Rail profile detection data analysis and grinding system

Abstract

The present application provides a kind of steel rail profile detection data analysis and polishing system, it is related to track maintenance data processing field.The system washes the noise point cloud of target steel rail to extract the measured profile multidimensional array of rail surface, and generates profile space characteristic deviation amount by registration;Extract constant frame set in non-cutting area and historical period data to construct reference alignment coordinate system;Compare time series span change characteristics and perform cross-physical domain mapping calculation to generate material removal impedance constraint factor;Use the factor to perform boundary clamping on dynamic cutting compensation tolerance model, generate the upper limit of revised extreme value;When over-limit, decouple to generate multidimensional polishing linkage strategy.The present application effectively overcomes the distortion and misplacement of point cloud caused by traveling excitation through multidimensional comparison linkage distribution, realizes full closed loop adaptive polishing.

Description

Technical Field

[0001] This invention relates to the field of track maintenance data processing technology, specifically a rail profile detection data analysis and grinding system. Background Technology

[0002] As maintenance of high-speed, heavy-haul railways evolves towards digitalization and precision, the core logic of rail profile grinding systems is shifting from extensive mechanical cutting to data-driven intelligent closed-loop control. In this process, achieving high-fidelity 3D topographic reconstruction under complex service environments and accurately translating it into execution constraints for the underlying electro-hydraulic servo mechanism has become a key path for technological breakthroughs in this field.

[0003] Existing rail grinding systems still face significant limitations under actual operating conditions: the high-frequency mechanical vibration of the vehicle chassis combined with the long-wave irregularities of the track causes severe distortion in the raw point cloud collected by the photoelectric sensor array. For example, the existing technology with announcement number CN111809464B mainly relies on polar coordinate profile and corrugation detection data to automatically output recommended grinding strategies, but this solution does not adequately isolate transient vibrations in the data processing link. Forcibly importing the undecoupled noisy point cloud into the rule evaluation component can easily lead to feature space misalignment and matching pipeline congestion, thereby generating false contour feature deviations. Existing business strategy generation components rely heavily on manual experience to set static cutting compensation tolerances, ignoring the distribution of the micro-metallurgical hardened layer formed by the rail under long-term rolling. When facing sections with high material removal resistance, a single geometric optimization logic can cause the grinding component to blindly cut too deeply, causing oscillation of the underlying push rod and thermodynamic damage. Summary of the Invention

[0004] The purpose of this invention is to provide a rail profile inspection data analysis and grinding system. This system constructs an immune-excitation absolute reference frame by extracting a constant topology from the non-cutting zone in the spatial domain, and introduces historical inspection cycle morphology into the time domain to perform cross-cycle pure-state morphology feature comparison. Furthermore, the extracted microscopic material impedance parameters are used as feedforward constraint operators to implement nonlinear clamping on the initial geometric cutting tolerance. This addresses the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A rail profile inspection data analysis and grinding system, specifically comprising:

[0007] The perception and evaluation component is configured to acquire noisy point cloud data of the target rail surface, perform feature cleaning on the noisy point cloud data to extract a multi-dimensional array of the measured rail surface morphology containing three-dimensional spatial coordinates and mileage timestamps; call a preset standard profile template, and perform spatial registration on the multi-dimensional array of the measured rail surface morphology based on the standard profile template in the data space to generate a profile spatial feature deviation.

[0008] The spatiotemporal feature extraction component is configured to extract first data representing the geometrically invariant topological features of the non-cutting zone from the multi-dimensional array of the measured morphology of the track surface, wherein the first data consists of a discrete set of spatial feature points; and to obtain second data representing the historical periodic morphological state of the same mileage section.

[0009] The bypass spatiotemporal comparison component is configured to receive the first data, the second data and the contour spatial feature deviation, construct a reference alignment coordinate system based on the cross-period topological mapping relationship between the first data and the second data, map the second data to the reference alignment coordinate system, and generate a historical morphology array after spatial reference alignment.

[0010] Extract the temporal span variation features of the deviation between the historical topography array after spatial reference alignment and the contour spatial feature, perform cross-physical domain state mapping calculation on the temporal span variation features, and generate the material removal impedance constraint factor;

[0011] The tolerance dynamic correction component is configured to receive the material removal impedance constraint factor, use the material removal impedance constraint factor to perform boundary clamping on the preset dynamic cutting compensation tolerance model, and generate the corrected upper limit of the dynamic cutting compensation tolerance allowable extreme value.

[0012] The closed-loop instruction generation component is configured to generate a multi-dimensional grinding linkage strategy control instruction when it is determined that the deviation of the contour space features exceeds the upper limit of the allowable extreme value of the corrected dynamic cutting compensation tolerance. The multi-dimensional grinding linkage strategy control instruction is used to trigger the target bottom-level grinding component to execute the target state geometric compensation control sequence.

[0013] Compared with the prior art, the beneficial effects of the present invention are:

[0014] This invention uses a sensing and evaluation component to clean and extract a multi-dimensional array of measured track surface topography from noisy point clouds. It then utilizes a spatiotemporal feature extraction component to strip contact surface variables along the depth coordinate axis, extracting the geometrically invariant topology of the track waist as the first data. This mechanism constructs an absolutely aligned coordinate system for immune high-frequency excitation of the vehicle body, effectively eliminating feature misalignment caused by transient spatial distortion and improving the accuracy of contour spatial feature deviation extraction.

[0015] To address the limitations of relying solely on geometric dimensions for compensation, the bypass spatiotemporal comparison component performs discrete partial derivative and definite integral calculations based on the temporal span variation characteristics of the deviation between the historical topography array and the current contour to isolate background noise. Furthermore, it maps the high-frequency geometric fluctuation characteristics in space across domains into a material-free impedance constraint factor. This approach transforms the microscopic metallurgical lattice hardening state, which is difficult to measure directly, into a calculable digital index, providing an objective and quantitative data foundation for control strategies under extreme physical conditions.

[0016] By utilizing the tolerance dynamic correction component to receive impedance constraint factors, the linear servo gain term within the basic boundary function is proportionally compressed, generating a corrected upper limit for the dynamic cutting compensation tolerance. This closed-loop command generation mechanism not only decouples macroscopic physical dimensions into a multi-dimensional grinding linkage strategy based on nominal rotational speed and downward torque, but also dynamically tightens the tolerance boundary through feedforward impedance prediction. This optimizes the bottleneck of excessive cutting caused by reliance on manual experience in rail profile grinding and repair, avoids mechanical oscillation of the bottom push rod and grinding wheel overload failure, and enhances the system robustness of heavy-load optimization. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0018] Figure 2 Flowchart for perceptual data cleaning and spatiotemporal feature multidimensional alignment analysis;

[0019] Figure 3 This is a flowchart of cross-domain impedance mapping and tolerance dynamic clamping closed-loop control. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Example 1:

[0023] Please see Figures 1 to 3 The present invention provides a technical solution:

[0024] A rail profile inspection data analysis and grinding system, comprising:

[0025] The perception and evaluation component is configured to acquire noisy point cloud data of the target rail surface, perform feature cleaning on the noisy point cloud data to extract a multi-dimensional array of the measured rail surface morphology containing three-dimensional spatial coordinates and mileage timestamps; call a preset standard profile template, and perform spatial registration on the multi-dimensional array of the measured rail surface morphology based on the standard profile template in the data space to generate a profile spatial feature deviation.

[0026] The spatiotemporal feature extraction component is configured to extract first data representing the geometrically invariant topological features of the non-cutting zone from the multi-dimensional array of the measured morphology of the track surface, wherein the first data consists of a discrete set of spatial feature points; and to obtain second data representing the historical periodic morphological state of the same mileage section.

[0027] The bypass spatiotemporal comparison component is configured to receive the first data, the second data and the contour spatial feature deviation, construct a reference alignment coordinate system based on the cross-period topological mapping relationship between the first data and the second data, map the second data to the reference alignment coordinate system, and generate a historical morphology array after spatial reference alignment.

[0028] Extract the temporal span variation features of the deviation between the historical topography array after spatial reference alignment and the contour spatial feature, perform cross-physical domain state mapping calculation on the temporal span variation features, and generate the material removal impedance constraint factor;

[0029] The tolerance dynamic correction component is configured to receive the material removal impedance constraint factor, use the material removal impedance constraint factor to perform boundary clamping on the preset dynamic cutting compensation tolerance model, and generate the corrected upper limit of the dynamic cutting compensation tolerance allowable extreme value.

[0030] The closed-loop instruction generation component is configured to generate a multi-dimensional grinding linkage strategy control instruction when it is determined that the deviation of the contour space features exceeds the upper limit of the allowable extreme value of the corrected dynamic cutting compensation tolerance. The multi-dimensional grinding linkage strategy control instruction is used to trigger the target bottom-level grinding component to execute the target state geometric compensation control sequence.

[0031] The perception and evaluation component is further configured to: perform voxel filtering and discrete noise stripping operations on the noisy point cloud data at the edge computing node, and output the multi-dimensional array of the track surface measured topography; calculate the normal Euclidean distance vector from each node of the multi-dimensional array of the track surface measured topography to the surface of the standard profile template, using the origin of the coordinate system of the standard profile template as a reference, and aggregate the normal Euclidean distance vector and output it as the contour space feature deviation.

[0032] The spatiotemporal feature extraction component is further configured to: peel off the set of dynamic variable points representing the contact surface of the rail crown from the multidimensional array of measured rail surface morphology along the depth coordinate axis, extract the set of constant frame representing the geometry of the rail waist to generate the first data; and retrieve the morphology matrix of the previous inspection batch corresponding to the mileage stamp from the structured maintenance database according to the current spatial mileage sequence to generate the second data.

[0033] The bypass spatiotemporal comparison component is further configured to: construct a cross-period topological feature correlation matrix between the first data and the second data; perform singular value decomposition on the cross-period topological feature correlation matrix to solve for the rotation and translation tensor characterizing transient vibration offset; establish the reference alignment coordinate system using the rotation and translation tensor; and perform discrete partial derivative calculation between the contour space feature deviation and the historical morphology array after alignment with the spatial reference.

[0034] The high-frequency fluctuation residuals characterizing spurious transient excitations are extracted from the discrete partial derivative sequence. Combined with the current travel velocity characteristics, the high-frequency fluctuation residuals are subjected to cumulative definite integral calculation within the spatial mapping interval corresponding to the preset time-domain evaluation window to generate a spatiotemporal noise isolation threshold. When the transient absolute fluctuation amplitude corresponding to the extracted discrete partial derivatives exceeds the normalized noise boundary mapped by the spatiotemporal noise isolation threshold, a state mapping mechanism that converts spatial geometric features to the material removal impedance constraint factor is triggered.

[0035] The tolerance dynamic correction component has a preset reference material impedance constant and a basic boundary function, and is further configured to: perform normalized inverse proportional mapping processing on the material removal impedance constraint factor according to the preset reference material impedance constant to generate a proportional compensation coefficient.

[0036] The basic boundary function representing the dynamic mapping relationship between the extreme value of the basic cutting depth and the spatial deviation is retrieved. The proportional compensation coefficient is used as a multiplication operator to act on the linear follower gain term inside the basic boundary function of the dynamic cutting compensation tolerance, while maintaining the original rigid minimum cutting tolerance of the basic boundary function unchanged. When the material removal impedance constraint factor represents high material removal resistance, the upper limit of the allowable extreme value of the dynamic cutting compensation tolerance is proportionally tightened by the proportional compensation coefficient.

[0037] The closed-loop instruction generation component is further configured to: perform macroscopic physical dimension decoupling mapping on the residual amount of deviation exceeding the upper limit of the allowable extreme value of dynamic cutting compensation tolerance through a preset equipment load allocation matrix, generate decoupling state parameters covering the target nominal speed and the downward torque, encapsulate based on the decoupling state parameters and output multi-dimensional grinding linkage strategy control instructions to the electro-hydraulic servo mechanism.

[0038] Upon receiving the system degradation protection flag transmitted through the system data bus, the system degradation protection mechanism is triggered, forcibly resetting the proportional compensation coefficient to a unit constant, causing the corrected dynamic cutting compensation tolerance to revert to the preset dynamic cutting compensation tolerance. The system degradation protection flag is generated and triggered by the spatiotemporal feature extraction component when it determines that the number of spatial feature points in the extracted first data is lower than a preset convergence threshold.

[0039] The bypass spatiotemporal comparison component is further configured to: establish a mapping flow model containing a dynamic sliding feature window when performing the operation of mapping the discrete partial derivatives to the material removal impedance constraint factor;

[0040] The real-time evaluation interval is extracted from the discrete partial derivative sequence through the dynamic sliding feature window; the number distribution of effective mutation nodes exceeding the spatiotemporal noise isolation threshold within the real-time evaluation interval is statistically analyzed to generate peak clustering density;

[0041] Obtain the preset dimension conversion alignment coefficient, and use the dimension conversion alignment coefficient as a conversion multiplier to perform heterogeneous dimension alignment multiplication operation with the peak cluster density to generate the normalized reference material removal impedance constraint factor.

[0042] The bypass spatiotemporal comparison component is further configured to: before transferring the factor to the downstream component, perform a gated logic comparison between the peak cluster density and the periodic wave erosion extreme value established based on the statistical boundary deduction of the on-site macroscopic physical excitation.

[0043] If the peak cluster density is determined to be greater than the extreme value of the periodic erosion, then the internal short-wave high-frequency enhanced impedance mapping path is triggered and activated.

[0044] Under the shortwave high-frequency enhanced impedance mapping path, a preset nonlinear amplification weighting factor greater than the unit constant is extracted. The nonlinear amplification weighting factor is used as a quadratic multiplication operator and a dimension conversion alignment coefficient is multiplied together to generate a reconstructed material removal impedance constraint factor to replace the reference material removal impedance constraint factor. The reconstructed material removal impedance constraint factor is then output back to the tolerance dynamic correction component to replace the original input.

[0045] The multidimensional array of track surface measured morphology represents a structured multidimensional spatial feature data set. It contains a heterogeneous matrix of real three-dimensional geometric coordinate nodes and train travel time / mileage dual timestamps, and is mapped as a standard data base for the perception and evaluation components to flow to other downstream components. The multidimensional array of track surface measured morphology is input from external perception nodes and serves as a unified processing benchmark for subsequent spatial registration by the system. Specifically, a voxel filter mesh size parameter is preset. This size parameter is a physical metric representing the granularity of three-dimensional spatial downsampling; in this embodiment, the preferred value is calibrated to a 2mm three-dimensional cube side length. The 2mm setting is larger than the physical diameter of unstructured high-frequency scattering noise points such as rust and oil stains on the track surface, and smaller than the trough span of the small plastic deformations that the rail body needs to focus on, thus preserving the core geometric features without loss while stripping away discrete noise.

[0046] This embodiment sets a preset convergence threshold when performing physical boundary safety verification. Its preferred value is fixed at 500 spatial feature points per effective scan frame. When the number of extracted constant track frame points is lower than this safety threshold, it indicates that the track web area has been severely covered by ballast or heavily sludge-covered. Continuing spatial registration will lead to the risk of misalignment cutting due to insufficient alignment reference. A dimensional transformation alignment coefficient is introduced to characterize the mapping from spatial geometric abrupt change frequency to microscopic metallurgical hardness resistance, with a preferred value calibrated to 0.08 MPa / (peak count / m).

[0047] In this embodiment, a test track covering a stepped sample of light to heavy corrugation is selected in the design of a simulated working environment. A torque sensing center is deployed to synchronously capture the transient output torque of the servo grinding motor at a constant feed depth to map the actual material removal resistance. The photoelectric sensing module is activated to output a pure geometric peak clustering density sequence from the photoelectric sensing array. The mapping slope is determined by extracting the high-frequency jump feature extrema points aligned on the time axis and performing linear regression fitting. The dimension transformation alignment coefficient is extracted and verified. .

[0048] To address the frequent high-frequency, severe deformation and corrugation conditions of target rails in real-world applications, a pre-calibration of the periodic corrugation occurrence extreme value and a nonlinear amplification weighting factor is established. This periodic corrugation occurrence extreme value is based on the evolution of the statistical boundary of macroscopic physical excitation. A full set of excitation waveform data samples covering one major overhaul cycle of the line's history is acquired, and high-dimensional feature space clustering analysis is performed. After stripping away the long-wave smooth base distribution representing natural wear, the peak interval of the exponentially clustered quadratic anomaly distribution in the residual high-frequency vibration energy spectrum is identified. The lower limit of the physical tolerance boundary on the left side of the convergent distribution of this cluster interval is taken as the decision watershed. Periodic corrugation occurrence extreme value. To characterize the critical number constant of spatial abrupt changes in the occurrence of malignant high-frequency deformation, the quantization value of this extreme value is calibrated to 12 abrupt change nodes / meter. When the peak density exceeds the distribution point of this abrupt change node, the degree of work hardening caused by lattice slip in the rail surface material exhibits a step-like leap. At this point, it is necessary to forcibly block the conventional linear impedance mapping path to prevent equipment damage.

[0049] The specific logic for determining the nonlinear amplification weighting factor is as follows: On a semi-physical simulation platform with a set of preset severely corrugated sample sections, the downward torque of the servo motor is gradually increased in micro-steps, and the dead zone threshold of the stall current jump of the motor stator is captured in real time. When the cutting state transitions from "slipping without material removal" to the critical point of "effective cutting and producing smooth shavings," the torque ratio at this time is recorded, thereby deriving the true impedance amplification factor under this working condition. Based on this extreme test calibration, the nonlinear amplification weighting factor is obtained. .in To characterize the dimensionless multiplier of the resistance jump under extreme physical fields, in this embodiment its preferred value range is set to [1.5, 2.5], and the optimal quantization value is calibrated to 2.

[0050] A system degradation protection flag representing the underlying hardware control signaling is defined, and its value is limited to a binary logic state (0 or 1). When the system state is stable, the system degradation protection flag is set to logic 0 (silent and not triggered); when the above-mentioned convergence threshold bottom line anomaly is triggered, the system degradation protection flag flips to logic 1, which means that the system degradation protection mechanism is triggered, forcibly blocking the injection of the higher-order partial derivative feedforward compensation signal, and forcibly resetting the proportional compensation coefficient to a unit constant, so that the corrected dynamic cutting compensation tolerance falls back to the preset dynamic cutting compensation tolerance, ensuring that the equipment enters the static physical optimization pure geometric cutting fallback state.

[0051] To address the technical problem of data space misalignment caused by severe vibration under complex working conditions in a traveling grinding vehicle, this preferred embodiment executes the collaborative interaction and control of each component according to the following logic:

[0052] The perception and evaluation component is configured to acquire noisy point cloud data, which includes superimposed transient spatial distortion artifacts and unstructured discrete noise, via a high-frequency photoelectric sensor array rigidly mounted on the chassis of the vehicle being inspected or polished. The perception and evaluation component receives this noisy point cloud data at an edge computing node and uses a preset voxel filter mesh size parameter to perform spatial meshing resampling of the noisy point cloud data to perform discrete noise removal. This voxel filter mesh size parameter is a physical metric characterizing the granularity of three-dimensional spatial downsampling; in this embodiment, its preferred value is defined as a 2mm side length of a three-dimensional cube.

[0053] The perception and evaluation component extracts a pre-defined standard profile template and, within a unified digital data space, calculates the normal Euclidean distance vector from each node within the multi-dimensional array of the measured orbital topography to the template surface, using the origin of the template's coordinate system as the absolute alignment reference. By performing spatial aggregation operations on these distance vectors, the perception and evaluation component outputs a profile spatial feature deviation that characterizes the global spatial offset, thereby providing a deviation reference with absolute mathematical consistency for subsequent system pipelines.

[0054] In the underlying data space, the standard profile template is specifically represented as an ideal two-dimensional or three-dimensional digital coordinate grid matrix that is pre-imported and generated based on the nominal design drawings of national railways (such as the TB / T series standards). Each spatial node inside the template is given an absolute zero-deviation coordinate reference constant, which serves as the only legal reference base map for performing spatial normal Euclidean distance registration of the multi-dimensional array of track surface measured morphology.

[0055] The spatiotemporal feature extraction component is configured to receive a multi-dimensional array of measured track surface morphology and perform logical decoupling of the physical structure. This component performs layered cutting along the digital spatial depth coordinate axis, stripping away the variable point set representing the rail crown contact surface (the area frequently subjected to wheel-rail rolling and exhibiting dynamic variable wear); then, it extracts the constant frame set representing the rail waist geometry from the lower part and encapsulates it to generate the first data. Based on the spatial mileage sequence identifier of the current system, the spatiotemporal feature extraction component retrieves the historical morphology matrix of the previous inspection batch that perfectly corresponds to the current mileage stamp from the structured maintenance database via a network communication interface and generates the second data. By extracting the rail waist—a physical part less prone to wear—as an anchor point through variable stripping, the spatial reference drift problem during historical comparison is resolved.

[0056] The bypass spatiotemporal comparison component is configured to receive the aforementioned first data, second data, and contour spatial feature deviation. Based on the first data, it solves for the rotation and translation tensor representing the transient vibration shift between the two detections using a matrix singular value decomposition operator, and establishes a reference alignment coordinate system accordingly. Then, it maps the second data into this coordinate system to achieve absolute spatial reference alignment of the historical morphology array. Discrete partial derivative calculations are performed between the contour spatial feature deviation and the historical morphology array after spatial reference alignment. A definite integral is then performed on the high-frequency fluctuation residuals during the calculation process to obtain the spatiotemporal noise isolation threshold, thereby intercepting meaningless spurious excitation spikes.

[0057] The bypass spatiotemporal comparison component then calls the defined dimension conversion alignment coefficient as the conversion multiplier and performs an alignment multiplication operation with the peak cluster density. The bypass spatiotemporal comparison component performs a gated logic comparison between the currently calculated peak cluster density and the periodic erosion occurrence extreme value established based on the statistical boundary deduction of the macroscopic physical excitation in the field; if it is determined that the peak cluster density does not exceed the limit, the above product result is directly output as the normalized material removal impedance constraint factor; if it is determined that the peak cluster density is greater than the extreme value, the component triggers the shortwave high-frequency enhanced impedance mapping path, extracts the preset nonlinear amplification weighting factor, and uses it as a quadratic multiplication operator to directly apply to the dimension conversion alignment coefficient, thereby multiplying the system's prediction of the material removal impedance of the material hardening section caused by erosion. Upon receiving the aforementioned mapped material removal impedance constraint factor, the tolerance dynamic correction component is configured to: logically convert it into a proportional compensation coefficient; retrieve the fundamental boundary function characterizing the dynamic mapping relationship between the extreme value of the basic cutting depth and the spatial deviation; apply this proportional compensation coefficient as a multiplication operator to the linear follower gain term within the fundamental boundary function of the dynamic cutting compensation tolerance; and maintain the original rigid minimum cutting tolerance of the fundamental boundary function unchanged. When the material removal impedance constraint factor characterizes high material removal resistance, the upper limit of the allowable extreme value of the dynamic cutting compensation tolerance is proportionally tightened through the proportional compensation coefficient, forming a safety clamp.

[0058] 1) Specific description of the perception and evaluation components: In view of the technical problems that the traveling grinding vehicle is inevitably affected by the high-frequency mechanical vibration of the chassis, the long-wave irregularity of the track, and the scattered light from the external environment under real working conditions, this system makes the physical boundary explicit at the data acquisition source.

[0059] The perception and evaluation component uses a high-frequency photoelectric sensor array (configured as a 2D / 3D line laser profilometer or binocular vision camera) rigidly mounted on the chassis of the vehicle being inspected or polished to perform high-frequency dynamic scanning of the rail surface under real-world operating conditions, thereby acquiring noisy point cloud data. This noisy point cloud data inevitably contains transient spatial distortion artifacts and unstructured discrete noise.

[0060] The perception and evaluation component is further configured to: perform voxel filtering and discrete noise removal operations on the noisy point cloud data at the edge computing node; in this process, statistical distribution distance calculation is combined to remove discrete noise points that are free outside the main track surface; the component will strongly correlate and reorganize the cleaned retained point set with the mileage timestamp according to a unified spatial topological order, and output a multi-dimensional array of track surface measured morphology containing three-dimensional spatial coordinates and mileage timestamps.

[0061] Using the origin of the coordinate system of the preset standard profile template as a reference, this component calculates the normal Euclidean distance vector from each node of the multi-dimensional array of the measured track surface topography to the surface of the standard profile template. During this aggregation process, the perception and evaluation component, within the data space, calls the root mean square spatial scalarization aggregation operator to aggregate the normal Euclidean distance vector and output it as the profile space feature deviation. Specifically, it is defined as follows: ; where R in the system spatial feature mapping means the output contour spatial feature deviation, which is preferably in the range of 0.05mm to 5mm in this embodiment, with a preferred value of 0.1mm; K means the total number of valid physical space feature points contained in the current edge calculation voxel grid, and its value is a set of positive integers greater than zero; The measured normal Euclidean distance vector of the k-th node extracted by the system; This is the normal unit reference mapping vector of the standard profile template surface at the corresponding digital projection point. It should be noted that this embodiment can adaptively shrink with the base resolution of the photoelectric sensing array in actual engineering. By introducing a root-mean-square spatial scalarization aggregation operator, the artifact extremum amplification caused by occasional high-frequency local discrete noise is effectively suppressed, thereby improving the smoothness of the on-site entity perception features.

[0062] 2) Specific implementation instructions for the spatiotemporal feature extraction component: Under severe vibration conditions, conventional comparisons are prone to reference drift. The spatiotemporal feature extraction component includes separating cutting variables and physical constants from the multi-dimensional array of measured rail surface morphology. The specific execution steps are as follows: extract the standard rail profile geometric reference of the system and read the preset spatial depth segmentation threshold from it. In digital space, along the depth coordinate axis (Z-axis), the component performs layered cutting and peeling calculations with the spatial depth segmentation threshold as the absolute rigid boundary, removing the set of dynamic variable points representing the rail crown contact surface in the multi-dimensional array of measured rail surface morphology (this area has highly unmeasurable wear variables due to frequent wheel-rail rolling); then, a constant frame set representing the rail waist geometry is extracted from the lower part to generate the first data. Among them, the spatial depth segmentation threshold... To characterize the spatial coordinate axis scalar parameters of the physical interface between the effective geometric frame and the wear variable zone, in this embodiment, their preferred value is set at 25mm directly below the absolute highest reference plane of the rail top. The plastic rheology and severe vertical wear limit depth of the rail caused by heavy-load wheel rolling rarely exceed 15mm below the rail top, while the traditional smooth junction of the straight / circular sections of the rail web is generally located at a depth of less than 30mm from the rail top. Locking the segmentation threshold at this 25mm safety blind zone isolation zone allows for the maximum preservation of the true and constant geometric topological support framework below while eliminating nonlinear deformation interference data points at the top.

[0063] The spatiotemporal feature extraction component is also configured to: extract the spatial mileage sequence of the current system, retrieve the morphology matrix of the previous detection batch with the corresponding mileage stamp from the structured maintenance database based on the spatial mileage sequence of the current system, and generate the second data.

[0064] In the process of generating the second data, this embodiment takes into account the inevitable cumulative drift of the on-board odometer hardware between two periodic detections. It matches the data from the structured maintenance database and sets a bidirectional dynamic physical addressing tolerance window centered on the odometer stamp of the current spatial mileage sequence, specifically extending 10 meters forward and backward. The system retrieves all historical batch topology matrices within this tolerance window as a candidate pool at once. This ensures that subsequent singular value decomposition operators can extract a sufficient number of real homogeneous topological nodes from this candidate pool for coordinate system alignment, thereby generating the second data.

[0065] 3) The specific implementation logic of the bypass spatiotemporal comparison component is as follows: Addressing the "noise gradient explosion" limit problem easily triggered in cross-cycle calculus calculations, the bypass spatiotemporal comparison component is configured to: address the disordered point cloud misalignment problem caused by high-frequency excitation of the underlying chassis; based on the extracted first and second data, this component extracts the corresponding topological nodes of the two data in isomorphic physical space, thereby constructing a cross-cycle topological feature correlation matrix between the first and second data. Its decomposition processing flow is as follows: The local geometric space centroids of the first and second data point sets are calculated in parallel, and a centering and mean-removing spatial offset operation is performed on the two arrays; the two mean-removed node topological sequences are mapped in the system to physical space coordinate blocks with a dimension topology of N×3, where N is the total number of spatial frame feature points; the bypass spatiotemporal comparison component performs tensor inner product operations based on these two sets of data to construct a core state data layer representing the corresponding relationship of bearing features, characterized as a cross-cycle topological feature correlation matrix H. The cross-period topological characteristic correlation matrix H is constrained to be a 3×3 real symmetric matrix in the physical system architecture. Its nine numerical units objectively represent the spatial projection orthogonal cross-offset weights along the X, Y, and Z physical axes. Subsequently, the core covariance operator is invoked to perform orthogonal singular value decomposition (SVD) calculations on the cross-period topological characteristic correlation matrix H: The left singular matrix U and right singular matrix V, containing latent spatial structure features, were successfully extracted through the above matrix decomposition steps. Simultaneously, the extracted... The matrix is ​​a diagonal matrix containing the singular values ​​of matrix H. The numerical elements on its diagonal objectively represent the constant distribution of geometric deformation weights along the characteristic directions of each orthogonal space; this is achieved by performing physical pose algebraic mapping and reorganization. The physical attitude offset caused by the transient mechanical excitation of the underlying layer during the two detection periods is solved and represented as a rotation-translation tensor. The bypass spatiotemporal comparison component establishes a reference alignment coordinate system using rotation and translation tensors, maps the second data to the reference alignment coordinate system, generates a historical topography array after spatial benchmark alignment, and peels off the vehicle body attitude offset through physical-level overlap.

[0066] 4) The bypass spatiotemporal alignment component is also configured as follows:

[0067] The high-frequency fluctuation residual is defined as the absolute value of the deviation after subtracting the reference physical noise smoothing filter surface function from the discrete partial derivative sequence. This residual represents the unsteady high-frequency fluctuation component caused by transient excitation of the vehicle body and superimposed on the steady-state morphology of the track, and serves as the sole quantization input source for subsequent energy integration accumulation to determine the system noise envelope boundary.

[0068] The temporal span variation features are extracted between the historical topographic array after the contour space feature deviation is aligned with the spatial reference. Specifically, spatial state analysis is performed on the features to solve for the discrete partial derivatives represented by the span variation rate between adjacent spatial nodes. High-frequency fluctuation residuals representing spurious transient excitations are extracted from the discrete partial derivative sequence. Combined with the current system's travel velocity characteristics within the spatial mapping interval corresponding to a preset time-domain evaluation window, cumulative definite integrals are calculated on the high-frequency fluctuation residuals to generate a spatiotemporal noise isolation threshold. The specific generation logic is as follows: ;in The energy threshold parameter characterizes the spatiotemporal noise isolation threshold; R is the contour spatial feature deviation; s is the span parameter of the spatial node; The preset baseline physical noise smoothing filter surface function is used; its core logic implementation is configured as follows: ;

[0069] in The meaning is the baseline of the normal superposition of macroscopic orbital environment noise distribution at the current node coordinates s; W means the total number of discrete span nodes contained in the one-dimensional spatial sliding window; w is the historical discrete node index cursor parameter inside the sliding window, and its value range is a positive integer sequence from 1 to W. This refers to the partial derivative of the historical effective spatial coordinate span deviation extracted from the discrete partial derivative sequence.

[0070] This is the preset time-domain evaluation window constant. t is the real-time physical timestamp variable when the system is currently performing this definite integral derivation, and it is related to the time-domain evaluation window. Together, they constitute the absolute time boundary of the current dynamic sliding integral; dt is an infinitesimal integral element on the time axis. In this embodiment... The optimal value is set as the estimated time required for the current grinding vehicle to travel through two standard sleeper intervals. By integrating and accumulating the residual energy over the complete long-wave excitation physical cycle, artifacts generated by transient discrete mechanical impacts can be smoothed and absorbed to the greatest extent, ensuring the output's spatiotemporal noise isolation threshold. It can accurately reflect the rigid noise floor level of the physical environment in which the system is currently located, thus providing a solid logical judgment isolation lower limit for the effective identification of subsequent corrugation changes in real rail materials.

[0071] In existing technologies, directly equating geometric partial derivatives with physical impedance leads to dimensional inconsistencies and logical contradictions. This embodiment utilizes a numerical comparator embedded in the system to pre-execute the following dimensionality reduction mapping logic: ;in In the system control logic, it represents the normalized noise boundary scalar after dimensionality reduction and reconstruction. In this embodiment, its preferred range is locked to 0.01 to 0.05 mm / m depending on the distribution characteristics of the underlying filter function. The spatiotemporal noise isolation threshold is generated by the preorder integral; the system extracts the absolute value of the discrete partial derivative of the current node in real time as the transient absolute fluctuation amplitude, and determines whether the transient absolute fluctuation amplitude corresponding to the extracted discrete partial derivative crosses the normalized noise boundary mapped by the spatiotemporal noise isolation threshold. At that time, the bypass spatiotemporal comparison component is configured to: trigger a state mapping mechanism that transforms spatial geometric features into material-free impedance constraint factors; and establish a mapping flow model containing a dynamic sliding feature window when performing the operation of mapping discrete partial derivatives to material-free impedance constraint factors. The system obtains a preset time length constant, uses it as the time constraint dimension to establish the dynamic sliding feature window, and extracts a finite-length real-time evaluation interval from the sequence data stream of discrete partial derivatives through the dynamic sliding feature window. The service value of this time length constant is calibrated to 1.5s. At the nominal travel speed, this duration exactly covers the minimum spatial physical extension span of the local macroscopic material change of the rail, which avoids the high-frequency corrugation characteristics being diluted by the low-frequency smooth section due to an excessively large window, and also prevents the misjudgment of discrete accidental spikes induced by an excessively small window, thereby constructing the most basic real-time evaluation boundary of the data stream.

[0072] The distribution of the number of effective mutation nodes exceeding the spatiotemporal noise isolation threshold within the real-time evaluation interval is statistically analyzed. This distribution is then divided by the physical spatial span scalar corresponding to the current dynamic sliding feature window, thus performing dimensionality reduction and outputting a purely geometric "peak cluster density." The model for this local dimensionality reduction step is configured as follows: ;in Peak clustering density used as a reference input source for downstream impedance mapping; This represents the total number of valid mutation peak nodes extracted and accumulated by the system within the current time-domain feature window. This represents the objective physical spatial span scalar, mapped in the multi-dimensional array of measured topography and bound to the current domain evaluation window. Specifically, it represents the length of the window extending longitudinally along the rail. The value of this scalar is controlled by the vehicle's travel velocity vector and is dynamically calculated and distributed synchronously from the system's underlying odometer. This ensures that the generated density features are no longer affected by the total number of samples within a single window, providing a spatial corrugation evaluation scale with absolute comparability across intervals.

[0073] This component acquires a preset dimension conversion alignment coefficient, uses this coefficient as a conversion multiplier and performs a heterogeneous dimension alignment multiplication operation with the peak cluster density to generate a normalized baseline material removal impedance constraint factor. If the peak cluster density is determined to be greater than a preset extreme value for periodic erosion, the component triggers and activates the internal shortwave high-frequency enhanced impedance mapping path, extracts a preset nonlinear amplification weighting factor greater than a unit constant, uses this nonlinear amplification weighting factor as a quadratic multiplication operator and performs a multiplication operation with the dimension conversion alignment coefficient to generate a reconstructed material removal impedance constraint factor to replace the baseline material removal impedance constraint factor, and outputs the reconstructed material removal impedance constraint factor back to the tolerance dynamic correction component to replace the original input.

[0074] In this embodiment, cross-physical domain state mapping calculations are performed during bypass spatiotemporal alignment, and the following piecewise multi-source joint constraint equations are constructed to derive the material removal impedance constraint factor: ;in The material rejection impedance constraint factor is the final output of the system. The peak clustering density parameter is used for the statistical dimensionality reduction output within the dynamic sliding feature window. The pre-defined extreme value gate boundary for periodic wave motion is set. This is the preset unit conversion alignment factor; This is the nonlinear amplification weighting factor used to trigger the enhancement path. The system compares the previously isolated peak clustering density parameters using the aforementioned gating logic. Perform algebraic mapping and reorganize the physical property data carrier into a material-free impedance constraint factor that can be transferred to downstream tolerance-dynamically corrected components. This ensures that the "material resistance constraint factor" ultimately issued by the system represents the surge in real microscopic metallurgical hardness and material mechanical resistance caused by local lattice deformation of the rail, achieving hard synchronization between information layer characteristics and physical execution layer states.

[0075] 4) Specific implementation instructions for the tolerance dynamic correction component:

[0076] The tolerance dynamic correction component is configured to: receive the material removal impedance constraint factor characterizing the micro-metallurgical hardness, and perform normalized inverse proportional mapping processing on it according to the preset reference material impedance constant, thereby generating a proportional compensation coefficient. The tolerance dynamic correction component has a preset reference material impedance constant reflecting the conventional cutting resistance of standard non-destructive rails. Internally, the tolerance dynamic correction component calls a division operator module, using the reference material impedance constant as the numerator and the material removal impedance constraint factor generated in real-time mapping as the denominator, to construct a normalized mapping relationship: ;in To calculate the target proportional compensation coefficient; The material rejection impedance constraint factor is the input for real-time flow. The reference material impedance constant pre-stored in the system represents the standard baseline cutting compressive resistance of the rail base material in the target line section under actual physical conditions, without encountering microscopic fatigue hardening. In this embodiment, the reference material impedance constant is derived based on the measured metallographic mechanical hardness of the standard U71Mn heavy-haul rail at the factory. Its physical standardization and quantitative optimization range is defined as 350MPa to 450MPa, with an optimal value of 400MPa. It should be noted that this resistance constant can be adaptively refreshed and loaded in conjunction with the specific track maintenance level database. By introducing baseline feedforward constraint parameters controlled by the actual metallographic hardness boundary parameters, the logical engagement between spatial digital deviation and the actual execution resistance threshold of the underlying material is achieved, thereby cutting off ineffective anti-collapse blind deep cutting when encountering spalling and softening sections, and expanding the operational safety boundary of the system in complex material degradation road networks.

[0077] In this embodiment, the parallel saturation interception dead zone logic is as follows: when a decision is made... (When physically characterized as encountering a softened rail zone or an abnormal spalling section) the system will block the above calculations and force [the process to proceed]. The clamp is reset to a unit absolute constant of 1 to strictly prevent the equipment from over-amplifying and cutting downwards; when in the high-frequency wave-scraping section... When the ratio operator is in the range of 0, it will deterministically output a decimal factor within the closed interval (0, 1).

[0078] The component retrieves the basic boundary function that characterizes the dynamic mapping relationship between the extreme value of the basic cutting depth and the spatial deviation. The component uses the proportional compensation coefficient as a multiplication operator to apply the linear follower gain term inside the basic boundary function of the dynamic cutting compensation tolerance, and maintains the original rigid minimum cutting tolerance of the basic boundary function unchanged.

[0079] In the industrial physical control chain, the fundamental boundary function is a mathematical clamping baseline established based on the coupling of objective field cutting follow-up energy efficiency and equipment system assembly dead zone. The core control logic flow of the fundamental boundary function running inside this component is defined as follows: ;in R is the upper limit of the basic tolerance extreme value of the cutting compensation output to the subsequent stage; R is the contour space feature deviation parameter of the real-time feedforward input. The minimum cutting constant is set for the system to maintain the most basic physical contact state of the bottom grinding wheel, which is quantified as 0.2mm in this preferred scenario; To shield and absorb vibrations from the original mechanical assembly of the underlying chassis, the control dead zone tolerance range is preferably quantified to 0.1mm; To characterize the pure linear gain weight of the servo feedback tool cutting follow stiffness strength of the system, it is set to 0.85 in this scheme. By loading this determined mathematical control model, the system can automatically deduce the basic envelope clamping curve with rigid lower limit protection and linear follow-up slope based on the current objective residual size in a normal smooth operating section without high-frequency wave mill extreme impedance interference.

[0080] Furthermore, this component will include a proportional compensation coefficient. As a multiplication operator, it acts on the linear follower gain term within the aforementioned basic boundary function, while maintaining the original rigid minimum cutting tolerance of the basic boundary function. Under this constraint logic, the modified dynamic cutting compensation tolerance model, reconstructed by system feedforward intervention, is standardized as follows: ;in In the macro control link, this is represented as the upper limit of the modified dynamic cutting compensation tolerance allowable extreme value that is ultimately transferred to the downstream servo mechanism. The target proportional compensation coefficient, generated for decoupling the preceding system and used to adaptively suppress material resistance, acts directly and solely as a multiplier on the linear gain weights. By introducing a fixed-point isolation gain term feedforward constraint mechanism, a nonlinear adaptive tightening of the dynamic error compensation force and a static mechanical fit are achieved, resulting in architecture-level physical isolation that avoids the pressure loss oscillation of the underlying grinding wheel caused by abnormal proportional intervention, thus maintaining the strong robustness of the system under heavy load optimization under extreme working conditions. When the material removal impedance constraint factor represents the abnormal hardening region characterized by high material removal resistance, the allowable extreme value upper limit of the dynamic cutting compensation tolerance is proportionally tightened by the proportional compensation coefficient.

[0081] 5) Specific implementation description of the closed-loop instruction generation component: At the physical end of the control chain, the closed-loop instruction generation component is configured as follows: When the deviation of the contour space feature exceeds the upper limit of the dynamic cutting compensation tolerance limit that has been corrected and tightened, the system retrieves the preset equipment load allocation matrix. Through the preset equipment load allocation matrix, the system performs macroscopic physical dimension decoupling mapping on the residual amount of the deviation that exceeds the allowable upper limit, thereby generating heterogeneous state decoupling parameters covering the nominal target speed and downward torque required by the bottom execution end. Based on these decoupling state parameters, the closed-loop instruction generation component encapsulates them according to the fieldbus protocol to generate multi-dimensional grinding linkage strategy control instructions, and outputs multi-dimensional grinding linkage strategy control instructions to the bottom electro-hydraulic servo mechanism. The multi-dimensional grinding linkage strategy control instructions are used to trigger the target bottom grinding component to execute the target state geometric compensation control sequence.

[0082] The target state geometric compensation control sequence is generated by the clock generator built into the closed-loop instruction generation component. According to the system's preset underlying communication bus period of 10ms / frame, the discrete target speed calculated above is generated. ) and downward torque ( The state parameters are smoothly interpolated and packaged according to the time axis to form a continuous control time sequence that can drive the servo valve of the target bottom grinding component to perform continuous dynamic trajectory cutting.

[0083] In this deep cyber-physical system collaboration process, to avoid tolerance exceeding limits directly causing mechanical overload and stalling in a single execution dimension, this embodiment adopts the following state-space equation for smooth decoupling and amortization dispatch of the execution instruction dimension: ;in and These represent independently mapped and heterogeneous outputs: the target speed of the grinding wheel motor and the downward torque control signal of the hydraulic servo cylinder; and These represent the residual deviation monitored by the system and its rate of change with respect to the derivative with respect to time; M is the equipment load allocation matrix for the load control weight. In this embodiment, the first row of the matrix... In the physical architecture, the proportional attenuation parameter that is objectively represented as the mapping of the spatial deviation residual amount to the target speed of the grinding wheel motor is defined in the preferred range of 1.5 to 3, with a preferred value of 2.2. The differential damping parameter, characterized as the mapping of the deviation change rate to the target rotational speed, is preferably defined in the range of 0.1 to 0.5, with a preferred value of 0.3.

[0084] The second row of the matrix In the fault-tolerant control of the system, the objective characteristic is the normal proportional gain buffer multiplier of the electro-hydraulic servo feed cylinder that overcomes the mechanical damping of the original vibration of the bottom chassis. Its preferred range is defined as 0.5 to 1.2, and the preferred value is 0.8. The normal differential smooth multiplier of the feed cylinder is characterized by a preferred range of 0.05 to 0.2, with a preferred value of 0.1.

[0085] The formation of the above four sets of heterogeneous dynamic allocation parameters abandons manual trial and error. Instead, during the initial configuration cycle of the system, standard step test pulses are sent to the underlying electro-hydraulic servo mechanism, and the closed-loop torque decay steady-state echo is captured. Automated regression identification is performed to generate the parameters, which are then solidified in the system as a preset equipment load allocation matrix.

[0086] The spatiotemporal feature extraction component is also configured to: continuously monitor the number of spatial feature points of the first data; once it is determined that the number of extracted spatial feature points of the first data is lower than a preset convergence threshold, the spatiotemporal feature extraction component generates and triggers a flip-up system degradation protection flag. In railway maintenance scenarios, extreme and harsh working conditions can cause a sharp decrease in the effective physical echo acquired by the photoelectric sensor array. In order to prevent the subsequent spatial registration pipeline based on the singular value decomposition operator from inducing the risk of misalignment due to insufficient alignment reference, the system pre-positions this rigid blocking logic on the underlying hardware control signaling side. In this embodiment, the preset convergence threshold is preferably set to 500 effective spatial feature points / scan frame. The flag transition value is limited to a binary logic state (0 or 1). When the system is in a stable cruise and the number of spatiotemporal feature points is sufficient, the flag maintains logic 0; once the spatiotemporal feature extraction component determines an anomaly, the flag is immediately flipped to logic 1. The closed-loop instruction generation component is further configured to: upon receiving a system degradation protection flag transmitted through the system data bus, trigger the system degradation protection mechanism, forcibly intercept the high-order compensation link, reset the proportional compensation coefficient to a unit constant, and cause the corrected dynamic cutting compensation tolerance to revert to the preset dynamic cutting compensation tolerance. This fallback mechanism avoids equipment downtime and blind cutting loss of control during ongoing maintenance operations, ensuring that the system smoothly degrades to the basic static geometric optimization mode.

[0087] It should be further explained that: the target bottom-level grinding component refers to an electromechanical-hydraulic integrated physical execution terminal that receives multi-dimensional grinding linkage strategy control commands from the data bus and directly implements rigid physical cutting. In terms of hardware structure, this component consists of an orthogonally coupled electro-hydraulic servo feed mechanism and a cutting power drive unit. The electro-hydraulic servo feed mechanism includes a bottom-level push rod with an absolute displacement sensor and a proportional servo hydraulic cylinder. This mechanism is specifically configured to analyze the downward torque generated by the decoupling of the preceding matrix (…). The signaling overcomes the chassis's inherent vibration damping and executes a dynamic feed and rigid contact action along the Z-axis towards the rail surface; the cutting power drive unit includes a high-power variable frequency motor and a heavy-duty grinding wheel rigidly connected to its spindle. This unit is specifically configured to analyze the nominal speed generated by the decoupling of the preorder matrix ( The signaling executes a high-speed rotary cutting operation to overcome the micro-metallurgical hardening resistance of the rail. The trigger for this component to execute the target state geometric compensation control sequence indicates that, under the synchronization of the system's underlying hard real-time clock, the electro-hydraulic servo feed mechanism and the cutting power drive unit continuously and concurrently adjust the depth of cut and spindle cutting speed according to the continuously updated upper limit of the correction tolerance. This achieves a physical offset closed loop to correct the deviation of the physical trajectory from the digital profile, while avoiding mechanical vibration of the underlying push rod and overload failure of the grinding wheel.

[0088] The core control objective of this embodiment is to generate a modified upper limit for the dynamic cutting compensation tolerance through feedforward intervention. The control center for this output value is the target proportional compensation coefficient, whose value range converges to a closed interval greater than zero and less than or equal to one under the condition of severe material removal impedance.

[0089] When the output value approaches the maximum extreme value of 1: the material removal impedance constraint factor calculated at the system's bottom layer is less than the preset reference material impedance constant. The microscopic metallurgical lattice of the rail section where the grinding equipment is currently located has not undergone significant work hardening, or it has encountered a softened and peeling area of ​​the rail surface material. The technical response is to forcibly block the attenuation logic, making the upper limit of the extreme value completely equivalent to the static geometric optimization basic tolerance. Strictly prevent the equipment from over-amplifying and cutting down due to compensation failure in the softened section, forming a physically absolute anti-collapse protection.

[0090] When the output value approaches the minimum extreme value of 0: the real-time generated material removal impedance constraint factor exponentially exceeds the baseline of the reference material impedance constant. The actuator encounters a short-wave high-frequency abrasion lattice hardening zone with high material removal resistance. The technical response is to forcibly clamp the dynamic cutting compensation tolerance downwards nonlinearly, making it infinitely close to the system's set minimum cutting constant. This convergence trend establishes an insurmountable safety baseline, absolutely ensuring that the underlying electro-hydraulic servo mechanism will not suffer irreversible damage such as torque overload or stall burnout due to spatial dimension command mismatch in the physical field where cutting resistance soars.

[0091] A piecewise positive correlation nonlinear mapping relationship exists between the peak cluster density and the "material impedance constraint factor." This directly corresponds to the plastic rheology and work hardening phenomena of the rail surface material after being subjected to high-frequency wheel-rail impact. Designing this as a positive correlation mapping enables this method to achieve cross-physical domain quantitative tracking from "spatial high-frequency deformation" to "material metallurgical hardness." This positive correlation design establishes the data benchmark for system impedance feedforward. It ensures that when micro-erosion deteriorates in the road network, the system can generate a high-resistance prediction before the torque feedback of the servo motor, achieving an advancement in control dimension from "passive physical torque feedback" to "active data state prediction."

[0092] The nonlinear amplification weighting factor, acting as a discrete multiplication operator, is triggered only when the "cluster density of wave crests" exceeds the "extreme value of periodic wave grinding." It exhibits a step-wise positive correlation with the predicted material removal impedance. When the wave grinding node density exceeds the extreme threshold, the material's microscopic cutting resistance no longer conforms to the basic linear growth model but instead exhibits an exponential resistance transition due to extreme lattice dislocation. Introducing a weighting constant within a certain range to perform a quadratic multiplication operation provides an objective mathematical approximation and compensation for this nonlinear physical transition phenomenon at the system information control level. The direct injection of the nonlinear amplification weighting factor reconstructs the impedance safety boundary during computation. When facing extreme hardening conditions, the amplified intervention strength ensures that the tolerance clamping action can be tightened in one go, preventing instantaneous mechanical damage induced by the slow response of the linear mapping and maintaining the system's strong robustness during heavy-load optimization.

[0093] This embodiment is configured in a digital twin monitoring scenario for dynamic grinding of heavy-haul railway lines. This scenario synchronously transmits objective line conditions based on measured mileage data. The system receives discrete inputs containing multi-dimensional spatial coordinates and calculates the tolerance extremes of the underlying servo mechanism. The periodic corrugation occurrence extreme value gate constant is set to 12 abrupt change nodes / meter; the dimension transformation alignment coefficient is set to 15 MPa per unit peak density; the optimal value of the nonlinear amplification weighting factor is fixed at 2; the reference material impedance constant is set to 400 MPa; the minimum cutting constant is set to 0.2 mm; the control dead zone tolerance range is set to 0.1 mm; and the linear servo gain weight is set to 0.85. The real-time external inputs of the system include macroscopic spatial deviations and the peak clustering density output after dimensionality reduction. Table 1 below aims to verify the cross-physical domain constraint intervention effect of the "bypass spatiotemporal comparison component" and the "tolerance dynamic correction component" by inputting six sets of monitoring data representing different degrees of objective wear and physical hardening of rails. By demonstrating the nonlinear clamping and suppression effect of the excitation wave peak within the spatial frequency on the extreme value of dynamic cutting tolerance, this embodiment quantitatively reveals the decisive protection mechanism that prevents overload and mechanical fatigue of the underlying servo equipment in advance when encountering a malignant hardening section.

[0094] Table 1: Monitoring data on objective wear and physical hardening of rails

[0095]

[0096] In scenarios one through three, because the real-time generated material removal impedance constraint factor is lower than the reference material impedance constant, the system's underlying logic triggers dead zone interception, forcibly resetting the target proportional compensation coefficient clamp to the unit absolute constant 1.0, thus blocking overprotection. In scenarios four through six, because the peak clustering density exceeds the periodic ripple and the impedance crosses the baseline, the nonlinear amplification operator and inverse proportional compression calculation constraint are forcibly activated.

[0097] In Table 1, the tolerance compression convergence rate is calculated by subtracting the target proportional compensation coefficient generated by real-time mapping from the unit absolute constant 1 within the mathematical system. This parameter is used to quantitatively characterize the dynamic suppression physical strength of the system on the basic geometric cutting tolerance command space when facing high material removal resistance; the higher its value, the greater the anti-overload rigid clamping force performed by the system for extreme metallurgical fatigue regions.

[0098] After triggering the extreme value gating, the value of the material removal impedance constraint factor is obtained by multiplying the preset nonlinear amplification weighting factor, the dimension conversion alignment coefficient, and the real-time input peak clustering density. The target proportional compensation coefficient is calculated by dividing the system-fixed reference material impedance constant by the current material removal impedance constraint factor, to characterize the amount of reverse impedance suppression.

[0099] Independent performance verification was conducted by extracting data combinations from "Scenario Six: Extreme Mechanical Excitation Section": Under the existing simple geometric contour following technology path, the control system relies solely on a two-millimeter spatial deviation for simple static dimensional compensation. According to the fixed basic boundary flow logic, without the introduction of impedance feedforward constraints, the original tolerance output ultimately sent to the servo cylinder will reach 1.815 mm (the calculation logic is the effective spatial deviation residual multiplied by the linear follower gain plus a minimum constant). Under the same environmental input boundary, the cross-physical domain constraint framework constructed in this embodiment captures the objectively existing 30 peaks per meter topology within this spatial band, deduces the nonlinear deformation impedance of 900 MPa in this section, and clamps the final tolerance command flowing to the servo mechanism to 0.917 mm. Based on this objective data evolution, the dynamic correction mechanism of this embodiment successfully reduced the deep cutting command that could cause tool breakage by 49.5% when encountering extreme hardening conditions. The objective calculation basis for this compression ratio is as follows: the absolute safe reduction amount is obtained by subtracting the upper limit of the dynamic cutting compensation tolerance extreme value after correction of the original tolerance output value, and then the absolute safe reduction amount is divided by the original tolerance output value to obtain the final ratio. This comparison result confirms that constructing a low-level protection network that directly constrains the feedforward output command by the hard index of physical environment metallurgical resistance completely eliminates the systemic risks of servo mechanical stall and overcurrent burnout from the very front end of the physical link.

[0100] like Figure 1 The diagram illustrates the closed-loop operation of the rail profile inspection data analysis and grinding system under real heavy-haul railway maintenance conditions. The physical execution entity on the left represents the coordinated state of the system's underlying hardware, while the technical roadmap on the right corresponds to the core control link in this embodiment.

[0101] The first node's "Perception Assessment and Feature Extraction" corresponds to the perception and assessment component and the spatiotemporal feature extraction component. Its characterization system acquires noisy point cloud data and performs cleaning, outputting a multi-dimensional array of measured orbital surface morphology. It then extracts the first data of the geometrically invariant topology of the non-cutting area of ​​the orbital waist and the second data of the historical morphology, completing spatial registration.

[0102] The second node, "Bypass Spatiotemporal Topology Alignment," corresponds to the bypass spatiotemporal alignment component. This step constructs a cross-cycle reference alignment coordinate system based on the previously extracted data, extracts the temporal span variation characteristics, performs cross-physical domain state mapping calculations, and outputs the core material removal impedance constraint factor.

[0103] The third node, "Tolerance Dynamic Boundary Clamping," corresponds to the tolerance dynamic correction component. It receives the aforementioned material removal impedance constraint factor, performs a nonlinear boundary clamping operation on the preset dynamic cutting compensation tolerance model, and outputs the corrected upper limit of the allowable extreme value, forming a safety anti-collapse pocket.

[0104] The fourth node, "Multi-dimensional Grinding Closed-Loop Control," corresponds to the closed-loop command generation component. When the contour deviation is determined to exceed the limit, this module generates a multi-dimensional grinding linkage strategy control command based on the corrected extreme value, directly triggering and controlling the target bottom-level grinding component (grinding wheel array) in the left-hand diagram to perform adaptive geometric compensation cutting.

[0105] It should be noted that all computational logic in this application employs regression analysis, including but not limited to machine learning algorithms, to deeply analyze the collected parameters and identify their natural trends and interrelationships. In all computational formulas of this application, the parameters in each formula undergo dimensionless processing within a consistent range to ensure that different physical quantities are compared on the same scale. Dimensionless processing techniques include, but are not limited to, Min-Max-Normalization and Z-Score standardization. To decouple the core algorithm of this invention from specific application strategies and to ensure the configurability and ease of debugging of the technical solution, in the specific implementation path of this invention, all configurable operating parameters are read through a standardized "configuration interface." The data source of this configuration interface is a "data storage module" (e.g., a non-transitory computer-readable storage medium, such as a configuration file, database entry, or cloud configuration service), which is configured to store configuration data in key-value pair format.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A rail profile inspection data analysis and grinding system, characterized in that, Specifically, it includes: The perception and evaluation component is configured to acquire noisy point cloud data of the target rail surface, perform feature cleaning on the noisy point cloud data to extract a multi-dimensional array of the measured rail surface morphology containing three-dimensional spatial coordinates and mileage timestamps; call a preset standard profile template, and perform spatial registration on the multi-dimensional array of the measured rail surface morphology based on the standard profile template in the data space to generate a profile spatial feature deviation. The spatiotemporal feature extraction component is configured to extract first data representing the geometrically invariant topological features of the non-cutting zone from the multi-dimensional array of the measured morphology of the track surface, wherein the first data consists of a discrete set of spatial feature points; and to obtain second data representing the historical periodic morphological state of the same mileage section. The bypass spatiotemporal comparison component is configured to receive the first data, the second data and the contour spatial feature deviation, construct a reference alignment coordinate system based on the cross-period topological mapping relationship between the first data and the second data, map the second data to the reference alignment coordinate system, and generate a historical morphology array after spatial reference alignment. Extract the temporal span variation features of the deviation between the historical topography array after spatial reference alignment and the contour spatial feature, perform cross-physical domain state mapping calculation on the temporal span variation features, and generate the material removal impedance constraint factor; The tolerance dynamic correction component is configured to receive the material removal impedance constraint factor, use the material removal impedance constraint factor to perform boundary clamping on the preset dynamic cutting compensation tolerance model, and generate the corrected upper limit of the dynamic cutting compensation tolerance allowable extreme value. The closed-loop instruction generation component is configured to generate a multi-dimensional grinding linkage strategy control instruction when it is determined that the deviation of the contour space features exceeds the upper limit of the allowable extreme value of the corrected dynamic cutting compensation tolerance. The multi-dimensional grinding linkage strategy control instruction is used to trigger the target bottom-level grinding component to execute the target state geometric compensation control sequence.

2. The rail profile detection data analysis and grinding system according to claim 1, characterized in that: The perception and evaluation component is further configured to: perform voxel filtering and discrete noise stripping operations on the noisy point cloud data at the edge computing node, and output the multi-dimensional array of the measured orbital surface topography. Using the origin of the coordinate system of the standard profile template as a reference, calculate the normal Euclidean distance vector from each node of the multidimensional array of the measured shape of the track surface to the surface of the standard profile template, aggregate the normal Euclidean distance vector and output it as the profile space feature deviation.

3. The rail profile detection data analysis and grinding system according to claim 2, characterized in that: The spatiotemporal feature extraction component is further configured to: peel off the set of dynamic variable points representing the contact surface of the rail crown from the multidimensional array of measured rail surface morphology along the depth coordinate axis, extract the set of constant frame representing the geometry of the rail waist to generate the first data; and retrieve the morphology matrix of the previous inspection batch corresponding to the mileage stamp from the structured maintenance database according to the current spatial mileage sequence to generate the second data.

4. The rail profile detection data analysis and grinding system according to claim 3, characterized in that: The bypass spatiotemporal comparison component is further configured to: construct a cross-period topological feature correlation matrix between the first data and the second data; perform singular value decomposition on the cross-period topological feature correlation matrix to solve for the rotation and translation tensor characterizing transient vibration offset; establish the reference alignment coordinate system using the rotation and translation tensor; and perform discrete partial derivative calculation between the contour space feature deviation and the historical morphology array after alignment with the spatial reference. Extract the high-frequency fluctuation residuals that characterize spurious transient excitations from the discrete partial derivative sequence, and combine them with the current travel speed characteristics. Within the spatial mapping interval corresponding to the preset time domain evaluation window, perform cumulative definite integral calculation on the high-frequency fluctuation residuals to generate a spatiotemporal noise isolation threshold value. When the transient absolute fluctuation amplitude corresponding to the extracted discrete partial derivative exceeds the normalized noise boundary mapped by the spatiotemporal noise isolation threshold, a state mapping mechanism is triggered to transform the spatial geometric features into the material removal impedance constraint factor.

5. The rail profile detection data analysis and grinding system according to claim 4, characterized in that: The tolerance dynamic correction component has a preset reference material impedance constant and a basic boundary function, and is further configured to: perform normalized inverse proportional mapping processing on the material removal impedance constraint factor according to the preset reference material impedance constant to generate a proportional compensation coefficient. The basic boundary function representing the dynamic mapping relationship between the extreme value of the basic cutting depth and the spatial deviation is retrieved. The proportional compensation coefficient is used as a multiplication operator to act on the linear follower gain term inside the basic boundary function of the dynamic cutting compensation tolerance, while maintaining the original rigid minimum cutting tolerance of the basic boundary function unchanged. When the material removal impedance constraint factor represents high material removal resistance, the upper limit of the allowable extreme value of the dynamic cutting compensation tolerance is proportionally tightened by the proportional compensation coefficient.

6. The rail profile detection data analysis and grinding system according to claim 5, characterized in that: The closed-loop instruction generation component is further configured to: perform macroscopic physical dimension decoupling mapping on the residual amount of deviation exceeding the upper limit of the allowable extreme value of dynamic cutting compensation tolerance through a preset equipment load allocation matrix, generate decoupling state parameters covering the target nominal speed and the downward torque, encapsulate based on the decoupling state parameters and output multi-dimensional grinding linkage strategy control instructions to the electro-hydraulic servo mechanism. Upon receiving the system degradation protection flag transmitted through the system data bus, the system degradation protection mechanism is triggered, forcibly resetting the proportional compensation coefficient to a unit constant, causing the corrected dynamic cutting compensation tolerance to revert to the preset dynamic cutting compensation tolerance. The system degradation protection flag is generated and triggered by the spatiotemporal feature extraction component when it determines that the number of spatial feature points in the extracted first data is lower than a preset convergence threshold.

7. The rail profile detection data analysis and grinding system according to claim 6, characterized in that: The bypass spatiotemporal comparison component is further configured to: establish a mapping flow model containing a dynamic sliding feature window when performing the operation of mapping the discrete partial derivatives to the material removal impedance constraint factor; The real-time evaluation interval is extracted from the discrete partial derivative sequence through the dynamic sliding feature window; the number distribution of effective mutation nodes exceeding the spatiotemporal noise isolation threshold within the real-time evaluation interval is statistically analyzed to generate peak clustering density; Obtain the preset dimension conversion alignment coefficient, and use the dimension conversion alignment coefficient as a conversion multiplier to perform heterogeneous dimension alignment multiplication operation with the peak cluster density to generate the normalized reference material removal impedance constraint factor.

8. The rail profile detection data analysis and grinding system according to claim 7, characterized in that: The bypass spatiotemporal comparison component is further configured to perform gated logic comparison between the peak cluster density and the periodic wave erosion occurrence extreme value established based on the statistical boundary deduction of on-site macroscopic physical excitation; If the peak cluster density is determined to be greater than the extreme value of the periodic erosion, then the internal short-wave high-frequency enhanced impedance mapping path is triggered and activated. Under the shortwave high-frequency enhanced impedance mapping path, a preset nonlinear amplification weighting factor greater than the unit constant is extracted. The nonlinear amplification weighting factor is used as a quadratic multiplication operator and a dimension conversion alignment coefficient is multiplied together to generate a reconstructed material removal impedance constraint factor to replace the reference material removal impedance constraint factor. The reconstructed material removal impedance constraint factor is then output back to the tolerance dynamic correction component to replace the original input.

Images