A fluorescent magnetic powder image flaw detection method applied to pressure steel pipe crack diagnosis

Abstract

The present application relates to the technical field of nondestructive testing and image processing, and discloses a fluorescent magnetic powder image flaw detection method applied to pressure steel pipe crack diagnosis. The present application aims to solve the problems of discontinuous detection process, inconsistent operation conditions and difficulty in realizing quantitative evaluation due to reliance on subjective judgment in the prior art manual fluorescent magnetic powder flaw detection method. The present application continuously performs magnetization, spraying and synchronous image acquisition steps on the inner wall of the pressure steel pipe through an integrated mobile platform to obtain fluorescent images carrying position information; and adopts an image processing procedure including image preprocessing, candidate region segmentation based on color features and crack recognition based on morphological features, thereby realizing automatic diagnosis of cracks.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing and image processing technology, specifically, to a fluorescent magnetic particle imaging flaw detection method for diagnosing cracks in pressure steel pipes. Background Technology

[0002] Fluorescent magnetic particle testing, as a mature non-destructive testing method, is widely used to detect discontinuous defects such as cracks and inclusions on and near the surface of ferromagnetic materials (such as pressure steel pipes). Its basic principle is as follows: First, the workpiece to be tested is magnetized to generate an internal magnetic field; then, a suspension containing fluorescent magnetic powder particles is applied to the workpiece surface; if defects exist on the workpiece surface, a leakage magnetic field will be formed at that location, thereby attracting and accumulating the fluorescent magnetic powder; finally, under ultraviolet light excitation, the accumulated magnetic powder emits bright fluorescence, clearly revealing the location and shape of the defects. This testing method plays a crucial role in the safe operation and maintenance of large, critical pressure-bearing equipment such as pressure steel pipes in hydropower stations. Currently, most fluorescent magnetic particle testing operations inside such large pipelines still rely on traditional manual methods, requiring inspectors to enter the pipeline. This typically involves pre-constructing complex scaffolding, then using a handheld magnetic yoke to locally magnetize sections of the pipe wall, manually spraying the magnetic powder suspension, and visually observing and recording the results.

[0003] However, the aforementioned traditional manual inspection methods suffer from inherent drawbacks such as low inspection efficiency, high labor intensity, and harsh working environments. More importantly, the segmented and multi-step nature of manual operation (with separate steps for magnetization, spraying, and observation) and the high dependence of inspection results on the subjective judgment of the inspectors make it difficult to ensure consistency of inspection conditions (such as magnetization intensity, spraying amount, observation distance and angle) throughout the entire inspection process, as well as the objectivity of result interpretation. This results are often qualitative, difficult to quantify and compare precisely, and have poor repeatability. Therefore, how to implement an automated, continuous, and standardized inspection process for the inner wall of pressure steel pipes, and transform the inspection results dependent on subjective vision into objective, quantifiable data to achieve accurate diagnosis and quantitative assessment of cracks, has become a pressing technical problem to be solved in this field. Summary of the Invention

[0004] In view of the aforementioned problems with existing artificial fluorescent magnetic particle inspection methods, such as discontinuous detection processes, inconsistent operating conditions, and reliance on subjective judgment for quantitative evaluation, this invention provides a fluorescent magnetic particle imaging inspection method for crack diagnosis of pressure steel pipes. To solve the above-mentioned technical problems, this invention provides the following technical solution:

[0005] This invention provides a fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes, which includes the following steps:

[0006] S1: Magnetize the target detection area on the inner wall of the pressure steel pipe to establish a closed magnetic field within the pipe wall of the target detection area;

[0007] S2: Spray fluorescent magnetic powder suspension onto the target detection area after the magnetization treatment, so that the fluorescent magnetic powder accumulates at the leakage magnetic field generated by the crack in the target detection area.

[0008] S3: The target detection area is excited by an ultraviolet light source, and the fluorescence image of the target detection area is acquired by an image acquisition unit. The fluorescence image carries position and time information.

[0009] S4: The fluorescent image is processed to identify cracks in the pressure steel pipe. The processing includes: first, preprocessing the fluorescent image to correct geometric distortion and brightness unevenness; then, segmenting independent fluorescent foreground regions in a preset color space based on fluorescent color characteristics; and finally, calculating the morphological characteristic parameters of the independent fluorescent foreground regions to determine whether they are real cracks.

[0010] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to the present invention, in step S1, the specific operation is as follows: the arc-shaped inner wall of the pressure steel pipe is magnetized by a magnetization device carried on a mobile platform. During operation, the magnetic pole feet of the magnetization device are kept in close contact with the arc-shaped inner wall. The magnetization device is composed of a permanent magnet and a soft magnetic material yoke, and the contact surface of the magnetic pole feet is pre-processed into an arc surface that matches the curvature of the inner wall.

[0011] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, the S2 step specifically involves: driving the fluorescent magnetic particle suspension with a diaphragm pump, and uniformly spraying the fluorescent magnetic particle suspension onto the target detection area using a fan-shaped nozzle; simultaneously, acquiring the speed signal output by the encoder of the motor driving the mobile robot in real time, and synchronously adjusting the output flow rate of the diaphragm pump according to the speed signal at a preset ratio, thereby ensuring that the coverage of fluorescent magnetic particle suspension on the pipe wall per unit area remains consistent under different travel speeds.

[0012] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, the specific operation in step S3 is as follows: an industrial camera with a global shutter function is used to acquire images; and an ultraviolet light source in a ring around the camera lens is controlled simultaneously to ensure that the ultraviolet light source is lit and emits light only during the exposure of the industrial camera, thereby obtaining a clear fluorescent image with no motion blur and uniform excitation illumination.

[0013] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, the preprocessing in step S4 specifically involves: firstly, applying a pre-generated geometric distortion lookup table to process the fluorescent image to correct the image distortion introduced by the combined action of the wide-angle lens of the camera and the arc-shaped inner wall of the pressure steel pipe; then, applying a pre-generated brightness distribution lookup table to process the geometrically corrected image to compensate for the brightness differences caused by uneven light source distribution.

[0014] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, the process of segmenting the crack candidate region and filtering out background noise in step S4 specifically involves: converting the preprocessed image from the RGB color space to the HSV color space, and performing binarization processing according to preset hue and saturation thresholds; and performing morphological opening operations on the binarized image to effectively remove background noise formed by isolated magnetic particle particles.

[0015] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, the process of determining whether a crack is real in step S4 specifically involves: calculating the aspect ratio and linearity of each independent white region in the processed image; inputting the determined aspect ratio and linearity as a combination into a pre-trained crack feature recognition model, which then determines whether the independent white region is a real crack.

[0016] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes described in this invention, after determining that it is a real crack, the following operations are also included: analyzing the brightness and width of the real crack in the preprocessed fluorescent image, thereby quantifying and evaluating the real crack as a crack hazard index; and visually marking the real crack on the two-dimensional unfolded diagram and three-dimensional model of the pressure steel pipe by combining the location information carried by the fluorescent image.

[0017] This invention also provides a fluorescent magnetic particle imaging flaw detection device for diagnosing cracks in pressure steel pipes, used to perform the above-mentioned method, specifically including:

[0018] A magnetization device is used to magnetize the target detection area on the inner wall of a pressure steel pipe.

[0019] A fluorescent magnetic powder spraying system is used to spray a fluorescent magnetic powder suspension onto the target detection area that has been magnetized by the magnetization device.

[0020] The image acquisition unit includes an ultraviolet light source and an industrial camera, used to excite and acquire fluorescence images of the target detection area after the target detection area is sprayed with fluorescent magnetic powder suspension;

[0021] The processing unit, electrically connected to the image acquisition unit, is used to process the fluorescent image to identify cracks in the pressure steel pipe; wherein, along the travel direction of the moving platform, the magnetization device, the fluorescent magnetic powder spraying system, and the image acquisition unit are arranged in a fixed sequence.

[0022] As a preferred embodiment of the fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to the present invention, it further includes: a mobile platform for carrying the magnetization device, the fluorescent magnetic particle spraying system and the image acquisition unit, and capable of crawling on the inner wall of the pressure steel pipe; wherein the mobile platform includes a motor and a magnetic drive wheel, and is driven by differential rotation; the magnetic drive wheel is composed of alternating permanent magnet segments and knurled metal wheel segments.

[0023] The beneficial effects of this invention are as follows: By automating and standardizing the flaw detection process, this invention improves detection efficiency and reduces the subjectivity of manual visual inspection; at the same time, by performing multi-stage image processing and intelligent recognition on fluorescent images, it is possible to quantitatively assess the geometric morphology and severity of cracks, and combine location information to achieve precise defect localization, thereby improving the objectivity and traceability of diagnostic results. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a fluorescent magnetic particle imaging flaw detection method applied to the diagnosis of cracks in pressure steel pipes.

[0026] Figure 2 This is a flowchart of image processing and analysis.

[0027] Figure 3 Flowchart for the synchronous acquisition of location information and images.

[0028] Figure 4 This is a schematic diagram of the overall structure of the flaw detection robot. Detailed Implementation

[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0031] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0032] Example 1

[0033] Reference Figures 1-3 This is one embodiment of the present invention, which provides a fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes, including the following steps:

[0034] S1. Magnetize the target detection area on the inner wall of the pressure steel pipe to establish a closed magnetic field within the pipe wall of the target detection area.

[0035] The arc-shaped inner wall of the pressure steel pipe is magnetized by a magnetizing device mounted on a mobile platform. During operation, the magnetic pole feet of the magnetizing device are kept in close contact with the arc-shaped inner wall. The magnetizing device consists of a permanent magnet and a soft magnetic yoke, and the contact surface of the magnetic pole feet is pre-processed into an arc surface that matches the curvature of the inner wall.

[0036] In this process, a working magnetic field with a magnetic flux density that meets the sensitivity requirements of magnetic particle inspection is pre-established within the pipe wall in front of the robot's path.

[0037] The magnetization device forms a magnetic circuit system for guiding and confining magnetic flux. In this embodiment, the system mainly consists of permanent magnets and a soft magnetic material yoke. The permanent magnet is preferably a neodymium iron boron permanent magnet with a high magnetic energy product to provide a strong magnetomotive force within a compact volume; the soft magnetic material yoke is preferably an electrical pure iron (e.g., DT4 grade) with low coercivity and high permeability. Its function is to connect the magnetic poles of the two permanent magnets on the non-working surface (i.e., the side away from the steel pipe wall) to form a complete magnetic field loop, ensuring that the magnetic flux can be guided and confined with maximum efficiency.

[0038] The working principle of this magnetic circuit system is as follows: During operation, magnetic lines of force (i.e., magnetic flux) originate from one magnetic pole (e.g., the N pole) of the magnetization device, pass through the contact surface, and enter the steel pipe wall. Since steel itself is an excellent magnetic conductor, the magnetic flux flows parallel to the direction of the robot's movement within the pipe wall for a certain distance, forming a tangential magnetic field for detection. Subsequently, the magnetic flux exits the pipe wall from the other magnetic pole (S pole), returns to the magnetization device, and finally flows back to the initial magnetic pole through the soft magnetic yoke above, thus forming a complete closed magnetic circuit with the steel pipe wall as the core working section. When a crack exists in the pipe wall area covered by this closed magnetic circuit, because the magnetic resistance of the crack is much greater than that of the steel, some of the magnetic flux escapes from the surface of the steel pipe, crosses the crack, and then re-enters the steel pipe, thereby forming a leakage magnetic field with a high magnetic field gradient directly above the crack.

[0039] To ensure the stability and effectiveness of magnetization during the robot's dynamic crawling process, this invention employs an elastic floating suspension mechanism. The magnetizing device is not rigidly fixed to the robot chassis but is connected via this suspension mechanism. This mechanism utilizes a spring to provide a continuous and moderately sized preload, consistently pushing the magnetizing device against the tube wall; simultaneously, its floating structure allows the magnetizing device a certain degree of free travel in the direction perpendicular to the tube wall. When the robot encounters welds, pits, or other uneven surfaces during crawling, this elastic floating mechanism automatically compensates for these height changes, ensuring that the magnetic pole feet can flexibly and dynamically conform to the tube wall surface, thereby continuously maintaining a zero or minimal air gap at the contact surface during dynamic processes.

[0040] Furthermore, any air gap between the magnetic pole feet and the pipe wall creates significant magnetic resistance, severely reducing the effective magnetic flux entering the pipe wall. By perfectly matching the curvature of the contact surface with the curvature of the pipe wall, a transition from "point contact" or "line contact" to "surface contact" can be achieved. In practical applications, the magnetic pole feet of the magnetizing device are designed as replaceable modular components. Before inspecting a pressure steel pipe of a specific diameter, magnetic pole feet that perfectly match the curvature of the inner wall of the steel pipe must be selected or machined and installed on the magnetizing device to ensure optimal magnetic coupling in any inspection task.

[0041] S2. Spray fluorescent magnetic powder suspension onto the magnetized target detection area, so that the fluorescent magnetic powder accumulates at the leakage magnetic field generated by the crack in the target detection area.

[0042] A diaphragm pump drives a fluorescent magnetic powder suspension, which is then uniformly sprayed onto the target detection area by a fan-shaped nozzle. Simultaneously, the speed signal output by the encoder of the motor driving the mobile robot is acquired in real time, and the output flow rate of the diaphragm pump is adjusted synchronously according to a preset ratio based on the speed signal, thereby ensuring that the coverage of fluorescent magnetic powder suspension on the unit area of ​​the tube wall remains consistent under different travel speeds.

[0043] The spraying process follows the magnetization process, wetting the magnetized pipe wall surface and providing fluorescent magnetic powder particles that can be captured by the leakage magnetic field. When the suspension containing magnetic powder particles comes into contact with the pipe wall surface where a leakage magnetic field already exists, a physical aggregation process occurs: the microscopic leakage magnetic field with a high magnetic field gradient leaked from the crack defect immediately captures and attracts the magnetic powder particles in the liquid flowing through or in contact with the area, causing them to accumulate rapidly and in large quantities at the crack location, thus forming a magnetic trace indicator on a macroscopic scale that is consistent with the shape of the crack and is clearly visible under ultraviolet light.

[0044] In this embodiment, a diaphragm pump is selected as the power source for the spraying system. The diaphragm pump has no rotating parts that directly contact the liquid, enabling it to effectively transport liquid mixtures containing solid particles (i.e., magnetic powder). It is less prone to wear and clogging, ensuring reliability during long-term operation. Furthermore, the flow rate of the diaphragm pump can be precisely controlled by adjusting its operating parameters, forming an actuator that is linked to the robot's speed control. After being output from the diaphragm pump, the fluorescent magnetic powder suspension is transported through pipelines to a fan-shaped nozzle. This fan-shaped nozzle can form a flat liquid mist fan with a specific angle and width, which can more efficiently and evenly cover a rectangular area compared to a conical nozzle, ensuring that the sprayed area has neat edges and no blind spots during robot movement.

[0045] To achieve dynamic synchronization between the jet flow rate and the robot's travel speed, the synchronization control logic is as follows: Encoders installed on the drive motor shaft or wheels of the mobile robot continuously send pulse signals to the central controller with extremely high time resolution. The central controller calculates the number of pulses received per unit time, thus accurately determining the robot's current real-time linear velocity. The controller has a pre-defined function, which can be expressed by the following formula:

[0046]

[0047] in, for The target flow rate of the diaphragm pump at any given time; for The robot's real-time speed is constantly fed back from the motor encoder; This is a preset linkage coefficient. The volume of suspension required to be sprayed per unit distance traveled by the robot is determined by the effective coverage width of the nozzle and the desired liquid film thickness. As a concrete calculation example, the linkage coefficient... The value can be determined in the following ways:

[0048]

[0049] in, The effective coverage width (in meters) formed by the fan-shaped nozzle at the working distance from the pipe wall. The average thickness (in meters) of the fluorescent magnetic powder suspension liquid film to be formed on the tube wall.

[0050] For example, if the effective coverage width of the nozzle is 0.1 meters, and the optimal liquid film thickness desired in engineering is 0.0002 meters (i.e., 0.2 millimeters), then the theoretical value of the linkage coefficient K is... .

[0051] In practical applications, this theoretical value can be fine-tuned through on-site experiments to achieve the best magnetic trace display effect.

[0052] S3. The target detection area is excited by an ultraviolet light source, and the fluorescence image of the target detection area is acquired by the image acquisition unit. The fluorescence image carries position and time information.

[0053] An industrial camera with a global shutter function is used for image acquisition; and a ring of ultraviolet light sources surrounding the camera lens is controlled simultaneously to ensure that the ultraviolet light sources are lit and emit light only during the exposure of the industrial camera, thereby obtaining clear fluorescent images with no motion blur and uniform excitation light.

[0054] The imaging process is located at the rear end of the robotic production line, accurately and clearly capturing and recording the crack indication image formed by the aggregation of fluorescent magnetic powder after magnetization and spraying.

[0055] In this embodiment, the core component of the image acquisition unit is an industrial camera. Preferably, a CMOS or CCD industrial camera with a global shutter function is used. The global shutter function ensures that when capturing objects within the field of view of the mobile robot, the image is frozen by simultaneously acquiring the light signals of all pixels, thereby avoiding image tilting or geometric distortion that may occur due to line-by-line scanning exposure using rolling shutter technology.

[0056] To excite the fluorescent magnetic powder to emit light, multiple high-power ultraviolet LEDs are arranged in a ring around the camera as the excitation light source. A specific wavelength of ultraviolet light (e.g., 365nm) is selected to match the optimal excitation spectrum of the fluorescent magnetic powder used. Arranging multiple LEDs in a ring around the camera lens allows for cross-illumination of the detection area from different angles, forming a broad and uniform excitation light field. This arrangement effectively eliminates or reduces noticeable shadows caused by a single light source, preventing the missed crack indication due to shadow obstruction.

[0057] When the camera prepares to take an exposure, it sends an external trigger signal to the light source's driving circuit. Upon receiving this signal, the light source driving circuit immediately illuminates all ultraviolet LEDs at maximum power. After the camera completes its set exposure time (e.g., a few milliseconds), the trigger signal disappears, and the light source driving circuit immediately turns off all LEDs. This "exposure-as-flash" synchronization mechanism has two significant advantages: First, because the light source operates for only a very short exposure time, its average power consumption is extremely low, greatly saving the robot's energy consumption and extending its operation time; second, replacing continuous illumination with high-intensity, short-duration pulsed light can significantly reduce the impact of robot movement during the exposure time, thereby obtaining high-sharpness fluorescent images with higher contrast and cleaner backgrounds.

[0058] As each fluorescent image is successfully acquired and saved, the robot's positioning system will immediately package and store the current position and attitude data (e.g., the distance traveled based on the wheel odometry, the circumferential angle and slope calculated based on the inertial measurement unit IMU) along with a precise timestamp as the metadata of that image.

[0059] In this embodiment, the positioning system works through data fusion, and its specific principle is as follows:

[0060] 1. Distance calculation based on wheel odometer:

[0061] The system continuously reads the number of pulses generated by the photoelectric encoder installed on the robot's drive wheels. By using a calibrated coefficient of "distance traveled per unit number of pulses," the system accumulates and converts the pulse counts to accurately calculate the total distance the robot travels along the pipe wall surface.

[0062] 2. Attitude calculation based on inertial measurement unit:

[0063] The inertial measurement unit integrates a three-axis gyroscope and a three-axis accelerometer.

[0064] The gyroscope is responsible for measuring the robot's angular velocity in three axes. By integrating these angular velocity data over time, the robot's real-time attitude can be calculated, including its circumferential angle (i.e., the "clock" direction) within the cross-section of the pressure steel pipe and its pitch angle (i.e., the "slope") along the pipe's axis.

[0065] Accelerometers can measure the components of gravitational acceleration on three axes when the object is in a static or uniform motion. This can be used to calibrate and correct long-term drift errors caused by gyroscope integration.

[0066] 3. Data fusion:

[0067] To obtain accurate and drift-free pose information, the positioning system employs a data fusion algorithm (such as complementary filtering or Kalman filtering) to combine the relatively accurate short-term distance information provided by the wheel odometer with the angular velocity and acceleration information provided by the inertial measurement unit (IMU) that can sense global orientation and attitude. This algorithm can use accelerometer data to correct for gyroscope integral drift, and by combining the odometer readings, ultimately output a high-precision, complete positioning result that includes three-dimensional spatial position and attitude.

[0068] S4. Process the fluorescence image to identify cracks in the pressure steel pipe. The processing includes: first, preprocessing the fluorescence image to correct geometric distortion and brightness unevenness; then, segmenting independent fluorescence foreground regions based on fluorescence color characteristics within a preset color space; finally, calculating the morphological characteristic parameters of the independent fluorescence foreground regions to determine whether they are real cracks.

[0069] First, a pre-generated geometric distortion lookup table is applied to process the fluorescence image to correct the image distortion introduced by the combined action of the camera's wide-angle lens and the curved inner wall of the pressure steel pipe. Then, a pre-generated brightness distribution lookup table is applied to process the geometrically corrected image to compensate for the brightness differences caused by uneven light source distribution.

[0070] The pre-processed image is converted from the RGB color space to the HSV color space and binarized according to preset hue and saturation thresholds. Morphological opening operations are then performed on the binarized image to effectively remove background noise formed by isolated magnetic powder particles.

[0071] For each individual white region in the processed image, the aspect ratio and linearity of the region are calculated. The determined aspect ratio and linearity are combined and input into a pre-trained crack feature recognition model, which then determines whether the individual white region is a real crack.

[0072] The brightness and width of real cracks in preprocessed fluorescence images are analyzed to quantify and assess real cracks as crack hazard indexes. By combining the location information carried by the fluorescence images, real cracks are visualized on the two-dimensional unfolded diagram and three-dimensional model of the pressure steel pipe.

[0073] The image processing and analysis process is divided into several sequentially executed stages. Its function is to process the original fluorescence image to extract and confirm the real crack information, and finally to quantitatively evaluate the confirmed crack.

[0074] I. Image Preprocessing:

[0075] A pre-generated geometric distortion lookup table is applied to process the fluorescence image to correct the image distortion introduced by the combined effect of the camera's wide-angle lens and the curved inner wall of the pressure steel pipe. The geometric distortion lookup table is generated before formal inspection by photographing a standard calibration board (e.g., a checkerboard or dot array board) with feature points arranged in an array at known positions, and analyzing the mapping relationship between the positions of these feature points in the photographed image and their actual positions. This process can restore the distorted curved surface image to a flat rectangular image.

[0076] As a specific implementation, the geometric distortion lookup table is essentially two mapping matrices, which we call... and The dimensions of these two matrices are exactly the same as those of the image to be corrected. The values ​​stored in the image represent the corrected image coordinates. The pixel at that location should be taken from which part of the original distorted image? Use coordinates to retrieve data; similarly, The value stored in represents which one should be retrieved from. Coordinates are used for capture. During geometric correction, the processing unit iterates through every pixel of the newly generated target image. Then according to and The system uses the provided coordinates to find the corresponding pixel values ​​in the original image and fills them in, thus efficiently completing the entire image distortion removal and perspective transformation process.

[0077] The geometrically corrected image is then processed using a pre-generated brightness distribution lookup table to compensate for brightness differences caused by uneven light source distribution and reduced illumination due to camera lens optical characteristics. Specifically, due to the physical law of fourth cosine, the edge areas of an image captured by a lens receive significantly less illumination than the central areas. This phenomenon manifests visually as darker corners, commonly known as vignetting. The brightness distribution lookup table is generated by photographing a standard uniform grayscale plate and analyzing its brightness distribution. This process eliminates or reduces the lens vignetting effect, making the background brightness of the entire image more uniform.

[0078] A brightness distribution lookup table can be a floating-point matrix with the same size as the image; we call it... (Gain diagram). Each value in this matrix , representing the coordinates The pixel brightness needs to be multiplied by a compensation factor. For example, in brighter areas at the center of the image, this factor might be less than 1; while in darker areas around the edges, it might be greater than 1. During brightness equalization, the processing unit multiplies the brightness value of each pixel in the geometrically corrected image by a compensation factor. The compensation coefficients at the corresponding positions in the image are adjusted to achieve a uniform background brightness across the entire image.

[0079] II. Crack Candidate Region Segmentation and Filtering:

[0080] The pre-processed image is converted from the commonly used RGB (red, green, blue) color space in computers to the HSV (hue, saturation, lightness) color space, which is more in line with human visual perception. In the HSV space, a specific color (such as the fluorescent yellow-green in this case) has very concentrated hue and saturation values, while the lightness value, affected by illumination, can vary considerably. Therefore, by setting a precise hue and saturation threshold range based on pre-defined fluorescence color characteristics and performing binarization on the image, all pixels matching the fluorescence color characteristics can be effectively identified as foreground (marked as white), while background pixels such as pipe walls, rust, and oil stains can be suppressed as background (marked as black), thus generating a black and white binary image.

[0081] As a specific implementation method, the hue and saturation threshold ranges are not fixed, but are precisely determined through a calibration process for each detection task or each batch of fluorescent magnetic powder. This calibration process is as follows:

[0082] 1. Sample collection: Before formal testing, using current robots, fluorescent magnetic powder suspension and lighting conditions, a standard test block containing typical artificial cracks is photographed to obtain one or more high-resolution fluorescent images of the sample.

[0083] 2. Feature Statistics: In the sample image, clearly defined crack fluorescence indicator areas are manually or semi-automatically selected. Then, the HSV values ​​of all pixels within this area are statistically analyzed, and histograms of hue and saturation distributions are calculated separately.

[0084] 3. Threshold Determination: Based on the statistical histogram, determine the minimum tonal range that can cover more than 95% (or another preset percentage) of the crack pixels. and minimum saturation lower limit For example, if statistics show that the hue values ​​of cracks are mainly concentrated between 80 and 100, and the saturation values ​​are all higher than 0.7 (within the normalization range of 0-1), then the threshold range for this task is determined.

[0085] The subsequent binarization process can be described as follows:

[0086] Iterate through every pixel of the input image. Obtain its HSV value .

[0087] if and Then, at the corresponding position in the output binary image The value is assigned to 1 (white); otherwise,

[0088] At the corresponding position in the output binary image The value is assigned to 0 (black).

[0089] This calibration-then-processing method ensures the accuracy of color segmentation and adaptability to different working conditions.

[0090] Next, a morphological opening operation is performed on the binary image to filter out isolated noise points. The algorithm consists of two consecutive geometric transformation steps:

[0091] The first step is a shrinkage operation: This operation shrinks the boundaries of all white areas in the image inward by a specified number of pixels. As a result, tiny, isolated white areas formed by a single or a few magnetic powder particles disappear directly from the image because their size is less than or equal to the shrinkage amount. Meanwhile, continuous white lines of a certain length and width formed by real cracks, although their width narrows, retain their main structure.

[0092] The second step is the expansion operation: Immediately following the contraction operation, this operation expands the boundaries of all remaining white areas outward by the same number of pixels. This operation restores the crack lines that were narrowed in the first step to a width close to their original width. Isolated noise points that disappeared in the first step, no longer existing, cannot be expanded.

[0093] By combining the operations of contraction and expansion, random noise in the background can be effectively filtered out without significantly affecting the actual crack indication shape, thereby greatly purifying the image and retaining only the candidate crack regions with a certain shape and scale.

[0094] III. Crack Identification and Confirmation:

[0095] For each individual white region in the processed image, morphological feature parameters are calculated to determine its morphological characteristics. In this embodiment, these morphological feature parameters are the aspect ratio, which measures the thinness of the white region, and the linearity, which measures the flatness of the white region. The aspect ratio is obtained by calculating the ratio of the longer side to the shorter side of the smallest bounding rectangle of the region; the linearity is obtained by calculating the average distance from the pixels of the region to its principal axis or the goodness of fit. The specific calculation method is as follows:

[0096] Calculation of aspect ratio (AR): Calculate the ratio of the length L of the longest side to the length W of the shortest side of the smallest bounding rectangle that can completely enclose the white area.

[0097] Linearity (Lin) calculation: Perform linear regression analysis on the coordinates of all pixels within the white area to obtain a best-fit line. Calculate the average vertical distance of all pixels to this best-fit line. Linearity It is defined as a quantity inversely proportional to the average distance: Subsequently, the determined aspect ratio and linearity are combined (i.e., a feature vector) and input into a pre-trained crack feature recognition model (e.g., a Support Vector Machine (SVM) or a small convolutional neural network (CNN). This model is trained using a database containing a large number of labeled samples through a series of standard supervised machine learning procedures. The training process specifically includes the following steps:

[0098] 1. Data Acquisition and Labeling: First, thousands or even tens of thousands of fluorescent magnetic particle inspection images are acquired under different working conditions (including different pipe diameters, different lighting conditions, and different wall surface conditions). Then, experienced non-destructive testing experts manually interpret these images, accurately selecting all target areas in the images and labeling each area with tags such as "real crack," "scratch," "weld edge," and "background noise."

[0099] 2. Feature Extraction: For all labeled regions (including real cracks and various non-crack interferences), their aspect ratio and linearity are calculated using the aforementioned method. Thus, each labeled region corresponds to a set of feature data (aspect ratio, linearity) and a true label.

[0100] 3. Dataset Partitioning: Randomly divide all labeled feature data sets into training and test sets according to a certain ratio (e.g., 80%:20%). The training set is used to train the model, enabling it to learn the association between features and labels; the test set is not involved in training at all, and is only used to evaluate the model's generalization ability and final performance after training is completed.

[0101] 4. Model Training and Optimization: Select a suitable classifier model (e.g., Support Vector Machine (SVM) or Decision Tree) and train it using the training set data. The goal of training is to find an optimal decision boundary that can best distinguish the feature combinations of "true cracks" from feature combinations of other perturbations. During training, methods such as cross-validation can be used to adjust the model's hyperparameters (e.g., the penalty coefficient C and kernel parameter gamma in SVM) to prevent overfitting and obtain optimal classification performance.

[0102] 5. Model Evaluation and Deployment: After training, an independent test set is used to evaluate the model's performance metrics, such as accuracy, precision, recall, and F1 score. Only when the model's performance meets the preset engineering requirements is the model finalized and deployed into the actual flaw detection software.

[0103] The model learns that real fatigue cracks typically exhibit extremely high aspect ratios and linearity. Based on this, the model analyzes the feature combinations of each individual white region in the input and outputs a judgment result (e.g., yes / no, or a confidence score), thereby achieving accurate identification of real cracks.

[0104] IV. Quantitative Assessment and Visualization Report:

[0105] The brightness and width of the actual crack in the preprocessed fluorescence image are analyzed. The fluorescence brightness and display width of the crack indication are positively correlated with the intensity of the leakage magnetic field at that location, and the intensity of the leakage magnetic field indirectly reflects the severity information such as the crack depth and aperture. The system refers to a "fluorescence feature-crack severity" correspondence model established in advance through numerous tests on defects of known sizes. The analyzed brightness and width values ​​are comprehensively calculated to quantify and assess the actual crack as an intuitive crack hazard index. In this embodiment, the model can be a linear weighted model, and the crack hazard index H can be calculated using the following formula:

[0106]

[0107] in, The average fluorescence intensity value is calculated along the crack indicator line; This is the average indicated width value; and These are the weighting coefficients corresponding to brightness and width, respectively. .

[0108] The values ​​of these two weighting coefficients reflect the relative importance of fluorescence intensity and width when assessing the hazard of a crack. They are determined as follows: A large number of standard crack samples with known depths and openings are tested, and the statistical correlation between their fluorescence intensity and width and the actual hazard level (e.g., obtained through metallographic analysis) is analyzed. If a stronger correlation is found between intensity and hazard level, then intensity is assigned a higher weighting. A larger value (e.g., 0.7); conversely, if the width is more strongly correlated with the hazard level, assign a smaller value. A relatively large value. In a typical application, it can be set to... This indicates that, in the overall evaluation, fluorescence brightness is considered slightly more important than the indication width.

[0109] Finally, combining the location information carried by the fluorescent image, the actual crack was visualized on the two-dimensional unfolded diagram and three-dimensional model of the pressure steel pipe. This visualization not only accurately includes the location, direction, and length of the actual crack in the global coordinate system of the steel pipe, but also uses different colors (e.g., green represents low risk, yellow represents medium risk, and red represents high risk) or symbols to intuitively distinguish the hazard level defined by the crack hazard index, thus forming a comprehensive flaw detection report that includes the precise location, geometric shape, and quantified hazard level of the defect.

[0110] Example 2

[0111] Reference Figure 4 These are two embodiments of the present invention. This embodiment provides a fluorescent magnetic particle imaging flaw detection device for the diagnosis of cracks in pressure steel pipes. This device is used to perform the method in Embodiment 1.

[0112] Specifically, it includes:

[0113] A magnetization device is used to magnetize the target detection area on the inner wall of a pressure steel pipe.

[0114] A fluorescent magnetic powder spraying system is used to spray a suspension of fluorescent magnetic powder onto a target detection area that has been magnetized by a magnetization device.

[0115] The image acquisition unit, including an ultraviolet light source and an industrial camera, is used to excite and acquire fluorescence images of the target detection area after the target detection area is sprayed with fluorescent magnetic powder suspension.

[0116] The processing unit, electrically connected to the image acquisition unit, is used to process fluorescent images to identify cracks in the pressure steel pipe; wherein, along the travel direction of the moving platform, the magnetization device, the fluorescent magnetic powder spraying system, and the image acquisition unit are arranged in sequence.

[0117] The mobile platform is used to carry the magnetization device, the fluorescent magnetic powder spraying system and the image acquisition unit, and can crawl on the inner wall of the pressure steel pipe. The mobile platform includes a motor and a magnetic drive wheel, which is driven by differential rotation. The magnetic drive wheel is composed of alternating permanent magnet segments and knurled metal wheel segments.

[0118] The magnetization device consists of a permanent magnet with a high magnetic energy product (such as a neodymium iron boron magnet) and a soft magnetic yoke (such as electrical pure iron) for guiding and confining the magnetic flux. The contact surface of its magnetic pole feet is machined into a replaceable arc-shaped module that matches the curvature of the inner wall of the steel pipe being tested. Simultaneously, the magnetization device is connected to a robot platform via an elastic floating suspension mechanism to ensure that the magnetic pole feet can always flexibly and dynamically conform to the pipe wall during dynamic movement, maintaining a stable magnetization effect.

[0119] The fluorescent magnetic powder spraying system includes a storage tank for storing the suspension, a diaphragm pump, and fan-shaped nozzles. The system achieves precise synchronization between the spray flow rate and the robot's travel speed; its control law is designed as a linear relationship. in, for The target flow rate of the diaphragm pump at any given time. for The real-time movement speed of the robot. The linkage coefficient K has a clear physical meaning, and its value is determined by the effective coverage width of the nozzle. With the desired liquid film thickness product This is determined and preset in the robot's central controller. The controller continuously reads the speed... and multiply it by a coefficient This allows for dynamic calculation and command of the diaphragm pump to output the corresponding flow rate. .

[0120] The image acquisition unit includes an industrial camera with a global shutter function and multiple high-power ultraviolet LEDs arranged in a ring around its lens. Its control circuit is synchronized with the camera's exposure signal to achieve an "exposure-as-flash" working mode, thereby acquiring high-sharp fluorescence images without motion blur.

[0121] The processing unit can be a high-performance embedded computer mounted on a robot platform, or a ground-based industrial control computer connected to the robot via a data cable. The software program running within this unit implements a complete image processing and analysis workflow. This workflow sequentially performs the following operations on the acquired fluorescence images: First, geometric correction and brightness equalization are performed on the images; then, through color space conversion, threshold segmentation, and morphological operations, noise is separated and filtered from the background to obtain independent fluorescence foreground regions; next, the morphological feature parameters of the foreground regions are calculated and input into a crack feature recognition model to confirm whether they are real cracks; finally, the confirmed real cracks are quantitatively evaluated, and a comprehensive flaw detection report is generated based on the location information.

[0122] The aforementioned functional units are integrated and carried on a mobile platform. This mobile platform carries the magnetization device, the fluorescent magnetic powder spraying system, and the image acquisition unit, and can crawl along the inner wall of the pressure steel pipe. The mobile platform includes a motor and magnetic drive wheels, which are driven by differential rotation. The magnetic drive wheels are structurally composed of alternating permanent magnet segments and knurled metal wheel segments.

[0123] As one specific implementation, the mobile platform provides the entire device with the ability to stably adhere and reliably move within the inner wall of the steel pipe in all orientations (including horizontal, inclined, vertical, and even inverted).

[0124] The drive system employs a differential drive layout. The mobile platform includes motors and magnetic drive wheels. Differential control of the motors enables the platform to move forward, backward, and turn in place. In a preferred embodiment, the mobile platform includes four motors and four magnetic drive wheels, forming a one-to-one drive relationship between the four motors and the four magnetic drive wheels, thereby providing strong driving torque and flexible steering capability.

[0125] The magnetic drive wheel is structurally composed of alternating permanent magnet segments and knurled metal wheel segments. This arrangement allows the adsorption function and the driving function to be structurally separated.

[0126] The main function of the permanent magnet segment is to provide a strong, perpendicular suction force to the tube wall, firmly attaching the weight of the entire device to the inner wall of the steel tube, which is a safety guarantee to prevent the robot from falling.

[0127] The knurled metal wheel segment is in direct contact with the tube wall. Its surface is knurled to increase the coefficient of friction. This segment is driven to rotate by a motor, providing the robot with the traction force required for forward movement or turning.

[0128] Ensure that the robot can move efficiently and flexibly while possessing strong suction power.

[0129] In summary, this invention improves detection efficiency and reduces the subjectivity of manual visual inspection by automating and standardizing the flaw detection process. At the same time, by performing multi-stage image processing and intelligent recognition on fluorescent images, it can quantitatively assess the geometric shape and severity of cracks, and combine location information to achieve precise defect localization, thereby improving the objectivity and traceability of diagnostic results.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A fluorescent magnetic particle imaging method for diagnosing cracks in pressure steel pipes, characterized in that, include: S1: Magnetize the target detection area on the inner wall of the pressure steel pipe to establish a closed magnetic field within the pipe wall of the target detection area; S2: A fluorescent magnetic powder suspension is sprayed onto the target detection area after the magnetization treatment, causing the fluorescent magnetic powder to accumulate at the leakage magnetic field generated by the crack in the target detection area; the fluorescent magnetic powder suspension is driven by a diaphragm pump and uniformly sprayed onto the target detection area by a fan-shaped nozzle; at the same time, the speed signal output by the encoder of the motor driving the mobile robot is acquired in real time, and the output flow rate of the diaphragm pump is synchronously adjusted according to the speed signal according to a preset ratio, so as to ensure that the amount of fluorescent magnetic powder suspension coverage on the unit area of ​​the pipe wall remains consistent under different travel speeds; S3: The target detection area is excited by an ultraviolet light source, and a fluorescence image of the target detection area is acquired by an image acquisition unit. The fluorescence image carries position and time information. An industrial camera with a global shutter function is used for image acquisition. The ultraviolet light source in a ring around the camera lens is controlled synchronously so that the ultraviolet light source is lit and emits light only during the exposure of the industrial camera, thereby obtaining a clear fluorescence image with no motion blur and uniform excitation light. S4: The fluorescence image is processed to identify cracks in the pressure steel pipe. The processing includes: first, applying a pre-generated geometric distortion lookup table to the fluorescence image to correct the image distortion introduced by the combined action of the camera's wide-angle lens and the curved inner wall of the pressure steel pipe; then, applying a pre-generated brightness distribution lookup table to the geometrically corrected image to compensate for brightness differences caused by uneven light source distribution; then, segmenting independent fluorescent foreground regions based on fluorescence color characteristics within a preset color space; finally, calculating the morphological characteristic parameters of the independent fluorescent foreground regions to determine whether they are real cracks.

2. The fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to claim 1, characterized in that, In step S1, the specific operation is as follows: the arc-shaped inner wall of the pressure steel pipe is magnetized by a magnetizing device carried on a mobile platform. During operation, the magnetic pole feet of the magnetizing device are kept in close contact with the arc-shaped inner wall. The magnetizing device is composed of a permanent magnet and a soft magnetic material yoke, and the contact surface of the magnetic pole feet is pre-processed into an arc surface that matches the curvature of the inner wall.

3. The fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to claim 1, characterized in that, The process of segmenting an independent fluorescent foreground region based on the fluorescent color characteristics within a preset color space in step S4 is specifically as follows: the pre-processed image is converted from the RGB color space to the HSV color space, and binarized according to preset hue and saturation thresholds; and morphological opening is performed on the binarized image to effectively remove background noise formed by isolated magnetic powder particles, thereby obtaining the independent white region.

4. The fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to claim 3, characterized in that, The process of determining whether a region is a real crack in step S4 is as follows: Calculate the aspect ratio and linearity of each independent white region in the processed image; input the determined aspect ratio and linearity as a combination into a pre-trained crack feature recognition model, which then determines whether the independent white region is a real crack.

5. The fluorescent magnetic particle imaging flaw detection method for crack diagnosis of pressure steel pipes according to claim 4, characterized in that, After identifying the crack as a real crack, the following operations are also included: analyzing the brightness and width of the real crack in the preprocessed fluorescence image to quantify and evaluate the real crack as a crack hazard index; and visually marking the real crack on the two-dimensional unfolded diagram and three-dimensional model of the pressure steel pipe by combining the location information carried by the fluorescence image.

6. A fluorescent magnetic particle imaging flaw detector for crack diagnosis of pressure steel pipes, characterized in that, The apparatus is used to perform the method according to any one of claims 1 to 5, specifically comprising: A magnetization device is used to magnetize the target detection area on the inner wall of a pressure steel pipe. A fluorescent magnetic powder spraying system is used to spray a fluorescent magnetic powder suspension onto the target detection area that has been magnetized by the magnetization device. The image acquisition unit includes an ultraviolet light source and an industrial camera, used to excite and acquire fluorescence images of the target detection area after the target detection area is sprayed with fluorescent magnetic powder suspension; The processing unit, electrically connected to the image acquisition unit, is used to process the fluorescent image to identify cracks in the pressure steel pipe; wherein, along the travel direction of the moving platform, the magnetization device, the fluorescent magnetic powder spraying system, and the image acquisition unit are arranged in a fixed sequence.

7. The fluorescent magnetic particle imaging flaw detector for crack diagnosis of pressure steel pipes according to claim 6, characterized in that, Also includes: The mobile platform is used to carry the magnetization device, the fluorescent magnetic powder spraying system, and the image acquisition unit, and can crawl on the inner wall of the pressure steel pipe; wherein, the mobile platform includes a motor and a magnetic drive wheel, and is driven by differential rotation; the magnetic drive wheel is composed of alternating permanent magnet segments and knurled metal wheel segments.

Images