A method and system for detecting defects in wheel tread

Abstract

This invention discloses a method and system for detecting defects in wheel treads. The method includes: applying uniform excitation electromagnetic force to a moving wheel tread to induce heat; acquiring thermal images of the wheel tread in real time for each frame based on the induced heat to construct a first thermal image sequence; performing velocity compensation and splicing reconstruction on the first thermal image sequence based on the wheel tread's movement speed to obtain a second thermal image sequence; rearranging the second thermal image sequence by blocks to obtain an original image matrix; decomposing the original image matrix into a low-rank background matrix and a sparse defect matrix by finding an optimal set of parameters; and acquiring all pixels in the sparse defect matrix whose induced heat falls within the double threshold range based on a pre-constructed dual threshold range, and generating a defect distribution map of the wheel tread. The technical solution of this invention can dynamically detect defects in wheel treads.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a method and system for detecting defects in wheel treads. Background Technology

[0002] The wheel tread is a critical part that comes into direct contact with the rail, and the initiation and propagation of fatigue cracks on its surface can lead to serious safety accidents.

[0003] Fatigue microcracks in wheel treads are often small and difficult to detect in the early stages. Currently, the sensitivity of these microcrack detection methods is low, especially during high-speed dynamic testing, where the microcrack signal is easily drowned out by background noise, leading to missed detections.

[0004] Therefore, there is an urgent need to develop a solution that can more accurately detect microcracks in wheel treads. Summary of the Invention

[0005] This invention provides a method and system for detecting defects in wheel treads, enabling accurate detection of microcracks in wheel treads.

[0006] In a first aspect, the present invention provides a method for detecting defects in wheel treads, comprising:

[0007] By applying uniform excitation electromagnetic force to the wheel tread in motion, induced heat is generated on the wheel tread.

[0008] Based on the induced heat, thermal images of each frame of the wheel tread are acquired in real time to construct a first thermal image sequence; based on the movement speed of the wheel tread, the first thermal image sequence is speed compensated and stitched together to reconstruct a second thermal image sequence.

[0009] The second thermal image sequence is rearranged into blocks to obtain the original image matrix; by finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix.

[0010] Based on a pre-built dual threshold range, all pixels in the sparse defect matrix whose induced heat falls within the dual threshold range are obtained, and a defect distribution map of the wheel tread is generated.

[0011] Furthermore, the speed compensation process includes:

[0012] Stable feature points are extracted between adjacent frames in the first thermal image sequence;

[0013] Calculate the angular velocity between adjacent frames based on the spatial displacement of feature points between adjacent frames and the time interval between frames;

[0014] The rotational speed of the wheel is obtained, and the rotational speed of the wheel is fused and corrected with the angular velocity between adjacent frames to obtain the compensated angular velocity between adjacent frames.

[0015] Furthermore, the process of piecing together and reconstructing includes:

[0016] Each frame of thermal image is mapped onto an unfolded plane with circumferential angle and axial position of tread surface as independent variables, and the angular displacement between adjacent frames is calculated based on the compensated angular velocity difference between adjacent frames.

[0017] The thermal images mapped in the unfolded plane are sampled at equal angular intervals, and the unfolded image sequence is obtained based on the angular displacement between adjacent frames;

[0018] Select any frame of thermal image in the unfolded plane as the reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and then stitch together each registered frame of thermal image.

[0019] Each frame of the stitched thermal image is remapped back to the camera's field of view coordinate system to obtain the second thermal image sequence.

[0020] Furthermore, the parameter set includes: nuclear norm regularization weight, L1 norm regularization weight, block size, block overlap ratio, and double threshold coefficient;

[0021] Among them, the nuclear norm regularization weight is used to control the degree of low rank of the low-rank background matrix;

[0022] L1 norm regularization weights are used to control the sparsity of the sparse defect matrix;

[0023] The block size is used to split the original image matrix into multiple independently processed blocks;

[0024] The block overlap ratio is used to limit the percentage of the overlap area between adjacent blocks to the size of a single block;

[0025] The double threshold coefficient is used to distinguish elements in the low-rank background matrix and the sparse defect matrix.

[0026] Furthermore, the process of finding an optimal set of parameters specifically includes:

[0027] Each parameter in the parameter set is encoded according to a preset rule, and multiple candidate parameter individuals are randomly generated to construct an initial population;

[0028] Based on any individual with any parameter in the initial population, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix, and the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix are calculated.

[0029] Based on the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix, a comprehensive evaluation of the current parameter individual is carried out.

[0030] Select the individual with the highest overall evaluation in the initial population, and generate a new generation population based on this individual. Then, find the individual with the highest overall evaluation in the new generation population.

[0031] The process of iteratively executing the generation of a new generation of population based on the parameter individual with the highest comprehensive evaluation in the current population continues until the number of iterations reaches a preset upper limit or the comprehensive evaluation converges. At this point, the parameter individual with the highest comprehensive evaluation is taken as the optimal parameter set.

[0032] Furthermore, a comprehensive evaluation of the current individual parameters can be expressed as:

[0033]

[0034] in, This represents the comprehensive evaluation indicators. , , These are empirical weighting coefficients. and These represent the thermal contrast and signal-to-noise ratio in the sparse defect matrix, respectively. This represents the temperature standard deviation in the low-rank background matrix.

[0035] Furthermore, the process of constructing the dual threshold range includes:

[0036] Calculate the mean and standard deviation of induced heat in the sparse defect matrix, and based on the mean and standard deviation of induced heat, obtain the upper and lower limits of the dual threshold range.

[0037] Furthermore, after generating the defect distribution map of the wheel tread, the following steps are also included:

[0038] In the defect distribution map of the wheel tread, construct one or more target detection areas; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and mark the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

[0039] Furthermore, the target detection area includes defective and non-defective portions;

[0040] For any target detection area, the process of calculating thermal contrast includes: obtaining the thermal contrast of the target detection area based on the temperature of the defective part, the temperature of the non-defective part, and the ambient temperature.

[0041] For any target detection region, the process of calculating the signal-to-noise ratio includes: obtaining the signal-to-noise ratio of the target detection region based on the average temperature of the pixels in the entire target detection region, the average temperature of the pixels in the defective part, the average temperature of the pixels in the non-defective part, and the number of pixels in the target detection region.

[0042] Furthermore, for any target detection region, its thermal contrast and signal-to-noise ratio meet preset conditions, specifically including:

[0043] The thermal contrast of the target detection area is greater than or equal to the preset thermal contrast.

[0044] In addition, the signal-to-noise ratio of the target detection area is greater than or equal to the preset signal-to-noise ratio.

[0045] Furthermore, after defining the target detection area that meets the preset conditions for thermal contrast and signal-to-noise ratio as the defect area, the process also includes:

[0046] Extract the independent defect targets in the defect region and obtain the geometric features of the independent defect targets; map the geometric features of the independent defect targets onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

[0047] Furthermore, it also includes:

[0048] Pre-build the correspondence between thermal contrast, signal-to-noise ratio, and geometric features and defect types;

[0049] The thermal contrast and signal-to-noise ratio of the defect region are obtained, and the defect type of the independent defect target is determined by combining the geometric features of the independent defect target.

[0050] Furthermore, it also includes:

[0051] Quantitative relationship curves between different defect depths and thermal contrast are pre-constructed;

[0052] Obtain the evolution curve of thermal contrast in the defect area over time;

[0053] The peak value of the thermal contrast in the evolution curve is selected and substituted into the inverse function of the quantitative relationship curve to obtain the depth value of the independent defect target.

[0054] Secondly, the present invention provides a wheel tread defect detection system, comprising:

[0055] The induction unit is used to induce heat on the wheel tread by applying uniform excitation electromagnetic force to the wheel tread in motion.

[0056] The construction unit is used to collect thermal images of each frame of the wheel tread in real time based on the induced heat generated, and construct a first thermal image sequence; based on the movement speed of the wheel tread, the first thermal image sequence is subjected to speed compensation and stitching reconstruction to obtain a second thermal image sequence.

[0057] The decomposition unit is used to rearrange the second thermal image sequence into blocks to obtain the original image matrix; by finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix.

[0058] The generation unit is used to obtain all pixels in the sparse defect matrix whose induced heat is within the double threshold range based on a pre-constructed double threshold range, and generate a defect distribution map of the wheel tread.

[0059] Furthermore, the building unit includes:

[0060] Extraction subunits are used to extract stable feature points between adjacent frames in the first thermal image sequence;

[0061] The first calculation subunit is used to calculate the angular velocity between adjacent frames based on the spatial displacement of feature points between adjacent frames and the time interval between frames;

[0062] The fusion subunit is used to obtain the rotational speed of the wheel and fuse and correct the rotational speed of the wheel with the angular velocity between adjacent frames to obtain the compensated angular velocity between adjacent frames.

[0063] Furthermore, the building unit also includes:

[0064] The second calculation subunit is used to map each frame of thermal image onto an unfolded plane with circumferential angle and axial position of tread surface as independent variables, and to calculate the angular displacement between adjacent frames based on the compensated angular velocity difference between adjacent frames.

[0065] The sampling subunit is used to sample the thermal image mapped in the unfolded plane at equal angular intervals and obtain the unfolded image sequence based on the angular displacement between adjacent frames;

[0066] The stitching subunit is used to select any frame of thermal image in the unfolded plane as a reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and stitch together each registered frame of thermal image.

[0067] The mapping subunit is used to remap each frame of the stitched thermal image back to the camera's field of view coordinate system to obtain the second thermal image sequence.

[0068] Furthermore, the system also includes:

[0069] The calibration unit is used to construct one or more target detection areas in the defect distribution map of the wheel tread; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and calibrate the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

[0070] Furthermore, the system also includes:

[0071] A quantitative unit is used to extract independent defect targets in the defect region and obtain the geometric features of the independent defect targets; the geometric features of the independent defect targets are mapped onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

[0072] Thirdly, embodiments of the present invention provide an electronic device, the electronic device comprising:

[0073] At least one processor; and a memory communicatively connected to the at least one processor; wherein,

[0074] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the steps of the wheel tread defect detection method according to any embodiment of the present invention.

[0075] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of the wheel tread defect detection method of any embodiment of the present invention.

[0076] Compared with the prior art, the present invention has the following advantages:

[0077] The technical solution in this embodiment of the invention first applies a uniform excitation electromagnetic field to the wheel tread in motion to induce heat. Then, based on the induced heat, thermal images of the wheel tread are acquired in real time for each frame, constructing a first thermal image sequence. Based on the wheel tread's speed, the first thermal image sequence is speed-compensated and stitched together to obtain a second thermal image sequence. Next, the second thermal image sequence is rearranged in blocks to obtain the original image matrix. By finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix. Based on a pre-constructed dual-threshold range, all pixels in the sparse defect matrix whose induced heat falls within the dual-threshold range are obtained, and a defect distribution map of the wheel tread is generated. This solution, on the one hand, eliminates the dynamic blur of the original thermal image by speed compensation and stitching together the first thermal image sequence; on the other hand, by finding the optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix with maximum differentiation. Finally, by setting dual thresholds, pixels with abnormal heat are extracted from the sparse defect matrix, ensuring the sensitivity of defect detection during high-speed motion. Attached Figure Description

[0078] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0079] Figure 1 A schematic flowchart of a wheel tread defect detection method provided in an embodiment of the present invention;

[0080] Figure 2 This is a schematic diagram of the structure of a rectangular magnetic yoke electromagnetic induction sensor provided in an embodiment of the present invention;

[0081] Figure 3 A hardware schematic diagram for detecting wheel tread defects provided in an embodiment of the present invention;

[0082] Figure 4 A schematic diagram of a speed compensation process provided in an embodiment of the present invention;

[0083] Figure 5 This is a schematic diagram of a splicing and reconstruction process provided in an embodiment of the present invention;

[0084] Figure 6 This is a schematic diagram illustrating the decomposition of an original image matrix according to an embodiment of the present invention;

[0085] Figure 7A framework diagram for detecting wheel tread defects provided in an embodiment of the present invention;

[0086] Figure 8 This is a schematic diagram of a wheel tread defect detection system provided in an embodiment of the present invention;

[0087] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0088] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0089] In related technologies, the following are some commonly used wheel defect detection schemes:

[0090] 1) Ultrasonic testing detects defects by utilizing the propagation characteristics of ultrasonic waves within the metal material of the wheel. When ultrasonic waves encounter internal cracks or other defects, reflection and refraction occur. By capturing the signal characteristics of the reflected waves, the location and size of the defect can be determined. However, this method has a near-field blind zone and poor dynamic coupling, which can easily lead to missed detection of surface and near-surface microcracks.

[0091] 2) Magnetic particle inspection involves applying an external magnetic field to the wheel. The magnetic resistance at the defect location changes, creating a leakage magnetic field. Magnetic powder is then sprinkled on the wheel, and the leakage magnetic field attracts the powder, forming a clear magnetic trace, thus determining the defect's location and shape. However, this method is ineffective for deep defects, and the wheel surface must be cleaned before inspection; otherwise, the magnetic trace display will be affected.

[0092] 3) Penetrant testing involves applying a penetrant containing colored dyes or fluorescent agents to the wheel surface. The penetrant seeps into tiny cracks on the surface. A developer is then applied, which absorbs the penetrant within the cracks, revealing the defect outline. However, this method can only detect surface defects and cannot detect internal defects or non-open surface defects. The testing process is cumbersome, and its sensitivity is greatly affected by the surface finish.

[0093] 4) Electromagnetic ultrasonic testing: When a high-frequency pulsed current is passed through a coil, a current is induced on the wheel surface. This current, under the influence of the Lorentz force in the magnetic field, generates ultrasonic waves. The residual stress and defects of the wheel rim are calculated using the time difference of sound propagation. However, this method has room for improvement in the detection accuracy of deep, minute defects, and the equipment is expensive and susceptible to electromagnetic interference.

[0094] In summary, the relevant technologies still have shortcomings, making them unable to meet the current requirements for wheel tread defect detection.

[0095] To address at least one of the technical problems existing in the aforementioned related technologies, the present invention provides a method for detecting defects in wheel treads. Figure 1 This is a flowchart illustrating a wheel tread defect detection method provided in an embodiment of the present invention. This embodiment is particularly applicable to the detection of fatigue microcracks in wheel treads during high-speed movement. The method can be executed by a wheel tread defect detection system, which can be implemented in software and / or hardware and can be configured in an electronic device.

[0096] like Figure 1 As shown, the method includes:

[0097] S1, by applying uniform excitation electromagnetic force to the wheel tread in motion, induces heat in the wheel tread.

[0098] For example, an electromagnetic induction sensor with a rectangular magnetic yoke structure can be used to apply uniform excitation electromagnetic current to the wheel tread, thereby uniformly heating the surface of the wheel tread. Additionally, a power generator can provide the required excitation current to the electromagnetic induction sensor, for example, an excitation current of 200A at a frequency of 186kHz. At this frequency and current intensity, it is possible to ensure that the electromagnetic induction sensor generates sufficient eddy currents on the wheel tread, thereby exciting abnormal Joule heating at the crack area for crack and defect localization.

[0099] Figure 2 This is a schematic diagram of the structure of a rectangular magnetic yoke type electromagnetic induction sensor provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the electromagnetic induction sensor mainly consists of a magnetic yoke and a coil. The coil is divided into two parts and wound around both sides of the magnetic yoke, with the current directions of the coils wound on both sides of the magnetic yoke being opposite. The rectangular magnetic yoke structure not only effectively enhances the magnetic field strength in the sensing area, making the thermal signal in the crack area more prominent, but also has a more uniform magnetic field distribution than traditional coils.

[0100] The electromagnetic induction sensor is fixedly installed near the wheel tread. During installation, the height and angle of the sensor need to be finely adjusted to ensure that the sensor maintains the optimal working distance from the wheel tread and avoids contact friction. At the same time, by adjusting the position of the sensor, it is ensured that the sensor's scanning range can cover the entire wheel tread, especially areas prone to cracks.

[0101] Figure 3 This is a hardware schematic diagram of a wheel tread defect detection method provided in an embodiment of the present invention, as shown below. Figure 3As shown, the system includes at least: an electromagnetic induction sensor, a power generator, an infrared thermal imager, and a computer system. The power generator provides the necessary excitation current to the electromagnetic induction sensor; the infrared thermal imager is used to acquire real-time temperature changes on the wheel tread surface and generate thermal images; the computer system, as the core platform for data acquisition and processing, processes the thermal image data obtained from the infrared thermal imager, performing image reconstruction, motion compensation, crack detection, and quantitative analysis. Additionally, it may include: a water-cooling device, a synchronization trigger, and a mechanical transmission device. The water-cooling device ensures stable system operation, especially under high-power operation, preventing overheating; the synchronization trigger coordinates the operation of the power generator and the infrared thermal imager, ensuring time synchronization between the two; the mechanical transmission device enables the wheel to rotate at a preset speed, ensuring a comprehensive scan of the wheel tread surface while maintaining a stable relative position between the sensor and the wheel surface during dynamic operation.

[0102] S2, based on the induced heat, real-time acquisition of thermal images of each frame of the wheel tread to construct a first thermal image sequence; and based on the movement speed of the wheel tread, speed compensation and splicing reconstruction of the first thermal image sequence to obtain a second thermal image sequence.

[0103] The system utilizes an infrared thermal imager to acquire thermal images in real time at a preset frame rate, such as 200Hz. Each frame records the temperature distribution of the wheel tread. The infrared thermal imager can accurately detect areas of thermal anomalies in the wheel tread caused by eddy current excitation, thereby locating cracks and defects. Furthermore, a synchronous trigger coordinates the operation of the power generator and the infrared thermal imager, ensuring time synchronization between the two. Based on this, thermal images and excitation signals can be captured simultaneously during wheel rotation, resulting in accurate thermal imaging data.

[0104] It should be noted that since the wheel tread is in motion, a high-frequency frame rate needs to be set to complete the real-time acquisition of thermal images. At this time, there are overlapping parts between adjacent frames of thermal images. Therefore, speed compensation and stitching reconstruction are also required based on the movement speed of the wheel tread to eliminate the influence of overlapping areas.

[0105] Figure 4 This is a schematic diagram of a speed compensation process provided in an embodiment of the present invention, as shown below. Figure 4 As shown, the speed compensation process includes:

[0106] S2011, extract stable feature points between adjacent frames in the first thermal image sequence.

[0107] S2012, calculates the angular velocity difference between adjacent frames based on the spatial displacement of feature points between adjacent frames and the inter-frame time interval.

[0108] S2013, obtain the rotational speed of the wheel, and fuse and correct the rotational speed of the wheel with the angular velocity difference between adjacent frames to obtain the compensated angular velocity difference between adjacent frames.

[0109] The angular velocity between adjacent frames is estimated based on image features, while the wheel rotation speed is obtained from the speed acquisition device.

[0110] Specifically, the first thermal image sequence acquired by the infrared thermal imager is denoted as... Based on corner detection, gradient operators, or phase correlation methods, in adjacent frames and Extracting stable feature points between Based on the spatial displacement of feature points in two frames and inter-frame time interval The linear velocity between adjacent frames can be obtained:

[0111]

[0112] By combining the wheel radius R, the linear velocity between adjacent frames can be converted into angular velocity:

[0113]

[0114] Alternatively, the wheel speed can be obtained and recorded using an encoder or speed sensor installed on the mechanical transmission device. Using least squares fitting or robust regression, the image features are estimated... and After performing fusion correction, a smooth angular velocity is obtained after filtering out jitter and instantaneous fluctuations. This refers to the angular velocity difference between adjacent frames after compensation, in order to improve the stability and accuracy of velocity estimation.

[0115] Figure 5 This is a schematic diagram of a splicing and reconstruction process provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the process of splicing and reconstructing includes:

[0116] S2021, each frame of thermal image is mapped onto an unfolded plane with circumferential angle and axial position of tread surface as independent variables, and the angular displacement between adjacent frames is calculated based on the compensated angular velocity difference between adjacent frames.

[0117] S2022, the thermal image mapped in the unfolded plane is sampled at equal angular intervals, and the unfolded image sequence is obtained based on the angular displacement between adjacent frames.

[0118] S2023, select any frame of thermal image in the unfolded plane as the reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and stitch together each registered frame of thermal image.

[0119] S2024, each frame of the stitched thermal image is remapped back to the camera's field of view coordinate system to obtain the second thermal image sequence.

[0120] Specifically, the wheel tread is approximated as a cylindrical surface, and for each frame... Coordinate transformation is performed based on wheel geometry parameters and shooting calibration parameters, mapping to a coordinate system based on circumferential angles. The unfolded plane with the axial position z of the tread surface as the independent variable Based on the compensated angular velocity difference between adjacent frames. Calculate the angular displacement between adjacent frames:

[0121]

[0122] In a unified standard angle grid On (m=1,…,M), for all unfolded planes Resampling and interpolation are performed to transform the originally non-uniform sampling into an unfolded image sequence with equal angular intervals. .

[0123] In a unified angle domain, with a certain reference frame Based on this, other frames are compensated for by advance / lag angular displacement. Sub-pixel registration is performed along the circumferential angle, and the corresponding relationship is as follows:

[0124]

[0125] in, This represents the standard angular coordinates in the reference frame under a unified angular domain. This represents the actual sampling angle corresponding to this position in the k-th frame to be registered; For the selected reference frame index, This represents the estimated circumferential angular displacement increment from frame i to frame i+1 (obtained by integrating or accumulating the smoothed angular velocities from the previous frame). Therefore, the increment from the reference frame is given. The total rotation angle evolved to the current k-th frame is added to the reference angle. Get it This achieves the "advanced / delayed" compensation of the k-th frame along the circumferential direction to the angle relative to the reference frame. The aligned subpixel registration relationship provides a one-to-one coordinate mapping for subsequent multi-frame weighted stitching.

[0126] The registered multi-frame images are stitched together according to the weighted average or maximum response principle to generate a motion-compensated reconstructed image. During the weighted averaging process, the weights can be adaptively adjusted based on the noise level, local gradient, or SNR of each frame to further suppress the interference of low-quality frames on the reconstruction results.

[0127] This embodiment employs a speed compensation and stitching reconstruction method. Compared to traditional line scan reconstruction images and images reconstructed using 2×2 robust Gaussian filtering, it can improve the thermal differences between different regions of the wheel tread. Furthermore, the speed compensation process significantly reduces motion blur caused by speed fluctuations, resulting in a more refined reconstructed image. The edges of the medium crack are clearer, and both TT and SNR are significantly improved.

[0128] Ultimately, the image will be reconstructed. Remapping back to the camera's field of view coordinate system yields the second thermal image sequence after velocity compensation and stitching reconstruction. This sequence serves as input to the block sparse decomposition (PSD) algorithm, providing high-quality, low-ambiguity foundational data for subsequent background-defect separation and small-target crack enhancement.

[0129] S3, rearrange the second thermal image sequence to obtain the initial image matrix; by finding an optimal set of parameters, decompose the initial image matrix into a low-rank background matrix and a sparse defect matrix.

[0130] Figure 6 This is a schematic diagram of decomposing the original image matrix according to an embodiment of the present invention, as shown below. Figure 6 As shown, the original image matrix reflects the global temperature information in the thermal image after block rearrangement, while the low-rank background matrix reflects the local thermal field information of the wheel tread that is approximately uniform and slowly changing, and the sparse defect matrix reflects local thermal anomaly information such as cracks and spalling.

[0131] Specifically, after obtaining the second thermal image sequence, the velocity-compensated sequence is first rearranged into blocks. Then, using a sliding window technique, multiple local image patches are extracted from the rearranged thermal image. Each patch is vectorized into a column of the original image matrix, forming a set of feature vectors. The elements in each column represent the grayscale or temperature value of each pixel in the patch, reflecting the local features of that region. The difference between different columns lies in that they correspond to different local regions in the thermal image; that is, each column of features reflects the local temperature changes and crack features in different areas of the thermal image. These patches are used to construct an overall model of the image for subsequent processing.

[0132] In some embodiments, the parameter set includes: nuclear norm regularization weight, L1 norm regularization weight, block size, block overlap ratio, and double threshold coefficient.

[0133] Among them, the nuclear norm regularization weight is used to control the low rank of the low-rank background matrix; the L1 norm regularization weight is used to control the sparsity of the sparse defect matrix; the block size is used to split the original image matrix into multiple independently processed blocks; the block overlap ratio is used to limit the percentage of the overlapping area of ​​adjacent blocks to the single block size; and the double threshold coefficient is used to distinguish the elements in the low-rank background matrix and the sparse defect matrix.

[0134] Understandably, the kernel norm regularization weights and L1 norm regularization weights need to match the low rank of the background and the sparsity of defects; the block size and block overlap ratio need to be dynamically adapted according to the computing power or memory of the edge device to balance real-time performance and defect integrity; the dual threshold coefficients limit the upper and lower limits of a range, and by introducing adaptive dual thresholds for image segmentation, images within this range are considered possible defects, while those outside the range are considered background.

[0135] In some embodiments, the process of finding an optimal set of parameters specifically includes:

[0136] S3011, encode each parameter in the parameter set according to a preset rule, and randomly generate multiple candidate parameter individuals to construct an initial population.

[0137] S3012, based on any individual with any parameter in the initial population, decomposes the original image matrix into a low-rank background matrix and a sparse defect matrix, and calculates the thermal contrast and signal-to-noise ratio in the sparse defect matrix, as well as the temperature standard deviation in the low-rank background matrix.

[0138] It should be noted that throughout this text, thermal contrast ratio and signal-to-noise ratio refer to quantitative indicators between defective and non-defective portions within a specified area. For simplicity, the text will use "thermal contrast ratio and signal-to-noise ratio of a specified area," which represents the thermal contrast ratio and signal-to-noise ratio between defective and non-defective portions within that specified area.

[0139] S3013 comprehensively evaluates the current parameter individual based on the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix.

[0140] S3014: Select the individual with the highest overall evaluation in the initial population, and generate a new generation population based on this individual, and find the individual with the highest overall evaluation in the new generation population.

[0141] S3015 Iteratively executes the process of generating a new generation of population based on the parameter individual with the highest comprehensive evaluation in the current population, until the number of iterations reaches the preset upper limit or the comprehensive evaluation converges, and the parameter individual with the highest comprehensive evaluation at this time is taken as the optimal parameter set.

[0142] This embodiment employs an image reconstruction and defect enhancement method based on patch-based sparse decomposition (PSD): to address motion blur caused by dynamic scanning, the PSD algorithm is used to perform advanced processing on the second thermal image sequence to achieve defect separation in thermal images of complex surfaces.

[0143] To improve the detection capability of small-target defects such as microcracks, the original thermal image is split into two parts, a background matrix and a defect matrix, within the PSD decomposition framework. Parametric adaptive optimization and intelligent algorithms are then introduced, the principle of which is as follows:

[0144] First, the velocity-compensated thermal image sequence is rearranged into blocks to obtain the original image matrix D. According to the low-rank sparse decomposition model, this can be expressed as:

[0145]

[0146] in, The low-rank background matrix represents the approximately uniform and slowly varying thermal field of the wheel tread. is a sparse defect matrix, representing local thermal anomalies such as cracks and spalling; N is random noise; This is the set of parameters to be optimized.

[0147] In order to decompose the original image matrix using a set of parameters, the original image matrix D is reconstructed as follows:

[0148]

[0149] in, Let be the nuclear norm of the matrix. The positive weighting coefficients are... It is a constant. .

[0150] Preferably, the parameters in the parameter set include:

[0151] : Nuclear norm regularization weights, used to control the degree of low rank of the background matrix B;

[0152] L1 norm regularization weights are used to control the sparsity of the defect matrix C.

[0153] p: Block size (e.g., p×p);

[0154] o: Block overlap ratio;

[0155] k1, k2: Threshold coefficients in local adaptive double threshold segmentation, etc.

[0156] In a given At that time, principal component pursuit (PCP) or its accelerated variant is used to perform low-rank sparse decomposition of D, resulting in... and To measure a set of parameters To determine the merits and demerits, construct a comprehensive evaluation index. Its form is, for example:

[0157]

[0158] in, and The sparse defect matrix is ​​defined by the current parameters. The thermal contrast and signal-to-noise ratio of the extracted typical defect areas, The temperature standard deviation of the background area is used to reflect the smoothness of the background. , , These are empirical weighting coefficients.

[0159] To automatically find the optimal parameter parameters A genetic algorithm (GA) is used for global search, as follows:

[0160] 1) Encoding and initializing the population: Each parameter is encoded with a real number within a preset range, and several candidate parameter individuals are randomly generated to form the initial population;

[0161] 2) Fitness calculation: For each individual in the population Perform a low-rank sparse decomposition and defect extraction calculation to obtain the corresponding... , and Based on this, a comprehensive evaluation index is calculated. ;

[0162] 3) Selection, crossover and mutation: Individuals with high fitness are selected using methods such as roulette and tournaments. A new generation of parameter combinations is generated using single-point or multi-point crossover operators. Mutation is performed through small random perturbations to enhance search diversity.

[0163] 4) Termination and Output: When the number of iterations reaches the preset upper limit or the comprehensive evaluation index converges, the optimal individual is output. This is used as the optimal parameter set for PSD decomposition and small target detection.

[0164] Through the above-described parametric adaptive optimization process, the low-rank background matrix was optimized. and sparse defect matrix The adaptive splitting allows small-scale cracks to... The results show high contrast and high SNR sparse hotspots, while background noise and surface scratches are effectively suppressed.

[0165] S4. Based on the pre-built dual threshold range, obtain all pixels in the sparse defect matrix whose induced heat is within the dual threshold range, and generate a defect distribution map of the wheel tread.

[0166] The defect distribution map shows the distribution of each abnormal pixel in a discrete state.

[0167] In some embodiments, the process of constructing the dual threshold range includes:

[0168] Calculate the mean and standard deviation of induced heat in the sparse defect matrix, and based on the mean and standard deviation of induced heat, obtain the upper and lower limits of the dual threshold range.

[0169] Using a dual-threshold approach to extract defect pixels from a sparse defect matrix can effectively suppress noise such as surface scratches and corrosion. The dual-threshold representation is as follows:

[0170]

[0171] in, The upper limit of the double threshold. This is the lower limit of the dual threshold. and These are the mean and standard deviation of the induced heat in the sparse defect matrix, respectively. , and It is an empirical constant. It satisfies... and Pixels that are true are considered target pixels; otherwise, they are considered background pixels.

[0172] S5. Construct one or more target detection areas in the defect distribution map of the wheel tread; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and label the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

[0173] The target detection area includes both defective and non-defective parts.

[0174] It should be noted that thermal contrast is a quantitative indicator of the temperature difference between cracked and non-cracked areas in a thermal image. A higher thermal contrast value indicates a greater temperature difference between the cracked and non-cracked areas, making it easier for the detection system to identify cracks. Signal-to-noise ratio (SNR) is a metric used by the detection system to distinguish crack signals from background noise. A higher SNR value indicates that the crack signal is more prominent in the thermal image, and the noise has less impact on the detection results.

[0175] Based on the generated defect distribution map, in order to more clearly understand the quantitative information such as the type, size and location of defects, the target detection area is first selected, and then when the target detection area is determined to be a defect area, the defect situation within the defect area is quantitatively analyzed.

[0176] In some embodiments, the process of calculating thermal contrast for any target detection region includes:

[0177] The thermal contrast of the target detection area is obtained based on the temperature of the defective part, the temperature of the non-defective part, and the ambient temperature.

[0178] The formula for calculating thermal contrast ratio (TT) is as follows:

[0179]

[0180] in, The temperature of the defective part. The temperature of the non-defective portion. The ambient temperature.

[0181] In some embodiments, the process of calculating the signal-to-noise ratio for any target detection region includes:

[0182] The signal-to-noise ratio of the target detection area is obtained based on the average temperature of pixels in the entire target detection area, the average temperature of pixels in the defective part, the average temperature of pixels in the non-defective part, and the number of pixels in the target detection area.

[0183] The formula for calculating the signal-to-noise ratio (SNR) is as follows:

[0184]

[0185] in, It is the average temperature of the pixels in the entire target detection area. It is the average temperature of the pixels in the defective area. is the average temperature of pixels in the non-defective region, M is the number of pixels in the target detection region, and sqrt represents taking the square root of a given value. This is an intermediate parameter.

[0186] After analyzing thermal contrast (TT) and signal-to-noise ratio (SNR), the location and length of the crack can be determined by setting thresholds to filter out significant crack areas, and the crack location can be extracted using common image processing techniques such as connected component analysis and edge detection. The crack length is calculated by measuring the maximum extension of the crack boundary or using curve fitting methods. Higher thermal contrast and signal-to-noise ratio mean that it is easier to extract defect morphology and location information using algorithms.

[0187] S6. Extract each independent defect target in the defect region and obtain the geometric features of each independent defect target; map the geometric features of the independent defect targets onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

[0188] First, for the defect regions in the sparse defect matrix that satisfy the conditions of thermal contrast and signal-to-noise ratio, connected component analysis is performed within the defect regions to extract each independent defect target. For each individual defect target Calculate its area ,perimeter Spindle length Secondary axis length Direction angle Aspect Ratio Iso-geometric features.

[0189] Then, by using standard test blocks or calibration grids of known dimensions, a mapping relationship between thermal pixel coordinates and physical coordinates (mm) is established to obtain the scaling factor in the circumferential direction. (mm / pixel) and scaling factor in the radial (or axial) direction Using this calibration relationship, the principal axis length of the independent defect target is determined. and secondary axis length Convert to physical length ,width This allows us to obtain the length and width of the crack. Defect centroid coordinates After the same mapping, it is converted into the actual position on the wheel tread. This enables the quantitative output of crack location parameters.

[0190] In some embodiments, the method further includes: pre-constructing the correspondence between thermal contrast, signal-to-noise ratio, and geometric features and defect types; obtaining the thermal contrast and signal-to-noise ratio in the defect region, and determining the defect type of the independent defect target by combining the geometric features of the independent defect target.

[0191] Specifically, simulation and artificial defect specimen experiments can be used to establish the correspondence between thermal contrast, signal-to-noise ratio, geometric features, and defect types. For example:

[0192] when Larger (long and narrow), and and When both are high, the independent defect target will be... It was determined to be a typical fatigue crack;

[0193] when It is close to 1 (approximately blocky) and the area is relatively large, but Slightly below the level of a linear crack, it can be identified as a spalling or crushing defect;

[0194] When the area is very small and and At lower levels, scratches or surface noise tend to be removed using threshold rules.

[0195] In some embodiments, the above-mentioned multidimensional features ( , , , , Inputting a pre-trained multi-class decision tree or support vector machine (e.g.) enables automatic identification of defect types.

[0196] In some embodiments, the method further includes: pre-constructing quantitative relationship curves between different defect depths and thermal contrast; obtaining an evolution curve of thermal contrast in the defect region over time; selecting the peak value of thermal contrast in the evolution curve and substituting it into the inverse function of the quantitative relationship curve to obtain the depth value of the independent defect target.

[0197] Specifically, the quantitative relationship curves between different crack depths h and imaging thermal contrast TT can be obtained through COMSOL simulation. The calibration was performed using artificial defects of known depth in the experiment. In actual testing, each individual defect target... The evolution curve of TT over time was statistically analyzed. Take the peak value Substitute into the inverse function of the quantitative relationship curve The corresponding depth estimate is obtained. This enables quantitative assessment of crack depth.

[0198] Through the above steps, this invention not only uses TT and SNR to determine whether a defect exists, but also uses geometric features and simulation calibration to achieve comprehensive quantitative output of defect type, size, and location parameters (including length, width, depth, and spatial coordinates), meeting the engineering requirements for online monitoring and life assessment of EMU wheel tread cracks.

[0199] Based on the above embodiments, simulation experiments can also be constructed for verification.

[0200] (1) Simulation verification:

[0201] A coupled AC / DC and heat transfer model was established using COMSOL Multiphysics software to simulate the magnetic field distribution, eddy current distribution, and temperature field changes of cracks at different angles (0°, 45°, 90°) and of different sizes. During the simulation, the rationality of the sensor structure and the stability of the thermal imaging system under various dynamic conditions were verified by simulating the electromagnetic induction process of different crack morphologies.

[0202] (2) Experimental verification:

[0203] A simplified wheel specimen was selected, and artificial cracks with different parameters were machined. A dynamic detection experiment was conducted using a constructed detection system at three different speeds (25, 50, and 75 mm / s). The collected data were evaluated and compared to determine the effectiveness of the algorithm in detecting defects under rolling conditions.

[0204] Experimental verification, by comparing the location and size with actual cracks, ensures that the system can accurately identify and quantify minute cracks on the wheel surface under dynamic high-speed conditions.

[0205] (3) Result verification and elimination of false signals:

[0206] The test results are reviewed to eliminate possible spurious signals and ensure accuracy. By comparing experimental data with actual crack location information, the system's output crack location and size are verified to meet preset standards. The accuracy of crack identification and quantification is verified by comparing actual test results with simulation results. Simultaneously, a test report is generated, recording relevant crack information and providing repair suggestions.

[0207] Based on the above simulation experiments, Figure 7 A framework diagram for detecting wheel tread defects provided in an embodiment of the present invention is shown below. Figure 7 As shown, the entire defect detection process is divided into three stages.

[0208] The first stage involves data acquisition and system calibration: First, the system is powered on and performs a self-test. Then, the wheels are installed and positioned, and the transmission parameters are set. Next, the rectangular magnetic yoke sensor is installed and geometrically calibrated. Finally, the system is dynamically scanned and thermal images are acquired synchronously.

[0209] The second stage involves image enhancement and defect separation: First, the rotation speed between adjacent frames is estimated and curve-fitted with the wheel rotation speed. Then, thermal image unfolding and stitching reconstruction are performed based on the speed estimation. Next, blocks are generated and the image blocks are rearranged. Then, low-rank sparse decomposition and parameter adaptive optimization are adopted. Finally, adaptive double threshold segmentation and defect distribution map reconstruction are performed.

[0210] The third stage involves defect quantification and result output: First, defect areas are screened by calculating thermal contrast and signal-to-noise ratio. Then, the geometric size, location, and depth of independent defect targets are inverted. Finally, the results are verified, false signals are identified, and wheel repair suggestions are provided.

[0211] The beneficial effects of the technical solution in this embodiment are as follows:

[0212] On the one hand, by optimizing the magnetic field distribution through a rectangular magnetic yoke structure, a uniform magnetic field can be provided on complex surfaces, enhancing the ability to extract crack signals, especially maintaining high sensitivity during high-speed motion.

[0213] On the other hand, an image reconstruction algorithm based on motion compensation is used to improve the thermal contrast of cracks by eliminating dynamic blur. Through the calculation and analysis of thermal contrast and signal-to-noise ratio, the detection accuracy of cracks can be quantified, so that the size and depth of cracks can be assessed more accurately.

[0214] On the other hand, by applying the PSD image sparse decomposition method to the electromagnetic thermal imaging dynamic detection of wheel treads, the problem of motion blur at high speeds has been effectively overcome, and the sensitivity, anti-interference ability and quantification accuracy of microcrack detection have been significantly improved, providing strong technical support for the intelligent operation and maintenance of key components of the running gear of high-speed trains.

[0215] Figure 8 This is a schematic diagram of a wheel tread defect detection system provided in an embodiment of the present invention, as shown below. Figure 8 As shown, the system specifically includes: a sensing unit 100, a construction unit 200, a decomposition unit 300, and a generation unit 400.

[0216] The induction unit 100 is used to induce heat on the wheel tread by applying uniform excitation electromagnetic force to the wheel tread in motion.

[0217] The construction unit 200 is used to collect thermal images of each frame of the wheel tread in real time based on the induced heat generated, and construct a first thermal image sequence; based on the movement speed of the wheel tread, the first thermal image sequence is speed compensated and stitched together to reconstruct a second thermal image sequence.

[0218] Decomposition unit 300 is used to rearrange the second thermal image sequence into blocks to obtain the original image matrix; by finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix.

[0219] The generation unit 400 is used to obtain all pixels in the sparse defect matrix whose induced heat is within the double threshold range based on a pre-built double threshold range, and generate a defect distribution map of the wheel tread.

[0220] Furthermore, the construction unit 200 includes: an extraction subunit 2001, a first calculation subunit 2002, and a fusion subunit 2003.

[0221] Among them, the extraction subunit 2001 is used to extract stable feature points between adjacent frames in the first thermal image sequence;

[0222] The first calculation subunit 2002 is used to calculate the angular velocity between adjacent frames based on the spatial displacement of feature points between adjacent frames and the time interval between frames;

[0223] The fusion subunit 2003 is used to obtain the rotational speed of the wheel and fuse and correct the rotational speed of the wheel with the angular velocity between adjacent frames to obtain the compensated angular velocity between adjacent frames.

[0224] Furthermore, the construction unit 200 also includes: a second calculation subunit 2004, a sampling subunit 2005, a splicing subunit 2006, and a mapping subunit 2007.

[0225] The second calculation subunit 2004 is used to map each frame of thermal image onto an unfolded plane with circumferential angle and tread axial position as independent variables, and to calculate the angular displacement between adjacent frames based on the compensated angular velocity difference between adjacent frames.

[0226] The sampling subunit 2005 is used to sample the thermal image mapped in the unfolded plane at equal angular intervals and obtain the unfolded image sequence based on the angular displacement between adjacent frames;

[0227] The stitching subunit 2006 is used to select any frame of thermal image in the unfolded plane as a reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and stitch together each registered frame of thermal image.

[0228] The mapping subunit 2007 is used to remap each frame of the stitched thermal image back to the camera's field of view coordinate system to obtain the second thermal image sequence.

[0229] Furthermore, the system also includes a calibration unit 500.

[0230] The calibration unit 500 is used to construct one or more target detection areas in the defect distribution map of the wheel tread; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and calibrate the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

[0231] Furthermore, the system also includes a quantitative unit 600.

[0232] The quantitative unit 600 is used to extract independent defect targets in the defect area and obtain the geometric features of the independent defect targets; the geometric features of the independent defect targets are mapped onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

[0233] The quantitative unit 600 is also used to pre-build the correspondence between thermal contrast, signal-to-noise ratio, and geometric features and defect types; acquire the thermal contrast and signal-to-noise ratio in the defect area, and determine the defect type of the independent defect target by combining the geometric features of the independent defect target.

[0234] The quantitative unit 600 is also used to pre-construct quantitative relationship curves between different defect depths and thermal contrast; obtain the evolution curve of thermal contrast in the defect region over time; select the peak value of thermal contrast in the evolution curve and substitute it into the inverse function of the quantitative relationship curve to obtain the depth value of the independent defect target.

[0235] The beneficial effects produced by this embodiment can be found in the preceding text, and will not be repeated here.

[0236] Figure 9 This is a schematic diagram of the structure of an electronic device implementing the wheel tread defect detection method of this invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0237] like Figure 9 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0238] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0239] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as wheel tread defect detection methods.

[0240] In some embodiments, the wheel tread defect detection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the wheel tread defect detection method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the wheel tread defect detection method by any other suitable means (e.g., by means of firmware).

[0241] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0242] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0243] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0244] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0245] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0246] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0247] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0248] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting defects in wheel treads, characterized in that, include: By applying uniform excitation electromagnetic force to the wheel tread in motion, induced heat is generated on the wheel tread. Based on the induced heat, thermal images of each frame of the wheel tread are acquired in real time to construct a first thermal image sequence; based on the movement speed of the wheel tread, the first thermal image sequence is speed compensated and stitched together to reconstruct a second thermal image sequence. The second thermal image sequence is rearranged into blocks to obtain the original image matrix; by finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix. Based on the pre-constructed dual threshold range, all pixels in the sparse defect matrix whose induced heat falls within the dual threshold range are obtained, and a defect distribution map of the wheel tread is generated. The process of finding an optimal set of parameters includes: Each parameter in the parameter set is encoded according to a preset rule, and multiple candidate parameter individuals are randomly generated to construct an initial population; Based on any individual with any parameter in the initial population, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix, and the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix are calculated. Based on the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix, a comprehensive evaluation of the current parameter individual is carried out. Select the individual with the highest overall evaluation in the initial population, and generate a new generation population based on this individual. Then, find the individual with the highest overall evaluation in the new generation population. The process of iteratively executing the generation of a new generation of population based on the parameter individual with the highest comprehensive evaluation in the current population continues until the number of iterations reaches a preset upper limit or the comprehensive evaluation converges. At this point, the parameter individual with the highest comprehensive evaluation is taken as the optimal parameter set.

2. The method according to claim 1, characterized in that, The process of speed compensation includes: Stable feature points are extracted between adjacent frames in the first thermal image sequence; Calculate the angular velocity between adjacent frames based on the spatial displacement of feature points between adjacent frames and the time interval between frames; The rotational speed of the wheel is obtained, and the rotational speed of the wheel is fused and corrected with the angular velocity between adjacent frames to obtain the compensated angular velocity between adjacent frames.

3. The method according to claim 2, characterized in that, The process of splicing and reconstructing includes: Each frame of thermal image is mapped onto an unfolded plane with circumferential angle and axial position of tread surface as independent variables, and the angular displacement between adjacent frames is calculated based on the compensated angular velocity difference between adjacent frames. The thermal images mapped in the unfolded plane are sampled at equal angular intervals, and the unfolded image sequence is obtained based on the angular displacement between adjacent frames; Select any frame of thermal image in the unfolded plane as the reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and then stitch together each registered frame of thermal image. Each frame of the stitched thermal image is remapped back to the camera's field of view coordinate system to obtain the second thermal image sequence.

4. The method according to claim 1, characterized in that, The parameter set includes: nuclear norm regularization weight, L1 norm regularization weight, block size, block overlap ratio, and double threshold coefficient; Among them, the nuclear norm regularization weight is used to control the degree of low rank of the low-rank background matrix; L1 norm regularization weights are used to control the sparsity of the sparse defect matrix; The block size is used to split the original image matrix into multiple independently processed blocks; The block overlap ratio is used to limit the percentage of the overlap area between adjacent blocks to the size of a single block; The double threshold coefficient is used to distinguish elements in the low-rank background matrix and the sparse defect matrix.

5. The method according to claim 1, characterized in that, A comprehensive evaluation of the current parameters for each individual is expressed as follows: ； in, This represents the comprehensive evaluation indicators. , , These are empirical weighting coefficients. and These represent the thermal contrast and signal-to-noise ratio in the sparse defect matrix, respectively. This represents the temperature standard deviation in the low-rank background matrix.

6. The method according to claim 1, characterized in that, The process of constructing a dual threshold range includes: Calculate the mean and standard deviation of induced heat in the sparse defect matrix, and based on the mean and standard deviation of induced heat, obtain the upper and lower limits of the dual threshold range.

7. The method according to claim 1, characterized in that, After generating the defect distribution map of the wheel tread, the following steps are also included: In the defect distribution map of the wheel tread, construct one or more target detection areas; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and mark the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

8. The method according to claim 7, characterized in that, The target detection area includes defective and non-defective portions; For any target detection area, the process of calculating thermal contrast includes: obtaining the thermal contrast of the target detection area based on the temperature of the defective part, the temperature of the non-defective part, and the ambient temperature. For any target detection region, the process of calculating the signal-to-noise ratio includes: obtaining the signal-to-noise ratio of the target detection region based on the average temperature of the pixels in the entire target detection region, the average temperature of the pixels in the defective part, the average temperature of the pixels in the non-defective part, and the number of pixels in the target detection region.

9. The method according to claim 7, characterized in that, For any target detection region, its thermal contrast and signal-to-noise ratio meet preset conditions, specifically including: The thermal contrast of the target detection area is greater than or equal to the preset thermal contrast. In addition, the signal-to-noise ratio of the target detection area is greater than or equal to the preset signal-to-noise ratio.

10. The method according to claim 7, characterized in that, After defining the target detection area that meets the preset conditions for thermal contrast and signal-to-noise ratio as the defect area, the following steps are also included: Extract the independent defect targets in the defect region and obtain the geometric features of the independent defect targets; map the geometric features of the independent defect targets onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

11. The method according to claim 10, characterized in that, Also includes: Pre-build the correspondence between thermal contrast, signal-to-noise ratio, and geometric features and defect types; The thermal contrast and signal-to-noise ratio of the defect region are obtained, and the defect type of the independent defect target is determined by combining the geometric features of the independent defect target.

12. The method according to claim 10, characterized in that, Also includes: Quantitative relationship curves between different defect depths and thermal contrast are pre-constructed; Obtain the evolution curve of thermal contrast in the defect area over time; The peak value of the thermal contrast in the evolution curve is selected and substituted into the inverse function of the quantitative relationship curve to obtain the depth value of the independent defect target.

13. A wheel tread defect detection system, characterized in that, The system is configured to implement the method according to any one of claims 1-12, the system comprising: The induction unit is used to induce heat on the wheel tread by applying uniform excitation electromagnetic force to the wheel tread in motion. The construction unit is used to collect thermal images of each frame of the wheel tread in real time based on the induced heat generated, and construct a first thermal image sequence; based on the movement speed of the wheel tread, the first thermal image sequence is subjected to speed compensation and stitching reconstruction to obtain a second thermal image sequence. The decomposition unit is used to rearrange the second thermal image sequence into blocks to obtain the original image matrix; by finding an optimal set of parameters, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix. The generation unit is used to obtain all pixels in the sparse defect matrix whose induced heat is within the double threshold range based on a pre-built double threshold range, and generate a defect distribution map of the wheel tread. The process of finding an optimal set of parameters includes: Each parameter in the parameter set is encoded according to a preset rule, and multiple candidate parameter individuals are randomly generated to construct an initial population; Based on any individual with any parameter in the initial population, the original image matrix is ​​decomposed into a low-rank background matrix and a sparse defect matrix, and the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix are calculated. Based on the thermal contrast and signal-to-noise ratio in the sparse defect matrix and the temperature standard deviation in the low-rank background matrix, a comprehensive evaluation of the current parameter individual is carried out. Select the individual with the highest overall evaluation in the initial population, and generate a new generation population based on this individual. Then, find the individual with the highest overall evaluation in the new generation population. The process of iteratively executing the generation of a new generation of population based on the parameter individual with the highest comprehensive evaluation in the current population continues until the number of iterations reaches a preset upper limit or the comprehensive evaluation converges. At this point, the parameter individual with the highest comprehensive evaluation is taken as the optimal parameter set.

14. The system according to claim 13, characterized in that, The building unit includes: Extraction subunits are used to extract stable feature points between adjacent frames in the first thermal image sequence; The first calculation subunit is used to calculate the angular velocity between adjacent frames based on the spatial displacement of feature points between adjacent frames and the time interval between frames; The fusion subunit is used to obtain the rotational speed of the wheel and fuse and correct the rotational speed of the wheel with the angular velocity between adjacent frames to obtain the compensated angular velocity between adjacent frames.

15. The system according to claim 14, characterized in that, The building unit also includes: The second calculation subunit is used to map each frame of thermal image onto an unfolded plane with circumferential angle and axial position of tread surface as independent variables, and to calculate the angular displacement between adjacent frames based on the compensated angular velocity difference between adjacent frames. The sampling subunit is used to sample the thermal image mapped in the unfolded plane at equal angular intervals and obtain the unfolded image sequence based on the angular displacement between adjacent frames; The stitching subunit is used to select any frame of thermal image in the unfolded plane as a reference frame, register other frames of thermal images by advancing or delaying the angular displacement between adjacent frames, and stitch together each registered frame of thermal image. The mapping subunit is used to remap each frame of the stitched thermal image back to the camera's field of view coordinate system to obtain the second thermal image sequence.

16. The system according to claim 13, characterized in that, The system also includes: The calibration unit is used to construct one or more target detection areas in the defect distribution map of the wheel tread; calculate the thermal contrast and signal-to-noise ratio in the target detection areas; and calibrate the target detection areas whose thermal contrast and signal-to-noise ratio meet the preset conditions as defect areas.

17. The system according to claim 16, characterized in that, The system also includes: A quantitative unit is used to extract independent defect targets in the defect region and obtain the geometric features of the independent defect targets; the geometric features of the independent defect targets are mapped onto an unfolded plane with circumferential angle and tread axial position as independent variables to determine the physical position of the independent defect targets.

18. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to said at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the steps of the method according to any one of claims 1-12.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to perform the steps of the method according to any one of claims 1-12.

Images