Rail surface defect detection method and system

Abstract

The invention relates to a rail surface defect detection method and system, and the method comprises the steps: collecting a reflection response sequence of a rail surface under various illumination conditions through the controllable dynamic disturbance of incident angles and illumination intensity of different positions of the rail surface, and representing the microscopic optical response behavior of a rail surface material; for each tiny area on the surface of the rail, dynamic reflection stability is calculated, and areas with abnormal stability are marked as defect candidate areas; analyzing the difference between the defect candidate area and a normal wheel-rail contact trace, identifying whether abnormal residues or abnormal abrasion boundaries generated due to defect evolution exist or not by combining optical heterogeneity boundary characteristics, and verifying the authenticity of the defect; for the judged defect candidate region, judging whether the optical response of the adjacent region has a general change trend or not along the contact track direction and the normal micro-region fluctuation direction of the rail surface, and determining the space connectivity and the diffusion trend of the defect and a growth defect region; the system executes the steps of the method through a processor.

Description

TECHNICAL FIELD

[0001] The present application relates to the detection of defects on the surface of a rail, and in particular to a method and system for detecting defects on the surface of a rail. BACKGROUND

[0002] At present, the prior art such as Chinese patent CN116645371A, a steel rail surface defect detection method and system based on feature search, although it has been systematically designed in terms of feature extraction, convolutional network fusion, multi-scale receptive field construction, etc., and has good end-to-end detection capability, but still has obvious technical limitations and deficiencies in practical application. First, this scheme mainly relies on the convolutional feature extraction of gray-scale images and depth maps, constructs a multi-scale feature fusion to generate a disease area heat map, and then performs secondary filtering through the depth map to determine the defect area. Although it can effectively detect macroscopic defects such as rail surface spalling, chunking, and cracking, it is difficult to ensure high-precision identification for small or hidden defects such as microscopic cracks, grain distortion, and early fatigue cracks on the rail surface, as the feature extraction error is easily caused by surface texture, light interference, or pollution. Since this method is completely based on the convolution of image features (multi-layer ResNet50 stacking + MRF multi-scale receptive field), it is extremely sensitive to texture-like pseudo-defects, especially under the influence of water stains, oil stains, or local residual images, the disease heat map is easily contaminated, and the subsequent depth Figure Two Although the secondary filtering can filter out some false defects with small height differences, it lacks an effective mechanism to distinguish pseudo-defects with similar optical properties (such as water film reflection and oil film scattering), and is prone to false positives or missed detection.

[0003] Secondly, this scheme highly depends on depth map, but in actual railway scene, track surface high reflection, light spot, pollution, oil stain can cause depth sensor failure or error, and further affect the determination of disease area. The limitation of depth map leads to insufficient detection accuracy of small scale erosion and micro cracks. In addition, the detection logic of the invention mainly focuses on feature processing of single frame image, although it is called end to end, but it does not fully utilize the time sequence information, and cannot dynamically perceive the evolution trend or cumulative fatigue expansion of defects, cannot judge whether the crack is static residue or expanding evolution, lacks time sequence stability analysis and dynamic evolution monitoring capability. Thirdly, this method does not introduce polarization detection and multispectral technology, and has no perception to optical disturbance under the condition of surface wet and slippery. In the face of rainy day inspection, water film covering and track surface lubricating oil residue, it is easy to identify flowing water film reflection, high light or oil stain as erosion or crack, and cannot effectively distinguish dynamic environmental disturbance from real disease. In addition, the existing technology has poor environmental adaptability, and the defect features are too dependent on the distribution of training data of the model. When the field environment is different from the training data (such as different reflection, pollution and wear path), the model performance is easy to degrade, and the generalization ability is limited. The locality of convolution network features leads to insufficient boundary detail processing capability, especially for crack endpoints and spatial derivative risk of small pieces, which cannot realize trend prediction. Furthermore, although this method expands the perception range through multi-scale receptive field, the layering of receptive field leads to large model calculation, although it claims to process a single 2048x4096 image at a speed of 100ms, but in the actual large-scale track full-range high-frequency inspection scene, there is still a speed bottleneck, especially in the continuous sampling dynamic scene, which cannot realize real-time online dynamic monitoring.

[0004] Finally, the method of the patent mainly relies on two-stage filtering of feature heat map and depth map, although it improves the robustness to a certain extent, but does not construct dynamic threshold regulation and growing area expansion mechanism, it is difficult to determine the spatial connectivity and diffusion priority path of crack in real time, cannot effectively evaluate the future development trend of crack, and lacks the ability to actively intervene in risk crack. In system integration, this method mainly faces static image reasoning, and does not perform joint perception and decision for multi-modal data (such as polarization, multispectral and time dynamic disturbance) under complex track surface state, the system design has single data dependence, which limits the environmental adaptability in engineering deployment.

[0005] In summary, the existing steel rail surface defect detection method based on feature search has certain technical advantages in feature fusion and multi-scale perception, but has many disadvantages such as insufficient micro crack recognition ability, poor environmental robustness, lack of dynamic evolution trend monitoring, limited pseudo defect distinguishing ability, strong dependence on depth map, lack of boundary derivative recognition, lack of time sequence perception, and inability to realize defect automatic growth and spatial expansion prediction. SUMMARY

[0006] The purpose of the present application is to provide a rail surface defect detection method and system, so as to solve some of the problems and deficiencies pointed out in the background art.

[0007] The present application solves the above-mentioned technical problems by adopting the following technical solution: a rail surface defect detection method, comprising: collecting a reflection response sequence of the rail surface under multiple illumination conditions by using dynamic disturbance with controllable incident angle and illumination intensity at different positions on the rail surface, to represent the micro-optical response behavior of the rail surface material; based on the collected reflection response sequence, calculating the dynamic reflection stability for each micro area of the rail surface, and marking the area with abnormal stability as a defect candidate area through the difference in area stability; Analyzing the difference between the defect candidate area and the normal wheel-rail contact trace, combining the optical heterogeneity boundary feature, identifying whether there is abnormal residue or non-normal wear boundary caused by defect evolution, to verify the authenticity of the defect; For the determined defect candidate area, judging whether the optical response of the adjacent area has a common change trend along the contact track direction and the normal micro area fluctuation direction of the rail surface, determining the spatial connectivity and diffusion trend of the defect, and growing the defect area.

[0008] Further, in the dynamic disturbance collection process, the polarization direction of the incident light is modulated to capture the reflection polarization abnormality caused by surface micro-cracks or grain distortion under the same incident angle and light intensity change conditions; during the reflection response sequence collection, different waveband multispectral light sources are used to distinguish between surface defects and potential internal damage of the material.

[0009] Further, the calculation of the dynamic reflection stability includes multi-time period sliding window analysis on the reflection response sequence to identify the fluctuation trend of the micro area at different time scales, and to distinguish between short-term fluctuations caused by external environmental changes and persistent stability abnormalities caused by material damage; the stability abnormality determination combines the spatial autocorrelation of the reflection response, and uses the stability correlation degree between the micro area and the surrounding area to eliminate false abnormal markers caused by local shielding or pollution.

[0010] Further, the difference analysis between the defect candidate area and the normal wheel-rail contact trace uses bidirectional scanning sampling to judge whether there is asymmetric change in the optical response under different contact directions, for identifying abnormal wear boundaries caused by defect evolution; the analysis of the optical heterogeneity boundary feature includes boundary continuity detection of the local reflection gradient of the defect candidate area to judge whether there is boundary disturbance caused by foreign matter attachment or peeling.

[0011] Further, the identification of abnormal residues combines the response delay characteristics. Under dynamic light disturbance, it detects whether there is an optical response time lag phenomenon in the micro area to distinguish the surface residue from the material body reflection difference; in the diffusion analysis of the defect area along the contact track direction, combined with the historical data of the track wear direction, the priority path relationship of defect diffusion is established to preferentially judge the defect connectivity trend along the material mechanical fatigue direction.

[0012] Further, the diffusion judgment of the normal micro area fluctuation direction includes fluctuation synchronicity analysis of the micro area response of the upstream and downstream of the defect candidate area, which is used to detect the spatial derivation risk of crack end points or micro block edges; the defect growth process adopts a dynamic correlation threshold strategy, which dynamically adjusts the growth boundary based on the stability change rate of the candidate area.

[0013] Further, in the defect candidate area verification process, the optical reflection change characteristics of the surface wet slip state are combined to identify the water film coverage or oil stain state, and to distinguish environmental interference from real defect characteristics; the time sequence stability fluctuation characteristics of the defect candidate area are used to determine whether there is an accumulated fatigue crack evolution, and the crack propagation direction is inferred through the dynamic fluctuation trend; In the process of verifying the defect candidate area, the optical characteristic parameters of the surface state change are introduced, and a dynamic fluctuation function is constructed based on the response time sequence fluctuation behavior to identify the difference between real crack characteristics and environmental interference; in order to distinguish the real crack evolution characteristics of the rail surface from the pseudo defect signals caused by water film and oil stain factors, a response stability variation function integral model is adopted, which is defined as follows: ; Wherein: represents the cumulative dynamic fluctuation index of the reflection response at time , incident angle , which is used to describe the defect evolution intensity; is the unit reflection intensity response at time and angle ; is the second time derivative of the reflection response, which reflects the acceleration change of the reflection signal, i.e. the fluctuation intensity; is the response delay offset factor related to the incident angle, which is used to adjust the non-uniformity of light propagation caused by the wet slip state; is the time decay background function, which is used to describe the slow change trend of the overall reflection caused by the water film or oil stain, and has low-frequency stability characteristics; is the dynamic weight function, which reflects the weight priority at a specific angle and time, and is used to emphasize the abnormal enhancement area near the critical reflection angle; is the sampling start time; When continuously in a plurality of exceeds a dynamic threshold If the region exceeds the dynamic threshold in a plurality of If the region exceeds the dynamic threshold in a plurality of

[0014] Further, the optical reflection change characteristic of the surface wet and slippery state includes detecting the critical reflection angle response to the incident light, and identifying whether the surface has the water film or oil film attachment phenomenon when the reflection critical angle deviates; the environment interference distinguishing includes comparing the reflection decay rate of the surface micro area, analyzing the time decay mode of the reflection intensity in the continuous disturbance sampling, and distinguishing the static defect from the false abnormal response caused by the dynamic water film flow.

[0015] Further, the environment interference identification is to judge the short-wave infrared absorption characteristic change in the water film state by using the multi-spectrum contrast response; the oil stain state judgment combines the polarization state change of the reflected light, and uses the polarization rate reduction phenomenon caused by the surface oil film to distinguish the oil stain interference from the defect reflection characteristic of the rail surface body.

[0016] A rail surface defect detection system, comprising: A dynamic light disturbance module for projecting a light beam with adjustable incident angle, light intensity, polarization direction and multi-spectrum band to the rail surface; a reflection collection unit for collecting the reflection signal of the rail surface under different light conditions; A data processing unit composed of a processor and a memory, for calculating the reflection stability of each micro area and marking the stability abnormal area according to the reflection signal: 1) Calculate the reflection stability of each micro area and mark the stability abnormal area; 2) Compare the normal wheel-rail contact trace to verify whether the abnormal area is a real defect; 3) Analyze the spatial connectivity and diffusion trend of the defect; 4) Use the critical reflection angle deviation, short-wave infrared absorption change and reflection polarization rate change to exclude the water film and oil film environment interference; An output unit for displaying the defect position, diffusion trend and risk warning information; the processor is programmed to execute the steps of the method.

[0017] The beneficial effects of the present application: through the joint collection of dynamic light disturbance, multi-spectral response and polarization information, the micro-optical behavior of the rail surface and near-surface layer is comprehensively perceived, which can effectively identify early defects such as micro-cracks, grain distortion and denudation, and make up for the shortcomings of traditional single-vision or profile collection means in easily missing fine defects, especially in complex environmental conditions, still maintaining high stability and high sensitivity detection. By introducing critical reflection angle offset detection, multi-spectral reflection contrast, reflection attenuation rate analysis, polarization rate change perception and other multi-dimensional features, the dynamic environmental interference such as water film, oil stain and dust attachment and the real defect response can be effectively distinguished, reducing the false positives and missed detections caused by environmental factors, and greatly improving the adaptability of the system in the actual railway scene. Through the time sequence sliding window and dynamic fluctuation integral function, the cumulative evolution process of the crack can be monitored, and the growth trend of the defect can be dynamically determined, supporting early warning of fatigue crack propagation, denudation boundary evolution and other problems, realizing the transformation from static detection to dynamic monitoring, helping the operation and maintenance department to intervene in the early stage of defect development, and reducing the risk of accidents.

[0018] The synchronous analysis of the contact trajectory direction and the normal fluctuation direction is adopted, combined with dynamic correlation threshold adjustment, to realize the spatial connectivity determination and expansion determination of the defect area, which can automatically identify the main expansion direction and boundary range of the crack, reduce the manual judgment error, and reduce the post-processing workload. Through the joint intelligent discrimination of environmental perception, time sequence evolution and space growth, the dependence on traditional manual secondary confirmation is reduced, the inspection cost and manpower investment are reduced, the detection efficiency and data reliability are improved, and the scheme is suitable for high-frequency and long-distance rail inspection tasks. The scheme can be compatible with existing rail inspection vehicles, laser scanning equipment or vision systems, and is easy to integrate and deploy, has good system engineering landing performance, can be applied to online inspection, rail inspection vehicles, unmanned aerial vehicle inspection and other scenes, and supports all-weather and all-scene automatic defect detection. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The present application is a rail surface defect detection overall flowchart.

[0020] Figure 2 The present application is a rail surface multi-parameter defect detection function relationship diagram.

[0021] Figure 3 The present application is a defect and environmental interference discrimination flowchart.

[0022] Figure 4 The present application is a rail crack multi-modal detection and verification simplified flowchart of embodiment 1.

[0023] Figure 5 The present application is a rail multi-angle spectral detection and interference discrimination simplified flowchart of embodiment 2. DETAILED DESCRIPTION

[0024] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0025] In combination with the accompanying drawings Figure 1 , the present application provides a method for detecting defects on the surface of a rail, which realizes high-precision and dynamic adaptive detection of defects on the surface of a rail. By applying dynamic disturbance with controllable incident angle and illumination intensity at different positions on the surface of the rail, multi-angle and multi-intensity illumination sampling of the surface of the rail is realized. The dynamic disturbance includes changing the incident angle of the illumination light source and adjusting the illumination level of the light source, so that different reflection responses are generated at the same surface position under different illumination conditions, and then the micro-optical behavior data of the surface material are constructed. Different angles of projection can be realized by using multiple light sources, and the light source intensity is adjusted synchronously by using an electronic controller, and the illumination disturbance collection is carried out in a dynamic cycle during the track inspection process. Subsequently, based on the obtained multi-angle and multi-intensity optical reflection data, a reflection response sequence of different micro regions on the surface of the rail is established. The reflection response sequence of each micro region records the optical behavior under dynamic disturbance conditions, and can truly reflect the surface state and micro characteristics of the region material. Then, dynamic reflection stability calculation is performed on the reflection response sequence. The so-called dynamic reflection stability refers to whether the optical reflection value of a micro region on the surface of the rail is stable, regular or has abnormal fluctuations under the change of multiple illumination conditions. The calculation method includes: statistical analysis is performed on the reflection response sequence of each micro region, the light intensity fluctuation amplitude, response gradient change, time sequence correlation coefficient and other indicators of each sampling point are calculated, and the reflection stability value of the region is generated. For a perfect rail surface, it usually has a relatively stable reflection response sequence, and the fluctuation characteristics show continuity and consistency, while for a region with defects (such as cracks, corrosion, block shedding, etc.), the reflection response will produce obvious fluctuation abnormalities under dynamic light disturbance due to the change of micro morphology or material characteristics. By comparing the reflection stability of each micro region, using the region stability difference analysis method, the regions with significant deviation from the surrounding regions are screened out, and these regions with abnormal stability are marked as defect candidate regions. The judgment of stability abnormality can be based on the dynamic threshold of global stability distribution, or can be adaptively determined in combination with neighborhood correlation, so as to effectively avoid the influence of environmental noise or local light and shadow interference, and ensure the accuracy and robustness of defect detection.

[0026] For the defect candidate regions marked by dynamic light perturbation and reflection stability analysis in the early stage, further defect authenticity verification is carried out. For each defect candidate region, the optical characteristics of the normal wheel-rail contact area are collected and compared. Due to long-term wheel-rail contact of the rail surface, the perfect rail surface will form regular contact wear marks, which show relatively consistent directionality, wear width and reflection stability in optical reflection, and usually have continuous and smooth reflection characteristics. Defects such as fatigue cracks, spalling, and corrosion, due to the destruction of the material structure or the mutation of the surface topography, will cause obvious abnormalities in the local reflection characteristics. In order to distinguish whether this abnormality belongs to a real defect or an accidental interference or environmental factor, the analysis of optical heterogeneity boundary characteristics is combined in the implementation process. A local neighborhood window is established around the defect candidate region to detect the optical boundary difference between the region and the surrounding intact area. The so-called optical heterogeneity boundary refers to the mutation degree of the reflection response of the candidate region and the normal region in space under the condition of dynamic light perturbation, mainly in the change of reflection intensity gradient, the discontinuity of spectral response, and the alienation of surface scattering mode. By calculating the reflection gradient distribution at the boundary of the candidate region, it is judged whether the boundary presents the non-continuous boundary characteristics formed by non-natural wear. For example, the normal wheel-rail contact wear boundary usually presents a regular gradient transition, while the defect-induced boundary presents local mutation, random fracture, and disordered reflection direction. Further, in order to verify whether there are abnormal residues or non-normal wear boundaries caused by defect evolution, the implementation combines multi-angle scanning of reflection time series to collect the response of the defect candidate region under multiple light states, and by analyzing the delay phenomenon and saturation state of the reflection response, it is identified whether there are abnormal residues such as oil stains, dust, metal chips, etc. on the surface. Since these residues are usually attached to the micro concave, crack or edge corrosion of the defect area, their optical reflection characteristics are different from those of smooth surfaces, showing phenomena such as increased light scattering, reflection saturation, and local light intensity abnormal change.

[0027] For the defect candidate area marked and verified by dynamic light disturbance and reflection stability analysis in the early stage, further spatial connectivity and diffusion trend judgment of the defect is carried out. For each determined defect candidate area, based on the actual operation of the rail surface physical structure, a two-way analysis framework of contact track direction and normal micro area fluctuation direction is established, wherein the contact track direction refers to the running track direction formed by the long-term rolling of the train wheel on the rail surface, which usually extends along the length direction of the rail, and the normal micro area fluctuation direction refers to the change direction of the local ups and downs perpendicular to the track contact surface, which is used to describe the diffusion or boundary extension of the defect in the surface normal direction. In the implementation process, first, an extension window is established in the edge of the defect candidate area in the contact track direction and the normal fluctuation direction respectively, and the optical reflection response data of the adjacent area is collected, and the reflection stability, light intensity fluctuation amplitude, response time series change trend and other indexes of the adjacent area under the condition of dynamic light disturbance are monitored. Then, the optical response data of the adjacent area is compared with the determined defect area, and whether there is a common change trend is analyzed, that is, the synchronism change of the optical response of multiple adjacent micro areas in dynamic sampling, such as overall decrease of reflection intensity, weakening of optical stability consistency, consistent abnormality of spectral reflection boundary, etc. If the adjacent area and the defect candidate area show similar abnormal response mode under dynamic disturbance, it is judged that these areas belong to the spatial extension part of the same defect structure. Further, in order to accurately judge the spatial connectivity of the defect, the correlation degree calculation between micro areas is adopted in the implementation process, and the response cooperative change degree of each adjacent micro area is quantified as a correlation coefficient. If the correlation coefficient exceeds the preset threshold, it is judged that the area and the defect candidate area exist physical defect connection and belong to the diffusion area of the defect. Through continuous analysis in the contact track direction, the extension trend of the crack along the running direction can be recognized, and combined with the normal micro area fluctuation analysis, the outward expansion behavior of the block, erosion or wear edge can be found, so as to realize the dynamic judgment of the spatial growth of the defect. The method gradually expands the defect boundary through the growth mechanism of the defect area, ensures the complete identification of the overall situation of the actual defect, and avoids the defect detection truncation or omission caused by the limitation of initial marking. Finally, the defect area after spatial diffusion judgment is uniformly marked as the extended defect connected area, the spatial form and diffusion trend of the defect are output, the dynamic monitoring ability for complex defects such as crack propagation, surface erosion and growth type wear is formed, and the method is suitable for omnidirectional intelligent detection and trend warning of rail surface defects.

[0028] Combination with the attached Figure 2, in order to further enhance the detection ability of the small cracks on the surface of the rail, grain distortion and potential damage inside the material, in the dynamic disturbance collection process, the joint technical scheme of incident light polarization direction modulation and multi-spectral response collection is adopted. On the basis of dynamic light disturbance sampling, by modulating the polarization direction of the incident light, the sensitive capture of different optical response characteristics of the rail surface is realized. Polarization modulation includes the alternative switching of linear polarization and circular polarization, and under the conditions of fixed incident angle and light intensity, by adjusting the polarization direction, the optical reflection characteristics of the surface microstructure are more revealed. Due to the existence of micro-cracks, grain distortion, metal organization unevenness and other phenomena on the surface of the rail, these local defects will disturb the polarization state of light, causing the polarization direction of reflected light to shift or the polarization degree to decrease, therefore, in the implementation, the polarization sensing device is used to measure the polarization state of the reflected light, and the polarization response difference between the defect area and the normal area under different polarization incident conditions is recorded, especially for the surface micro-cracks, due to their anisotropic characteristics, the change of reflected polarization will be more sensitive, and this characteristic can effectively distinguish the normal wear reflection from the polarization anomaly caused by cracks. Secondly, in the process of collecting the reflection response sequence, combined with the synchronous irradiation of multi-band multi-spectral light source, specific including the multi-source switching sampling of visible light band, near-infrared band and even short-wave infrared band, different bands of light have different penetration ability for the surface and the inside of the material, surface defects such as cracks and denudation mainly affect the reflection characteristics of visible light band, while potential damage inside the material, such as micro fatigue cracks and organization degradation, leads to the change of near-infrared and short-wave infrared reflection. Therefore, in the implementation process, the reflection response sequence under each band is recorded respectively, the surface reflection anomaly and the internal scattering change are compared, through the joint analysis of multi-spectral response, a multi-dimensional recognition model of defects is established, and the differentiation and recognition of the micro-cracks on the surface of the rail and the potential damage inside the material are realized.

[0029] In order to distinguish the real defects caused by material damage from the pseudo anomalies caused by environmental factors or interference more accurately, a method based on dynamic reflection stability is proposed. In the process of dynamic light disturbance sampling on the surface of the rail, for the same micro area, the reflection response sequence of the micro area under multiple light states is collected to form a time series reflection data set, and the data set is input into a sliding time window for stability analysis. The sliding window includes multiple time scales, such as short-time window, medium-time window and long-time window, which are used to capture short-period environmental fluctuations and long-term material response changes. First, in the short-time window, the fluctuation amplitude, change rate and gradient of the reflection response are analyzed to determine whether there is a short-time abnormal fluctuation caused by light disturbance, temporary environmental factors such as dust, temporary obstacles, water stains and other factors. Then, in the long-time window, the cumulative fluctuation trend of the reflection response is detected to identify whether there is a stable and continuous decline or gradual amplification of the fluctuation mode. This long-term degradation of stability is usually related to material damage or surface fatigue crack propagation. By comparing the response characteristics of different time scales, the short-term accidental disturbance and material damage can be distinguished. Further, in order to avoid the misjudgment of isolated abnormality caused by local environmental interference, spatial autocorrelation analysis is introduced in the stability anomaly judgment. Specifically, the reflection stability of the target micro area and its surrounding adjacent areas is calculated, the dynamic reflection data of the adjacent areas is extracted by using the spatial sliding window, and the stability change trend of the target area and the surrounding area is compared. If the abnormal fluctuation of the target area has continuity or neighborhood synchronous change in space, it is determined that the possibility of real defect is high. On the contrary, if the fluctuation anomaly of the target area is irrelevant to the surrounding area, and the stability of the adjacent area is normal, it can be judged that the anomaly is caused by local pollution, attachment, light spot reflection and other factors, which belongs to pseudo anomaly. The spatial autocorrelation analysis quantifies the spatial logic of local anomaly by calculating the spatial correlation coefficient, fluctuation consistency index or stability gradient distribution, which significantly improves the accuracy of anomaly detection.

[0030] In order to further improve the identification accuracy of defect candidate regions and reduce the misjudgment rate, a technical solution combining bidirectional scanning sampling based on normal wheel-rail contact mark difference and optical heterogeneity boundary feature analysis is adopted. For each marked defect candidate region, optical response capture is performed using bidirectional scanning sampling. The so-called bidirectional scanning refers to dynamic light disturbance collection along the train running direction (forward direction) and the direction opposite to the running direction (reverse direction) for the same track surface region. The optical reflection response of the defect candidate region under different scanning directions is recorded, aiming to detect the influence of contact direction change on the reflection characteristics. Since the normal wheel-rail contact wear on the track surface has directionality, the optical reflection of the normal wear area shows symmetry or regular change under different scanning directions, reflecting the uniformity and predictability of the contact wear. However, if there is an abnormal wear boundary caused by defect evolution, such as erosion, spalling, crack propagation, etc., it will lead to non-symmetrical change of optical response in bidirectional scanning, for example, sudden change of light intensity in forward scanning and slow change in reverse scanning, or local reflection enhancement in one direction and reflection weakening in the other direction. This non-symmetrical feature can be used as an important basis for identifying abnormal wear boundaries. Further, in the optical heterogeneity boundary feature analysis, for the above detected non-symmetrical response area, local reflection gradient continuity detection is implemented, which specifically includes spatial sampling of the light intensity gradient at the boundary of the defect candidate region and analyzing the smoothness and continuity of the gradient change. The normal wheel-rail wear boundary usually shows a stable gradual gradient, reflecting the natural transition of material wear, while the boundary caused by foreign matter attachment or local spalling usually shows a sudden gradient, a jump in reflection intensity or a local extreme value. This boundary disturbance phenomenon can be quantified by the discontinuity index of the local gradient curve, such as calculating the number of sudden points of the first derivative or the extreme values of the second derivative. By comprehensively judging the optical reflection asymmetry under bidirectional scanning and the continuity feature of the boundary gradient, the real abnormal wear boundary caused by material defects can be effectively distinguished from the surface optical abnormalities caused by attachments, contaminants and spalling, further verifying the authenticity of the defects and the boundary range, and avoiding misjudgment and missed detection.

[0031] In order to effectively distinguish the surface residues from the actual defects of the material body, and improve the spatial connectivity judgment accuracy of defect detection, a technical scheme based on optical response delay characteristics is adopted to identify abnormal residues and analyze the defect diffusion priority path based on the historical data of the rail wear direction. In the dynamic light disturbance sampling process, the reflection response sequence of the candidate area and its neighborhood area of the rail surface defect is collected under different light conditions, and the optical response time lag phenomenon of the micro area is detected in the response sequence. The response delay refers to the fact that when the light intensity changes or the incident angle is disturbed, the optical reflection of the normal rail surface will immediately change, showing high response synchronicity. If there are abnormal residues such as oil stains, dust, water film and other residues on the surface, these attached layers will cause the reflected light to appear delayed response when the disturbance changes, that is, the response curve of light intensity change lags behind the time of the disturbance control signal. By comparing the synchronicity and delay curve of the optical response, it can be accurately distinguished whether the abnormal reflection behavior is caused by the material body or the surface residues. In addition, in order to further analyze the spatial connectivity and diffusion trend of the defect, the priority path relationship of the defect diffusion is established along the rail contact track direction of the defect area combined with the historical wear data of the rail. The historical wear data of the rail is analyzed, the fatigue stress concentration area and wear deviation trend of the rail are extracted, and a fatigue direction priority model is established. The model is based on the long-term stress characteristics of the rail, and the deviation between the main direction of material mechanics fatigue and the track direction is associated, and then the extension tendency of the crack or block defect is inferred. In implementation, the micro area dynamic optical response cooperative analysis is carried out along the historical wear main path direction at the edge of the determined defect candidate area, and it is judged whether there is abnormal synchronous response in the adjacent micro area. If the adjacent micro area along the fatigue direction shows the continuity or tendency change of optical abnormality, these areas are further included in the defect connectivity area, and the spatial growth and expansion judgment of the defect is realized.

[0032] In order to accurately determine the spatial diffusion trend of rail surface defects, especially the growth behavior of crack endpoints or micro spalling edges, a joint determination mechanism of normal micro-area fluctuation synchronism analysis and dynamic correlation threshold adjustment is proposed. In the marked defect candidate area, the normal micro-area fluctuation direction is selected as the analysis axis. The normal micro-area fluctuation direction is perpendicular to the local high and low fluctuation change direction of the rail contact surface, which is orthogonal to the contact trajectory direction of the rail surface. Through the collection of dynamic optical reflection response of the upstream and downstream (i.e. normal adjacent area) of the candidate area, the micro-area time sequence fluctuation curve is established. In the implementation process, the collected normal micro-area reflection response is analyzed in time sequence, the reflection intensity change, fluctuation amplitude and stability decay trend of the candidate area and its upstream and downstream adjacent micro-area are compared, and whether these areas exist response synchronism is judged. If it is found that the adjacent micro-area shows similar optical abnormal synchronous fluctuation under the dynamic light disturbance, especially the local fluctuation of the crack endpoint or the spalling edge is intensified, it is indicated that there is a spatial derivation risk, i.e. the defect continues to expand along the normal direction. In order to avoid the over-expansion of the defect growth boundary caused by environmental disturbance or data noise, a dynamic correlation threshold strategy is further adopted to control the defect growth process. According to the stability change rate of the dynamic reflection of the candidate area, the determination threshold of the defect growth is adjusted in real time. If the stability change rate is large, it indicates that the defect area appears rapid fluctuation anomaly in a short time, and the system reduces the growth threshold to allow the defect boundary to expand moderately, so as to capture the potential rapid evolution risk; otherwise, if the stability change rate is small, it indicates that the defect area is in a relatively stable or slow development stage, and the system increases the growth threshold to inhibit the overgrowth of the boundary and prevent the non-defect area from being mistakenly integrated into the defect connected area.

[0033] Combined with the Figure 3 In order to accurately distinguish the real crack evolution characteristics of the rail surface from the optical pseudo-abnormalities caused by environmental factors (such as water film covering or oil stain adhesion), a defect candidate area verification mechanism based on dynamic optical disturbance sampling is proposed, which uses the time stability fluctuation characteristics combined with the optical response difference of the surface wet and slippery state for comprehensive judgment. In the defect candidate area verification process, multi-angle and multi-time reflection response data are collected, and the surface wet and slippery state is modeled as a key interference factor. Under the wet and slippery state, the reflection light intensity change of the rail surface usually shows overall slow decay or response delay, which is easy to be confused with the reflection abnormality of real defects such as micro-cracks and fatigue erosion. Therefore, the optical characteristic parameters of surface state change are introduced, and the dynamic fluctuation function is constructed based on the collected multi-time data to identify the cumulative evolution behavior of real defects. Specifically, the response stability variation function integral model is used to process the reflection data, which is defined as follows: ; Where, represents the time time, incident angle The cumulative dynamic fluctuation index of the reflection response is used to characterize the defect evolution intensity. is the time and the angle unit reflection intensity response; is the second-order time derivative of the reflection response, reflecting the acceleration change of the reflection signal, and embodying the severity of the fluctuation; is the response delay offset factor related to the incident angle, used to adjust the light propagation path offset caused by the water film or oil film, so that the model adapts to the wet and slippery state response characteristics of different angles; is the time decay background function, used to represent the overall optical reflection slow change caused by the water film or oil stain flowing on the surface, usually with low-frequency stability; is the dynamic weight function, used to weight the sensitivity of the reflection fluctuation at a specific angle and time, especially near the critical reflection angle, which can be used to emphasize the local reflection abnormal change; is the sampling start time. Through the above model, the high-frequency local response fluctuation caused by real defects can be effectively distinguished from the low-frequency overall reflection trend caused by water film and oil stain. When the fluctuation exceeds the dynamic threshold for a plurality of different incident angle directions , it is determined that the region has real cumulative fatigue crack evolution behavior, indicating that the material shows consistent fluctuation aggravation phenomenon under multi-angle disturbance, with typical defect expansion characteristics; while when it presents periodic decay in time or is not concentrated in the distribution of each angle direction, showing randomness or flow characteristics, it is inferred that the optical interference false response is caused by water film flow or oil stain adhesion, thereby effectively avoiding false positives.

[0034] In order to effectively distinguish the real defects on the surface of the rail from the optical interference caused by environmental factors (such as water film, oil film), a multi-angle dynamic judgment mechanism based on the optical reflection characteristics of the surface wet and slippery state is proposed, which specifically includes the critical reflection angle response detection of incident light and the time series analysis of reflection decay rate. First, in the dynamic disturbance sampling process, by gradually adjusting the angle of incident light, the reflection intensity change of the rail surface under different incident angle conditions is detected, and the response characteristics of the critical reflection angle are particularly concerned. The so-called critical reflection angle refers to the specific angle at which the light from the air into the liquid layer occurs total reflection or reflection rate mutation when there is liquid coverage on the interface. Because the refractive index of water film or oil film is different from that of dry rail surface, when there is liquid coverage on the surface, the critical angle position will shift obviously. By comparing the critical angle response changes of different areas, it can be quickly judged whether there is water film or oil film attached. For example, in the air-metal interface, the reflection changes regularly with the change of incident angle, while in the air-liquid-metal multi-interface, the reflection rate will produce inflection point or shift at a specific angle. In implementation, the change is obtained by high-precision angle scanning to realize real-time identification of the wet and slippery state. Secondly, in the process of distinguishing environmental interference, the reflection decay rate of the surface micro area is further used for dynamic analysis. Specifically, in the continuous optical disturbance sampling, the change trend of the reflection intensity of each micro area with time is recorded, and the time series decay curve is constructed. For static defects such as cracks and blocks, the abnormal optical reflection shows stable reflection intensity mutation or fixed local dark area, and the decay rate is usually zero or stable. And for dynamic environmental factors such as water film and oil film, due to their fluidity or evaporation effect, the surface reflection intensity will gradually decay or fluctuate with time, showing a slow changing reflection decay mode. In specific implementation, the reflection intensity change rate of each micro area is calculated. If a sustained decay trend or periodic fluctuation is detected, and there is synchronous low-frequency change with the adjacent area, it can be judged as a pseudo abnormal response caused by environmental interference.

[0035] In order to accurately identify the false defect signals caused by environmental factors, especially the optical anomalies caused by water film and oil stain adhesion, a joint environmental interference identification method based on multispectral contrast response and polarization state analysis is proposed. In the dynamic optical sampling process, a multispectral light source is used to irradiate the rail surface, including visible light band, near-infrared band and short-wave infrared band, and the reflection signals of each band are collected and compared simultaneously. Because the water film has obvious absorption characteristics in the short-wave infrared band, compared with the dry rail surface or solid defect area such as crack, the water film covered area will show significant attenuation of short-wave infrared reflection. By comparing the reflection intensity changes of the same surface area under different spectra, especially the abnormal absorption phenomenon in the short-wave infrared region, it can be quickly judged whether there is water film coverage. For example, in the visible light band and near-infrared band, the reflection changes of water film and surface micro-defects are similar, but in the short-wave infrared band, the absorption of water film to light leads to a significant decrease in reflection intensity, which can be used to effectively distinguish water film interference from real surface defects. In order to further distinguish the oil stain state from the reflection characteristics of real defects on the rail surface, on the basis of multispectral sampling, combined with the dynamic analysis of the polarization state of reflected light, specifically: the linearly polarized or circularly polarized light is incident on the rail surface, and the polarization ratio change of the reflected light is collected. Because the oil film adhesion will cause multiple scattering and interference of surface light, causing the polarization ratio of reflected light to decrease, the metal defects or wear areas of the surface body usually maintain a high polarization retention, especially at a specific incident angle, such as near Brewster angle, the polarization characteristics of metal reflection are more stable. When implemented, by measuring the polarization parameters of reflected light, such as degree of polarization, extinction ratio or polarization direction offset, it is judged whether there is an abnormal decrease in polarization ratio caused by oil stain, if the degree of polarization is detected to decrease significantly and the spatial distribution overlaps with the abnormal area of visible light reflection, it can be inferred that the area is an oil film interference, rather than a crack or wear of the material body.

[0036] Example 1: Combined with the drawings Figure 4In this embodiment, a certain railway bureau is set to inspect a heavy-load freight rail section using a multi-modal detection device. The inspection equipment is equipped with a multi-spectral imaging module with polarization modulation function. When the inspection vehicle travels at a speed of 40 km / h, it dynamically samples the rail surface. During on-site detection, the inspection system finds a local reflection abnormal area in a certain section of the track. The system first uses visible light polarization modulation to control the incident light. The modulation method uses linear polarization 0°, 45°, and 90° switching in turn, and the light intensity remains fixed. The incident angle is set to 20° to exclude strong reflection interference. The polarization retention rate of the normal track surface reflection changes at different polarization angles, and the polarization reflection ratio fluctuates within ±3%. In the abnormal area, the polarization reflection ratio is 18% at 0° polarization, drops to 11% at 45°, and rises to 16% at 90°, with a fluctuation amplitude of more than 7%, which is significantly higher than the normal track surface polarization change range. After comparative analysis, it is preliminarily judged that there is a microstructure abnormality in this area, such as grain distortion or fine cracks. In addition, to further verify the type of the abnormal area, the system collects multi-spectral response of the same micro area, records the reflection response under visible light (wavelength 550 nm), near-infrared (wavelength 850 nm), and short-wave infrared (wavelength 1450 nm) three wavebands. The sampling results show that in the visible light band, the reflectivity of the abnormal area is 36%, which is similar to that of the surrounding normal area; in the near-infrared band, the reflectivity of the abnormal area decreases to 21%, which is 7% different from the 28% reflectivity of the normal area; in the short-wave infrared band, the reflectivity of the abnormal area further decreases to 12%, while the normal area remains at about 24%. According to the gradual attenuation characteristics of multi-spectral response, it can be inferred that the abnormal area not only has surface reflection abnormalities, but also has potential damage inside the material, such as grain distortion, metal fatigue layer, or subsurface micro-cracks. Because these internal defects will cause a significant decrease in infrared and short-wave infrared reflection, combined with the dual evidence of polarization abnormality and multi-band attenuation, the system finally determines that the area is a defect area with crack evolution risk. Further manual review using ultrasonic detection verification indeed detects a micro-crack initiation point in the area, with a crack length of about 4 mm and a depth of about 0.3 mm. The entire detection process starts from polarization dynamic disturbance collection, and through real-time optical response analysis combined with multi-spectral comparison, surface micro-cracks and potential internal damage of the material can be accurately distinguished, achieving precise identification of defect type and distribution, significantly reducing the influence of environmental interference, and improving the sensitivity of early crack identification.

[0037] After the preliminary detection based on polarization modulation and multi-spectral reflection response, the inspection team continued to conduct a more in-depth dynamic reflection stability analysis on the abnormal area to verify whether the crack was in a cumulative expansion state and to rule out the influence of environmental interference. The detection system implemented multi-time period sliding window analysis on the reflection response sequence of the area, setting short-time window (0.5 seconds), medium-time window (2 seconds) and long-time window (5 seconds) respectively. In the short-time window, the detection found that the reflection intensity had small fluctuations, with a fluctuation amplitude of ±2%, which was attributed to environmental interference such as sunspot or windblown dust. In the long-time window, the reflection intensity of the abnormal area continuously decreased, with an average decrease of 9%, accompanied by a gradual decrease in stability index, indicating the existence of persistent abnormalities caused by material structure changes.

[0038] To further rule out false abnormal signals, the system conducted spatial autocorrelation analysis on the area, comparing the stability change trend of the abnormal micro-area with that of the adjacent 5 micro-areas. It was found that the stability fluctuation synchronization correlation coefficient of the adjacent areas was 0.92, indicating that the anomaly was not isolated, but had spatial consistency of stability degradation, further confirming the existence of the crack. Subsequently, the system used bidirectional scanning sampling to analyze the difference in the direction of wheel-rail contact of the abnormal area. The inspection vehicle implemented forward and reverse sampling on the same track section. When moving forward, the detection found that the reflection intensity of the area suddenly changed to 14%, while moving backward, the reflection intensity changed smoothly to 19%, with a difference of 5%, which was significantly higher than the average difference (only 1%) of the surrounding normal wear area. Combined with the analysis of the track mechanics model, it was judged that the difference was caused by the abnormal wear boundary, which meant that the crack or erosion area formed an irregular boundary reflection characteristic.

[0039] To further confirm the boundary state, the system extracts the local reflection gradient and finds that the reflection gradient of the boundary of this area changes from the normal 0.3% per millimeter to a sudden jump of 1.2% per millimeter of light intensity, indicating that the boundary has obvious optical heterogeneity, which is consistent with the boundary disturbance characteristics of material erosion or crack propagation. At the same time, the system also detects the influence of abnormal residues and uses the optical response time delay characteristics for dynamic analysis. Under continuous light disturbance, the response delay of the normal rail surface area is 0.02 seconds, while in the abnormal area, the local micro-area response delay reaches 0.11 seconds, with obvious lag. Combined with on-site observation, there is oil stain attachment in this area, but the reflection attenuation of the oil stain area shows a decrease in all wavebands without obvious boundary gradient change in multispectral sampling, while the crack area shows local polarization reflection anomaly and boundary light intensity mutation, so the system successfully distinguishes oil stain residues from the body defects through time delay characteristics. Subsequently, combined with track maintenance history data and fatigue path analysis, the system establishes a defect diffusion priority path model. Track wear data shows that the wheel-rail contact in this section is concentrated in the left area of the rail head center, and long-term stress causes the crack to expand left along the contact trajectory. The system accordingly prioritizes scanning for defect connectivity along the fatigue direction and finds that the crack extends 7.5 mm along the contact trajectory direction, while the normal extension is only 2.1 mm, verifying the priority expansion trend of the fatigue main path.

[0040] To address the risk of crack endpoints and micro-spalling edges, the system also implements normal micro-area wave fluctuation synchronization analysis and compares the stability fluctuations of the upstream and downstream areas of the crack. It is found that the fluctuation correlation coefficient of the adjacent areas in the normal direction reaches 0.87, indicating that there is a spatial derivative risk in the endpoint area. To prevent over-expansion of defect recognition, the system uses a dynamic correlation threshold strategy based on stability change rate to dynamically adjust the growth boundary. When the crack stability change rate exceeds 10%, the system reduces the growth threshold and expands the detection range, while when the edge area change rate is less than 3%, the system shrinks the growth boundary to avoid incorporating environmental noise into the defect area. Finally, the detection system outputs the spatial topology distribution map of the crack, records the total length of the crack as 8.2 mm, with the main crack direction extending 7.5 mm along the track running direction and the normal extension being 2.1 mm. The crack depth is estimated to be 0.35 mm, and the system marks this area as a moderate risk defect and recommends rail grinding and crack sealing repair within 3 days.

[0041] Example 2: In combination with the attached Figure 5After the inspection vehicle completed polarization modulation, multi-spectral identification, dynamic stability analysis, and spatial expansion determination, to further verify whether the crack area had cumulative fatigue crack evolution and to eliminate optical interference caused by residual water film after rain and oil contamination from track lubricants, technicians used the response stability variation function integral model to make high-precision determinations of abnormal areas. The inspection period was from 9:10 to 9:11 on the same day, and a total of 298 sets of optical response sequence data were collected. 、 、 ) and perform multi-time series dynamic fluctuation evaluation on the reflection response R(t,θ).

[0042] This method is used in each The second-order time derivative of the reflected response is calculated as follows: ; set up seconds, the sampling interval is 50ms, and the reflected acceleration change of each set of data can be obtained by the above formula; in order to eliminate the delay error caused by the slippery state, the system introduces Related response delay factors In this test, according to the actual water film refractive index, the reflection interference analysis at different angles is carried out, and the values ​​are as follows: 、 、 At the same time, the low-frequency background function is introduced Simulating the attenuation trend under water film flow, the function uses the approximate model as , the on-site fitting results , , which is used to represent the slow response in the slippery area.

[0043] The entire dynamic fluctuation function Defined as: ; in, is a dynamic weight function, which is constructed in the present invention as a Gaussian distribution set for the critical reflection angle (about 39°) to emphasize Optical anomalies near the critical angle: ; Substitute the above parameters into the formula to Taking the incident angle direction as an example, the calculation is done in the time interval The dynamic response fluctuation integral within is numerically integrated to obtain , and in and 45° directions, respectively , . Compare the dynamic threshold set by this system , the result shows that in more than two angle directions , the identification condition about the real evolution of fatigue crack is met, so this area is confirmed to be in the state of continuous crack propagation.

[0044] On the contrary, the inspector made a comparative calculation at another abnormally high reflection point about 8 meters away from the defect area. Due to the accumulation of water after the rain the night before, there was a large area of water film on the surface. The calculation of the same angle The result is: 15° direction is 0.91, 30° direction is 1.03, 45° direction is 0.94, although there are occasional local disturbances, but the overall presents low amplitude, periodic fluctuations, and in any direction does not exceed , the spatial distribution is not concentrated, and the system determines that this point is a water film interference false response, which does not constitute a structural defect.

[0045] Finally, after artificial crack detection review, the first area has fatigue initiation point and has expanded to 8.6mm length, the crack edge is highly consistent with the profile drawn by the system, with an error of less than 1mm; the second area has no crack or damage, only local surface residual wet marks are found.

[0046] The on-site detection continues to the in-depth judgment link of wet and slippery environment interference. Especially after the rain, water film residues and oil stains formed by vehicle lubricating oil droplets appear in some local areas along the railway. If not distinguished, it is easy to lead to crack missed detection or false report. Therefore, the suspected environmental interference area was detected by critical reflection angle response and reflection attenuation rate analysis. Taking the track number A-07 section as an example, the inspection system collected the reflection intensity change curve from the incident angle 0° to 60°, and found that in the normal dry rail surface, the reflection intensity and the incident angle change relationship showed a stable linear increasing curve, without obvious inflection point, and the maximum reflectivity reached 40% at 45°; but in the abnormal area, the inspection equipment recorded that the reflection intensity suddenly changed at 38°, from 31% to 18%, which was judged as the critical reflection angle shift phenomenon. Combined with the optical principle, when the surface of the rail is dry, the critical angle of the air-steel interface should be close to the high absorption state without total reflection, and when the surface is covered with water film, forming an air-water-steel multi-layer interface, the reflection characteristics will change. According to the reflection model, the critical angle corresponding to the on-site data decreases from the theoretical 39° to 38°, which meets the characteristics of thin water film interference.

[0047] Further, to determine whether it is a dynamic water film interference, the inspection system implements continuous disturbance sampling on the micro area, sets a time window of 5 seconds, a sampling frequency of 20 Hz, and records the time decay curve of the reflection intensity. In the dry crack area, the reflection intensity fluctuates by only ±1% over time, while in the water film area, the reflection intensity continuously decays from the initial 30% to 22%, with an average decay rate of 1.6% per second. The following decay rate model is used to calculate on site: ; wherein, , , , the calculation is as follows: ; The rate is much higher than the normal track surface, indicating that there is optical decay caused by liquid flow on the surface, further verifying that it is a dynamic water film interference.

[0048] To avoid relying solely on a single angle, the system also introduces multi-spectral reflection data for auxiliary verification. In the same area, the reflection intensities at 550 nm (visible light), 850 nm (near-infrared), and 1450 nm (short-wave infrared) are collected. In the dry crack area, the reflectance changes at the three wavelengths are 36%, 29%, and 25%, respectively; while in the water film covered area, the visible light reflectance is 32%, the near-infrared reflectance drops to 24%, and the short-wave infrared reflectance sharply drops to 11%. Based on the short-wave infrared absorption characteristics of water, it is confirmed that this significant decay conforms to the water film absorption model and does not conform to the crack reflection characteristics, so the area is finally determined to be a water film covered pseudo anomaly.

[0049] In addition, to distinguish oil pollution interference, the system uses polarization state analysis, uses polarized light sources for illumination, and records the polarization ratio of reflected light. The reflectance at polarization angles of 0° and 90° for the normal track surface is 34% and 36%, respectively, and the polarization preservation rate (PPR) is: ; In the oil pollution area, the polarization reflection collected is 29% and 30%, respectively, and the PPR is: ; The polarization ratio is significantly reduced, close to the non-polarized reflection state, which conforms to the scattering depolarization phenomenon caused by oil film. Combined with the site environment, it is found that this area is exactly located in the train track lubrication section, where there is lubricating oil residue. Through the reduction of the polarization ratio, the system accurately determines it to be oil pollution interference rather than a material defect.

[0050] According to the above data, the final output inspection report classifies the reflection anomaly of the A-07 section, the water film interference area is 2.4 meters, the oil stain interference area is 0.8 meters, and the real crack detection area is an 8.6 mm crack belt, completely eliminating false positives caused by environmental factors, and ensuring high precision and high reliability of track defect detection.

[0051] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for detecting defects on the surface of a rail, characterized in that: include: By applying dynamic perturbations with controllable incident angles and light intensities at different locations on the rail surface, the reflectance response sequences of the rail surface under various lighting conditions are collected to characterize the microscopic optical response behavior of the rail surface material. Based on the collected reflection response sequence, the dynamic reflection stability is calculated for each tiny area on the rail surface. Based on the regional stability differences, the areas with abnormal stability are marked as defect candidate areas. Analyze the differences between the candidate defect area and the normal wheel-rail contact trace, and combine the optical heterogeneity boundary characteristics to identify whether there are abnormal residues or abnormal wear boundaries caused by defect evolution, so as to verify the authenticity of the defect; For the identified defect candidate areas, along the contact track direction and the normal micro-area fluctuation direction of the rail surface, it is judged whether the optical responses of adjacent areas have a common change trend, and the spatial connectivity and diffusion trend of the defects, as well as the growth defect area, are determined.

2. A method for detecting defects on a rail surface according to claim 1, characterized in that: During the dynamic disturbance acquisition process, the polarization direction of the incident light is modulated to capture reflection polarization anomalies caused by surface microcracks or grain distortion under the same conditions of incident angle and light intensity change; when collecting the reflection response sequence, a multi-spectral light source with different wavelengths is used to distinguish surface defects from potential damage inside the material.

3. A method for detecting defects on a rail surface according to claim 2, characterized in that: The calculation of the dynamic reflection stability includes performing a multi-period sliding window analysis on the reflection response sequence to identify the fluctuation trends of small areas on different time scales and distinguish between short-term fluctuations caused by changes in the external environment and persistent stability anomalies caused by material damage. The determination of the stability anomaly is combined with the spatial autocorrelation of the reflection response and the degree of correlation between the stability of the small area and the surrounding area to eliminate pseudo-anomaly marks caused by local occlusion or contamination.

4. A method for detecting defects on a rail surface according to claim 3, characterized in that: The difference analysis between the defect candidate area and the normal wheel-rail contact trace adopts bidirectional scanning sampling to determine whether there is asymmetric change in the optical response under different contact directions, which is used to identify abnormal wear boundaries caused by defect evolution; the analysis of optical heterogeneity boundary characteristics includes boundary continuity detection of the local reflection gradient of the defect candidate area to determine whether there is boundary disturbance caused by foreign matter adhesion or peeling.

5. A method for detecting defects on the surface of a rail according to claim 4, characterized in that: The identification of abnormal residues is combined with the response delay characteristics to detect whether there is an optical response time lag phenomenon in the micro area under dynamic light perturbation, so as to distinguish the difference in reflection between surface residues and the material body; In the diffusion analysis of the defect area along the contact track, combined with the historical data of the track wear direction, the defect diffusion priority path relationship is established, and the defect connectivity trend along the material mechanical fatigue direction is preferentially judged.

6. A method for detecting defects on the surface of a rail according to claim 5, characterized in that: The diffusion judgment of the normal micro-area fluctuation direction includes performing a fluctuation synchronization analysis on the micro-area responses upstream and downstream of the defect candidate area to detect the spatial derivative risk of crack endpoints or micro-block edges; the defect growth process adopts a dynamic correlation threshold strategy to dynamically adjust the growth boundary based on the stability change rate of the candidate area.

7. The method for detecting defects on the surface of a rail according to claim 1, characterized in that: During the defect candidate area verification process, the optical reflection change characteristics of the surface wet state are combined to identify the water film coverage or oil pollution state, and distinguish between environmental interference and real defect characteristics; The temporal stability fluctuation characteristics of the defect candidate area are used to determine whether cumulative fatigue crack evolution occurs, and the crack propagation direction is inferred from the dynamic fluctuation trend.

8. A method for detecting defects on the surface of a rail according to claim 7, characterized in that: The optical reflection change characteristics of the surface in a wet and slippery state include detecting the critical reflection angle response of the incident light, and identifying whether there is a water film or oil film attached to the surface when the critical reflection angle shifts; the differentiation of environmental interference includes using the reflection attenuation rate comparison of surface micro-areas, analyzing the time decay pattern of the reflection intensity in continuous disturbance sampling, and distinguishing between static defects and pseudo-abnormal responses caused by dynamic water film flow.

9. A method for detecting defects on a rail surface according to claim 8, characterized in that: The identification of the environmental interference is to use multi-spectral contrast response to judge the changes in short-wave infrared absorption characteristics under the water film state; the judgment of the oil pollution state is combined with the changes in the polarization state of the reflected light, and the polarization rate reduction phenomenon caused by the surface oil film is used to distinguish the oil pollution interference from the defective reflection characteristics of the rail surface itself.

10. A rail surface defect detection system, characterized in that: include: A dynamic light perturbation module is used to project a light beam with adjustable incident angle, light intensity, polarization direction and multi-spectral band onto the rail surface; Reflection collection unit, used to collect reflection signals from the rail surface under different lighting conditions; a data processing unit, composed of a processor and a memory, for processing data based on the reflected signal; 1) Calculate the reflection stability of each micro-area and mark the areas with abnormal stability; 2) Compare the normal wheel-rail contact traces to verify whether the abnormal area is a real defect; 3) Analyze the spatial connectivity and diffusion trend of defects; 4) Eliminate environmental interference from water and oil films by utilizing critical reflection angle shift, short-wave infrared absorption changes, and reflection polarization rate changes; An output unit is used to display defect location, diffusion trend and risk warning information; the processor is programmed to execute the steps of any one of claims 1-9.

Images