Pipeline magnetic flux leakage - residual magnetic defect detection device and inversion method

Abstract

This invention discloses a pipeline leakage magnetic field and residual magnetic field defect detection device and inversion method, comprising: a detection shell, a drive shell, a detection module, and a drive module; the detection shell and the drive shell are connected by a universal joint to facilitate turning at pipeline bends; several detection modules are circumferentially spaced inside the detection shell for detecting leakage magnetic field and residual magnetic field signals of pipeline defects; four drive modules are circumferentially spaced inside the drive shell to provide speed for pipeline crawling, enabling dynamic and stable movement on the pipeline during defect detection. Therefore, this invention integrates leakage magnetic field and residual magnetic field detection into a single detection device, acquiring both types of detection data simultaneously in a single detection, eliminating the need for secondary detection, simplifying the detection process, improving efficiency, reducing pipeline downtime, and lowering equipment investment and labor costs.

Description

Technical Field

[0001] This invention relates to the fields of mechanical structure design and non-destructive testing, specifically to a pipeline leakage magnetic flux-residual magnetic defect detection device and inversion method. Background Technology

[0002] Currently, pipeline / tank defect detection mainly relies on single sensing technologies. For example, traditional magnetic detection uses fluxgate sensors to identify abnormalities in metal structures, or analyzes ultrasonic guided waves or traditional impact vibration signals to determine defects.

[0003] However, existing technologies have the following limitations: 1) Single detection dimension: Traditional magnetic detection is not sensitive enough for micro-cracks, ultrasonic guided wave detection relies on coupling agent, and the hammer of traditional vibration detection relies on free fall, which is too large and difficult to achieve multi-dimensional data complementarity when used independently.

[0004] 2) Poor adaptability: Pipe inspection devices are mostly designed to wrap around the pipe, making it difficult to adapt to various pipe diameters. Wheel sets are mostly designed in one direction or fixed angle, lacking flexible posture adjustment capabilities. Summary of the Invention

[0005] The present invention aims to at least solve one of the technical problems existing in the prior art. Therefore, in response to the above-mentioned problems, the object of the present invention is to provide a pipeline leakage magnetic flux-residual magnetic flux defect detection device and inversion method capable of dynamically detecting defects in buried pipelines.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a pipeline leakage magnetic flux-residual magnetic flux defect detection device, comprising a detection housing, a drive housing, a detection module, and a drive module; The detection housing and the drive housing are connected by a universal joint, which facilitates changing direction and turning at pipe bends; The detection housing is provided with several detection modules spaced apart circumferentially inside, for detecting leakage magnetic field signals and residual magnetic field signals of pipeline defects; The drive housing has four drive modules spaced circumferentially inside, providing speed for the external crawling of the pipeline, enabling the detection device to move dynamically and smoothly on the pipeline when detecting defects.

[0007] In some possible implementations, each of the detection modules includes a probe holder, two support bases, two permanent magnets with opposite polarities, two detection probes, and two support wheels, wherein: Two permanent magnets are fixedly mounted on the probe bracket. A first detection probe and a second detection probe are located between the two permanent magnets and at their tails. The first detection probe is used to detect the leakage magnetic field signal of the pipeline defect, and the second detection probe is used to detect the residual magnetic field signal of the pipeline defect. Two support seats are spaced apart on the top of the probe bracket for fixing to the inner wall of the detection housing. Two support wheels are also provided at the bottom of the probe bracket.

[0008] In some possible implementations, both the first detection probe and the second detection probe employ triaxial Hall sensors to acquire leakage magnetic flux X, Y, Z triaxial signals and residual magnetic flux X, Y, Z triaxial signals of pipeline defects.

[0009] In some possible implementations, each of the drive modules includes a mounting frame, a drive motor, an axle, a timing belt, a wheel, two fixed supports, two V-shaped axle fixing rods, and two shock absorbers, wherein: The two V-shaped wheel axle fixing rods are connected to the wheel via the wheel axle. The drive motor and the wheel axle are both mounted on the fixing frame. The drive motor drives the wheel to rotate via the synchronous belt and the wheel axle. One end of each V-shaped wheel axle fixing rod is connected to the corresponding fixing support, and the other end of each V-shaped wheel axle fixing rod is used to install the wheel axle. At the corner of each V-shaped wheel axle fixing rod, a shock absorber is connected to the corresponding fixing support to form a wheel axle limiting structure. When in the working position, the shock absorber is in a certain compressed state. When encountering a defect or protrusion, the shock absorber will extend or compress to keep the wheel pressed tightly against the pipe and maintain forward friction. The two V-shaped wheel axle fixing rods and the wheel axle form a crank-rocker mechanism to ensure that the power of the drive motor can be stably transmitted to the wheel when the wheel crosses an obstacle.

[0010] In some possible implementations, the front end of the inner wall of the detection housing is provided with several rollers at intervals to facilitate the movement of the detection housing on the pipeline.

[0011] Secondly, the present invention also provides a method for detecting pipeline magnetic flux leakage and residual magnetic defects, comprising: The pipeline leakage magnetic field-residual magnetic field defect detection device is fitted onto the pipeline to be inspected; The leakage magnetic field and residual magnetic field signals of pipeline defects are obtained through the detection module; The acquired leakage magnetic field-residual magnetic field signal is converted into a two-dimensional image of the magnetic signal using GAF; By inputting a two-dimensional image of the magnetic signal into the defect classification model, pipeline defects can be classified. By inputting two-dimensional images of magnetic signals of known defect categories into a size inversion network, the three-dimensional dimensions of pipeline defects can be inverted.

[0012] In some possible implementations, the defect classification model uses a ResNet101 network to identify pipeline defects, and the process is as follows: Two-dimensional images of magnetic signals are obtained by converting the leakage magnetic flux (X, Y, Z axes) and residual magnetic flux (X, Y, Z axes) signals of pipeline defects into six single-channel two-dimensional images via GAF. Six single-channel two-dimensional images are stitched together along the channel dimension to form a multi-channel fused feature map, which is then input into the ResNet101 network. The process of pipeline defect identification using the ResNet101 network is as follows: The ResNet101 network consists of five convolutional stages, Conv1 to Conv5. In Conv1, the number of kernel channels is increased from 3 to 6. Conv2x to Conv5x are all composed of a series of bottleneck residual blocks. Each block contains three convolutional layers of 1×1, 3×3, and 1×1, which are used to distinguish the deep discriminative features of groove defects, hole defects, and crack defects in the pipeline. After passing through a global average pooling layer, it is connected to a final fully connected classification layer with 3 output neurons, corresponding to the three defect categories of the pipeline. The Softmax function is used to output high-confidence defect category labels.

[0013] In some possible implementations, a two-dimensional image of the magnetic signal of a known defect type is input into a size inversion network to realize the inversion of the three-dimensional size of the pipe defect. The process is as follows: The ConvNeXt network was chosen as the size inversion network. The input to the size inversion network is a multi-channel fused feature map that is the same as the ResNet101 network. The multi-channel fused feature map corresponds to the leakage magnetic flux X, Y, Z axis signals and the residual magnetic flux X, Y, Z axis signals of the pipe defect, respectively. The predicted dimensions of pipe defects are output through a size inversion network. The process is as follows: First, the multi-channel fused feature map is downsampled and embedded into high-dimensional features by a Patchify layer; Subsequently, the process is carried out through four phases of ConvNeXt blocks, with downsampling performed before each phase to gradually expand the perception of abstract high-level semantic information. Finally, after global average pooling, the data is fed into the multilayer perceptron regression head, which outputs the predicted size of the pipeline defect.

[0014] Thirdly, the present invention also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the method described thereon.

[0015] Fourthly, the present invention also provides a computer-readable storage medium for storing one or more programs, characterized in that the one or more programs include computer instructions for causing a computer to perform the method.

[0016] Because the present invention adopts the above technical solution, it has the following characteristics: 1. Currently, most buried pipeline defect detection uses single Hall effect detection. This invention, targeting minute defects, combines residual magnetic field detection and magnetic flux leakage detection to make defect identification more accurate and reduce misjudgments. Single magnetic flux leakage detection mainly determines the existence of defects by the amount of magnetic field leakage, but it is difficult to accurately distinguish the type of defects. Magnetic flux leakage detection can quickly locate the defect location and provide quantitative information such as the depth and length of the defect, but it cannot effectively determine whether the defect is corrosion, crack, or mechanical scratch. Residual magnetic field detection can help determine the nature of defects by analyzing the differences in the residual magnetic field distribution of pipeline materials. For example, the stress concentration caused by cracks will lead to local residual magnetic anomalies, while the residual magnetic field change pattern of corrosion defects is different. Combining the two can significantly reduce the probability of misjudging cracks as corrosion. With a wider coverage of defect types and no blind spots, the two technologies complement each other in their detection principles, enabling them to cover tiny or special defects that are difficult to identify with magnetic flux leakage detection alone. Magnetic flux leakage detection is highly sensitive to volumetric defects (such as large-area corrosion and pores), but has a lower detection rate for shallow and narrow microcracks (such as stress corrosion cracks). Remanent magnetization detection is more sensitive to damage such as internal stress concentration and microcracks in materials. Even if the defect does not form an obvious "volume loss", it can capture the signal through changes in remanent magnetization. The combination of the two technologies can achieve full coverage of all types of defects from macroscopic corrosion to microscopic cracks.

[0017] 2. This invention uses a drive motor to dynamically detect defects in buried pipelines. The drive module is equipped with four wheels with independent suspension systems. Each wheel has an independent suspension system and is not directly connected to other suspension systems. Compared with the traditional axle suspension system, it has better suspension stability, better pipeline surface adaptability, and better handling. The independent suspension system can significantly reduce bumps and swaying when dealing with uneven surfaces, improve grip and stability, and make it more stable when turning and changing lanes.

[0018] 3. This invention combines magnetic flux leakage and residual magnetism detection, which can reduce repetitive operations and lower detection and time costs. If a suspected defect is found by single magnetic flux leakage detection, it is often necessary to use other detection technologies (such as ultrasonic testing) for verification, which increases the number of detection steps and time. Magnetic flux leakage and residual magnetism detection are integrated into the same detection device, and two types of detection data can be obtained at the same time in one test, eliminating the need for secondary testing, simplifying the detection process, improving efficiency, reducing pipeline downtime, and lowering equipment investment and labor costs.

[0019] In summary, this invention can be widely applied to non-destructive testing of pipeline defects. Attached Figure Description

[0020] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings: Figure 1 This is a schematic diagram of the pipe leakage magnetic flux-residual magnetic flux defect detection device according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the overall structure of the detection module of the pipeline leakage magnetic flux-residual magnetic flux defect detection device according to an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the detection module structure according to an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of the overall structure of the driver module according to an embodiment of the present invention.

[0024] Figure 5 This is a schematic diagram of the driver module structure according to an embodiment of the present invention.

[0025] Figure 6 The image shows two-dimensional GAF feature images of the groove-shaped defect in three directions according to an embodiment of the present invention, wherein (a) is the leakage magnetic axis; (b) is the leakage magnetic circumferential direction; (c) is the leakage magnetic radial direction; (d) is the remanent magnetic axis; (e) is the remanent magnetic axis; and (f) is the remanent magnetic axis.

[0026] Figure 7 The two-dimensional GAF feature images of the hole-shaped defect in three directions are shown in the embodiment of the present invention, wherein (a) is the leakage magnetic axis; (b) is the leakage magnetic circumferential direction; (c) is the leakage magnetic radial direction; (d) is the remanent magnetic axis; (e) is the remanent magnetic axis; and (f) is the remanent magnetic axis.

[0027] Figure 8 The image shows two-dimensional GAF feature images of crack-type defects in three directions according to an embodiment of the present invention, wherein (a) is the leakage magnetic axis; (b) is the leakage magnetic circumferential direction; (c) is the leakage magnetic radial direction; (d) is the remanent magnetic axis; (e) is the remanent magnetic axis; and (f) is the remanent magnetic axis.

[0028] Figure 9 This is a schematic diagram of the structure of three ResNet101 feature fusion models in an embodiment of the present invention.

[0029] Figure 10 This is a schematic diagram of the defect multimodal detection process according to an embodiment of the present invention. Detailed Implementation

[0030] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0031] Although terms such as first, second, third, etc., may be used in this document to describe multiple elements, components, regions, layers, and / or segments, these elements, components, regions, layers, and / or segments should not be limited by these terms. These terms may be used only to distinguish one element, component, region, layer, or segment from another. Unless the context clearly indicates otherwise, terms such as "first," "second," and other numerical terms used herein do not imply order or sequence. Therefore, the first element, component, region, layer, or segment discussed below may be referred to as the second element, component, region, layer, or segment without departing from the teachings of the exemplary embodiments.

[0032] For ease of description, spatial relative terms may be used in the text to describe the relationship of one element or feature relative to another element or feature as shown in the figure. These relative terms include, for example, "inside," "outside," "middle," "outer," "below," "above," etc. Such spatial relative terms are intended to include different orientations of the device in use or operation, other than those depicted in the figure.

[0033] Existing pipeline defect detection methods suffer from limitations such as limited detection dimensions and poor adaptability to movement. This invention provides a pipeline magnetic flux leakage and residual magnetic field defect detection device and inversion method, comprising a detection housing, a drive housing, detection modules, and drive modules. The detection housing and drive housing are connected by a universal joint, facilitating directional changes at pipeline bends. Several detection modules are spaced circumferentially within the detection housing to detect the magnetic flux leakage and residual magnetic field signals of pipeline defects. Four drive modules are spaced circumferentially within the drive housing to provide speed for pipeline crawling, enabling the detection device to move dynamically and smoothly along the pipeline during defect detection. Therefore, this invention integrates magnetic flux leakage and residual magnetic field detection into a single device, acquiring both types of data in a single detection, reducing pipeline downtime and lowering equipment investment and labor costs.

[0034] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0035] Example 1: As Figure 1 , Figure 2 As shown, the pipeline leakage magnetic flux-residual magnetic flux defect detection device provided in this embodiment includes a detection housing 1, a drive housing 2, a detection module 3, and a drive module 4.

[0036] Both the detection housing 1 and the drive housing 2 are made of aluminum alloy 7075 through sheet metal processing, turning, milling, drilling, welding and other processes. The detection housing 1 and the drive housing 2 are connected by a universal joint 5, which facilitates turning at pipe bends.

[0037] Several detection modules 3 are arranged at intervals along the circumference inside the detection housing 1 to detect the leakage magnetic field signal and residual magnetic field signal of pipeline defects; At least four drive modules 4 are arranged circumferentially inside the drive housing 2 to provide speed for the external crawling of the pipeline, so that it can perform dynamic and stable detection on the pipeline when detecting defects.

[0038] In a preferred embodiment of the present invention, such as Figure 3 As shown, each detection module 3 includes a probe bracket 31, two permanent magnets 32, two detection probes 33, two support bases 34, and two support wheels 35.

[0039] Two permanent magnets 32 with opposite polarities are fixedly mounted on the probe bracket 31. A first detection probe 33 and a second detection probe 33 are located between the two permanent magnets 32 and at their tails. The first detection probe 33 is used to detect the leakage magnetic field signal of the pipeline defect, and the second detection probe 33 is used to detect the residual magnetic field signal of the pipeline defect. Support seats 34 are spaced apart on the top of the probe bracket 31 for fixing it to the inner wall of the detection housing 1. The probe bracket 31 is also equipped with two support wheels 35. The function of the support wheels 35 is to maintain a constant lifting distance between the detection module and the outer wall of the pipeline, preventing the detection probe from colliding with the pipeline wall or causing signal distortion due to unevenness of the pipeline surface (such as welds, corrosion pits, protrusions). The probe bracket 31 and the support seats 34 are mainly used to provide stable positions for the permanent magnets 32 and the detection probes 33.

[0040] Furthermore, the two permanent magnets 32 can be U-shaped permanent magnets, and the magnetic pole spacing between the two permanent magnets 32 is 60mm. This is an example, but not limited to this.

[0041] Furthermore, the distance between the detection module 3 and the pipe wall is 2-5mm to prevent the detection module 3 from colliding and wearing down when passing through weld seams or protruding defects during operation.

[0042] Furthermore, both the first detection probe 33 and the second detection probe 33 can be equipped with an MXL90393 triaxial Hall sensor to acquire leakage magnetic flux X, Y, Z axis and residual magnetic flux X, Y, Z axis signals of pipeline defects.

[0043] It should be noted that each detection module 3 can perform independent detection, facilitating timely replacement of damaged modules. In use, the detection module 3 can be viewed as a small vehicle. The permanent magnet 32 ​​is used to generate an induced magnetic field in the pipeline. In the direction of travel, a triaxial Hall sensor receives the induced magnetic field generated by the pipeline to determine pipeline defects, enabling the detection of pipelines of different sizes. The front end of the detection module 3 is the magnetic flux leakage detection section, and the rear end is the residual magnetism detection section. The magnetic field required for magnetic flux leakage detection is generated by two permanent magnets arranged in opposite directions. The triaxial Hall sensor is installed between the two permanent magnets to perform magnetic flux leakage detection. The triaxial Hall sensor for residual magnetism detection is installed at the rear of the detection module. After the permanent magnets in the magnetic flux leakage system pass a defect, the residual magnetic signal generated by the defect can be detected.

[0044] In a preferred embodiment of the present invention, a plurality of rollers 6 are provided at intervals on the front end of the inner wall of the detection housing 1 to facilitate movement on the pipeline.

[0045] In a preferred embodiment of the present invention, such as Figure 4 , Figure 5As shown, drive module 4 mainly utilizes a four-bar linkage to act as a shock absorber and buffer, preventing the detection device from stopping when encountering pipeline defects. Each drive module 4 includes a fixed frame 41, a drive motor 42, a timing belt 43, an axle 44, a wheel 45, a V-shaped axle fixing rod 46, a fixed support 47, and a shock absorber 48. The drive motor 42 and the axle 44 are both mounted on the fixed frame 41, and the drive motor 42 is connected to the wheel 45 via the timing belt 43. The two V-shaped axle fixing rods 45 are connected to the wheel via the axle. One end of each V-shaped axle fixing rod 46 is connected to the fixed support 47, and the other end is fitted with the wheel 45. Specifically, one end of the V-shaped axle fixing rod 46 has a groove with the same width as the diameter of the axle 44, allowing the axle 44 to slide within the groove. At the corner of each V-shaped axle fixing rod 46, it is connected to the fixed support 47 via a shock absorber 48 to form a fixed position. When in the working position, the shock absorber 48 is under a certain compression state. When encountering defects or protrusions, the shock absorber 48 will extend or compress to keep the wheel pressed tightly against the pipe and maintain forward friction. Among them, the V-shaped wheel axle fixing rod 46 and the shock absorber 48 serve as wheel axle limiters, and the two V-shaped wheel axle fixing rods 46 and the wheel axle 44 form a crank-rocker mechanism to ensure that the power of the drive motor can be stably transmitted to the wheel when the wheel crosses an obstacle.

[0046] Furthermore, each wheel 45 uses an aluminum alloy rim with an outer polyurethane coating, which ensures grip while also providing shock absorption and wear resistance.

[0047] Furthermore, the drive motor 42 can be a 3510 brushless motor with a gearbox.

[0048] Example 2: The pipeline leakage magnetic flux-residual magnetic flux defect detection device provided in this example obtains the leakage magnetic flux-residual magnetic flux signal of the defect. Further judgment of the size and type of the pipeline defect is needed. Therefore, a three-dimensional inversion of the pipeline defect is performed based on the triaxial leakage magnetic flux-residual magnetic flux signal. The specific process is as follows: S1. The one-dimensional magnetic signal of the triaxial leakage magnetic field-residual magnetic field signal is converted into a two-dimensional image signal of this model using the Gramian Angular Field (GAF). S2. Input the two-dimensional image of the magnetic signal into the defect classification model to achieve preliminary defect classification; S3. Use the two-dimensional image of the magnetic signal of the known defect category as the input of the size inversion network to achieve high-precision inversion of the three-dimensional dimensions of the pipeline defect in length, width and depth.

[0049] In summary, the two-stage architecture effectively combines the complementary advantages of visual and magnetic signals, achieving accurate quantization inversion from recognition to quantization. The 3D inversion performance under different networks is compared, with mean absolute error (MAE) and mean relative error (MRE) used as evaluation metrics for 3D inversion.

[0050] In a preferred embodiment of the present invention, the vast majority of data acquired in nondestructive testing is one-dimensional time-domain signal. However, one-dimensional signals can only provide very limited feature information, and the detection process requires interpreting univariate time-series data from other dimensions. For example, when using MFL technology for pipeline magnetic flux leakage detection, it is necessary to convert time features into spatial features to better observe defect features. GAF can convert one-dimensional signals into two-dimensional images. GAF retains both the complete information of the signal and its time dependence during the signal-to-image conversion process. Specifically, the process of GAF converting a one-dimensional signal into a two-dimensional image is as follows: S11. Transform the time-series data to the interval [-1, 1]. Assume the time-domain data is... The normalized value is denoted as To ensure that all values ​​fall within the domain of the arccosine function, let the normalized one-dimensional time-series signal be... Time domain data Normalized to [ The formula for the interval [1, 1] is:

[0051] S12. Convert the normalized values ​​to polar coordinates:

[0052] in, Polar angle, Polar radius, For timestamps, N This is the number of all time points contained in the time series data. This polar coordinate transformation is bijective, meaning that in... Within the interval, polar angle With numerical values It is a one-to-one correspondence. Among them, the polar angle... It encodes the amplitude and polarity of the original signal. This preserves the temporal position of the time point in the sequence proportionally. The polar coordinate transformation encoding process includes both of these pieces of information without any loss of information.

[0053] S13. GAF defines a special inner product form, which generates the final two-dimensional image matrix G by calculating the cosine of the sum of the polar angles at different time points, where: the normalized GAF ​​matrix is ​​shown below:

[0054] Specifically, the first in the matrix i line, number j The elements of a column are defined as follows:

[0055] in, This represents the first... in polar coordinate space. i The time point and the j The "total deflection angle" at each point in time relative to the origin.

[0056] From formulas (2) and (3), we can obtain: (6); (7); Then, using trigonometric identities:

[0057] Combining (5), (6), and (7), we get: (9).

[0058] Substituting into (4) yields the complete two-dimensional image matrix G.

[0059] Then, the normalized values Encoded as angles in polar coordinates timestamp Encoded as radius r i The definition of GAF is: (10); in, timestamp representing a point , N It is the number of all time points contained in the time series data. Each time series data point contains two pieces of information. One is the normalized value of the data point. The other is its temporal position (polar radius). This preserves the temporal position of each time point in the sequence proportionally. The polar coordinate transformation encoding process includes both of this information without any loss. The inner product between two time points is the cosine of the sum of the polar angles of their polar coordinate transformations.

[0060] Equation (4) is the result of GAF, where time is encoded as the geometric dimension of a square matrix, and further decomposition of the inner product yields the expression: (12).

[0061] Compared to the original time series data, GAF adds an extra dimension of information, transforming each point in the time data into its correlation with other points over time.

[0062] As the above theory shows, GAF can transform one-dimensional time-series signals into two-dimensional image signals. To extract depth features from defect magnetic signals, the triaxial magnetic signals obtained from experimental measurements and simulations are converted into corresponding two-dimensional GAF feature maps to construct a dataset for three-dimensional inversion of magnetic signal defects. Two-dimensional GAF feature images of different types of defects in three directions (axial, radial, and circumferential) are shown below. Figures 6-8 As shown. By Figures 6-8 It can be seen that, compared with one-dimensional leakage and residual magnetic data signals, two-dimensional GAF images can obtain more information about different defects. The information content of two-dimensional images in three directions of each defect is richer, and the features are evenly distributed in the whole image. Two-dimensional images can better present the defect feature vector.

[0063] In a preferred embodiment of the present invention, the defect classification model is obtained by training a ResNet network; the training process is not detailed here. ResNet networks have better deep network construction capabilities, avoiding gradient vanishing and gradient explosion caused by excessive network depth, and also solving the degradation problem that occurs with increasing network depth. Compared to other ResNet networks, this embodiment uses ResNet101, which has a deeper network structure; its "bottleneck" design reduces the computation and number of parameters in the network.

[0064] This embodiment uses ResNet101 as the feature fusion network for triaxial magnetic flux leakage GAF images of pipe defects. The ResNet101 feature fusion model structures for groove defects, hole defects, and crack defects are as follows: Figure 9 As shown.

[0065] Specifically, the input to the ResNet101 network is the generated fused GAF ​​image. The leakage magnetic flux (X, Y, Z axes) and remanent magnetic flux (X, Y, Z axes) signals from the pipe defect are transformed into six single-channel two-dimensional images using Gram angular field transformation. These images are then stitched together along the channel dimension to form a multi-channel fused feature map of size 224×224×6, which serves as the network's unified input. Complementary information from the two physical fields is integrated at the input level. To adapt to the six-channel input, the first convolutional layer of the original ResNet101 is modified, increasing the number of channels in its kernel from 3 to 6. The main body of the ResNet101 network consists of five convolutional stages (Conv1 to Conv5), containing a total of 101 weighted layers. Conv2_x to Conv5_x are all composed of a series of bottleneck residual blocks stacked together. Each block contains three convolutional layers: 1×1, 3×3, and 1×1. First, the channel dimension is compressed to reduce computational cost. Then, feature extraction is performed in the low-dimensional space, and finally, it is expanded back to the high-dimensional space, improving computational efficiency while maintaining feature extraction capabilities. The deep network ensures effective gradient propagation through cross-layer identity shortcut connections, enabling it to learn deep discriminative features to distinguish three types of defects—grooves, holes, and cracks—from complex textures. After a global average pooling layer, the network connects to a final fully connected classification layer with 3 output neurons corresponding to the three defect categories. The Softmax function is used to output the probability of belonging to each category. The ResNet101 network is trained using weights pre-trained on the ImageNet dataset to accelerate convergence and improve generalization through transfer learning. The optimizer used is Adam, with an initial learning rate set to 1e-4 and adjusted using a multinomial decay strategy. The classification module is the key decision-making link in the entire defect quantization pipeline. Its output high-confidence defect category labels will be used as input to the subsequent size inversion network, enabling the inversion network to load more targeted inversion model parameters for the corresponding category, thereby achieving accurate defect inversion from identification.

[0066] In a preferred embodiment of the present invention, after accurately classifying the defect type, to achieve high-precision inversion of the three-dimensional dimensions of the defect, the inversion network of this embodiment adopts the ConvNeXt network, which is specifically optimized for regression tasks, and constructs a feature extraction and regression network with ConvNeXt as the core architecture. The input is a multi-channel GAF image corresponding to a defect of a known category determined by the defect classification model, aiming to directly regress the physical dimensions of the defect from the fused leakage-residual magnetism signals. The overall process is as follows: Figure 10As shown, the process employs deep feature extraction → feature fusion → regression prediction. The input to the inversion network is a six-channel GAF image, identical to the defect classification model, with a size of 224×224×6, corresponding to the leakage magnetic field X, Y, Z and the remanent magnetic field X, Y, Z signals, respectively. The inversion network first passes through a "Patchify" layer (i.e., a 4×4 convolution with a stride of 4), downsampling the image and embedding it into high-dimensional features. Subsequently, the features are processed through four phased ConvNeXt blocks, with downsampling performed before each phase to progressively expand the perception of abstract high-level semantic information. Finally, after global average pooling, the features are fed into a multilayer perceptron regression head, outputting the predicted size of the defect, including the corresponding model defect and its length, width, and height dimensions. The training process of the inversion network is not detailed here. The core of the inversion network is the ConvNeXt feature extraction backbone network, whose detailed parameter configuration is shown in Table 1. The network consists of four stages (Stage 1-4), with the number of channels increasing in each stage (96, 192, 384, 768) and the spatial resolution decreasing in each stage (56×56, 28×28, 14×14, 7×7).

[0067] Table 1. Main parameters of the ConvNeXt feature extraction backbone network

[0068] Compared to traditional convolutional networks and the visual Transformer, the ConvNeXt feature extraction backbone network adopts a modern design through pure convolutional operations. While maintaining the inherent advantages of convolutional locality and translational invariance, it also draws on the successful advantages of the Transformer. Employing designs such as large-kernel depthwise separable convolutions, inverse bottleneck structures, the GELU activation function, and layer normalization, it achieves an excellent balance between feature extraction capability and computational efficiency, making it suitable for learning robust regression features from GAF images with complex textures.

[0069] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In the description of this specification, the terms "a preferred embodiment," "furthermore," "specifically," "in this embodiment," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments in this specification. 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 can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A pipeline leakage magnetic flux-residual magnetic flux defect detection device, characterized in that, Includes a detection housing, a driver housing, a detection module, and a driver module; The detection housing and the drive housing are connected by a universal joint, which facilitates changing direction and turning at pipe bends; The detection housing is provided with several detection modules spaced apart circumferentially inside, for detecting leakage magnetic field signals and residual magnetic field signals of pipeline defects; The drive housing has four drive modules spaced circumferentially inside, providing speed for the external crawling of the pipeline, enabling the detection device to move dynamically and smoothly on the pipeline when detecting defects.

2. The pipeline leakage magnetic flux-residual magnetic flux defect detection device according to claim 1, characterized in that, Each of the aforementioned detection modules includes a probe holder, two support bases, two permanent magnets with opposite polarities, two detection probes, and two support wheels, wherein: Two permanent magnets are fixedly mounted on the probe bracket. A first detection probe and a second detection probe are located between the two permanent magnets and at their tails. The first detection probe is used to detect the leakage magnetic field signal of the pipeline defect, and the second detection probe is used to detect the residual magnetic field signal of the pipeline defect. Two support seats are spaced apart on the top of the probe bracket for fixing to the inner wall of the detection housing. Two support wheels are also provided at the bottom of the probe bracket.

3. The pipeline leakage magnetic flux-residual magnetic flux defect detection device according to claim 2, characterized in that, Both the first and second detection probes employ triaxial Hall sensors to acquire leakage magnetic flux X, Y, Z triaxial signals and residual magnetic flux X, Y, Z triaxial signals of pipeline defects.

4. The pipeline leakage magnetic flux-residual magnetic flux defect detection device according to claim 2, characterized in that, Each of the aforementioned drive modules includes a fixed frame, a drive motor, an axle, a timing belt, a wheel, two fixed supports, two V-shaped axle fixing rods, and two shock absorbers, wherein: Both the drive motor and the axle are mounted on the fixed frame. The drive motor drives the wheel to rotate via the synchronous belt and the axle. The two V-shaped axle fixing rods are connected to the wheel via the axle. One end of each V-shaped axle fixing rod is connected to the corresponding fixed support, and the other end of each V-shaped axle fixing rod is used to mount the axle. At the corner of each V-shaped axle fixing rod, a shock absorber is connected to the corresponding fixed support to form an axle limiting structure. When in the working position, the shock absorber is in a certain compressed state. When encountering a defect or protrusion, the shock absorber will extend or compress to keep the wheel pressed tightly against the pipe and maintain forward friction. The two V-shaped axle fixing rods and the axle form a crank-rocker mechanism to ensure that the power of the drive motor can be stably transmitted to the wheel when the wheel crosses an obstacle.

5. The pipeline leakage magnetic flux-residual magnetic flux defect detection device according to claim 1, characterized in that, The front end of the inner wall of the detection housing is provided with several rollers at intervals to facilitate the movement of the detection housing on the pipeline.

6. A method for detecting magnetic flux leakage and residual magnetic defects in pipelines, characterized in that, include: The pipeline leakage magnetic flux-residual magnetic flux defect detection device as described in any one of claims 3 to 5 is fitted onto the pipeline to be inspected; The leakage magnetic field and residual magnetic field signals of pipeline defects are obtained through the detection module; The acquired leakage magnetic field-residual magnetic field signal is converted into a two-dimensional image of the magnetic signal using GAF; By inputting a two-dimensional image of the magnetic signal into the defect classification model, pipeline defects can be classified. By inputting two-dimensional images of magnetic signals of known defect categories into a size inversion network, the three-dimensional dimensions of pipeline defects can be inverted.

7. The pipeline leakage magnetic flux-residual magnetic flux defect detection method according to claim 6, characterized in that, The defect classification model uses a ResNet101 network to identify pipeline defects. The process is as follows: Two-dimensional images of magnetic signals are obtained by converting the leakage magnetic flux (X, Y, Z axes) and residual magnetic flux (X, Y, Z axes) signals of pipeline defects into six single-channel two-dimensional images via GAF. Six single-channel 2D images are stitched together along the channel dimension to form a multi-channel fused feature map, which is then input into the ResNet101 network. The process of pipeline defect identification using the ResNet101 network is as follows: The ResNet101 network includes five convolutional stages Conv1 to Conv5. In the convolutional stage Conv1, the number of channels in the convolutional kernel is adjusted from 3 to 6. The convolutional stages Conv2x to Conv5x are all composed of a series of bottleneck residual blocks stacked together. Each block contains three layers of convolutions: 1×1, 3×3, and 1×1, which are used to distinguish the deep discriminative features of groove defects, hole defects, and crack defects in pipelines. After passing through a global average pooling layer, a final fully connected classification layer is connected. The number of output neurons is 3, corresponding to the three defect categories of the pipeline, and the Softmax function is used to output high-confidence defect category labels.

8. The pipeline leakage magnetic flux-residual magnetic flux defect detection method according to claim 7, characterized in that, The two-dimensional image of the magnetic signal with known defect categories is input into the size inversion network to realize the inversion of the three-dimensional size of the pipeline defect. The process is as follows: The ConvNeXt network was chosen as the size inversion network. The input to the size inversion network is a multi-channel fused feature map that is the same as the ResNet101 network. The multi-channel fused feature map corresponds to the leakage magnetic flux X, Y, Z axis signals and the residual magnetic flux X, Y, Z axis signals of the pipe defect, respectively. The size prediction value of the pipeline defect is output by the size inversion network. The process is as follows: First, the multi-channel fused feature map is downsampled and embedded into high-dimensional features through a Patchify layer; then, it is processed through four stages of ConvNeXt blocks, with downsampling performed before each stage to gradually expand the perception of abstract high-level semantic information; finally, after global average pooling, it is fed into the multilayer perceptron regression head to output the size prediction value of the pipeline defect.

9. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include computer instructions for causing a computer to perform the method according to any one of claims 1-7.

Images