A fully automatic magnetic particle detection device and a defect identification method for defect identification

Abstract

The application discloses a kind of full-automatic magnetic particle detection device and defect identification method for defect identification, it is related to nondestructive testing technical field.The present application comprises: mobile platform, the bottom of the mobile platform is equipped with multiple groups of driving assembly, for moving on the surface of the workpiece to be measured;Multiple magnetization components, the magnetization component is arranged at the bottom of mobile platform, for generating rotating magnetic field to the workpiece to be measured magnetization;Magnetic suspension liquid spraying assembly, the magnetic suspension liquid spraying assembly is installed in the bottom of the mobile platform, and magnetic suspension liquid spraying assembly is located between adjacent magnetization component, for spraying magnetic suspension liquid to the surface of the workpiece to be measured.The present application realizes the physical isolation of detection personnel and dangerous scene, eliminates the subjective error of artificial detection, and is suitable for high altitude, closed, toxic, radiation and other complex working conditions of ferromagnetic material surface and near-surface defect detection.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a fully automated magnetic particle testing device and method for defect identification. Background Technology

[0002] Magnetic particle testing is a widely used method in the field of industrial nondestructive testing, mainly used to detect defects (such as cracks, folds, and inclusions) on and near the surface of ferromagnetic materials. Its basic principle is to use the leakage magnetic field generated at the defect site to attract magnetic particles, forming a magnetic trace display, thereby determining the location and shape of the defect.

[0003] Traditional magnetic particle detectors typically include a magnetizing power supply, a magnetizing probe (or magnetic yoke), and a magnetic suspension spraying device. In practical applications, they mainly suffer from the following shortcomings: Most existing testing instruments are handheld or close-range operated, requiring inspectors to be close to the workpiece being tested. This is extremely inconvenient when inspecting workpieces located at heights, in enclosed containers, in toxic environments, or in radiation zones. Existing testing instruments have a relatively limited operating mode. They either require manual testing or can only perform automated testing without being able to switch between modes. In addition, they require dedicated fluorescent lamps for both conventional magnetic particle testing and fluorescent magnetic particle testing, which cannot meet different testing needs and has poor applicability.

[0004] Therefore, this invention proposes a fully automated magnetic particle inspection device and a defect identification method for defect identification. Summary of the Invention

[0005] The purpose of this invention is to provide a fully automated magnetic particle inspection device and method for defect identification, so as to solve the problems mentioned in the background art.

[0006] According to a first aspect of the present invention, in order to achieve the above-mentioned objective, the present invention provides the following technical solution: a fully automated magnetic particle inspection device for defect identification, comprising: A mobile platform, wherein multiple sets of drive components are installed at the bottom of the mobile platform for moving on the surface of the workpiece being measured; Multiple sets of magnetization components are disposed at the bottom of the moving platform to generate a rotating magnetic field to magnetize the workpiece under test. A magnetic suspension spraying assembly is installed at the bottom of the mobile platform and located between adjacent magnetization assemblies, for spraying magnetic suspension liquid onto the surface of the workpiece being tested. The high-definition camera assembly includes a front-facing camera and a high-definition camera. The front-facing camera is located on the front side of the mobile platform and is used to capture images of the environment in front of the detector. The high-definition camera is located on the bottom of the mobile platform and is used to capture magnetic trace images of the detection area.

[0007] Furthermore, the drive assembly includes a fixed bracket fixedly installed at the bottom of the mobile platform. A magnetic pulley is rotatably installed on one side of the fixed bracket, and a reduction gearbox is detachably installed on the other side. The output shaft of the reduction gearbox is fixedly connected to the magnetic pulley, and a drive motor is fixedly installed on the input shaft of the reduction gearbox, and the drive motor is fixed on the outer wall of the reduction gearbox.

[0008] Furthermore, a photoelectric encoder is installed inside the reduction gearbox to record the displacement data of the magnetic chuck in real time.

[0009] Furthermore, the magnetization component is configured as a magnetic yoke, and an auxiliary wheel is installed at the bottom of the magnetic yoke. The number of magnetic yokes is set to four, and the four magnetic yokes are arranged in a rectangular shape at the bottom of the mobile platform. The auxiliary wheel and the lower surface of the magnetic attraction wheel are kept flush.

[0010] Furthermore, the magnetic suspension spraying assembly includes a nozzle bracket fixedly installed at the bottom of the mobile platform, a magnetic suspension nozzle rotatably mounted on the nozzle bracket, a magnetic suspension delivery hose connected to the magnetic suspension nozzle, and the magnetic suspension delivery hose extending through the mobile platform.

[0011] Furthermore, the magnetic suspension nozzle is connected to the nozzle bracket via a locking bolt, which is used to install and adjust the angle of the magnetic suspension nozzle.

[0012] Furthermore, a battery is installed on the top of the mobile platform, and a lighting lamp and an ultraviolet lamp are installed on the bottom of the mobile platform. The lighting lamp is used to meet the lighting requirements of conventional magnetic particle detection, and the ultraviolet lamp is used to meet the lighting requirements of fluorescent magnetic particle detection.

[0013] Furthermore, a control module is installed on the mobile platform. The control module specifically includes a working mode switching switch, a power switch, a magnetic suspension injection switch, a magnetization and motor start switch, and a control circuit. The working mode switching switch is used to switch between manual mode and remote control mode.

[0014] Furthermore, it also includes a remote control platform, which is connected to the detection device via wireless communication and is used to receive image data collected by the high-definition camera component, send control commands, and store detection data.

[0015] According to a second aspect of the present invention, the present invention provides a method for identifying magnetic trace image defects based on multi-scale feature fusion, integrated within a remote control platform, comprising the following steps: S1: Perform preprocessing and enhancement operations on the original acquired magnetic trace image to generate an enhanced image. The operations include illumination equalization, noise reduction, and sharpening. S2: Input the enhanced image into a pre-constructed defect detection network to detect and locate defect regions in the image, and output the defect category, bounding box, and pixel-level mask. The defect detection network includes a multi-scale feature extraction backbone, a feature fusion pyramid, and a lightweight detection head. S3: Based on the output of the defect detection network, quantitatively measure and calculate the size of the detected defects, and automatically assess the defect level according to the built-in industry standards; S4: Utilize the displacement data generated during the inspection process of the mobile magnetic particle inspection equipment to perform time-series information fusion on the defect detection results of multiple consecutive frames of images, so as to achieve deduplication and stitching of the same defect and generate a defect distribution heat map.

[0016] Furthermore, the illumination equalization in step S1 employs an improved homomorphic filtering algorithm to decompose the image into illumination and reflection components, thereby compressing low-frequency illumination unevenness and enhancing high-frequency defect edge information; the denoising employs a nonlocal mean filtering algorithm; the sharpening is performed based on the Laplacian operator, and then adaptive histogram equalization is used to limit local contrast amplification.

[0017] Furthermore, the multi-scale feature extraction backbone in the defect detection network is constructed using an improved MobileNet-V3 architecture, and the receptive field is expanded by introducing dilated convolution. The feature fusion pyramid is embedded with a coordinate attention mechanism to preserve positional information when fusing features at different scales. The lightweight detection head is used to decouple the output defect category probability, bounding box position, and pixel-level contour.

[0018] This invention has at least the following beneficial effects: 1. This invention, through the introduction of a remote control mode, achieves physical isolation between the operator and the testing site. Testing personnel can operate the equipment from a safe area via a remote control platform to complete operations in high-altitude, enclosed, toxic, or radiation environments, eliminating the personal risks of falls, poisoning, and radiation associated with traditional manual climbing and close-range handling of magnetization devices. The equipment's autonomous movement and automatic testing capabilities avoid continuous exposure of personnel to hazardous environments, extending the safety boundary of non-destructive testing from accessibility constraints to controllability constraints.

[0019] 2. This invention, through the combined design of magnetic suction wheels and magnetic yoke auxiliary wheels, enables the equipment to stably adhere to and move on curved surfaces, inclined surfaces, and complex workpiece surfaces with weld seam excess. The magnetic suction wheels provide overall adsorption force and driving force, while the magnetic yoke auxiliary wheels maintain a constant air gap in the magnetic circuit. Together, they adapt to surface undulations, ensuring that the magnetization effect is not affected by lift-off effects. The dual-light source configuration of the illumination lamp and ultraviolet lamp allows the same equipment to be switched between conventional magnetic particle inspection and fluorescent magnetic particle inspection, eliminating the need to change external lighting devices or carry multiple sets of equipment, thus simplifying on-site equipment configuration.

[0020] 3. This invention enables rapid switching between manual and remote control modes via a working mode switch. Manual mode is suitable for convenient testing or equipment debugging and maintenance in simple working conditions, while remote control mode is suitable for unmanned operation in complex working conditions. The two modes are interlocked by hardware to ensure clear ownership of control authority. Remote control is further subdivided into semi-automatic and fully automatic sub-modes. The former retains manual decision-making power regarding magnetization and spraying timing, while the latter enables fully autonomous execution of the testing process, adapting to different process specifications and on-site management needs.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a three-dimensional schematic diagram of the overall structure of the present invention from a first-view perspective; Figure 2 This is a two-dimensional schematic diagram of the overall structure of the present invention from a second perspective; Figure 3 This is a flowchart illustrating the method described in Embodiment 2 of the present invention.

[0024] Figure label: 1. Magnetic roller; 2. Drive motor; 3. Reduction gearbox; 4. Fixed bracket; 5. Fixing bolt; 6. Battery holder; 7. Battery; 8. Handle; 9. Working mode switch; 10. Front-facing camera; 11. Working mode indicator light; 12. Power indicator light; 13. Power switch indicator light; 14. Moving platform; 15. Magnetic yoke; 16. Magnetization and motor start switch; 17. Magnetic suspension spray switch; 18. Output shaft; 19. Nozzle bracket; 20. Magnetic suspension nozzle; 21. Magnetic suspension delivery hose; 22. Locking bolt; 23. Lighting lamp; 24. High-definition camera; 25. Ultraviolet lamp; 26. Magnetic suspension delivery hose. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Example 1: Please see Figures 1-2 This invention provides a technical solution: a fully automated magnetic particle inspection device for defect identification, comprising: The mobile platform 14 has multiple sets of drive components installed at its bottom for moving on the surface of the workpiece being measured. Multiple magnetization components are set at the bottom of the moving platform 14 to generate a rotating magnetic field to magnetize the workpiece under test. A magnetic suspension spraying assembly is installed at the bottom of the mobile platform 14 and is located between adjacent magnetization assemblies. It is used to spray magnetic suspension liquid onto the surface of the workpiece being tested. The high-definition camera assembly includes a front-facing camera 10 and a high-definition camera 24. The front-facing camera 10 is located on the front side of the mobile platform 14 and is used to capture images of the environment in front of the detector. The high-definition camera 24 is located on the bottom of the mobile platform 14 and is used to capture magnetic trace images of the detection area.

[0027] Regarding the technical solution of this embodiment, the drive assembly includes a fixed bracket 4 fixedly installed at the bottom of the mobile platform 14. A magnetic chuck 1 is rotatably installed on one side of the fixed bracket 4, and a reduction gearbox 3 is detachably installed on the other side. The output shaft 18 of the reduction gearbox 3 is fixedly connected to the magnetic chuck 1. A drive motor 2 is fixedly installed on the input shaft of the reduction gearbox 3, and the drive motor 2 is fixed on the outer wall of the reduction gearbox 3. The drive motor 2, the reduction gearbox 3, and the magnetic chuck 1 are arranged sequentially along the axis and integrated on one side of the fixed bracket 4 to form a compact drive unit. The output shaft 18 of the drive motor 2 is directly fixedly connected to the input shaft of the reduction gearbox 3, and the output shaft 18 of the reduction gearbox 3 is directly fixedly connected to the magnetic chuck 1. There are no intermediate transmission links such as couplings, belts, or chains in between. This structure eliminates the energy loss and gap of intermediate transmission, making the power transmission more direct and the start-stop response more rapid. It is especially suitable for the requirements of low speed, stability, and precise positioning of the mobile platform 14 in magnetic particle testing. Furthermore, the reduction gearbox 3 is detachably installed, so that the drive motor 2, the reduction mechanism, and the traveling wheels form an independent power unit. When the magnetic roller 1 needs to be replaced due to wear, or when the reduction gearbox 3 malfunctions, the entire mobile platform 14 does not need to be disassembled. The unit can be replaced simply by removing the fixing bolts 5, which significantly reduces the complexity of on-site maintenance.

[0028] In this embodiment, a photoelectric encoder is installed inside the reduction gearbox 3 to record the displacement data of the magnetic chuck 1 in real time. The photoelectric encoder is directly installed inside the reduction gearbox 3, forming an integrated structure with the gear transmission system. It can collect the actual rotation angle and number of revolutions of the wheel after reduction in real time. Compared with the conventional solution of installing the encoder on the motor shaft, this layout eliminates the influence of factors such as transmission backlash, gear meshing error, and wheel slippage of the reduction gearbox 3 on the accuracy of displacement measurement. This ensures that the recorded displacement data truly reflects the actual travel distance of the magnetic chuck 1 on the surface of the workpiece being measured, thus providing a reliable spatial coordinate reference for the accurate positioning of defects.

[0029] In this embodiment, the magnetization component is a magnetic yoke 15, with auxiliary wheels mounted on its bottom. Four magnetic yokes 15 are arranged in a rectangular pattern on the bottom of the moving platform 14, forming two pairs of orthogonal magnetic poles. When alternating current with a 90-degree phase difference is applied to the two pairs of yokes 15, the combined magnetic field generates a rotational effect on the workpiece surface, causing magnetic lines of force to sweep across the detection area in a circular trajectory. Compared to magnetization in a single direction, the rotating magnetic field can simultaneously excite defect leakage magnetic flux in all directions. Regardless of the crack's orientation, magnetic powder can be effectively attracted and revealed, eliminating blind spots caused by the magnetization direction being parallel to the defect's orientation.

[0030] Regarding the technical solution of this embodiment, the lower surfaces of the auxiliary wheel and the magnetic suction wheel 1 are kept flush. The auxiliary wheel is installed at the bottom of the magnetic yoke 15 to maintain a constant air gap between the pole shoe of the magnetic yoke 15 and the surface of the workpiece. The auxiliary wheel bears part of the weight of the platform, avoiding fluctuations in the air gap of the magnetic circuit caused by the direct pressure on the magnetic yoke 15, and ensuring stable and controllable magnetic resistance. When the platform crosses the weld seam height or surface undulations, the rolling support of the auxiliary wheel causes the pole shoe of the magnetic yoke 15 to follow and adhere, maintaining the continuity of magnetic flux density and preventing magnetic field attenuation caused by the lift-off effect.

[0031] Regarding the technical solution of this embodiment, the magnetic suspension spraying assembly includes a nozzle bracket 19 fixedly installed at the bottom of the mobile platform 14. A magnetic suspension nozzle 20 is rotatably mounted on the nozzle bracket 19, and a magnetic suspension delivery hose 26 is connected to the magnetic suspension nozzle 20. The magnetic suspension delivery hose 26 passes through the mobile platform 14 and is connected to a storage tank for storing magnetic suspension. A magnetic suspension motor pump is installed inside the storage tank, forming a long-distance liquid supply relationship with the nozzle on the mobile platform 14. The pump body does not move with the platform, avoiding interference from vibration sources to the image acquisition mechanism. At the same time, it reduces the mass of the platform body, which is beneficial for maintaining stable adsorption on weakly magnetic or thin-walled workpieces. The storage tank can be fixedly placed in a safe area, facilitating large-capacity liquid storage and frequent liquid replenishment operations without interrupting the detection process.

[0032] Regarding the technical solution of this embodiment, the magnetic suspension nozzle 20 is connected to the nozzle bracket 19 via a locking bolt 22. The locking bolt 22 is used to realize the installation and angle adjustment of the magnetic suspension nozzle 20. The nozzle bracket 19 and the magnetic suspension nozzle 20 are rotatably connected, so that the nozzle outlet direction can be adjusted within a certain range. For workpiece surfaces with different curvatures or welds at different positions, the operator can manually adjust the nozzle tilt angle so that the magnetic suspension jet impacts the detection area at the optimal angle, ensuring uniform coverage of magnetic powder and avoiding excessive splashing. This adjustability allows the same spraying mechanism to adapt to various geometric conditions such as planes, curved surfaces, and fillet welds without the need to replace special nozzles.

[0033] In this embodiment, a battery 7 is installed on the top of the mobile platform 14 via a battery holder 6. A lighting lamp 23 and an ultraviolet lamp 25 are installed at the bottom of the mobile platform 14. The lighting lamp 23 is used to meet the lighting requirements of conventional magnetic particle inspection, and the ultraviolet lamp 25 is used to meet the lighting requirements of fluorescent magnetic particle inspection. The lighting lamp 23 and ultraviolet lamp 25 are positioned at the bottom of the platform, close to the inspection area, shortening the optical path from the light source to the workpiece surface. This close arrangement reduces interference from stray ambient light and improves the contrast between defect magnetic marks and background contrast. For fluorescent magnetic particle inspection, short-distance ultraviolet irradiation reduces excitation light attenuation, making it easier for the image acquisition mechanism to capture the fluorescence of weak defects. The battery 7 is installed on the top of the mobile platform 14, which raises the overall center of gravity of the equipment. When the magnetic roller 1 is attached to the surface of the workpiece, the higher center of gravity creates a swing effect, which enhances the platform's ability to resist overturning torque. When the platform crosses a weld protrusion or encounters lateral disturbance, the restoring torque generated by the extended gravity arm helps to maintain posture stability and reduce the risk of tipping over.

[0034] It should be further noted that a handle 8 is installed on the top of the mobile platform 14, which makes it convenient for staff to operate by holding the handle 8.

[0035] Regarding the technical solution of this embodiment, a control module is installed on the mobile platform 14. The control module specifically includes a working mode switching switch 9, a power switch, a magnetic suspension liquid injection switch 17, a magnetization and motor start switch 16, and a control circuit. The working mode switching switch 9 is used to switch between manual mode and remote control mode.

[0036] Furthermore, the detection device also includes a remote control platform, which is connected to the detection device via wireless communication and is used to receive image data acquired by the high-definition camera component, send control commands, and store detection data.

[0037] Specifically, when switching to manual mode, the side with the front-facing camera 10 should face the inspection direction, and the inspection instrument should be placed above the weld to be inspected. Switch the working mode switch 9 to manual mode, press the power switch, and the switch indicator light will illuminate. Simultaneously, the working mode indicator light 11 will remain constantly lit. At this time, press the start switch, and the lighting lamp 23 will illuminate. The drive motor 2 will move slowly, and the magnetic yoke 15 will generate a rotating magnetic field, initiating magnetization. During the motor's movement, press the magnetic suspension injection switch 17, and the internal magnetic suspension pump will begin operating, allowing the magnetic suspension to pass through the rear... The magnetic suspension delivery hose 26 begins to draw the prepared magnetic suspension, and the magnetic suspension nozzle 20 begins to spray the magnetic suspension. In manual mode, the detection direction is adjusted manually. During the detection process, the detector automatically identifies and records defects through the high-definition camera 24, and can measure the size of the defects. When a defect is detected, the detector's working mode indicator 11 will flash at 0.5-second intervals to issue an alarm message. At the same time, the drive motor 2 will automatically stop. After manual confirmation, the drive motor 2 will continue to move slowly after the start switch is pressed again. When switching to remote control mode, align the side with the front-facing camera 10 towards the detection direction, switch the working mode switch 9 to remote control mode, press the power switch, the power indicator light will illuminate, and the working mode indicator light 11 will flash at 1-second intervals. Turn on the remote control platform; the platform displays two video feeds: the real-time video feed transmitted from the front-facing camera 10 and the detection feed displayed by the high-definition camera 24 located below the detector. Control the rotation of the drive wheels using the forward, backward, left, and right buttons on the remote control platform. When the left turn button is pressed, the right drive wheel of the front wheel rotates faster than the left, while the two rear drive wheels maintain the same speed, achieving left turn control. When the right turn button is pressed, the right drive wheel rotates slower than the left, while the two rear drive wheels maintain the same speed, achieving right turn control. When the forward button is pressed, all drive wheels rotate simultaneously at the same speed in the forward direction, and the detector moves linearly in that direction. When the reverse button is pressed, all drive wheels rotate simultaneously at the same speed in the reverse direction, and the detector moves linearly in that direction. Through the real-time image displayed on the front-facing camera 10 and the detector's motion control buttons on the control platform, the detector can be remotely controlled to the weld position of the workpiece being inspected. The remote control of the magnetization process is divided into two control modes: remote semi-automatic control and fully automatic control. The specific working process is as follows: (1) Remote semi-automatic control of magnetization process After the detector is remotely controlled to the weld position of the workpiece being inspected via the control platform, the corresponding mode indicator on the control platform illuminates when the operating mode is selected. Pressing the start switch button on the control platform illuminates the lighting lamp 23, and the drive motor 2 moves slowly. Simultaneously, the magnetic yoke 15 generates a rotating magnetic field, initiating magnetization. During motor movement, pressing the magnetic suspension spray button on the control platform automatically sprays the magnetic suspension. While moving, the encoder records the displacement data from the current position in real time, allowing for the positioning of the detector. When the detector detects a defect, it automatically stops, sends an alarm signal to the control platform, and the corresponding indicator light on the control platform flashes. The displayed inspection screen automatically locks the specific location and type of the defect, calculates its size, and stores the data in real time. After manual confirmation, pressing the start switch button on the control platform again resumes the automatic movement of the detector. In this control mode, the magnetization process is similar to manual control, except that it is remotely controlled manually.

[0038] (2) Remote fully automatic control of magnetization process After the inspection instrument is remotely controlled to the weld location of the workpiece to be inspected via the control platform, the corresponding mode indicator on the control platform will illuminate. By pressing the start switch button on the control platform, the inspection instrument will begin working in fully automatic mode. At this time, the inspection instrument will automatically start moving and then automatically spray magnetic suspension at 1-second intervals (the default time interval, which can be adjusted according to the situation). When a defect is detected, the instrument will automatically stop and send an alarm signal, the alarm light on the control platform will flash, and the locked defect information will be displayed on the real-time display screen (as above), while the data will be stored. After manual confirmation on the control platform, pressing the start button again will allow the inspection instrument to continue in fully automatic mode.

[0039] Example 2: like Figure 3 As shown, based on the above embodiment one, this embodiment further integrates a magnetic trace image defect recognition algorithm based on multi-scale feature fusion within the remote control platform. This algorithm is not a simple image binarization or threshold segmentation, but a deep learning recognition model specifically designed based on multi-scale feature fusion and attention mechanism, targeting the characteristics of low signal-to-noise ratio, diverse defect morphology, and complex background texture in magnetic particle detection magnetic trace images. The specific steps include: S1. Image Preprocessing and Enhancement Due to the influence of ambient lighting conditions (conventional white light or ultraviolet light) and the reflective properties of the workpiece surface, the original acquired magnetic trace images may have problems such as uneven lighting, low contrast, and noise interference. After receiving the image data transmitted from the high-definition camera (24), the remote control platform first performs a preprocessing procedure: Adaptive illumination equalization: An improved homomorphic filtering algorithm is used to decompose the image into illumination and reflection components. While compressing low-frequency illumination unevenness, it enhances high-frequency defect edge information, allowing the details of magnetic traces in dark areas to be revealed.

[0040] Non-local means denoising: To address the discrete powder clump noise that may be generated after the magnetic suspension is sprayed, a non-local means filtering algorithm is used. This algorithm can remove noise while preserving the continuity of linear defects such as cracks to the greatest extent, avoiding the edge blurring caused by traditional Gaussian filtering; Image sharpening and contrast stretching: Image sharpening is performed based on the Laplacian operator to enhance the contrast between the magnetic traces and the background material, and adaptive histogram equalization (CLAHE) is used to limit excessive amplification of local contrast and prevent artifact interference; S2. Construct a defect detection network based on multi-scale feature fusion The core algorithm in this embodiment is a lightweight convolutional neural network, specifically designed for real-time detection of magnetic trace defects. The network structure mainly consists of three parts: Multi-scale feature extraction backbone: An improved MobileNet-V3 is used as the backbone to extract features at different levels of magnetic trace images. The shallow feature maps have high resolution and contain fine magnetic trace edge and texture information, which are used to locate micro-cracks. The deep feature maps have a large receptive field and contain semantic information, which are used to identify long strip or large area defects. By introducing dilated convolution, the receptive field is expanded exponentially without increasing the number of parameters, effectively capturing the overall direction of slender cracks. Feature Fusion Pyramid (CA-FPN): A feature pyramid with a coordinate attention mechanism was designed. Traditional FPNs often lose positional information when fusing features at different scales. The CA-FPN module embeds positional information into the channel attention, enabling the network to focus not only on "whether magnetic marks exist" but also on "where the magnetic marks are." This is crucial for distinguishing between pseudo magnetic marks and real cracks at weld edges.

[0041] Lightweight inspection head: The decoupled inspection head outputs the defect category probability, bounding box position and pixel-level contour of magnetic traces respectively. The output defect categories include at least the following common defect types in magnetic particle inspection: linear cracks, circular pores, strip inclusions, dense pores, etc. S3. Defect Quantification and Dimensional Measurement Algorithm When the network detects a defect, the system automatically triggers the measurement function, solving the pain point of traditional manual inspection that requires manual estimation or measurement with the aid of tools: Size calibration based on pixel equivalent: Combining the displacement data of the magnetic wheel (1) recorded by the photoelectric encoder with the fixed installation height of the high-definition camera (24), the system automatically calculates the pixel equivalent (mm / pixel) at the current detection distance, without the need for manual placement of a scale. Morphological analysis: skeleton extraction and endpoint detection are performed on the detected defect mask; Length measurement: For linear defects (cracks, lack of fusion), calculate the actual length along the skeleton curve.

[0042] Area measurement: For circular defects (pores, inclusions), count the total number of pixels in the mask and convert it into the actual area (or convert it into the equivalent diameter). Rating determination: Based on the built-in industry standards (such as NB / T 47013.4-2015), the system automatically compares the length, spacing and quantity of defects and gives the defect level assessment results (Level I, Level II, Level III, etc.) in real time. S4. Defect tracking and trajectory reconstruction based on time-series information Unlike static image detection, the device in Example 1 is mobile. Leveraging this characteristic, the algorithm further incorporates a temporal information fusion module. Defect deduplication and stitching: When the equipment moves along the weld seam, the same defect may appear continuously in multiple frames of images. The algorithm uses the displacement data of the photoelectric encoder and feature matching (such as SIFT feature point matching) to determine whether the defect detected in the current frame overlaps with the previous frame. If they overlap, they are automatically merged into the same defect, and their cumulative length is updated to avoid the same defect being recorded repeatedly. Defect Distribution Heat Map: After inspection, the system automatically generates a defect distribution heat map of the inspected workpiece based on the displacement coordinates recorded by the photoelectric encoder and the defect detection results. This map visually displays the dense areas and severity of defects on the weld, providing visual data support for subsequent rework decisions. S5. Edge computing and lightweight model deployment Considering the real-time requirements in remote control mode, the aforementioned MFFD-Net model needs to be optimized using the TensorRT (or OpenVINO) inference acceleration engine and subjected to INT8 quantization before deployment. After optimization, the model's single-frame inference time on embedded GPUs (such as Jetson Xavier NX) can be controlled to <50ms, meeting the real-time requirements of motion detection.

[0043] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0044] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the claims of this application.

Claims

1. A fully automated magnetic particle inspection device for defect identification, characterized in that, include: A mobile platform (14) is provided with multiple sets of drive components installed at its bottom for moving on the surface of the workpiece being measured. Multiple sets of magnetization components are disposed at the bottom of the moving platform (14) to generate a rotating magnetic field to magnetize the workpiece under test; A magnetic suspension spraying assembly is installed at the bottom of the mobile platform (14) and is located between adjacent magnetization assemblies for spraying magnetic suspension onto the surface of the workpiece being tested. The high-definition camera assembly includes a front-facing camera (10) and a high-definition camera (24). The front-facing camera (10) is located on the front side of the mobile platform (14) and is used to collect images of the environment in front of the detector. The high-definition camera (24) is located on the bottom of the mobile platform (14) and is used to collect magnetic trace images of the detection area.

2. The fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: The drive assembly includes a fixed bracket (4) fixedly installed at the bottom of the mobile platform (14). A magnetic chuck (1) is rotatably installed on one side of the fixed bracket (4), and a reduction gearbox (3) is detachably installed on the other side. The output shaft (18) of the reduction gearbox (3) is fixedly connected to the magnetic chuck (1). A drive motor (2) is fixedly installed on the input shaft of the reduction gearbox (3), and the drive motor (2) is fixed on the outer wall of the reduction gearbox (3).

3. The fully automated magnetic particle inspection device for defect identification according to claim 2, characterized in that: The reduction gearbox (3) is equipped with a photoelectric encoder, which is used to record the displacement data of the magnetic chuck (1) in real time.

4. The fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: The magnetization component is configured as a magnetic yoke (15), and an auxiliary wheel is installed at the bottom of the magnetic yoke (15). The number of magnetic yokes (15) is set to four, and the four magnetic yokes (15) are arranged in a rectangular shape at the bottom of the moving platform (14). The auxiliary wheel and the lower surface of the magnetic suction wheel (1) are flush.

5. The fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: The magnetic suspension spraying assembly includes a nozzle bracket (19) fixedly installed at the bottom of the mobile platform (14), a magnetic suspension nozzle (20) is rotatably installed on the nozzle bracket (19), a magnetic suspension delivery hose (26) is connected to the magnetic suspension nozzle (20), and the magnetic suspension delivery hose (26) is laid through the mobile platform (14).

6. The fully automated magnetic particle inspection device for defect identification according to claim 5, characterized in that: The magnetic suspension nozzle (20) is connected to the nozzle bracket (19) by a locking bolt (22), which is used to realize the installation and angle adjustment of the magnetic suspension nozzle (20).

7. The fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: A battery (7) is installed on the top of the mobile platform (14), and a lighting lamp (23) and an ultraviolet lamp (25) are installed on the bottom of the mobile platform (14). The lighting lamp (23) is used to meet the lighting requirements of conventional magnetic particle detection, and the ultraviolet lamp (25) is used to meet the lighting requirements of fluorescent magnetic particle detection.

8. The fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: The mobile platform (14) is equipped with a control module, which specifically includes a working mode switching switch (9), a power switch, a magnetic suspension injection switch (17), a magnetization and motor start switch (16), and a control circuit. The working mode switching switch (9) is used to switch between manual mode and remote control mode.

9. A fully automated magnetic particle inspection device for defect identification according to claim 1, characterized in that: It also includes a remote control platform, which is connected to the detection device via wireless communication and is used to receive image data collected by the high-definition camera component, send control commands, and store detection data.

10. A method for identifying magnetic trace image defects based on multi-scale feature fusion, integrated into a remote control platform, characterized in that... Includes the following steps: S1: Perform preprocessing and enhancement operations on the original acquired magnetic trace image to generate an enhanced image. The operations include illumination equalization, noise reduction, and sharpening. S2: Input the enhanced image into a pre-constructed defect detection network to detect and locate defect regions in the image, and output the defect category, bounding box, and pixel-level mask. The defect detection network includes a multi-scale feature extraction backbone, a feature fusion pyramid, and a lightweight detection head. S3: Based on the output of the defect detection network, quantitatively measure and calculate the size of the detected defects, and automatically assess the defect level according to the built-in industry standards; S4: Utilize the displacement data generated during the inspection process of the mobile magnetic particle inspection equipment to perform time-series information fusion on the defect detection results of multiple consecutive frames of images, so as to achieve deduplication and stitching of the same defect and generate a defect distribution heat map.

11. The magnetic trace image defect recognition method based on multi-scale feature fusion according to claim 10, characterized in that, The illumination equalization in step S1 employs an improved homomorphic filtering algorithm to decompose the image into illumination and reflection components, thereby compressing low-frequency illumination unevenness and enhancing high-frequency defect edge information; the denoising employs a nonlocal mean filtering algorithm; the sharpening is performed based on the Laplacian operator, and then adaptive histogram equalization is used to limit local contrast amplification.

12. The magnetic trace image defect recognition method based on multi-scale feature fusion according to claim 10, characterized in that, The multi-scale feature extraction backbone of the defect detection network is constructed using an improved MobileNet-V3 architecture, and the receptive field is expanded by introducing dilated convolution. The feature fusion pyramid is embedded with a coordinate attention mechanism to preserve positional information when fusing features at different scales. The lightweight detection head is used to decouple the output defect category probability, bounding box position, and pixel-level contour.

Images