGB). In addition, the exponents of the exponentiation values are not limited to natural numbers, and may be real numbers (for example, an exponent of 1.5) other than the natural numbers. In addition, the server 1 may generate a monochromatic image in which magnetic characteristic values at respective positions in the magnetic tube are stored in respective pixels without calculating the exponentiation values. In addition, the server 1 may express the magnetic characteristic values as other pixel values (for example, gray scales) instead of colors. In this manner, the server 1 may generate images to which pixel values of respective pixels are allocated in correspondence with magnetic characteristic values at respective positions in the magnetic tube, and a generation method thereof is not particularly limited.
In learning, the server 1 generates a synthetic image with respect to all combinations when allocating respective exponentiation values to respective hues in order to prevent over-learning due to hues. That is, the server 1 generates six patterns of synthetic images such as a synthetic image in which the value with an exponent of 0 is allocated to G, the value with an exponent of 1 is allocated to B. and the square value is allocated to R, a synthetic image in which the value with an exponent of 0 is allocated to B, the value with an exponent of 1 is allocated to R, and the square value is allocated to G, . . . in addition to the synthetic image in which the value with an exponent of 0 is allocated to R, the value with an exponent of 1 is allocated to G, and the square value is allocated to B. The server 1 causes the estimation model 50 to learn the synthetic images of respective combinations, thereby preventing over-learning due to hues.
Returning to
That is, the server 1 acquires an estimation value of wall thickness information by inputting a synthetic image for training to the estimation model 50, and compares the estimation value and a correct value. The server 1 updates parameters such as a weight between neurons so that the estimation value and the correct value approximate each other. The server 1 performs sequential learning by using each pair of the synthetic image and the correct value, and finally generates the estimation model 50 in which the weight and the like are optimized.
In this embodiment, the server 1 learns measurement data in correspondence with the magnetic tube that becomes a measurement target in measurement data for training, and generates a plurality of the estimation models 50 corresponding to various magnetic tubes. Specifically, the server 1 learns the measurement data separately in correspondence with magnetic tube information relating to the magnetic tube that is a measurement target, and generates the plurality of estimation models 50 corresponding to the magnetic tube information.
The magnetic tube information is information representing a state or an attribute of the magnetic tube, and examples thereof include a thickness reduction aspect, dimensions, and a material of the magnetic tube. Note that, the pieces of magnetic tube information are illustrative only, and may be other pieces of information. The thickness reduction aspect is a state of the thickness reduction portion of the magnetic tube, and is defined by a shape pattern of the thickness reduction portion (for example, whether a cross-sectional shape of the thickness reduction portion is a rectangular shape, hemispherical shape, or the like). The dimensions are defined by a diameter and a length of the magnetic tube. The material is defined by the kinds of a magnetic substance that forms the magnetic tube.
The server 1 learns the training data separately in correspondence with the pieces of magnetic tube information, and generates the estimation model 50 corresponding to the thickness reduction aspect, the dimensions, and the like of the magnetic tube. Note that, in this embodiment, it is assumed that a plurality of the estimation models 50 are generated, but the wall thickness information may be estimated by a single estimation model 50 by causing the one estimation model 50 to learn the training data regardless of a difference in the magnetic tube information.
Actually, in a case of estimating the wall thickness information by using the estimation model 50, the server 1 acquires measurement data of the magnetic tube which is measured by a user from the terminal 2, and inputs the measurement data to the estimation model 50 to estimate the wall thickness information. Specifically, the server 1 acquires the measurement data, and accepts a designation input of the magnetic tube information relating to the magnetic tube that is a measurement target from a user though the terminal 2. The server 1 selects any one among a plurality of the estimation models 50 in correspondence with designated magnetic tube information. The server 1 generates hue images of respective hues from the acquired measurement data, and generates a synthetic image obtained by synthesizing the hue images. The server 1 inputs the synthetic image to the selected estimation model 50 to estimate the wall thickness of respective portions obtained by partitioning the magnetic tube per a certain length along the longitudinal direction by inputting. The server 1 returns the estimated wall thickness information to the terminal 2 and displays the information on the terminal 2.
Note that, details of an operation (display screen and the like) of the terminal 2 when estimating the wall thickness information will be described in Embodiment 2.
As described above, according to this embodiment, it is possible to appropriately estimate the thickness reduction state (wall thickness information) of the magnetic tube by using the estimation model 50, and it is possible to present a user with the thickness reduction state.
The control unit 11 performs preprocessing with respect to the measurement data for training (step S12). Specifically, the control unit 11 normalizes the magnetic characteristic values so that a magnetic characteristic value at the outside of the magnetic tube, and a magnetic characteristic value at a sound portion match each other in respective channels. In addition, the control unit 11 specifies base lines of respective channels by taking a moving average of the magnetic characteristic values along the longitudinal direction, and by making average values of the base lines of the respective channels match each other, thereby correcting the magnetic characteristic values to magnetic characteristic values in a case where the inspection probe 4 passes through the central axis of the magnetic tube. In addition, the control unit 11 adds summary statistics such as an average value and a standard deviation to the measurement data.
The control unit 11 expands the number of data of the measurement data (step S13). Specifically, the control unit 11 shifts channel numbers of data measured in respective channels to generate a plurality of patterns of measurement data. In addition, the control unit 11 adds a plurality of patterns of predetermined noise (Gaussian noise or the like) to the measurement data to generate a plurality of patterns of measurement data. In addition, the control unit 11 performs linear interpolation between two points continuous in the longitudinal direction to generate a plurality of patterns of measurement data.
The control unit 11 converts the measurement data of which the number of data is expanded into an image in which the vertical axis (first axis) of the image is set to a position along the longitudinal direction, the horizontal axis (second axis) is set to a position on a cross-section, and pixels values of respective pixels are allocated in correspondence with magnetic characteristic values at respective position inside the magnetic tube (step S14). Specifically, the control unit 11 calculates exponentiation values of a plurality of patterns of magnetic characteristic values which are different in an exponent, allocates the exponentiation values to hues different from each other, and generates hue images of respective hues. In addition, the control unit 11 generates a synthetic image obtained by synthesizing the respective hue images. The control unit 11 generates the synthetic image with respect to all combinations when allocating respective exponentiation values to respective hues, and sets the synthetic image as training data.
The control unit 11 generates the estimation model 50 that estimates wall thickness information in a case where the measurement data is input on the basis of the generated synthetic image and a correct value of the wall thickness information (step S15). For example, the control unit 11 generates a neural network such as CNN as the estimation model 50. The control unit 11 inputs the synthetic image generated in step S14 to the estimation model 50 to acquire an estimation value of the wall thickness information, and compares the acquired estimation value and the correct value to optimize parameters such as the weight of the estimation model 50 so that the estimation value and the correct value approximate each other. The control unit 11 terminates a series of processing.
The control unit 11 acquires measurement data of the magnetic tube from the terminal 2 (step S32). The control unit 11 performs preprocessing with respect to the acquired measurement data (step S33). The control unit 11 converts the measurement data after the preprocessing into an image (step S34). Specifically, the control unit 11 allocates exponentiation values of the magnetic characteristic values which are different in an exponent to respective hues, and generates a synthetic image obtained by synthesizing respective hue images.
The control unit 11 inputs the image converted in step S34 to the estimation model 50 to estimate wall thickness information of the magnetic tube (step S35). For example, the control unit 11 estimates the wall thickness of respective portions obtained by partitioning the magnetic tube per a certain length along the longitudinal direction. The control unit 11 transmits the estimated thickness information to the terminal 2 (step S36), and terminates a series of processing.
As described above, according to Embodiment 1, it is possible to appropriately estimate a state of the magnetic tube.
In Embodiment 1, description has been given of an aspect in which the measurement data of the magnetic tube is converted into an image, and the image is input to the estimation model 50. In Modification Example 1, description will be given of an aspect in which the original measurement data is used as is as an input to the estimation model 50. Note that, the same reference numerals will be given to redundant contents as in Embodiment 1, and description thereof will be omitted.
In this modification example, the server 1 constructs a model that can handle original measurement data (a magnetic characteristic value at each position inside the magnetic tube) as is instead of an image as the estimation model 50. For example, the server 1 constructs a model according to k-nearest neighbor algorithm (kNN) as the estimation model 50.
Note that, in this modification example, description will be given on the assumption that the estimation model 50 is a model according to kNN, but the estimation model 50 may be a model that can handle original measurement data, and may be, for example, SVM, random forest, linear classifier, or the like.
The server 1 generates the estimation model 50 on the basis of the same training data as in Embodiment 1. Note that, it is preferable that the server 1 performs main component analysis or the like on the measurement data for training to reduce the number of dimensions. When reducing the number of dimensions of the measurement data that is input to the estimation model 50, it is possible to reduce a processing load of the server 1. Note that, a user may arbitrarily select whether or not to perform reduction of the number of dimensions (main component analysis).
Here, the server 1 may input magnetic characteristic values at all positions along the longitudinal direction to the estimation model 50 as is, but in this embodiment, only measurement data of a part is extracted and is set as an input to the estimation model 50. Specifically, the server 1 extracts data of a peak portion corresponding to thickness reduction portion, and the data is set as an input to the estimation model 50.
A method of extracting data of the peak portion is not particularly limited, and for example, the server 1 specifies a base line (flat waveform) by taking a moving average of an output voltage, and specifies the peak portion in correspondence with the base line. Specifically, the server 1 specifies a portion where a difference from the base line is equal to or more than a predetermined threshold value as the peak portion. Note that, as illustrated in
The server 1 extracts data of the specified peak portion from measurement data of the entirety of the magnetic tube, and sets the data as an input to the estimation model 50. According to this, it is possible to appropriately estimate wall thickness information of the thickness reduction portion while reducing the amount of data to be processed by the estimation model 50.
Note that, extraction of data of the peak portion in this modification example may be applied to Embodiment 1, only the data of the peak portion may be converted into an input image, and the input image may be used as an input to the estimation model 50 according to Embodiment 1.
The server 1 extracts data of the peak portion from measurement data for training, and generates the estimation model 50 on the basis of the extracted data of the peak portion, and a correct value of wall thickness information at a position corresponding to the peak portion. That is, the server 1 maps the data of the peak portion to which a correct value (a correct label representing a correct wall thickness) of the wall thickness information is applied in a feature space. Then, the server 1 sets any value k according to the k-nearest neighbor algorithm to construct a data set for estimating the wall thickness information from the data of the peak portion.
In a case of estimating the wall thickness information, the server 1 extracts data of the peak portion from the measurement data acquired from the terminal 2, and maps the data in a feature space. Then, the server 1 identifies k pieces of data (data of the peak portion to which the correct label is applied) located in the vicinity of a mapped position, and estimates the wall thickness information of the magnetic tube that is an estimation target from a correct value of the wall thickness information which is applied to the k pieces of data.
The control unit 11 generates the estimation model 50 on the basis of the data of the peak portion which is extracted in step S201, and the correct value of the wall thickness information (step S202). Specifically, the control unit 11 generates a model according to the kNN. The control unit 11 maps the data of the peak portion to which a correct value (a correct label) of the wall thickness information is applied in a feature space, and sets any k value to generate a data set for estimating the wall thickness information from the data of the peak portion as the estimation model 50. The control unit 11 terminates a series of processing.
As described above, according to Modification Example 1, it is also possible to estimate the wall thickness information in an original measurement data state without converting the measurement data into an image. Particularly, in this modification example, it is possible to appropriately estimate the wall thickness information of the thickness reduction portion by extracting data of the peak portion which corresponds to the thickness reduction portion.
In this modification example, description will be given of an aspect in which the estimation model 50 is generated by using an ensemble learning method.
In this modification example, the server 1 generates the estimation model 50 by using the ensemble learning method. Specifically, the server 1 generates a model according to a decision tree (for example, LightGBM) by using a gradient boosting method as the estimation model 50.
The gradient boosting is a kind of the ensemble learning, and is a method of generating a final model by sequentially generating a plurality of weak identifiers (models). The server 1 generates one weak identifier (decision tree) by using the training data, and generates a subsequent weak identifier on the basis of a residual between an estimation value obtained by the generated weak identifier and a correct value. The server 1 sequentially generates weak identifiers with reference to a gradient of loss function defined by the residual between the estimation value and the correct value so as to consider a learning result of a previous weak identifier, and generates a final identifier, that is, the estimation model 50.
Note that, in this modification example, the gradient boosting is used as the ensemble learning, but other ensemble learning methods such as bagging in which a plurality of weak identifiers are generated in parallel may be used. In addition, the ensemble learning may be applied to the estimation model 50 according to Modification Example 1, and the ensemble learning may be performed when generating a model such as kNN and SVM.
The server 1 may set the data of the peak portion described in Modification Example 1 (or measurement data of the entirety of the magnetic tube) as an input to the estimation model 50, but in this modification example, the peak portion is further partitioned per a data section having a predetermined length, and data of each data section is input to the estimation model 50.
The server 1 sequentially shifts data sections set as an extraction target along the longitudinal direction, and extracts data of each of the data sections from the peak portion. For example, the server 1 extracts data of 10 points from an initiation portion of the peak portion where a difference from a base line is equal to or more than a threshold value as data of a first data section. The server 1 shifts an initiation point of a data section set as a measurement target by a predetermined number (for example, one), and extracts data of each data section. According to this, the server 1 extracts data of a plurality of the data sections from the data of one peak portion.
When estimating the wall thickness information with respect to one peak portion, the server 1 inputs the extracted data of the plurality of data sections to the estimation model 50 to estimate the wall thickness information. According to this, it is possible to estimate the wall thickness of the thickness reduction portion more appropriately.
The control unit 11 generates the estimation model 50 on the basis of the data of the plurality of data sections which is extracted in step S302, and a correct value of wall thickness information at the peak portion (thickness reduction portion) (step S303). Specifically, the control unit 11 generates a decision tree by using the gradient boosting method. The control unit 11 generates one weak identifier (decision tree) by using training data (data of the plurality of sections and a correct value of the wall thickness information), and generates a subsequent weak identifier on the basis of a residual between an estimation value of wall thickness information obtained by the generated weak identifier and a correct value. The control unit 11 sequentially generates weak identifiers in correspondence with a gradient of loss function defined by the residual between the estimation value and the correct value, and generates a final identifier, that is, the estimation model 50. The control unit 11 terminates a series of processing.
As described above, according to Modification Example 2, it is also possible to generate the estimation model 50 by using the ensemble learning according to the gradient boosting. In addition, the data of the peak portion is divided into data of the plurality of data sections and is input to the estimation model 50, and thus it is possible to estimate the wall thickness information of the peak portion (thickness reduction portion) more appropriately.
Description will be given of an example in which an ensemble learning method different from Modification Example 2. As typical ensemble learning, it is known that the same data set is input to a plurality of models different in a learning mechanism, and an average of outputs from the plurality of models, or a majority output from the plurality of models is set as a learning result.
In Modification Example 3, differently from the learning, data to be input is divided into a plurality of sets, the respective divided sets are respectively input to other corresponding models, outputs of the corresponding models are further input other models, and outputs of the other models are set as a learning result.
For example, the thin wall specialized model 50a is a model for 90% of data with a thickness of 0.1 to 0.8 mm, and 10% of data with a thickness exceeding 0.8 mm among all pieces of data, and the thick wall specialized model 50b is a model for remaining data that is not used by the thin wall specialized model 50a among the all pieces of data. A weighting relationship for both the models 50a and 50b in this example is illustrated in
Note that, a reference thickness for classifying the thin wall specialized model 50a and the thick wall specialized model 50b is set to 0.8 mm, but the reference thickness value is not illustrative only, and other numerical values may be employed. In addition, it is assumed that 90% of data with a thickness equal to or less than a predetermined value, and 10% of data with a thickness exceeding a predetermined value are employed as a target of the thin wall specialized model 50a, but a ratio of 90%:10% is illustrative only, and may be other ratios.
The control unit 11 acquires second training data including measurement data mainly at a thick wall as described above among the plurality of pieces of measurement data of the magnetic tube, and a correct value of wall thickness information associated with the measurement data (step S403). The control unit 11 generates the thick wall specialized model 50b that estimates and outputs second thickness information in a case where the measurement data is input on the basis of the acquired second training data (the measurement data and the correct value of the wall thickness information) (step S404).
The control unit 11 acquires third training data including the first wall thickness information output from the thin wall specialized model 50a, the second wall thickness information output from the thick wall specialized model 50b, and the correct value of the associated wall thickness information (step S405). The control unit 11 generates the final model 50c that estimates the wall thickness information and outputs the wall thickness information in a case where the first wall thickness information and the second wall thickness information are input on the basis of the acquired third training data (the first wall thickness information, the second wall thickness information, and the correct value of the wall thickness information) (step S406).
The control unit 11 acquires measurement data of the magnetic tube from the terminal 2 (step S412). The control unit 11 inputs the acquired measurement data to the thin wall specialized model 50a and the thick wall specialized model 50b (step S413). Next, the control unit 11 inputs an output from the thin wall specialized model 50a and an output from the thick wall specialized model 50b to the final model 50c to estimate the wall thickness information of the magnetic tube (step S414). The control unit 11 transmits the estimated wall thickness information to the terminal 2 (step S36), and terminates a series of processing.
Note that, in Modification Example 3, an average of the output from the thin wall specialized model 50a and the output from the thick wall specialized model 50b may be obtained without using the final model 50c, and the obtained average value may be set as the estimated wall thickness information of the magnetic tube.
According to Modification Example 3, since different appropriate models are applied to the thin wall side and the thick wall side, it is possible to estimate more accurate wall thickness information. Generally, the amount of data may be small on the thin wall side in many cases, but in any case, it is possible to improve estimation accuracy in the wall thickness information on the thin wall side where the amount of data is small.
Modification Example 4 is an example relating to a method of correcting measurement data that is used in learning of the estimation model by using a dynamic time warping method in order to reduce an influence of a hall element voltage due to disturbance of a magnetic flux in the MFR method.
First, the influence of the hall element voltage due to disturbance of the magnetic flux in the MFR method will be described.
To solve the above-described problem, measurement data that is used in learning of the estimation model is corrected by using the dynamic time warping method.
In Modification Example 4, as illustrated in
In Modification Example 4, since the measurement data that is used in learning of the estimation model is corrected by using the dynamic time warping method, it is possible to reduce an influence of the hall element voltage due to distortion of a magnetic flux in the above-described MFR method, and it is possible to estimate the thickness reduction shape of the magnetic tube with high accuracy.
Modification Example 5 is an example of increasing the number of data by using the main component analysis method.
The server 1 may use raw measurement data measured by the inspection probe 4 including eight channels (CH1 to CH8), and first group data obtained by performing preprocessing including arithmetic operation processing on the raw measurement data (including a moving average of a plurality of positions (10 to 50 points) along the longitudinal direction of the magnetic tube, a moving standard deviation, the degree of moving distortion, moving kurtosis, and a moving difference in the example shown in
Note that, in the example shown in
In Modification Example 5, since the number of data is increased by also adding the main component analysis results (in the example shown in
In Embodiment 1, description has been given of the defect estimation system that estimates defects such as the thickness reduction of the magnetic tube by using the estimation model 50. In this embodiment, description will be given of an embodiment when a user performs inspection for defects of the magnetic tube by actually using this system.
The model selection column 171 is an input column for selecting the estimation model 50 that is used for estimation of wall thickness information, and is an input column for accepting a designation input of magnetic tube information relating to the magnetic tube that is an estimation target (measurement target). For example, the model selection column 171 includes a thickness reduction aspect designation column 1711, a dimension designation column 1712, and a material designation column 1713. The terminal 2 accepts designation inputs of a thickness reduction aspect, dimensions, and a material of the magnetic tube through the respective designation columns.
The upload button 172 is a button for transmitting (uploading) measurement data to the server 1. In a case of accepting an operation input to the upload button 172, the terminal 2 transmits the measurement data to the server 1.
The provision availability selection column 173 is an input column for selecting provision (usage) availability of the measurement data to this system. The terminal 2 accepts a selection input according to provision availability of the measurement data in correspondence with an operation input to the provision availability selection column 173. In a case of accepting a selection input indicating provision of the measurement data, the server 1 stores the measurement data acquired from the terminal 2 in the measurement DB 143. As to be described later, the measurement data provided from a user is used for relearning (updating) of the estimation model 50.
Note that, the measurement data acquired from each terminal 2 may be used for relearning of the estimation model 50 regardless of presence or absence of provision availability selection.
The magnetic tube selection column 181 is an input column for selecting a magnetic tube of which wall thickness information is to be displayed. The terminal 2 displays wall thickness information of the magnetic tube selected in the magnetic tube selection column 181.
The wall thickness graph 182 is a graph showing wall thickness, which is estimated on the basis of the estimation model 50, of each portion of the magnetic tube. In the wall thickness graph 182, the horizontal axis represents a position of the magnetic tube in the longitudinal direction, and the vertical axis represents wall thickness. For example, the terminal 2 displays an average value of the wall thickness of each channel in the wall thickness graph 182. Note that, the terminal 2 may display the wall thickness for every channel in the wall thickness graph 182.
The cross-sectional image 184 is an image simulating a cross-section orthogonal to the magnetic tube longitudinal direction, and is an image that reproduces the wall thickness at each position on the cross-section in correspondence with the wall thickness information. As described above, the server 1 estimates the wall thickness at each channel (each position on the cross-section) at each position of the magnetic tube along the longitudinal direction. The server 1 compares the estimated wall thickness (a base line of the wall thickness graph 182) of each channel and wall thickness of a sound portion with each other to detect a thickness reduction portion where the thickness is smaller than that of the sound portion. In addition, the server 1 calculates a thickness reduction width that is a difference in the wall thickness between the sound portion and the thickness reduction portion. For example, the terminal 2 displays a cross-section of the magnetic tube in which CH1 is set to a 0 o'clock direction, and reproduces and displays the thickness reduction portion (concave portion) corresponding to the calculated thickness reduction width at a position corresponding to a channel (CH1 in
For example, the terminal 2 displays a line 183 on the wall thickness graph 182. The terminal 2 accepts a designation input for designating a position of the magnetic tube of which the cross-sectional image 184 is displayed in the longitudinal direction by accepting an operation of moving the line 183 along the horizontal axis. The terminal 2 displays the cross-sectional image 184 of the magnetic tube which corresponds to the position of the line 183.
In addition, the terminal 2 displays the number of the thickness reduction portions, a maximum value of the thickness reduction width, an average value, and the like which are detected on the basis of the wall thickness information.
Note that, the wall thickness display screen illustrated in
In addition, in this embodiment, it is assumed that the terminal 2 displays the wall thickness information on a dedicated screen, but the terminal 2 may only download a file (for example, a CSV file) storing the wall thickness information estimation result, and may not display the wall thickness information.
The download button 185 is a button for downloading the file storing the wall thickness information estimation result. In a case of accepting an operation input to the download button 185, the terminal 2 acquires the file storing wall thickness information estimation result from the server 1.
The second measurement data upload button 186 is a button for uploading second measurement data that is measurement data other than the above-described measurement data (measurement data based on magnetic flux resistance) to the server 1. In a case of accepting an operation input to the second measurement data upload button 186, the terminal 2 transmits measurement data obtained by an internal rotary inspection system (IRIS) to the server 1 as the second measurement data. The second measurement data will be described later.
As described above, the server 1 acquires the measurement data of the magnetic tube from the terminal 2 of each user, and estimates the wall thickness information. Then, the server 1 transmits the wall thickness information to the terminal 2 that is an acquisition source of the measurement data, and displays the wall thickness information on the terminal 2. In this embodiment, the server 1 performs relearning on the basis of the measurement data acquired from the terminal 2, and updates the estimation model 50.
Specifically, in a case of accepting a selection input indicating provision of the measurement data by the provision availability selection column 173, the server 1 stores the measurement data acquired from the terminal 2 in the measurement DB 143. The server 1 uses the measurement data stored in the measurement DB 143 as training data for relearning. The server 1 updates the estimation model 50 on the basis of the measurement data stored in the measurement DB 143, and a correct value of the wall thickness information corresponding to the measurement data.
For example, the correct value of the wall thickness information at the time of the relearning may be manually set through the screen in
The IRIS is a method of inspecting the wall thickness of the tube (magnetic tube) by using ultrasonic waves, and measures the wall thickness by inserting a probe provided with an ultrasonic probe into the tube filled with water, and by detecting a reflected wave of an ultrasonic beam generated from the probe by using the ultrasonic probe. The IRIS is known, and thus detailed description will be omitted in this embodiment. In the IRIS, an inspection speed is slower in comparison to the inspection method such as the magnetic flux resistance according to this embodiment, RFECT, and MFL which use magnetic characteristics, but measurement accuracy is relatively high.
In this embodiment, in a case of being provided with measurement data according to the IRIS from a user, the server 1 uses the data as the correct value of the wall thickness information. That is, the server 1 sets a pair of the measurement data based on the magnetic flux resistance and the second measurement data based on the IRIS as the training data for relearning.
The server 1 compares an estimation value of the wall thickness information based on the measurement data provided from a user, and the wall thickness indicated by the second measurement data, that is, the correct value with each other. Then, the server 1 updates parameters such as a weight between neurons so that both the values approximate each other. The server 1 performs relearning with respect to the measurement data of the magnetic tube which is provided from the user, and updates the estimation model 50.
The server 1 may cause one estimation model 50 to releam the measurement data of all users, but it is preferable that the estimation model 50 separately learns measurement data acquired from the terminal 2 of each user, and updates the estimation model 50 for every user. That is, the server 1 separately applies the measurement data provided from each user to the estimation model 50, and constructs the estimation model 50 for each user. According to this, it is possible to individually tune the estimation model 50 in correspondence with a tendency of the magnetic tube for which measurement (inspection) is performed by each user.
As described above, it is possible to improve the estimation accuracy of the estimation model 50 through the operation of this system.
The server 1 imposes a usage fee (charge) for the estimation model 50 on each user who uses this system. The usage fee may be, for example, a fixed amount (subscription) for each certain period, but in this embodiment, the usage fee is determined in correspondence with the amount of calculation in the processing of estimating the wall thickness information based on the estimation model 50. For example, the amount of calculation that becomes a reference of the usage fee determination may be the number of times of estimation of the wall thickness information (the number of magnetic tubes), a calculation time necessary for the estimation processing, the amount of data of the measurement data, or the like. In a case of estimating the wall thickness information by acquiring the measurement data from the terminal 2, the server 1 determines the usage fee imposed on the user in correspondence with the amount of calculation in the estimation. The server 1 stores the determined usage fee in the user BD 141, and finally charges the user the usage fee.
In this embodiment, the server 1 changes the usage fee in correspondence with presence or absence of provision (usage) of the measurement data. Specifically, in a case of being provided with measurement data (measurement data based on the magnetic flux resistance) that is a source of the wall thickness information estimation, the server 1 decreases the usage fee. In addition, in a case of being provided with the second measurement data (measurement data based on the IRIS) corresponding to the measurement data, the server 1 further decrease the usage fee. According to this, it is possible to efficiently perform collection of the training data for relearning.
The control unit 21 of the terminal 2 accepts a selection input of the estimation model 50 that uses the wall thickness information for estimation (step S501). Specifically, the control unit 21 accepts a designation input of magnetic tube information relating to a magnetic tube such as a thickness reduction aspect, dimensions, a material, and the like of the magnetic tube which are set as an estimation target. In addition, the control unit 21 accepts measurement data usage (provision) availability selection input related to updating of the estimation model 50 (step S502). The control unit 11 transmits measurement data of the magnetic tube to the server 1 in combination with the selection contents in steps S501 and S502 to the server 1 (step S403). In a case of acquiring the measurement data from the terminal 2, the control unit 11 of the server 1 executes processing in steps S33 to S35, and transmits wall thickness information estimated on the basis of the estimation model 50 to the terminal 2 (step S36).
In a case of acquiring the wall thickness information from the server 1, the control unit 21 of the terminal 2 displays the estimated wall thickness information on the display unit 24 (step S504). Specifically, the control unit 21 displays the wall thickness at each position (portion) of the magnetic tube along the longitudinal direction by the wall thickness graph 182, or the like, and displays the wall thickness at each position (channel) on a cross-section at any position in the longitudinal direction by the cross-sectional image 184 or the like.
The control unit 21 transmits the second measurement data obtained by measuring the wall thickness of the magnetic tube by a method other than the magnetic flux resistance to the server 1 in correspondence with an operation input from a user (step S505). For example, the second measurement data is measurement data according to the IRIS, and is data obtained by inserting a probe provided with an ultrasonic probe into the magnetic tube filled with water, and by detecting a reflected wave of an ultrasonic beam generated from the probe by using the ultrasonic probe.
After executing the processing in step S36, the control unit 11 of the server 1 determines a usage fee (charge) imposed on the user (step S506). Specifically, the control unit 11 determines the usage fee in correspondence with the amount of calculation (the number of times calculation, a calculation time, the amount of data of measurement data, and the like) of processing of estimating the wall thickness information based on the estimation model 50. Further, the control unit 11 decreases the usage fee in correspondence with presence or absence of provision of the measurement data and the second measurement data (usage or acquisition).
In a case of accepting a selection input indicating that the measurement data can be provided from a user, the control unit 11 stores the measurement data transmitted in step S403 in the measurement DB 143 (step S507). Further, in a case where the second measurement data is transmitted in step S405, the control unit 11 stores the second measurement data in the measurement DB 143 in association with the measurement data.
The control unit 11 updates the estimation model 50 on the basis of the measurement data and the second measurement data stored in the measurement DB 143 (step S508). That is, the control unit 11 sets the measurement data measured by a user as input data for training, uses a measurement value of the wall thickness in the second measurement data as a correct value of the wall thickness information, and updates parameters such as a weight between neurons. For example, the control unit 11 individually learns the measurement data acquired from the terminal 2 of each user, and updates the estimation model 50 for each user. The control unit 11 terminates a series of processing.
As described above, according to Embodiment 2, this system can be appropriately implemented, and the estimation model 50 can be optimized through the operation of this system.
It is to be noted that, as used herein and in the appended claims, the singular forms “a”, “an”, and “the” include plural referents unless the context clearly dictates otherwise.
The embodiments are illustrative only in all aspects, and it should be considered that the embodiments are not restrictive. The scope of the invention is indicated by the appended claims rather than the above-described meaning, and includes all modifications within the meaning and ranges equivalent to the appended claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2021-042847 | Mar 2021 | JP | national |
| 2021-042848 | Mar 2021 | JP | national |
The present application is the national phase under 35 U. S. C. § 371 of PCT International Application No. PCT/JP2022/009901 which has an International filing date of Mar. 8, 2022 and designated the United States of America, and claiming priority on Patent Application No. 2021-042847 filed in Japan on Mar. 16, 2021 and Patent Application No. 2021-042848 filed in Japan on Mar. 16, 2021.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2022/009901 | 3/8/2022 | WO |
