Two-dimensional visual residual magnetism detection damage evaluation method based on neural network

Abstract

The invention provides a two-dimensional visual residual magnetism detection damage evaluation method based on a neural network. On the basis of traditional one-dimensional residual magnetism detection, original magnetic field data in x, y and z directions in a space magnetic field are collected and further converted into dx, dy and dz component data, magnetic field intensity is synthesized, magnetic field data under seven different characteristic values are converted into 0-255 two-dimensional residual magnetism grey-scale maps through MATLAB, digital speckle images of corresponding areas are collected at the same time, and the residual magnetism detection accuracy is improved. The method comprises the following steps: performing mapping segmentation on two-dimensional residual magnetism grey-scale maps under different characteristics to form sample pairs, inputting the sample pairs into a convolutional neural network for training, outputting strain values under different two-dimensional residual magnetism grey-scale maps, being suitable for two-dimensional damage assessment of the ferromagnetic material, and being capable of realizing overall damage information of the surface of the ferromagnetic material and improving the accuracy of the two-dimensional damage assessment of the ferromagnetic material. The problems of missing detection and false detection in the one-dimensional residual magnetism detection process are effectively reduced, and implementation and popularization are facilitated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ferromagnetic material damage detection, and particularly relates to a two-dimensional visual residual magnetization detection damage evaluation method based on a neural network. BACKGROUND

[0002] With the continuous development of global manufacturing industry, ferromagnetic materials are widely used in aerospace, rail transportation, oil and gas pipeline engineering fields. Due to the effect of external load on the ferromagnetic material during service, stress concentration phenomenon occurs in the material. Therefore, it is of great significance to determine the high stress concentration area by using non-destructive testing method in engineering field.

[0003] Traditional ferromagnetic material damage detection methods mainly include ultrasonic detection, magnetic powder detection, ray detection and machine vision. Ultrasonic detection is not sensitive to small cracks and needs complex coupling agent; magnetic powder detection can only detect surface and near-surface defects and cannot detect internal damage; ray detection has radiation hazards and the equipment is expensive and complex to operate; machine vision can realize automatic damage evaluation when detecting material surface defects, but it cannot accurately judge the stress damage and strain inside the material, and machine vision is easily affected by environmental light. In addition, the existing common non-destructive testing methods have large technical errors, high cost, complex device and low efficiency, which reduces the economic benefits of manufacturing industry.

[0004] At present, the damage evaluation of ferromagnetic materials is mostly limited to the data of one detection line. However, one-dimensional residual magnetization detection mainly relies on the wave peak and valley characteristics to judge the defects, but due to edge effect and some complex characteristics, the wave peak and valley may appear, causing false judgment. At the same time, the residual magnetization signal on the one-dimensional detection line will lose a lot of damage information, resulting in poor measurement result accuracy. Therefore, only by detecting the residual magnetization signal on several detection lines cannot fully cover the actual surface of the parts, and the small detection range may cause missed detection. The existing non-destructive testing method cannot accurately evaluate the damage of ferromagnetic materials. SUMMARY

[0005] 1. The technical problem to be solved is:

[0006] How to accurately judge the stress damage and strain inside the material.

[0007] 2. Technical scheme:

[0008] In order to solve the above problems, the present application provides a two-dimensional visual residual magnetization detection damage evaluation method based on a neural network, which comprises the following steps:

[0009] S1: Obtain a residual magnetism two-dimensional gray scale image; collect residual magnetism signals on the surface of a ferromagnetic material; and convert the collected three-axis magnetic signal data into a two-dimensional residual magnetism gray scale image under different characteristic values by using a MATLAB program.

[0010] S2: Obtain a digital speckle image; and collect a digital speckle image on the surface of a ferromagnetic material in real time by using a camera.

[0011] S3: Image interpolation processing; select the origin of the two-dimensional residual magnetism gray scale image and the digital speckle image as a starting point to make the images coincide; and perform image interpolation processing on the image with fewer pixel points.

[0012] S4: Image segmentation; locate the maximum strain value and the maximum gray scale value of the residual magnetism two-dimensional gray scale image and the digital speckle image in the same region; perform image segmentation from the maximum value region of the residual magnetism two-dimensional gray scale image and the digital speckle image to the periphery of the framed region in sequence, and output a two-dimensional residual magnetism gray scale image block and a digital speckle image block.

[0013] S5: Construct a sample pair ; and take the average value of the gray scale values of the two-dimensional residual magnetism gray scale image block of the x, y, z, dx, dy, dz or comprehensive magnetic field intensity as ; and extract the average strain value of the digital speckle image block .

[0014] S6: Strain damage prediction; input the sample pair into a CNN convolutional neural network for training, take the two-dimensional gray scale image under the x, y, z, dx, dy, dz and comprehensive magnetic field intensity as a multi-channel input, output a strain distribution visualization image, and realize the prediction of residual magnetism signals to strain values.

[0015] Further, in S1, the method for obtaining a residual magnetism two-dimensional gray scale image is as follows: collect residual magnetism signals on the surface of a ferromagnetic material, and convert the collected three-axis magnetic signal data into a two-dimensional residual magnetism gray scale image under different characteristic values.

[0016] The specific process of collecting residual magnetism signals on the surface of a ferromagnetic material is as follows: divide the detection region by a fixed step, set a stable delay after the execution of each motion instruction, and the motion interval time must be met. , wherein is the stable delay, The single data reading time of the sensor is triggered after the mechanical vibration is fully attenuated. When the sensor reaches the target position (Xi, Yj), the motion controller sends a trigger pulse to the magnetic sensor, and simultaneously starts the data collection of the three-axis magnetic sensor (Bx, By, Bz). The sensor transmits the digital signal to the industrial computer through the SPI interface, and the position coordinate label and timestamp are embedded in the data packet. The data collected by the sensor is transmitted to the single-chip microcomputer, and the parity bit remains unchanged when the number of "1" in the 8-bit register is satisfied. The compiled ASCII code only contains plus and minus signs and numbers. A set of magnetic signal data of X, Y, Z three axes must not be all 0, otherwise the sensor is controlled to measure the magnetic field signal at the position again until the collected signal is correct. The sensor is moved to the next position point by the detection platform to repeat the above sampling steps.

[0017] The specific method of converting the collected three-axis magnetic signal data into a two-dimensional residual magnetism gray scale graph under different characteristic values is as follows: the two-dimensional residual magnetism gray scale graph under different characteristic values is directly extracted from the x, y, z three-direction magnetic signals collected by the sensor as the original signal; the diff function is used to calculate the difference in the row direction and the difference in the column direction of the magnetic signal in each direction respectively, and the differences in the two directions are combined to obtain the gradient two-dimensional residual magnetism gray scale graph of dx, dy, dz in different directions respectively; the original signal is squared and summed, and then the square root is taken, that is, , to obtain the two-dimensional residual magnetism gray scale graph.

[0018] Further, in S1, the residual magnetism signal on the surface of the ferromagnetic material is collected using a two-dimensional residual magnetism scanning platform, and the data collected by the two-dimensional residual magnetism scanning platform is converted into a hue value of a pixel point 0-255 and a two-dimensional image is generated; a MATLAB program is written to convert the collected three-axis magnetic signal data into a two-dimensional residual magnetism gray scale graph under different characteristic values. During the MATLAB data reading process, the mean interpolation method based on four neighborhoods is used to fill in the missing values.

[0019] Further, in S2, a CCD camera is used to collect digital speckle images on the surface of the ferromagnetic material in real time. The scanning area of the digital speckle image is the same as the residual magnetism scanning area, and the actual pixel size of the DIC image matches the residual magnetism scanning step.

[0020] Further, in S2, before collecting the digital speckle images on the surface of the ferromagnetic material in real time, the surface of the ferromagnetic material with stress concentration is sequentially polished, cleaned, dried, and painted, and a digital speckle patch is attached.

[0021] Further, in S3, the image interpolation processing adopts Lanczos interpolation, for each pixel (u, v) in the target image, its corresponding position (x, y) in the original image is found by inverse mapping, and then the weighted sum of pixels within the surrounding a x a range is calculated using the Lanczos kernel function as the pixel value of (u, v).

[0022] Further, in S4, the image segmentation adopts a bimodal threshold segmentation method, based on the two-dimensional remanence grayscale image, the signal intensity threshold is set to 80% of the maximum remanence value of the image, the high remanence anomaly area is extracted by global threshold segmentation, and the strain threshold of the DIC digital speckle image is set to 80% of the maximum strain value, the high strain area is segmented by Otsu dynamic threshold method; the pixel-level coordinate mapping relationship is established, the two types of threshold segmentation results are spatially superimposed, the Jaccard similarity algorithm is used to calculate the overlap area ratio of the two modal segmentation areas, and when the overlap rate is greater than 75%, the damage area is located; the damage area is finely segmented, and the area outside the damage area is roughly segmented.

[0023] Further, in the process of image segmentation, the upper left corner of the same area is selected as the origin, a two-dimensional rectangular coordinate system is established, the horizontal direction is defined as the x-axis and the vertical direction is defined as the y-axis, and the two images are spatially located based on the pixel coordinates; in the unified coordinate system, the common area of the remanence two-dimensional grayscale image and the digital speckle image is grid divided in the following manner:

[0024] For the located damage area, the horizontal block number is , the vertical block number is , and the block size is , wherein , , , are the horizontal and vertical coordinate ranges of the damage area, , are positive integers and is greater than the global horizontal block number M, is greater than the global vertical block number N.

[0025] For other areas outside the damage area, the horizontal block number is , the vertical block number is , and the block size is , wherein , are positive integers and is less than the global horizontal block number M, is less than the global vertical block number N.

[0026] Further, the two-dimensional remanence gray scale blocks and the digital speckle image blocks are given two-dimensional indexes to realize spatial mapping; for the sub-blocks in the damage area, the coordinate range is: , and the two-dimensional index (i, j) is given, wherein i [1, ], j [1, ]; for the sub-blocks outside the damage area, numbering starts from the coordinate range 1)], and the two-dimensional index is given according to the global block logic to realize one-to-one correspondence of the two image sub-blocks in the same physical area.

[0027] Further, in S6, the convolutional neural network model is used to statistically obtain the features of the two-dimensional gray scale blocks of remanence; a full-connection neural network is used: to perform module training to output the predicted value of the strain of the ferromagnetic material in different characteristic two-dimensional gray scale blocks.

[0028] 3. Beneficial effects:

[0029] The two-dimensional visual remanence damage evaluation method based on a neural network disclosed in the application adopts a two-dimensional remanence detection platform, uses a three-axis magnetic sensor to collect remanence signals in different directions for a damage area, converts the collected original remanence signals in x, y and z directions and the remanence signals under different characteristic values into two-dimensional remanence gray scale blocks with a value range of 0-255 by using MATLAB. The damage information of the whole surface of the ferromagnetic material can be obtained, and the problems of missed detection and false detection in the one-dimensional remanence detection process can be effectively reduced. A CCD camera is used to collect digital speckle images of the surface of the ferromagnetic material in real time to observe strain information. The two-dimensional remanence gray scale blocks under different characteristic values and the digital speckle images are subjected to interpolation processing and one-to-one mapping, the maximum strain value and the maximum gray scale value of the two-dimensional remanence gray scale blocks and the digital speckle images in the same area are quickly located, and the images are segmented from the maximum value area to the periphery of the framed area in turn; the average value of the gray scale values of the two-dimensional remanence gray scale blocks of x, y, z, dx, dy, dz or the comprehensive magnetic field intensity is taken as ; the average strain value of the digital speckle image blocks is extracted ; a sample pair is constructed , the sample pair is input into a convolutional neural network (CNN) for training, the two-dimensional remanence gray scale blocks under different characteristic values in x, y and z directions are taken as multi-channel inputs, and a strain distribution visualization map is output to realize prediction of the remanence signal to the strain value; the method only needs to collect a plurality of remanence signal detection lines on the basis of the traditional one-dimensional remanence detection, convert the remanence signal detection lines into two-dimensional remanence gray scale blocks with a value range of 0-255 by using MATLAB, collect the corresponding digital speckle images, map and segment the two-dimensional remanence gray scale blocks under different characteristics to form a sample pair, and The input convolutional neural network (CNN) is trained, and the output is the strain value under different two-dimensional residual magnetism gray scale maps, which is suitable for two-dimensional damage evaluation of ferromagnetic materials and is convenient for implementation and promotion. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0031] The present application will be described in detail below in conjunction with the drawings and examples.

[0032] As Figure 1 shown, a two-dimensional visual residual magnetism detection damage evaluation method based on a neural network comprises the following steps:

[0033] S1: obtaining a residual magnetism two-dimensional gray scale map; collecting residual magnetism signals on the surface of a ferromagnetic material; and converting the collected three-axis magnetic signal data into a two-dimensional residual magnetism gray scale map under different characteristic values.

[0034] In one embodiment, the specific process of collecting residual magnetism signals on the surface of a ferromagnetic material is as follows: dividing the detection area by a fixed step size, setting a stable delay after each motion instruction is executed , and meeting the condition of motion interval time : , wherein, is the stable delay, is the single data reading time of the sensor, and the collection is triggered after the mechanical vibration is fully attenuated; when the sensor reaches the target position (Xi, Yj), the motion controller sends a trigger pulse to the magnetic sensor, and the data collection of the three-axis magnetic sensor (Bx, By, Bz) is started synchronously; the sensor transmits the digital signal to the industrial computer through the SPI interface, and the position coordinate label and the time stamp are embedded in the data packet; the data collected by the sensor is transmitted to the single-chip microcomputer, and the parity bit of the number of "1" in the 8-bit register remains unchanged; the compiled ASCII code only contains plus and minus signs and numbers; a group of magnetic signal data of the x, y, and z axes must not all be 0, otherwise the sensor is controlled to measure the magnetic field signal at the position again until the collected signal is correct; the sensor is moved to the next position point by the detection platform, and the above sampling steps are repeated.

[0035] In one embodiment, the residual magnetism signals on the surface of a ferromagnetic material are collected using a two-dimensional residual magnetism scanning platform, the data collected by the two-dimensional residual magnetism platform is converted into hue values of pixel points 0-255 and a two-dimensional image is generated; during the MATLAB data reading process, the missing values are filled by using a mean interpolation method based on four neighborhoods.

[0036] In one embodiment, the two-dimensional residual magnetism gray scale map under different characteristic values is to directly extract the two-dimensional residual magnetism gray scale map in different directions by taking the magnetic signals in three directions collected by the sensor as the original signal; the diff function is used to calculate the difference in the row direction and the difference in the column direction of the magnetic signal in each direction respectively, the differences in the two directions are combined to obtain the gradient two-dimensional residual magnetism gray scale map of dx, dy and dz in different directions respectively; the original signal is squared and summed, and then the square root is taken, that is , to obtain the two-dimensional residual magnetism gray scale map.

[0037] S2: Obtain a digital speckle image; a CCD camera is used to collect a digital speckle image on the surface of a ferromagnetic material in real time.

[0038] In one embodiment, in the process of collecting the digital speckle image on the surface of the ferromagnetic material in real time by using the CCD camera, the scanning area of the digital speckle image is the same as the residual magnetism scanning area, and the actual pixel size of the DIC image matches the residual magnetism scanning step.

[0039] In one embodiment, before collecting the digital speckle image on the surface of the ferromagnetic material in real time, the surface of the stress-concentrated ferromagnetic material is sequentially polished, cleaned, dried, and lacquered, and a digital speckle patch is attached.

[0040] S3: Image interpolation processing; the origin of the two-dimensional residual magnetism gray scale map and the digital speckle image is selected as the starting point to make the pictures coincide; image interpolation processing is performed on the image with fewer pixels.

[0041] In one embodiment, the image interpolation processing uses the Lanczos interpolation, for each pixel (u, v) in the target image, the corresponding position (x, y) in the original image is found by inverse mapping, and then the weighted sum of the pixels within the a x a range around is calculated by using the Lanczos kernel function as the pixel value of (u, v).

[0042] S4: Image segmentation; the maximum strain value and the maximum gray value of the residual magnetism two-dimensional gray scale map and the digital speckle image in the same area are located, and the image segmentation is sequentially performed from the maximum value area of the residual magnetism two-dimensional gray scale map and the digital speckle image to the periphery of the framed area, to output the two-dimensional residual magnetism gray scale map block and the digital speckle image block.

[0043] In one embodiment, the image segmentation adopts a bimodal threshold segmentation method, based on a two-dimensional remanence grayscale image, sets a signal intensity threshold value to 80% of the maximum remanence value of the image, extracts a high remanence anomaly area through global threshold segmentation, synchronously sets a strain threshold value to 80% of the maximum strain value of a DIC digital speckle image, and adopts an Otsu dynamic threshold method to segment a high strain area; a pixel-level coordinate mapping relationship is established, the two types of threshold segmentation results are spatially superimposed, a Jaccard similarity algorithm is used to calculate the overlapping area proportion of the two modal segmentation areas, and when the overlapping rate is greater than 75%, a damage area is located; the damage area is finely segmented, and the area outside the damage area is roughly segmented.

[0044] In one embodiment, in the process of image segmentation, the upper left corner of the same area is selected as the origin, a two-dimensional rectangular coordinate system is established, the horizontal direction is defined as the x-axis and the vertical direction is defined as the y-axis, and the two images are spatially positioned based on pixel coordinates; in the unified coordinate system, the common area of the remanence two-dimensional grayscale image and the digital speckle image is grid divided in the following manner:

[0045] For the located damage area, the horizontal block number is , the vertical block number is , and the block size is , wherein , , , are the horizontal and vertical coordinate ranges of the damage area, , are positive integers and is greater than the global horizontal block number M, is greater than the global vertical block number N;

[0046] For other areas outside the damage area, the horizontal block number is , the vertical block number is , and the block size is , wherein , are positive integers and is less than the global horizontal block number M, is less than the global vertical block number N.

[0047] In one embodiment, the two-dimensional remanence grayscale image block and the digital speckle image block are given a two-dimensional index to realize spatial mapping; for the sub-blocks in the damage area, the coordinate range is: , and a two-dimensional index (i, j) is given, wherein i ∈ [1, ], j ∈ [1, ]; for the sub-blocks outside the damage area, the coordinate range is 1) Start numbering, give two-dimensional index to global block logic, realize one-to-one correspondence of two image sub-blocks in the same physical area.

[0048] S5: Construct sample pairs ; the average value of the two-dimensional remanent gray scale block gray value of x, y, z, dx, dy, dz or the comprehensive magnetic field intensity . ; extract the strain average value of the digital speckle image block ,

[0049] S6: Strain damage prediction, input the sample pair into the CNN convolutional neural network for training, input the two-dimensional gray scale under x, y, z, dx, dy, dz and the comprehensive magnetic field intensity as multi-channel input, output the strain distribution visualization map, realize the prediction of the remanent magnetic signal to the strain value.

[0050] In one embodiment, by inputting the two-dimensional remanent gray scale under x, y, z, dx, dy, dz and characteristics as multi-channel input, the remanent two-dimensional gray scale block statistical characteristics are obtained; a fully connected neural network is used, the module is trained to output the predicted value of the strain of the ferromagnetic material in the different characteristic two-dimensional gray scale.

[0051] The application collects the original magnetic field data in x, y, z directions in the space magnetic field, and further converts it into dx, dy, dz component data, and the comprehensive magnetic field intensity , the magnetic field data under 7 different characteristic values are converted into two-dimensional remanent gray scale of 0-255 by MATLAB, and the digital speckle image of the corresponding area is collected, and the two-dimensional remanent gray scale under different characteristics is mapped and segmented to form a sample pair, the sample pair is input into the convolutional neural network (CNN) for training, and the strain value under different two-dimensional remanent gray scale is output, which is suitable for accurate evaluation of two-dimensional damage of ferromagnetic materials, and is convenient for implementation and promotion.

[0052] Although the present application has been disclosed as above with reference to the preferred embodiments, they are not intended to limit the present application, and any person skilled in the art can make various changes or modifications without departing from the spirit and scope of the present application, therefore the protection scope of the present application should be defined by the protection scope of the claims of the present application.

Claims

1. A two-dimensional visualization-based method for damage evaluation using remanent magnetization based on neural networks, characterized in that: Includes the following steps: S1: Obtain a two-dimensional grayscale image of remanence; collect remanence signals from the surface of ferromagnetic materials; write a MATLAB program to convert the collected triaxial magnetic signal data into two-dimensional grayscale images of remanence under different eigenvalues; S2: Acquire digital speckle images; Use a camera to acquire digital speckle images of the ferromagnetic material surface in real time; S3: Image interpolation processing; select the origin of the two-dimensional remanent grayscale image and the digital speckle image as the starting point to make the images coincide; perform image interpolation processing on images with fewer pixels; S4: Image segmentation; Locate the maximum strain value and maximum gray value of the remanent magnetization two-dimensional grayscale image and digital speckle image in the same area, and perform image segmentation from the maximum and minimum value areas of the remanent magnetization two-dimensional grayscale image and digital speckle image to the periphery of the selected area, and output two-dimensional remanent magnetization grayscale image block and digital speckle image block. S5: Constructing Sample Pairs ; respectively x, y, z, dx, dy, dz or comprehensive magnetic field strength The average gray value of the two-dimensional remanent grayscale patch is used as Extract the strain average value of digital speckle image patches. ; S6: Strain damage prediction; The samples are input to train a CNN convolutional neural network, which takes x, y, z, dx, dy, dz, and the overall magnetic field strength as inputs. The two-dimensional grayscale image below is used as a multi-channel input, and the output is a visualization of the strain distribution, realizing the prediction of strain value from the remanent magnetization signal.

2. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 1, characterized in that: In S1, the method for obtaining the remanent magnetization two-dimensional grayscale image is as follows: the remanent magnetization signal on the surface of the ferromagnetic material is collected, and the collected triaxial magnetic signal data is converted into a two-dimensional remanent magnetization grayscale image under different feature values. The specific process for acquiring the remanent magnetization signal on the surface of the ferromagnetic material is as follows: the detection area is divided with a fixed step size, and a stable delay is set after each motion command is executed. And the exercise interval time must be met. Conditions: ,in, To stabilize latency, The sensor's single data reading time is defined as follows: data acquisition is triggered after the mechanical vibration has sufficiently decayed. When the sensor reaches the target position (Xi, Yj), the motion controller sends a trigger pulse to the magnetic sensor, simultaneously initiating data acquisition from the triaxial magnetic sensors (Bx, By, Bz). The sensor transmits digital signals to the industrial computer via the SPI interface, with the data packet embedding position coordinate tags and a timestamp. The data transmitted from the sensor to the microcontroller ensures that the parity bit in the 8-bit register remains unchanged, and the compiled ASCII code contains only plus and minus signs and numbers. A set of magnetic signal data for the X, Y, and Z axes must not be all zeros; otherwise, the sensor is controlled to measure the magnetic field signal at that position again until the acquired signal is correct. The detection platform then moves the sensor to the next position and repeats the above sampling steps. The specific method for converting the acquired triaxial magnetic signal data into two-dimensional remanent grayscale images under different eigenvalues ​​is as follows: The two-dimensional remanent grayscale images under different eigenvalues ​​are obtained by directly extracting the two-dimensional remanent grayscale images in different directions from the magnetic signals acquired by the sensor in the x, y, and z directions as the original signals; the diff function is used to calculate the difference in the row direction and the difference in the column direction of the magnetic signal in each direction for the magnetic signal values ​​in the x, y, and z directions respectively; the differences in the two directions are combined to obtain the gradient two-dimensional remanent grayscale images of dx, dy, and dz in different directions respectively; the square roots of the original signals are then calculated by taking the square root of the sum of the squares. Two-dimensional remanent grayscale images were obtained.

3. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 2, characterized in that: In S1, a two-dimensional remanent magnetization scanning platform is used to collect remanent magnetization signals on the surface of ferromagnetic materials. The data collected by the two-dimensional remanent magnetization scanning platform is converted into hue values ​​of pixels from 0 to 255 and a two-dimensional image is generated. A MATLAB program is written to convert the collected triaxial magnetic signal data into two-dimensional remanent magnetization grayscale images under different feature values. During the MATLAB data reading process, the missing values ​​are filled by the mean interpolation method based on the four neighborhood.

4. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 3, characterized in that: In S2, a CCD camera is used to acquire digital speckle images of the ferromagnetic material surface in real time. The scanning area of ​​the digital speckle image is the same as the remanent magnetization scanning area, and the actual pixel size of the DIC image matches the remanent magnetization scanning step size.

5. The two-dimensional visualization remanent magnetization detection damage evaluation method based on neural network as described in claim 4, characterized in that: in step S2, before acquiring the digital speckle image of the ferromagnetic material surface in real time, the surface of the ferromagnetic material with stress concentration is sequentially polished, cleaned, dried, and painted, and a digital speckle patch is attached.

6. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 4, characterized in that: In S3, the image interpolation process uses upper Lanczos interpolation. For each pixel (u,v) in the target image, its corresponding position (x,y) in the original image is found through reverse mapping. Then, the weighted sum of the pixels in the surrounding a×a range is calculated using the Lanczos kernel function as the pixel value of (u,v).

7. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 6, characterized in that: In S4, the image segmentation adopts a dual-modal thresholding method. Based on the two-dimensional remanent magnetization grayscale image, the signal intensity threshold is set to 80% of the maximum remanent magnetization value of the image. High remanent magnetization anomaly regions are extracted through global thresholding. Simultaneously, the strain threshold of the DIC digital speckle image is set to 80% of the maximum strain value, and the Otsu dynamic thresholding method is used to segment high strain regions. A pixel-level coordinate mapping relationship is established, and the two types of thresholding results are spatially superimposed. The Jaccard similarity algorithm is used to calculate the overlap area ratio of the two-modal segmentation regions. When the overlap rate is greater than 75%, the damage region is located. The damaged area is finely segmented, while the area outside the damaged area is roughly segmented.

8. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 7, characterized in that: In the image segmentation process, the upper left corner of the same region is selected as the origin to establish a two-dimensional rectangular coordinate system. The horizontal direction is defined as the x-axis and the vertical direction as the y-axis. The two images are spatially located based on pixel coordinates. Under the unified coordinate system, the common region of the remanent magnetization two-dimensional grayscale image and the digital speckle image is divided into grids in the following manner: The located damage area is divided into a mesh, with the number of horizontal blocks being [number missing]. The number of vertical blocks is Block size ,in , , , The x and y coordinates represent the range of the damaged area. , All are positive integers and Greater than the global horizontal block size M, Greater than the global vertical block count N; For areas outside the damaged region, a mesh is created, with the number of horizontal blocks being [number missing]. The number of vertical blocks is Block size ,in , All are positive integers and Less than the global horizontal block size M, It is less than the number of global vertical blocks N.

9. The two-dimensional visualization method for damage evaluation based on neural networks according to claim 8, characterized in that: The two-dimensional remanent magnetization grayscale patch and digital speckle image patch are assigned a two-dimensional index to achieve spatial mapping; for sub-blocks within the damaged area, the coordinate range is: Assign a two-dimensional index (i,j), where i∈[1,j]. ],j∈[1, For sub-blocks outside the damaged area, from the coordinate range 1)] Start numbering, assign a two-dimensional index using global block logic to achieve a one-to-one correspondence between two image sub-blocks in the same physical region.

10. The two-dimensional visualization remanence detection damage evaluation method based on neural networks according to claim 9, characterized in that: In S6, the convolutional neural network model uses statistical features of remanent magnetization two-dimensional grayscale patches; it employs a fully connected neural network structure. Module training is performed to output predicted values ​​of strain for ferromagnetic materials in different feature 2D grayscale images.

Images