Steel rail electromagnetic thermal imaging detection system and method based on double-layer spiral magnetic conductive coil

Abstract

The invention discloses a steel rail electromagnetic thermal imaging detection system and method based on a double-layer spiral magnetic conductive coil. The system comprises an excitation coil assembly, an excitation source, water cooling equipment, a thermal infrared imager, a synchronous control module and a computer. The excitation coil assembly comprises a double-layer spiral concentric-square-shaped excitation coil, the double-layer spiral concentric-square-shaped excitation coil is of a two-turn closed structure, each turn of coil is in a rectangular spiral shape, the upper layer and the lower layer are arranged in parallel, and a plurality of U-shaped magnetizers arranged in parallel are arranged on the two sides of each layer of the double-layer spiral concentric-square-shaped excitation coil. The double-layer spiral concentric-square-shaped excitation coil is connected with the excitation source and the water cooling device, the thermal infrared imager is arranged above the excitation coil assembly and connected with the computer, and the synchronous control module is connected with the excitation source and the thermal infrared imager. Under the large lift-off distance, the eddy current excitation efficiency and the thermal signal response intensity can be improved, effective imaging and recognition of defects in dynamic advancing are guaranteed, and automatic detection and quantitative evaluation are achieved.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic thermal imaging nondestructive testing technology, and in particular to an electromagnetic thermal imaging inspection system and method for rails based on a double-layer helical magnetic coil. Background Technology

[0002] To ensure the safe operation of rails and trains, regular non-destructive testing (NDT) is required for in-service rails. Among various NDT methods, Eddy Current Pulsed Thermography (ECPT) is a multi-physics-based NDT technique. Compared to traditional methods such as ultrasound and X-ray, ECPT offers advantages such as high resolution, fast detection speed, and intuitive results. Furthermore, it is easily integrated with advanced image processing and intelligent algorithms to achieve automatic defect detection and identification.

[0003] In recent years, with the widespread application of artificial intelligence, research has attempted to introduce models such as convolutional neural networks and Transformers into thermal image recognition. However, the amount of data collected during electromagnetic thermal imaging rail inspection is typically enormous, with the vast majority being defect-free background images and only a very small number of frames containing defect information. If supervised learning methods such as object detection models are used, not only is extensive manpower required for defect annotation, but a large amount of defect-free data remains unutilized, resulting in resource waste and making it difficult to adapt to the high-frequency, low-defect-rate scenarios in actual engineering inspections. In contrast, anomaly detection methods that do not require annotation are more practically valuable, especially in the face of complex interference backgrounds and weak defect responses.

[0004] Meanwhile, in the dynamic inspection of large workpieces such as rails, electromagnetic thermal imaging technology faces the challenge of the lift-off effect. Lift-off refers to the distance between the sensor coil and the surface being measured. As the lift-off distance increases, the magnetic flux coupled into the specimen decreases, the induced eddy currents weaken, and the thermal signal response of the defect is significantly reduced. To obtain sufficient induction heating intensity, a very small lift-off distance (generally <10mm) is usually required. In static or low-speed scanning inspection, the coil can be placed close to the surface to improve eddy current heating efficiency. However, for high-speed moving rail inspection vehicles, the rail surface is difficult to keep flat under on-site conditions and is subject to vibration and bumps. If the lift-off is too small, the coil assembly may collide with the rail surface, posing a safety hazard. Therefore, in dynamic inspection, the lift-off must be increased to ensure safety, which significantly weakens the induced eddy current heating effect and increases the inspection difficulty.

[0005] Furthermore, introducing high-permeability materials as the magnetic core can improve the magnetic flux focusing ability of the coil, thereby enhancing the eddy current excitation effect. However, at high frequencies, some permeable materials generate significant self-heating, producing a bright background in infrared thermal images and masking the true thermal signal of the defect. This leads to a decrease in the signal-to-noise ratio of the detection system, reducing the accuracy of defect identification. In addition, most rail crack defects are located in the edge regions on both sides of the rail head, but the magnetic field distribution of traditional planar coils is concentrated in the central region, without special magnetic field focusing optimization for the rail edge areas. Therefore, under conventional design, the defect response at the rail head edge is weak, which cannot meet the thermal imaging detection requirements for crack defects mainly located at the edge. Summary of the Invention

[0006] To address the aforementioned problems in the dynamic detection of rail electromagnetic thermal imaging, this invention proposes a rail electromagnetic thermal imaging detection system and method based on a double-layer helical magnetic coil. This system significantly improves eddy current excitation efficiency and thermal signal response intensity at greater lift-off distances, ensuring effective imaging and identification of rail defects during dynamic movement. Furthermore, by employing an unsupervised anomaly detection algorithm applicable to electromagnetic thermal imaging infrared sequence images, it achieves automatic detection and quantitative assessment of defects, thereby resolving the aforementioned issues.

[0007] This application discloses a rail electromagnetic thermal imaging detection system based on a double-layer spiral magnetic coil, characterized in that it includes an excitation coil assembly, an excitation source, a water cooling device, an infrared thermal imager, a synchronous control module, and a computer; The excitation coil assembly includes a double-layer spiral-type excitation coil, which is a two-turn closed structure. Each turn of the coil is rectangular spiral-shaped, and the upper and lower layers are arranged in parallel. On both sides of each layer of the double-layer spiral-type excitation coil, there are multiple parallel U-shaped magnetic conductors, and the U-shaped opening of the U-shaped magnetic conductors faces the surface of the rail. The double-layer spiral excitation coil is connected to the excitation source and water cooling equipment. The infrared thermal imager is installed above the excitation coil assembly and connected to the computer. The synchronous control module is connected to the excitation source and the infrared thermal imager.

[0008] Preferably, the width of the double-layer spiral excitation coil covers the top surface of the rail head, the length extends along the rail direction, and the vertical lifting distance between the plane of the double-layer spiral excitation coil and the top surface of the rail is ≥30mm.

[0009] Preferably, the double-layer spiral excitation coil is wound with a hollow copper tube.

[0010] Preferably, the material of the U-shaped magnetic conductor is an iron-based nanocrystalline soft magnetic alloy.

[0011] Preferably, the excitation source is a high-frequency inverter power supply with an output frequency of 250~300kHz and a power of ≥3kW.

[0012] Preferably, the water-cooling device is a hollow copper tube with a double-layer spiral excitation coil through which circulating coolant is introduced, including a cooling pump, heat exchanger, and pipelines, forming a closed-loop circulation.

[0013] Preferably, the infrared thermal imager is a mid-wave cooled thermal imager with a wavelength of 3~5 µm.

[0014] Preferably, the synchronization control module includes a high-speed data acquisition card and a digital signal processor, used to coordinate the output of the excitation power supply with the synchronization of the infrared thermal imager acquisition.

[0015] This application also discloses a rail electromagnetic thermal imaging detection method based on a double-layer helical magnetic coil, implemented based on the aforementioned rail electromagnetic thermal imaging detection system, including the following steps: S1. Installation and positioning: Fix the excitation coil assembly to the bottom of the rail inspection vehicle, keeping the plane of the double-layer spiral excitation coil parallel to the top surface of the rail and the lifting distance ≥30mm; S2. Electromagnetic excitation: During the movement of the rail inspection vehicle, a high-frequency alternating current is supplied to the double-layer spiral excitation coil using an excitation source, and a focused eddy current is induced on the surface of the rail through a U-shaped magnetic conductor. S3. Thermal image acquisition: Use an infrared thermal imager to synchronously acquire the temperature field of the rail surface and continuously acquire thermal infrared image sequences at a preset frame rate. S4. Image Processing and Defect Recognition: An unsupervised anomaly detection algorithm is used to process the thermal image sequence and output the defect location.

[0016] Preferably, step S4 includes the following steps: S41. Divide the single-frame thermal image into image blocks and add positional encoding, then input the block into the Transformer encoder to extract the final feature vector. ; S42. For defect-free training images Add noise to generate a noisy image. :

[0017] in, For diffusion time step, It is Gaussian white noise. For the front The cumulative product of the step diffusion rates; S43. Train the conditional diffusion denoising network to minimize noise prediction error:

[0018] in, For network parameters, For the loss of the entire network, The expectation is given for all training samples, sampling noise, and time steps. For network prediction noise, This represents the squared error between the noise predicted by the network and the actual noise. S44. Reconstruct the image and calculate the error. After adding noise to the image to be tested, input it into the trained conditional diffusion denoising network, output the reconstructed image, and calculate the residual heatmap:

[0019] in, This indicates the degree of abnormality for each pixel. Two-dimensional pixel coordinates of the image To reconstruct the image; S45. When the mean value of the residual heatmap exceeds the threshold obtained from the statistical analysis of the residuals of the defect-free samples, a defect alarm is triggered, and the defect area is located through connected component analysis.

[0020] The beneficial effects of this invention are: 1. The double-layer spiral excitation coil of the present invention adopts a two-turn closed structure. The multiple turns of overlapping winding significantly enhance the excitation of the eddy current field. The double-layer coil forms more overlapping magnetic flux paths under the same power, generating a stronger induced magnetic field and current, thereby improving the eddy current heating efficiency.

[0021] 2. The present invention effectively concentrates the magnetic lines of force generated by the coil by arranging a U-shaped magnetic conductor array on both sides of the double-layer spiral excitation coil and guides the magnetic flux to the surface of the rail to be inspected, thus significantly improving the magnetic flux focusing capability.

[0022] 3. The coil of this invention uses a hollow copper tube conductor and is cooled by water, which greatly improves the heat dissipation capacity of the coil. The iron-based nanocrystalline soft magnetic alloy conductor has very low iron loss and very little self-heating under high-frequency alternating magnetic field. Therefore, it will not produce significant overheating background interference in infrared thermal images, thus ensuring the contrast of defect thermal signals.

[0023] 4. The rectangular geometric dimensions of the double-layer spiral excitation coil 101 of the present invention match the cross-section of the rail. The coil width can cover the width of the top surface of the rail, and the coil length is extended to achieve full lateral coverage excitation of the rail surface. The heating effect of the coil lasts for a longer time and can accumulate sufficient defect thermal signals under dynamic detection conditions, thereby improving the imaging clarity.

[0024] 5. The infrared thermal imaging unsupervised anomaly detection algorithm of the present invention utilizes defect thermal image signals to achieve automatic identification of defects, thereby improving the reliability and accuracy of defect detection. Even in complex field environments, it can still accurately distinguish between real defect thermal anomalies and noise artifacts, reducing false alarm and missed alarm rates. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the excitation coil assembly structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the U-shaped magnetic conductor structure according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the induced current density and magnetic flux distribution on the rail surface according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the infrared thermal imaging results in actual detection according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments.

[0027] This application discloses a rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil, the structure of which is as follows: Figure 1 As shown, the system includes an excitation coil assembly 1, an excitation source 2, a water-cooling device 3, an infrared thermal imager 4, a synchronization control module 5, and a computer 6. A double-layer spiral excitation coil 101 is connected to the excitation source 2 and the water-cooling device 3. The infrared thermal imager 4 is positioned above the excitation coil assembly 1 and connected to the computer 6. The synchronization control module 5 is connected to the excitation source 2 and the infrared thermal imager 4.

[0028] like Figure 2 As shown, the excitation coil assembly 1 includes a "double-layer" planar double-loop spiral excitation coil 101 and a U-shaped magnetic conductor 102. The double-layer spiral excitation coil 101 is a rectangular spiral coil wound with a 6mm hollow copper tube, which is a two-turn closed structure. Each turn of the coil is rectangular and spiral-shaped, arranged in a double-layer planar layout with the upper and lower layers arranged in parallel. The double-layer spiral excitation coil 101 can cover the entire rail surface, and the specific size can be adjusted according to different types of rails. In this embodiment, the width of the double-layer spiral excitation coil 101 covers the top surface of the rail head, and the length extends along the rail direction, growing as much as possible within the allowable range of the excitation equipment to cover sufficient heating threads and extend the heating time. Since rail defects are common on both sides of the rail head, multiple such magnets are provided on both sides of each layer of the double-layer spiral excitation coil 101. Figure 3The parallel U-shaped magnetic conductors 102 shown are made of a high-saturation-permeability, low-loss iron-based nanocrystalline soft magnetic alloy, with dimensions of approximately 20mm × 20mm × 15mm. The U-shaped openings face the rail surface to enhance the magnetic field at the railhead edge. This structure is equivalent to adding two rows of parallel U-shaped magnetic conductors above the double-layer spiral excitation coil 101 to form a "double U-shaped" arrangement. This can converge the magnetic lines of force generated by the double-layer spiral excitation coil 101 and guide the magnetic flux to the area to be inspected on the rail surface, achieving enhanced focusing of the local magnetic field.

[0029] The entire excitation coil assembly 1 can be installed at the bottom of the rail inspection vehicle and fixed by steel clamps to ensure that the plane of the double-layer spiral excitation coil 101 is lifted vertically from the top surface of the rail by a distance of ≥30mm during operation, so as to meet the safety distance requirements of dynamic detection, avoid bumps and collisions, and at the same time generate sufficient induction heating effect at a safe distance.

[0030] In this embodiment, the two-turn, double-layer spiral excitation coil 101 provides a higher magnetomotive force (MTF) compared to a single-turn coil. According to Ampere's circuital law, the product of the number of turns N and the current I determines the magnetic field strength. Therefore, a double-turn coil provides approximately twice the MMF under the same current, and the superimposed magnetic flux path significantly increases the magnetic flux density below the coil. This means that under the same excitation power, a two-turn coil can generate a stronger magnetic field and eddy current effect than a single-turn coil. Furthermore, compared to coils with more than two turns, a two-turn coil balances excitation efficiency with coil structural complexity: if the number of turns is too high, the coil's resistance and inductance increase, leading to a decrease in current at high frequencies and difficulty in sufficient driving, thus reducing the magnetic field gain. Simultaneously, multi-layer coils are difficult to manufacture and cool, and interlayer coupling may introduce additional losses. Therefore, the double-turn spiral structure achieves the goal of enhancing the magnetic field while avoiding the drawbacks of too many coils.

[0031] By arranging an array of U-shaped magnetic conductors 102 on both sides of the double-layer spiral excitation coil 101, the magnetic lines of force generated by the coil can be effectively converged, and the magnetic flux can be directed to the surface of the rail inspection area. Figure 4The diagram illustrates the induced current density and magnetic flux distribution on the rail surface in this embodiment, showing a significant enhancement in the magnetic field strength on both sides of the rail head. Compared to traditional coreless coil structures, the added U-shaped magnetic conductor array 102 substantially increases the alternating magnetic flux density, enabling sufficient eddy current heating to be induced on the rail surface even at lift-off distances greater than 30mm. This magnetic flux focusing setup improves the magnetic field rate under large lift-off conditions, widens the effective distance of induction heating, and provides a reliable guarantee for dynamic non-destructive testing of the rail. By increasing the eddy current heating area and employing magnetic flux focusing measures, the induction heating intensity at the defect location is significantly improved, making the temperature rise in the defect area more significant relative to the background. Simultaneously, the U-shaped magnetic conductor array 102 reduces magnetic flux leakage and background noise in irrelevant areas, lowering the overall background temperature rise. Consequently, the defect thermal image signal captured by the infrared thermal imager is clearer and more prominent, improving the signal-to-noise ratio. This provides favorable conditions for subsequent algorithms to extract defect features.

[0032] Meanwhile, the rectangular geometry of the double-layer spiral excitation coil 101 matches the cross-section of the rail, and the coil width can cover the width of the top surface of the rail, achieving full lateral coverage excitation of the rail surface. This ensures that there are no blind spots in the infrared imaging area and that defects at the edges are not missed due to insufficient coil size. Furthermore, the extended coil length allows the heating effect of the coil on the defect to continue for a longer period when the inspection vehicle passes over it. Therefore, this embodiment can accumulate sufficient defect thermal signals even under dynamic detection conditions, improving imaging clarity.

[0033] Excitation source 2 provides a high-frequency, high-current alternating current source for the double-layer spiral excitation coil 101. In this embodiment, a water-cooled high-frequency inverter with a power of not less than 3kW is selected, and the output frequency is set in the range of 250~300kHz. After resonant compensation, the high-frequency AC power is output to the double-layer spiral excitation coil 101. Under the synergistic effect of the double-layer spiral excitation coil 101 and the U-shaped magnetic conductor 102, a strong alternating eddy current field is induced on the surface of the rail. Based on Faraday's law of electromagnetic induction, the alternating magnetic flux generates an induced electromotive force. Higher frequencies and more turns N can increase the magnetic flux density, thereby increasing the intensity of the induced eddy current. However, if the frequency is too high, it will be limited by the skin depth and the capability of the power devices. Therefore, a suitable frequency needs to be selected according to the application requirements.

[0034] The water-cooling device 3 is used to circulate coolant through the hollow copper tube of the double-layer spiral excitation coil 101, removing the heat generated by the coil due to current and iron loss. In this embodiment, the water-cooling device 3 consists of a cooling pump, a heat exchanger, and piping, forming a closed-loop circulation. This ensures that the temperature rise of the double-layer spiral excitation coil 101 and the U-shaped magnetic conductor 102 remains low during long-term operation, preventing overheating that could lead to performance degradation or thermal image background interference. The double-layer spiral excitation coil 101 uses a hollow copper tube conductor and is cooled by water, significantly improving its heat dissipation capacity. During continuous operation, the coolant removes the heat generated by the double-layer spiral excitation coil 101 and the U-shaped magnetic conductor 102, suppressing temperature accumulation and preventing the resistance of the double-layer spiral excitation coil 101 from increasing due to overheating and reducing excitation efficiency. Furthermore, the selected iron-based nanocrystalline magnetic conductor has very low iron loss and minimal self-heating under high-frequency alternating magnetic fields. Therefore, it does not generate significant overheating background interference in the infrared thermal image, ensuring the contrast of the defect thermal signal. This enables the double-layer spiral excitation coil 101 to operate stably for extended periods, meeting the reliability requirements of online rail inspection equipment.

[0035] Infrared thermal imager 4 is used to acquire real-time images of the temperature distribution on the rail surface. In this embodiment, a mid-wave cooled thermal imager with a wavelength of 3~5 µm and high frame rate and high resolution is used. During installation, the infrared thermal imager 4 is fixed above the rail inspection vehicle and aligned with the rail surface in the area of ​​action of the double-layer spiral excitation coil 101. The mid-wave cooled thermal imager has higher thermal sensitivity and response speed in dynamic inspection, and can clearly capture the temperature rise signal caused by minor defects. By adjusting the lens field of view, it is ensured that the thermal image covers the excitation area of ​​the double-layer spiral excitation coil 101 with a certain margin, so as to comprehensively monitor the inspected area.

[0036] The synchronization control module 5 includes a high-speed data acquisition card and a digital signal processor (DSP) to coordinate the output of the excitation power supply 2 with the acquisition by the infrared thermal imager 4. The synchronization control module 5 sends synchronization trigger signals to the excitation source 2 and the infrared thermal imager 4, ensuring strict alignment of the infrared thermal image frames with the electromagnetic excitation timing, thereby enabling the acquisition of the corresponding temperature response sequence in time. This allows for a continuous mode of simultaneous excitation and imaging. Synchronous acquisition eliminates information misalignment caused by motion and time differences, ensuring that subsequent image processing is based on accurate timing data.

[0037] Another embodiment of this application discloses a rail electromagnetic thermal imaging detection method based on a double-layer helical magnetic coil, implemented based on the above-mentioned rail electromagnetic thermal imaging detection system, including the following steps: S1. Installation and Positioning: Securely install the excitation coil assembly 101 on the bottom of the inspection platform of the rail inspection vehicle, ensuring the coil plane is parallel to the top surface of the rail. Use adjustable clamps to fix the distance between the double-layer spiral excitation coil 101 and the top surface of the rail at ≥30mm, ensuring the width of the double-layer spiral excitation coil 101 covers the entire rail head. Maintain a safe distance between the coil and the rail surface during the movement of the rail inspection vehicle to avoid collisions.

[0038] S2. Electromagnetic Excitation: Start excitation source 2, continuously supplying high-frequency alternating current to the double-layer spiral excitation coil 101 while the rail inspection vehicle moves. Under the influence of the current in the double-layer spiral excitation coil 101 and the U-shaped magnetic conductor 102, strong and locally focused eddy currents are induced on the surface of the rail below the double-layer spiral excitation coil 101. According to Ampere's law and Faraday's law in Maxwell's equations, the magnetic field generated by the alternating current induces a current in the rail, the magnitude of which is related to the rate of change of the magnetic field of the double-layer spiral excitation coil 101 and the conductivity σ of the specimen. Guided by the U-shaped magnetic conductor 102, more magnetic flux is concentrated and coupled into the rail, increasing the magnetic flux density on the rail surface. It can be considered that the U-shaped magnetic conductor 102 increases the permeability μ of the equivalent magnetic circuit, allowing the spatial magnetic field to more effectively close through the rail. Therefore, even with a lift of more than 30mm, a sufficiently strong eddy current thermal effect can still be induced on the rail surface. These induced currents flow on the surface of the rail (especially in areas where defects may exist). Due to the resistance of the material, electromagnetic energy is rapidly converted into heat energy, i.e., Joule heating. According to Joule's law, the greater the current density in the concentrated eddy current region, the higher the local Joule heating power, resulting in a more significant temperature rise in that region.

[0039] S3. Thermal Image Acquisition: The infrared thermal imager 4 synchronously monitors the surface temperature field of the excited area of ​​the rail and continuously acquires a sequence of thermal infrared images at a preset frame rate (e.g., 5 frames per second in this embodiment, the specific setting depends on the scanning speed). During the uniform movement of the rail inspection vehicle, the infrared thermal imager 4 acquires an image showing the distribution of the rail surface temperature over time (temperature-time series). This thermal image data is uploaded in real-time to a host computer or embedded processor for storage and processing via high-speed communication. During acquisition, care should be taken to avoid strong sunlight, rain, and other environmental factors directly affecting temperature measurement. Noise filtering and non-uniformity correction can be performed on the original thermal image sequence as needed to improve the accuracy of subsequent defect identification.

[0040] S4. Image Processing and Defect Recognition: This embodiment proposes an unsupervised anomaly detection algorithm for infrared thermal images based on the fusion of Transformer and conditional diffusion models (Transformer-diffusion probability model) to process thermal image sequences, enabling automatic identification and anomaly judgment of rail defects and outputting defect locations. This method possesses good image generalization ability and robustness, and can adapt to complex thermal image data under dynamic and large-lift conditions. The specific detection process includes the following steps: S41. Image Segmentation and Feature Extraction. First, each frame of the infrared thermal image is preprocessed and divided into several small image patches of equal size. These patches are then flattened into image tokens, and spatial location information is added. This sequence of image tokens is input into a multi-layer Transformer encoder (in this embodiment, L=6, each layer containing 8 multi-head attention heads) to extract global features that fuse spatial structure information. The final feature vector output by the Transformer (i.e., the prior image features) is denoted as... This vector represents the global structural prior distribution of the input image and serves as a guide for subsequent image reconstruction.

[0041] S42. Constructing Noise Perturbation Samples. To improve the model's generalization ability to "normal states," a random perturbation strategy is adopted during the training phase: a frame is randomly selected from defect-free images. And randomly sample the diffusion time steps Generate noisy images :

[0042] in, For diffusion time step, It is Gaussian white noise. For the front The cumulative product of the diffusion rate. This process simulates an image gradually being eroded by noise from a clear state, providing the model with a large number of "normal + noise" samples, effectively expanding the distribution range of the original image, and enhancing the network's ability to understand and reconstruct normal structures.

[0043] S43. Training a conditional diffusion denoising network. The model aims to denoise noisy images by diffusing the denoising network. Recover the noise component This is done in order to deduce the original, clear image. The core of training is constructing a neural network function. From the input image Current time step Image prior features extracted by Transformer The network outputs a predicted value for the noise, and the optimization objective is to minimize the mean squared loss function of the prediction error.

[0044] in, For network parameters, For the loss of the entire network, The expectation is given for all training samples, sampling noise, and time steps. This indicates that the neural network (with parameters) The predicted noise, This represents the squared error between the noise predicted by the network and the actual noise. The Adam optimizer is used during training with a learning rate of 1e-4 until convergence. Since no defective images are introduced during training, the model learns to reconstruct normal heatmaps, exhibiting good unsupervised detection properties.

[0045] S44. Reconstruct the image and calculate the error. During the detection phase, the thermal image to be tested is also first randomly noise-added to form the observation image. Then, it is fed into the trained denoising network, combined with the prior image features extracted by the Transformer. Complete the inverse diffusion reconstruction and output the reconstructed image. By comparing the differences between the input image and the reconstructed image, the residual heatmap can be calculated.

[0046] in, This indicates the degree of abnormality for each pixel. Given the two-dimensional pixel coordinates of the image, normal regions can usually be accurately reconstructed with residual values ​​close to zero; however, regions containing defects are difficult to reconstruct due to their structure being inconsistent with the training distribution, resulting in significantly increased residuals.

[0047] S45. Anomaly Detection and Output. The system determines whether a frame of image is in an abnormal state by statistically analyzing the mean, variance, and other indicators of the residual map. When the mean of the residual heatmap exceeds a threshold obtained from the statistical analysis of residuals of defect-free samples (e.g., mean plus k times the standard deviation), a defect alarm is triggered, and the defect area is located through connected component analysis.

[0048] The unsupervised anomaly detection algorithm for infrared thermal imaging in this embodiment utilizes defect thermal image signals to achieve automatic defect identification, significantly improving the reliability and accuracy of defect detection compared to traditional methods. Even in complex field environments, the system can accurately distinguish between genuine defect thermal anomalies and noise artifacts, reducing false alarm and false negative rates.

[0049] To further verify the validity of this application, a dynamic testing experiment was conducted under actual field conditions. The test specimen used was a standard 60 kg / m U75V type steel rail, whose surface was machined with six artificial crack defects (numbered #1~#6) by wire electrical discharge machining, and then subjected to long-term train running and compaction to approximate real working conditions. The geometric parameters of the defects are shown in Table 1: Table 1 Specific geometric information of the defects

[0050] In this experiment, the distance between the double-layer spiral excitation coil 101 and the rail was set to 30 mm, the operating frequency of the excitation power supply was 280 kHz, the power was set to 3 kW, the rail inspection vehicle moved at a constant speed of 3.6 km / h, the infrared thermal imager 4 was set to a frame rate of 100 Hz, and the resolution was 640×512.

[0051] When the vehicle-mounted rail electromagnetic thermal imaging detection system based on double-layer spiral magnetic coil passes through a section of rail to be inspected, the following steps are taken: (1) First, ensure that the excitation coil assembly 1 is correctly positioned and maintains a distance of 30mm from the rail surface; (2) Turn on the excitation source 2 to continuously excite the rail; (3) The infrared thermal imager 4 synchronously collects the temperature field change sequence of the rail surface and transmits the data to the host computer; (4) The host computer runs an unsupervised anomaly detection algorithm based on the Transformer-diffusion probability model to process the infrared thermal image sequence in real time and identify defects.

[0052] Specifically, the computer first preprocesses the acquired thermal image sequence, and then inputs the sequence into a pre-trained unsupervised anomaly detection algorithm based on the Transformer-diffusion probability model for analysis. The model automatically denoises and reconstructs each frame of the image, compares it with the original image, and calculates an anomaly map (such as a residual heatmap in this example).

[0053] Figure 5 The infrared thermal imaging results of this invention in actual detection are presented. It can be seen that when a side crack defect exists in the rail, the temperature rise signal in the defect area is significantly higher than that in the surrounding defect-free area after a period of excitation. The algorithm proposed in this application... Figure 5 After processing the infrared sequence shown, the crack defect was successfully detected and located: the defect area showed a significant anomaly in the normal background of the model reconstruction, and the brightness of the corresponding position in the residual heat map far exceeded the threshold. The computer determined that it was a real defect and triggered an alarm.

[0054] Experimental results show that even under conditions of 30mm lifting and dynamic operation of the rail inspection vehicle, the proposed solution can still clearly image the cracks and defects on the rail surface and accurately identify the defect locations through the algorithm. This verifies the effectiveness of the proposed system and method, demonstrating its significant advantages in improving dynamic detection capabilities during large lifting and in enhancing the intelligence of defect identification.

[0055] In summary, this application achieves efficient non-contact dynamic electromagnetic thermal imaging detection of rails by combining a "double-layer spiral coil + double U-shaped magnetic conductor" excitation structure with an unsupervised anomaly detection algorithm based on the Transformer-diffusion probability model. The thermal imaging detection system proposed in this application can still produce clear thermal images of surface and near-surface defects of rails under large lifting conditions, and efficiently extracts defect features using the algorithm. It combines the advantages of high detection sensitivity, high reliability, and device safety and stability, and has significant engineering application value.

[0056] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil, characterized in that, It includes an excitation coil assembly (1), an excitation source (2), a water cooling device (3), an infrared thermal imager (4), a synchronization control module (5), and a computer (6); The excitation coil assembly (1) includes a double-layer spiral excitation coil (101). The double-layer spiral excitation coil (101) has a two-turn closed structure. Each turn of the coil is rectangular spiral. The upper and lower layers are arranged in parallel. On both sides of each layer of the double-layer spiral excitation coil (101), there are multiple parallel U-shaped magnetic conductors (102). The U-shaped opening of the U-shaped magnetic conductor (102) faces the surface of the rail. The double-layer spiral excitation coil (101) is connected to the excitation source (2) and the water cooling device (3). The infrared thermal imager (4) is set above the excitation coil assembly (1) and connected to the computer (6). The synchronous control module (5) is connected to the excitation source (2) and the infrared thermal imager (4).

2. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 1, characterized in that, The width of the double-layer spiral excitation coil (101) covers the top surface of the rail head, and its length extends along the rail direction. The plane of the double-layer spiral excitation coil (101) is lifted perpendicularly from the top surface of the rail by a distance ≥30mm.

3. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 2, characterized in that, The double-layer spiral excitation coil (101) is wound with a hollow copper tube.

4. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 3, characterized in that, The material of the U-shaped magnetic conductor (102) is an iron-based nanocrystalline soft magnetic alloy.

5. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 4, characterized in that, The excitation source (2) adopts a high-frequency inverter power supply with an output frequency of 250~300kHz and a power of ≥3kW.

6. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 5, characterized in that, The water-cooling equipment (3) is a hollow copper tube of a double-layer spiral excitation coil (101) through which circulating coolant is introduced, including a cooling pump, heat exchanger, and pipeline, forming a closed-loop circulation.

7. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 6, characterized in that, The infrared thermal imager (4) is a medium-wave cooled thermal imager with a wavelength of 3~5 µm.

8. The rail electromagnetic thermal imaging detection system based on a double-layer helical magnetic coil according to claim 7, characterized in that, The synchronization control module (5) includes a high-speed data acquisition card and a digital signal processor, which are used to coordinate the output of the excitation power supply (2) with the synchronization of the infrared thermal imager (4).

9. A rail electromagnetic thermal imaging detection method based on a double-layer helical magnetic coil, characterized in that, Based on the rail electromagnetic thermal imaging system according to any one of claims 1-8, the system includes the following steps: S1. Installation and positioning: Fix the excitation coil assembly (101) to the bottom of the rail inspection vehicle, keeping the plane of the double-layer spiral excitation coil (101) parallel to the top surface of the rail and the lifting distance ≥30mm; S2, Electromagnetic excitation: During the movement of the rail inspection vehicle, a high-frequency alternating current is supplied to the double-layer spiral loop excitation coil (101) using an excitation source, and a focused eddy current is induced on the surface of the rail through the U-shaped magnetic conductor; S3. Thermal image acquisition: Use an infrared thermal imager to synchronously acquire the temperature field of the rail surface and continuously acquire thermal infrared image sequences at a preset frame rate. S4. Image Processing and Defect Recognition: An unsupervised anomaly detection algorithm is used to process the thermal image sequence and output the defect location.

10. The rail electromagnetic thermal imaging detection method based on a double-layer helical magnetic coil according to claim 9, characterized in that, S4 includes the following steps: S41. Divide the single-frame thermal image into image blocks and add positional encoding, then input the block into the Transformer encoder to extract the final feature vector. ; S42. For defect-free training images Add noise to generate a noisy image. : in, For diffusion time step, It is Gaussian white noise. For the front The cumulative product of the step diffusion rates; S43. Train the conditional diffusion denoising network to minimize noise prediction error: in, For network parameters, For the loss of the entire network, The expectation is given for all training samples, sampling noise, and time steps. For network prediction noise, This represents the squared error between the noise predicted by the network and the actual noise. S44. Reconstruct the image and calculate the error. After adding noise to the image to be tested, input it into the trained conditional diffusion denoising network, output the reconstructed image, and calculate the residual heatmap: in, This indicates the degree of abnormality for each pixel. Two-dimensional pixel coordinates of the image To reconstruct the image; S45. When the mean value of the residual heatmap exceeds the threshold obtained from the statistical analysis of the residuals of the defect-free samples, a defect alarm is triggered, and the defect area is located through connected component analysis.

Images