Railway sleeper crack detection method and system based on three-dimensional feature construction

Abstract

The invention provides a railroad sleeper crack detection method and system based on three-dimensional feature construction, and the method comprises the steps: obtaining sleeper data: obtaining a sleeper image and a thermal image; a crack sleeper screening step: screening the images of the sleepers with cracks, and calling the thermal images of the sleepers; a three-dimensional feature construction step: extracting a plane crack area, calculating the crack depth according to the thermal image through an inversion strategy, and adding the crack depth to the depth of the plane crack area to form a three-dimensional crack; a crack safety evaluation step: calibrating a main crack and a branch crack diffused with the main crack in the three-dimensional crack, analyzing a crack safety coefficient according to the size data of the main crack and the size data of the branch crack through an evaluation strategy, and judging whether the sleeper needs to be replaced or not according to the crack safety coefficient; the method has the advantages that a quantitative safety coefficient evaluation model is established through three-dimensional size analysis of the main crack and the branch cracks, and the structural stability of the sleeper can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of railway sleeper detection, and more specifically to a railway sleeper crack detection method and system based on three-dimensional feature construction. Background Art

[0002] As a core component of the track structure, railway sleepers play a key role in transmitting train loads, maintaining gauge stability, and dispersing stresses. During long-term service, sleepers are prone to cracks due to cyclic loads, environmental erosion, and construction defects.

[0003] Existing detection techniques include visual inspection and ultrasonic inspection. The defects of relying solely on visual inspection are as follows: ①. Depth blind spot: It can only detect surface opening cracks, and the missed detection rate of internal closed cracks (depth > 3mm) reaches 100%; ②. Direction dependence: Vertical cracks (such as on the side of the sleeper) are easily misjudged as defect-free due to the shadow effect, while the width measurement error of oblique cracks (angle < 30°) reaches ±40%; ③. Resolution bottleneck: The pixel limit of industrial cameras results in blurred edges of cracks at the 0.1mm level. The sub-pixel algorithm (error ±0.02mm) needs to be combined, but the computational complexity increases.

[0004] The defects of ultrasonic inspection are as follows: ①. Multi-interface scattering: The steel bar network inside the sleeper (spacing 150mm) causes the acoustic wave reflectivity to fluctuate by ±30%, and phased array focusing technology needs to be adopted (cost increases by 200%); ②. Geometric occlusion: A 10mm blind spot is formed behind the bolt hole (diameter 32mm), and multi-angle scanning (such as a 45° / 90° combination) is required, resulting in a 60% decrease in detection efficiency; ③. Mode confusion: The difference in the propagation speeds of shear waves and longitudinal waves in concrete (3.5km / s vs 5.9km / s) causes echo overlap, and time-frequency analysis (such as wavelet transform) is needed for separation, with a computational delay of up to 50ms.

[0005] In summary, the current sleeper crack detection is difficult to achieve the safety assessment of sleepers, there are detection bottlenecks, and it cannot meet the purposes of railway intelligent and safety detection. Summary of the Invention

[0006] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a railway sleeper crack detection method and system based on three-dimensional feature construction, which breaks through the limitations of traditional two-dimensional detection based on multi-modal data fusion, improves the crack recognition accuracy, and establishes a quantitative safety factor evaluation model through the three-dimensional size analysis of main cracks and branch cracks, and can accurately judge the structural stability of sleepers.

[0007] To achieve the above purpose, the present invention provides the following technical solutions: A railway sleeper crack detection method based on three-dimensional feature construction, including the following steps: A sleeper data acquisition step, which is to acquire sleeper images taken by a visual camera on the bottom of the train when the train is running and thermal images captured by an infrared camera; a cracked sleeper screening step, screening the sleeper images with cracks in the sleeper images by target detection, and retrieving the thermal imaging image of the sleeper; A three-dimensional feature construction step, constructing a two-dimensional coordinate system of the sleeper surface according to the screened sleeper image, and extracting the plane crack area of ​​the crack in the sleeper image in the two-dimensional coordinate system, and then calculating the crack depth through an inversion strategy according to the thermal imaging image, constructing the z-axis coordinate on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and adding the crack depth to the depth of the plane crack area in the three-dimensional coordinate system to form a three-dimensional crack; The crack safety assessment step is to calibrate the main crack and the branch cracks that spread from the main crack in the three-dimensional crack, analyze the crack safety factor of the main crack size data and the branch crack size data through an assessment strategy, and determine whether the sleeper needs to be replaced based on the crack safety factor.

[0008] Furthermore, the crack safety assessment step includes a crack classification strategy, and the crack classification strategy includes a crack segment differentiation step and a crack segment connection judgment step. The crack segment distinguishing step includes enclosing each continuous crack segment as a single crack in the three-dimensional crack by an encircling ring; The crack connection judgment step records the number of connections between each single crack and other single cracks as the connection number of the single crack, and the single crack with the largest number of connections is taken as the main crack, and the other single cracks are all branch cracks.

[0009] Furthermore, the crack classification strategy also includes a crack classification verification step. The crack classification verification step obtains the main crack size data and the branch crack size data output by the crack connection judgment step in the three-dimensional crack, compares the aspect ratio of the main crack with the aspect ratio of the branch crack or compares the depth of the main crack with the depth of the branch crack, and verifies whether the main crack and branch crack output by the crack connection judgment step are correct based on the comparison result. If not, the crack segment differentiation step is repeated.

[0010] Further, the evaluation strategy includes marking the end point of the branch crack far from the main crack as the first connection point, marking the connection point of the branch crack and the main crack as the second connection point, connecting the first connection point and the second connection point as the inclined crack of the branch crack, connecting the second connection point of the branch crack and the second connection point of the next branch crack as the straight crack of the main crack, calculating the included angle value between the inclined crack and the straight crack, and then calculating the crack safety factor by the evaluation algorithm through the length, width and depth of the main crack, the number of branch cracks, the length, width and depth of each branch crack, and the included angle value between each branch crack and the main crack.

[0011] Further, the evaluation algorithm is configured as: , , , , wherein, is the main crack reference risk parameter; is the position coordinate along the length of the main crack; is the length of the main crack; is the depth of the main crack; is the width of the main crack; is the length of the i-th branch crack; is the width of the i-th branch crack; is the depth of the i-th branch crack; is the included angle value between the i-th branch crack and the main crack; ; is the influence factor of the i-th branch crack; is the series order; is the factorial function; is the number of branch cracks; is the main crack risk correction term; is the comprehensive influence factor of branch cracks; is the crack safety factor, is the fractal series function; is the standardized logic function.

[0012] Further, in the three-dimensional feature construction step, a two-dimensional coordinate system is constructed in the sleeper image in the length direction and width direction of the sleeper, the outer contour points of the cracks in the sleeper image are screened according to the edge detection algorithm, the outer contour points are mapped in the two-dimensional coordinate system, and the adjacent outer contour points are connected to form a plane crack region.

[0013] Further, the inversion strategy includes a crack location step, a temperature data analysis step, and a heat conduction inversion step. In the crack positioning step, each edge contour point of the planar crack region in the sleeper image is mapped in the thermal imaging map in the form of coordinate points as the crack position; In the temperature data analysis step, multiple points are selected from the crack positions to record the temperature change curves over time, and the temperature difference between the crack positions and the normal positions of the sleepers is analyzed; In the heat conduction inversion step, a heat conduction model is constructed, and the temperature difference is substituted into the heat conduction model for inversion calculation to obtain the crack depth.

[0014] Furthermore, the inversion strategy also includes a vibration noise filtering step, which eliminates the jitter of the thermal imaging map to update the temperature difference between the crack position and the normal position of the sleeper, and obtains the vibration amplitude and correction coefficient to correct the heat conduction model to obtain the filtered heat diffusion rate. The crack depth is recalculated by inversion using the filtered diffusion rate and the updated temperature difference between the crack position and the normal position of the sleeper.

[0015] Furthermore, the crack safety assessment step includes a priority judgment sub-step. In the priority judgment sub-step, when there are two main cracks or two independent three-dimensional cracks in the three-dimensional feature construction step, the sleeper is directly output as a failed sleeper, and a replacement instruction is output.

[0016] A railway sleeper crack detection system based on three-dimensional feature construction, including a sleeper data acquisition module, which acquires the sleeper image taken by a visual camera carried under the vehicle body during train operation and the thermal imaging map captured by an infrared camera; A cracked sleeper screening module, which screens the sleeper images with cracks through object detection in the sleeper images and retrieves the thermal imaging map of the sleeper; A three-dimensional feature construction module, which constructs a two-dimensional coordinate system on the surface of the sleeper according to the selected sleeper image, extracts the planar crack region of the crack in the sleeper image in the two-dimensional coordinate system, and then calculates the crack depth through the inversion strategy according to the thermal imaging map. A z-axis coordinate is constructed on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and the crack depth is added to the depth of the crack region in the three-dimensional coordinate system to form a three-dimensional crack; A crack safety assessment module, which calibrates the main crack and the branch cracks spreading from the main crack in the three-dimensional crack, analyzes the crack safety factor through the evaluation strategy with the size data of the main crack and the size data of the branch cracks, and determines whether the sleeper needs to be replaced according to the crack safety factor.

[0017] Advantages of the present invention: By combining visual images with infrared thermal imaging, multi-dimensional feature extraction of cracks is achieved, breaking through the limitations of traditional two-dimensional detection, improving the crack recognition accuracy. Through a two-dimensional coordinate system and depth inversion strategy, a three-dimensional spatial distribution model of cracks is constructed to accurately quantify the crack depth (in millimeters) and spatial morphology, providing a reliable basis for safety assessment. And through the three-dimensional size analysis of main cracks and branch cracks, a quantitative safety factor assessment model is established, which can accurately judge the structural stability of the sleeper and provide scientific decision-making support for preventive maintenance. In addition, it is applicable to train driving scenarios under different working conditions, and a single analysis is carried out for different train driving in real time to ensure real-time monitoring of the sleeper and also improve the detection robustness in complex environments. Description of the Drawings

[0018] Figure 1 is the overall flowchart in the present invention; Figure 2 is the flowchart for inverting the crack depth from the thermal imaging diagram in the present invention; Figure 3 is the flowchart for crack classification in the present invention; Figure 4 is a partial diagram for crack recognition in the present invention; Figure 5 is the connection diagram of the detection system module in the present invention. Detailed Embodiment

[0019] The present invention will be further described in detail below in conjunction with the drawings and embodiments. The same components are denoted by the same reference numerals. It should be noted that the terms "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to the directions in the drawings, and the terms "bottom surface" and "top surface", "inner" and "outer" refer to the directions towards or away from the geometric center of a specific component, respectively.

[0020] At present, it is difficult to achieve the safety assessment of sleepers in the detection of sleeper cracks, there are detection bottlenecks, and it cannot meet the purpose of railway intelligent and safety detection. Therefore, the present invention designs a railway sleeper crack detection method based on three-dimensional feature construction, as Figure 1 shown, including the following steps: Sleeper data acquisition step: acquiring the sleeper images taken by the line-scan camera carried under the vehicle body during train driving and the thermal imaging diagrams captured by the infrared camera. The line-scan camera and the infrared camera are triggered synchronously to ensure the spatio-temporal matching of the visual and infrared images of the same sleeper. Among them, dynamic compensation technology is used to eliminate blurring, making the captured images more stable.

[0021] Cracked sleeper screening step: as Figure 4 shown, screening the sleeper images with cracks in the sleeper images through target detection and retrieving the thermal imaging diagrams of the sleeper.

[0022] Among them, the steps of identifying cracks in the sleeper image include: ① Data collection and preprocessing. A large number of sleeper images with different scenarios, lighting conditions, and crack types are collected, which can be taken on-site from the railway or obtained through simulation experiments; ② Data annotation. Use annotation tools to annotate the cracks in the sleeper image, clarify the location and boundary of the cracks, and the annotation format is usually XML; ③ Data augmentation: To increase the diversity of data and the generalization ability of the model, perform data augmentation operations on the original images. Common augmentation methods include rotation, flipping, scaling, brightness adjustment, contrast adjustment, etc.; ④ Dataset division: Divide the annotated dataset into a training set, a validation set, and a test set, with a ratio of usually 7:2:1. The training set is used for model training, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the model.

[0023] When 10 cracked sleeper images are selected from 1000 sleeper images, the numbering of the sleepers will be marked. The marking method is that the train marks 1 for the first sleeper passing through the starting point, and counts and marks when passing through subsequent sleepers. And the sleepers with numbered marks and all the captured sleeper images are transmitted to the database for storage.

[0024] Steps for constructing three-dimensional features: Construct a two-dimensional coordinate system on the sleeper surface based on the selected sleeper image, and extract the planar crack region of the crack in the two-dimensional coordinate system in the sleeper image. Specifically, in the sleeper image, construct a two-dimensional coordinate system in the length direction and width direction of the sleeper. According to the edge detection algorithm (object detection model), screen out the outer contour points of the crack in the sleeper image, map the outer contour points in the two-dimensional coordinate system, and connect adjacent outer contour points to form a planar crack region. Then, calculate the crack depth through the inversion strategy based on the thermal imaging map. Construct the z-axis coordinate (height direction of the sleeper) on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and attach the crack depth to the depth of the crack region in the three-dimensional coordinate system to form a three-dimensional crack; In railway detection, the application of thermal imaging technology in the detection of sleeper crack depth has become a common method, and its principle is temperature inversion calculation, such as Figure 2 shown. The inversion strategy in the present invention includes a crack positioning step, a temperature data analysis step, and a heat conduction inversion step. Crack positioning step: Map each edge contour point of the planar crack region in the sleeper image to the thermal imaging map in the form of coordinate points as the crack position. The purpose is to identify the crack position in the thermal imaging map. According to the heat situation of the thermal imaging map, the crack position can also be known. However, by combining the crack coordinates in the visual image with the thermal imaging map, the purpose of mutual correspondence and verification can be achieved, making the crack positioning more accurate; Temperature data analysis steps: Select multiple points at the crack location and record the temperature change curve over time, and analyze the temperature difference between the crack location and the normal position of the sleeper; Let the average temperature in the crack area be , and the average temperature in the normal area (normal position of the sleeper) be , then the temperature difference is: , and the time when the train passes through the sleeper can be calculated by the train speed and the sleeper width : , and the temperature difference and the time are the key input parameters for the subsequent heat conduction model; Heat conduction inversion steps: Construct a heat conduction model, substitute the temperature difference into the heat conduction model for inversion calculation to obtain the crack depth. The heat conduction satisfies the one-dimensional Fourier equation: , where: is the thermal diffusivity of the sleeper material ( ), is the coordinate in the depth direction. By the method of separation of variables or Laplace transform, the relationship between the crack depth and the temperature difference can be obtained: , where, is the heat flux density per unit area ( ), is the material density ( ), is the specific heat capacity ( ). When the crack is relatively deep, , approximately . For the parameters of concrete sleepers: , , , assuming , , , then .

[0025] Since when the train is running, vibration will occur after the wheel axle contacts the track, resulting in deviations in the thermal imaging map and the heat conduction model. Therefore, the inversion strategy also includes a vibration noise filtering step to eliminate jitter of the thermal imaging map to update the temperature difference between the crack location and the normal position of the sleeper. First, it is necessary to analyze the vibration frequency and amplitude through Fourier transform, , , is the vibration signal obtained by the sensor, usually set on the train, and of course it can also be set on the sleeper, is the vibration period, is the number of acquisition points, motion compensation, estimating the inter-frame displacement through feature point matching, performing a translation transformation on the image, using multi-frame averaging technology to suppress vibration noise, eliminating image blur caused by vibration, and the anti-shake image is used to improve the temperature difference of the measurement accuracy, and obtaining the vibration amplitude and correction coefficient to correct the heat conduction model to obtain the filtered thermal diffusivity. The filtered temperature difference is: , where is the Fourier transform, is the filter transfer function, heat conduction model correction, vibration causes the diffusivity to change, and a predetermined correction coefficient is introduced : , and the crack depth is re-inversely calculated through the filtered diffusivity and the temperature difference between the updated crack position and the normal position of the sleeper. , assuming that , , the filtered temperature difference , the corrected diffusivity , .

[0026] Crack safety assessment steps: Calibrate the main crack and the branch cracks spreading from the main crack in the three-dimensional crack, analyze the crack safety factor through the evaluation strategy with the size data of the main crack and the size data of the branch cracks, and judge whether the sleeper needs to be replaced according to the crack safety factor. Among them, a threshold range is set. The threshold range includes that the safety factor between {0.5, 1} is the warning state, between {0, 0.5} is the failure state, and between {1, 2} is the safety state.

[0027] Specifically, as Figure 3 shown, the crack safety assessment steps include a crack classification strategy. The crack classification strategy includes a crack segment distinction step. Usually, the crack is vertically cracked downward on the surface, so only the two-dimensional map of the crack plane area needs to be analyzed. However, for more accurate crack analysis, it is analyzed in the three-dimensional crack. Each continuous crack segment is surrounded by a surrounding circle to obtain a single crack. Assuming the crack is a long strip crack, and many branch cracks extend from the long strip crack, and the long strip crack is cracked along the length direction of the sleeper, then taking one end point of the long strip crack as the starting point, find the end point that is continuously connected to the starting point as the end point of the crack. The crack between the starting point and the end point is a single crack, and then surround other cracks in the same way; Crack connection judgment step: Record the connection quantity between each single crack and other single cracks as the connection number of this single crack. Take the single crack with the most connection numbers as the main crack, and other single cracks are all branch cracks. Usually, branch cracks are all connected to the main crack. This method of judging the main crack and branch cracks through connection can accurately identify the main-branch relationship in the crack network, and the main crack positioning accuracy rate reaches 95%.

[0028] In order to improve the analysis accuracy of crack types, the crack classification strategy in the present invention further includes a crack classification verification step. Obtain the main crack size data and the size data of branch cracks output by the crack connection judgment step in the three-dimensional crack. The size data mainly includes the length, brightness, and depth of the crack. Compare the aspect ratio of the main crack with the aspect ratio of the branch crack or compare the depth of the main crack with the depth of the branch crack. According to the comparison results, verify whether the main crack and branch cracks output by the crack connection judgment step are correct. If not, re-perform the crack segment distinction step. Usually, the aspect ratio of the main crack is greater than that of the branch crack, and the depth of the main crack is greater than that of the branch crack.

[0029] Specifically, the evaluation strategy includes marking the endpoint of the branch crack far from the main crack as the first connection point, marking the connection point of the branch crack and the main crack as the second connection point, connecting the first connection point and the second connection point as the oblique crack (straight line segment) of this branch crack, connecting the second connection point of this branch crack and the second connection point of the next branch crack as the straight crack of the main crack, calculating the included angle value between the oblique crack and the straight crack, and then calculating the crack safety factor through the evaluation algorithm by using the length, width, and depth of the main crack, the number of branch cracks, the length, width, and depth of each branch crack, and the included angle value between each branch crack and the main crack. Since the width of the crack may be partially wider, the width of the crack in the evaluation formula is calculated based on the widest value of this crack.

[0030] The evaluation algorithm includes a main crack benchmark risk calculation formula, a single branch crack influence factor calculation formula, a branch crack comprehensive influence calculation formula, and a safety factor calculation formula. Among them, the main crack benchmark risk calculation formula is configured as: , Among them, is the main crack benchmark risk parameter. Through the integral operation of the Gaussian kernel function and the error function ( ), the comprehensive influence of the geometric characteristics (length, width, depth) of the main crack on the structural safety is quantified. The larger the value, the more serious the threat of the main crack to the structural integrity; is the position coordinate along the length of the main crack. Taking one endpoint (detection starting point) of the main crack as the coordinate origin ( ), the position parameter in the one-dimensional coordinate system established along the derivative direction of the main crack, with the integration interval of 0, directly corresponding to the full length range of the main crack; is the length of the main crack. The end point of the main crack far from the starting point is the end point, and the distance between this end point and the starting point is the length of the main crack; is the depth of the main crack, which is the deepest depth of the crack perpendicular to the sleeper surface; is the width of the main crack, which is the maximum transverse dimension of the crack opening, is the Gaussian error function; the integral term reflects the risk distribution in the crack length direction.

[0031] The calculation formula configuration of the single-branch crack influence factor is: , where, is the length of the i-th branch crack; is the width of the i-th branch crack; is the gamma function, and the input is the depth of the branch crack , and its purpose is to amplify the threat of deep cracks. When , , when , , reflecting the characteristic of "doubling the depth, non-linear growth of risk"; is the depth of the i-th branch crack; is the hyperbolic tangent function, converts the included angle to radians, and its function is to limit the angle influence within the interval {0,1}; is the series order, which is the summation loop variable and is used to construct the polynomial influence of the branch crack length ; is the factorial function; is used to punish wide and shallow branches ( ); is the included angle value between the i-th branch crack and the main crack, is the included angle whose absolute value of cosine controls the weight of the series terms; is the influence factor of the i-th branch crack. Through multi-parameter non-linear coupling, it quantifies the comprehensive influence of the geometric characteristics (length, width, depth) of a single branch crack and its included angle with the main crack on the structural safety, The larger the value, the more significant the threat of this branch crack to the safety of the structure.

[0032] The calculation formula configuration of the comprehensive influence of branch cracks is: , where, is the number of branched cracks; is the main crack risk correction term, , to prevent from the division-by-zero error when it is 0, taking the main crack reference risk as the normalization reference for the influence of branched cracks. When the main crack risk , , then the influence of branched cracks is suppressed; is the comprehensive influence factor of branched cracks, coupling the threats of all branched cracks through a geometric series, reflecting the synergistic effect of the multi-crack system. When (no branched cracks), , the more branched cracks, that is or a single , the smaller the value (the threat increases).

[0033] The safety factor calculation formula is configured as: , where is the crack safety factor, is the fractal series function, , when s = 1 / 2, this function characterizes the fractal characteristics of crack propagation, and its convergence domain is , input inverts the main crack risk parameter into an inhibition factor ( ), reflecting the non-linear attenuation of the main crack risk on the safety factor. When , (the convergence value before the divergence critical point). When , , the series converges rapidly (such as when , ); is the comprehensive influence factor of branched cracks, and its square root reduces the numerical sensitivity, avoiding the denominator from fluctuating violently caused by small values (high threats), reflecting the sub-linear growth of the multi-crack coupling effect. is the normalized logic function, , then , weighting the influence of branched cracks (the coefficient 3 is a set number, and the equivalent threat of the branched crack group is usually on the order of 1 / 3 of the main crack, and it needs to be amplified by 3 times to match the main-secondary risk balance). When the branched threat exceeds the main crack reference risk, it triggers a rapid decay of the safety factor.

[0034] Data calculation example: Main crack: , , ; Branched crack 1: , , , ; Branch crack 2: , , , ; Main crack reference risk: ; Integral kernel is a normal distribution curve centered at 30mm with a standard deviation of 15mm, and the integration interval is {0,60}. Through numerical integration: ; Error function: (Saturation value of the error function); ; Branch crack 1: ; , , ; Series term: ; ; After the series term is corrected: ; Exponential term: ; ; Similarly, for branch crack 2: ; Comprehensive influence of branch cracks: ; For : ; For : ; Exponential term: ; ; Safety factor: ; Fractal series: ; ; ; ; Key factor analysis: 1. Main crack dominance: Maximal ( ), resulting in a sharp attenuation of the fractal series term; 2. Weak branch influence: The long branch crack due to the gamma function and is strongly inhibited; 3, Threshold triggering: makes the denominator approach 1, but the numerator has decayed to a dangerous level.

[0035] The crack safety assessment step includes a priority judgment sub-step. In the priority judgment sub-step, when there are two main cracks or two independent three-dimensional cracks in the three-dimensional feature construction step, the sleeper is directly output as a failed sleeper, and a replacement instruction is output.

[0036] Based on the crack detection method, a railway sleeper crack detection system based on three-dimensional feature construction is configured, such as Figure 5 shown, including: A sleeper data acquisition module that acquires the sleeper image captured by the visual camera carried under the vehicle body during train operation and the thermal imaging map captured by the infrared camera; A cracked sleeper screening module that screens out the sleeper images with cracks in the sleeper image through target detection and retrieves the thermal imaging map of the sleeper; A three-dimensional feature construction module that constructs a two-dimensional coordinate system on the sleeper surface according to the screened sleeper image, extracts the planar crack region of the crack in the sleeper image in the two-dimensional coordinate system, then calculates the crack depth through an inversion strategy based on the thermal imaging map, constructs the z-axis coordinate on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and attaches the depth of the crack region to the crack depth in the three-dimensional coordinate system to form a three-dimensional crack; A crack safety assessment module that calibrates the main crack and the branch cracks spreading from the main crack in the three-dimensional crack, analyzes the crack safety factor through an evaluation strategy with the size data of the main crack and the size data of the branch cracks, and determines whether the sleeper needs to be replaced according to the crack safety factor.

[0037] The above detection method and system both combine visual images and infrared thermal imaging to realize multi-dimensional feature extraction of cracks, break through the limitations of traditional two-dimensional detection, improve the crack recognition accuracy, construct a three-dimensional space distribution model of cracks through a two-dimensional coordinate system and a depth inversion strategy, accurately quantify the crack depth (millimeter level) and spatial form, and provide a reliable basis for safety assessment; and through the three-dimensional size analysis of the main crack and the branch cracks, establish a quantitative safety factor assessment model, which can accurately judge the structural stability of the sleeper, provide scientific decision-making support for preventive maintenance, and is also applicable to train running scenarios under different working conditions, perform a single analysis for different train runs in real time to ensure real-time monitoring of the sleeper, and can also improve the detection robustness in complex environments.

[0038] In addition, the sleeper crack detection method and detection system of the present invention are not only applicable to concrete sleepers, but also applicable to traditional wooden sleepers.

[0039] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A railway sleeper crack detection method based on three-dimensional feature construction, characterized in that: The steps include: A sleeper data acquisition step, which is to acquire sleeper images taken by a visual camera on the bottom of the train when the train is running and thermal images captured by an infrared camera; a cracked sleeper screening step, screening the sleeper images with cracks in the sleeper images by target detection, and retrieving the thermal imaging image of the sleeper; A three-dimensional feature construction step, constructing a two-dimensional coordinate system of the sleeper surface according to the screened sleeper image, and extracting the plane crack area of ​​the crack in the sleeper image in the two-dimensional coordinate system, and then calculating the crack depth through an inversion strategy according to the thermal imaging image, constructing the z-axis coordinate on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and adding the crack depth to the depth of the plane crack area in the three-dimensional coordinate system to form a three-dimensional crack; The crack safety assessment step is to calibrate the main crack and the branch cracks that spread from the main crack in the three-dimensional crack, analyze the crack safety factor of the main crack size data and the branch crack size data through an assessment strategy, and determine whether the sleeper needs to be replaced based on the crack safety factor.

2. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 1 is characterized in that: The crack safety assessment step includes a crack classification strategy, which includes a crack segment distinction step and a crack segment connection judgment step. The crack segment distinguishing step includes enclosing each continuous crack segment as a single crack in the three-dimensional crack by an encircling ring; The crack connection judgment step records the number of connections between each single crack and other single cracks as the connection number of the single crack, and the single crack with the largest number of connections is taken as the main crack, and the other single cracks are all branch cracks.

3. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 2 is characterized in that: The crack classification strategy also includes a crack classification verification step. The crack classification verification step obtains the main crack size data and the branch crack size data output by the crack connection judgment step in the three-dimensional crack, compares the aspect ratio of the main crack with the aspect ratio of the branch crack or compares the depth of the main crack with the depth of the branch crack, and verifies whether the main crack and branch crack output by the crack connection judgment step are correct based on the comparison result. If not, the crack segment differentiation step is repeated.

4. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 1 is characterized in that: The evaluation strategy includes marking the end point of the branch crack away from the main crack as the first connection point, marking the connection point between the branch crack and the main crack as the second connection point, taking the line connecting the first connection point and the second connection point as the oblique crack of the branch crack, taking the line connecting the second connection point of the branch crack and the second connection point of the next branch crack as the straight crack of the main crack, calculating the angle between the oblique crack and the straight crack, and then calculating the length, width and depth of the main crack, the number of branch cracks, the length, width and depth of each branch crack, and the angle between each branch crack and the main crack through an evaluation algorithm to obtain the crack safety factor.

5. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 4 is characterized in that: The evaluation algorithm is configured as follows: ， ， ， ， in, is the benchmark risk parameter for the main crack; is the position coordinate along the length of the main crack; is the main crack length; is the main crack depth; is the main crack width; is the length of the i-th branch crack; is the width of the i-th branch crack; is the depth of the i-th branch crack; is the angle between the i-th branch crack and the main crack; ; is the influence factor of the i-th branch crack; is the series order; is the factorial function; is the number of branch cracks; is the main crack risk correction item; is the comprehensive influencing factor of branch cracks; is the crack safety factor, is a fractal series function; is the normalized logistic function.

6. The railway sleeper crack detection method based on three-dimensional feature construction according to any one of claims 1 to 4, characterized in that: The three-dimensional feature construction step constructs a two-dimensional coordinate system in the sleeper image in the length direction and the width direction of the sleeper, selects the outer contour points of the crack in the sleeper image according to the edge detection algorithm, maps the outer contour points in the two-dimensional coordinate system, and connects adjacent outer contour points to form a planar crack area.

7. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 6 is characterized in that: The inversion strategy includes a crack location step, a temperature data analysis step, and a heat conduction inversion step. The crack locating step maps each edge contour point of the plane crack area in the sleeper image in the form of coordinate points in the thermal imaging image as the crack position; The temperature data analysis step is to select multiple points at the crack position to record the temperature change curve over time, and analyze the temperature difference between the crack position and the normal position of the sleeper; The heat conduction inversion step constructs a heat conduction model, substitutes the temperature difference into the heat conduction model, performs inversion calculation and obtains the crack depth.

8. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 7 is characterized in that: The inversion strategy also includes a vibration noise filtering step, which eliminates jitter on the thermal image to update the temperature difference between the crack position and the normal position of the sleeper, and obtains the vibration amplitude and correction coefficient to correct the heat conduction model to obtain the filtered thermal diffusivity. The crack depth is re-inverted and calculated using the filtered diffusivity and the updated temperature difference between the crack position and the normal position of the sleeper.

9. The railway sleeper crack detection method based on three-dimensional feature construction according to claim 1 is characterized in that: The crack safety assessment step includes a priority judgment sub-step. In the priority judgment sub-step, when there are two main cracks or two independent three-dimensional cracks in the three-dimensional feature construction step, the sleeper is directly output as a failed sleeper and a replacement instruction is output.

10. A railway sleeper crack detection system based on three-dimensional features, characterized in that: include: The sleeper data acquisition module acquires sleeper images taken by the visual camera on the bottom of the train and thermal images captured by the infrared camera when the train is running; A cracked sleeper screening module is used to screen cracked sleeper images in the sleeper images through target detection, and retrieve thermal imaging images of the sleepers; A three-dimensional feature construction module, which constructs a two-dimensional coordinate system of the sleeper surface according to the screened sleeper image, extracts the plane crack area of ​​the crack in the sleeper image in the two-dimensional coordinate system, calculates the crack depth through an inversion strategy according to the thermal image, constructs the z-axis coordinate on the basis of the two-dimensional coordinate system to form a three-dimensional coordinate system, and adds the crack depth to the depth of the crack area in the three-dimensional coordinate system to form a three-dimensional crack; The crack safety assessment module calibrates the main crack and the branch cracks that spread from the main crack in the three-dimensional crack, analyzes the crack safety factor of the main crack size data and the branch crack size data through an assessment strategy, and determines whether the sleeper needs to be replaced based on the crack safety factor.

Images