Wheel center positioning method and system of railway vehicle

Abstract

The invention provides a wheel center positioning method and system of a railway vehicle. Wherein the first detection unit and the second detection unit are mounted on two sides of the wheel, and the chord length of the wheel is measured through the first detection unit. When the second unit triggers detection, the second unit is controlled to move for a semi-chord distance in the same direction, and initial positioning of the wheel center is determined. Dynamic contact spot deformation and rim degradation data are fused, a collaborative analysis model is constructed, the nonlinear relation between the abrasion gradient and contact spot offset is extracted through deep learning, and dynamic correction is generated. And millimeter-level dynamic compensation is carried out on the positioning center based on the dynamic correction, stability control over the wheel-rail contact center in the transverse direction, the vertical direction and the rotating dimension is achieved, and stress concentration caused by rim abrasion is restrained. According to the technical scheme provided by the invention, the wheel-rail contact center positioning precision and stability are improved.

Description

Technical Field

[0001] The present application relates to the technical field of wheel center positioning, and particularly to a method and system for positioning the wheel center of a rail vehicle. Background Art

[0002] In the railway transportation system, the real-time and accurate positioning of the wheel-rail contact center is crucial for the running stability of the train, the control of wheel flange wear, and the improvement of the track life. Due to the geometric deformation and the change of contact stress generated by the long-term dynamic interaction between the wheel and the track, the existing technologies need to meet the real-time adaptive positioning ability under dynamic working conditions, and at the same time, avoid the problems of contact stress concentration and abnormal wear caused by positioning deviation.

[0003] One currently adopted solution is a dynamic monitoring system for wheel flange geometric parameters based on multi-sensor fusion. The wheel position data is collected in real time through laser ranging and an inertial sensing unit, the wheel-rail contact center is calculated by combining a preset geometric model, and the positioning mechanism is driven by a servo motor for position compensation.

[0004] However, relying on the preset geometric model for positioning calculation, but not deeply analyzing the influence of the deformation and stress distribution of the wheel-rail material on the offset of the contact center during the dynamic contact process, resulting in insufficient adaptability of the correction amount to the mechanical characteristics of the actual working conditions. During long-term operation, the positioning accuracy is easily interfered by factors such as material creep, causing the problem of low wheel center positioning accuracy. Summary of the Invention

[0005] The present application provides a method and system for positioning the wheel center of a rail vehicle to solve the problem of low wheel center positioning accuracy in the existing technology.

[0006] In a first aspect, the present application provides a method for positioning the wheel center of a rail vehicle, including: Pre-install a first detection unit and a second detection unit on both sides of the wheel; Determine the chord length of the wheel according to the detection process of the first detection unit for the wheel; When the second detection unit detects the wheel, calculate half of the chord length, and control the second detection unit to move half of the chord length in the same direction; When controlling the second detection unit to move half of the chord length in the same direction, determine the current position of the second detection unit as the preliminary positioning position of the wheel center; Based on the collaborative learning framework of the dynamic contact patch deformation characteristics and the geometric degradation of the wheel flange, generate an offset correction amount for the preliminary positioning position of the wheel center, and drive an actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center.

[0007] Optionally, the collaborative learning framework based on the dynamic contact patch deformation characteristics and wheel flange geometric degradation generates an offset correction amount for the preliminary positioning position of the wheel center, including: Obtain the geometric morphology of the wheel surface with different wear degrees through 3D laser scanning, generate a set of static parameters including the thickness of the wheel flange and the wear depth of the tread surface, and establish a spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface based on the thickness of the wheel flange and the wear depth of the tread surface; Distribute a millimeter-wave radar network longitudinally along the track to capture the spatial coordinates of the dynamic contact patch in the wheel-rail contact area in real time, and analyze the deformation characteristics of the dynamic contact patch in combination with the spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface; Input the wheel flange geometric degradation characteristics and the dynamic contact patch deformation characteristics in the static parameter set into the collaborative learning framework, and generate an offset correction amount for the preliminary positioning position of the wheel center through the interaction of local model parameter encryption iteration and global parameter aggregation; Driving the actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center includes: Based on the offset correction amount, combined with the spatial distribution law of the deformation characteristics of the dynamic contact patch, drive the actuator to adjust the preliminary positioning position, and maintain the positioning stability of the wheel-rail contact center during high-speed operation through the real-time matching of the wheel flange geometric characteristics and the contact patch coordinates.

[0008] Optionally, inputting the wheel flange geometric degradation characteristics and the dynamic contact patch deformation characteristics in the static parameter set into the collaborative learning framework, and generating an offset correction amount for the preliminary positioning position of the wheel center through the interaction of local model parameter encryption iteration and global parameter aggregation includes: Decompose the geometric deformation gradient tensor of the wheel flange geometric degradation characteristics, extract the degradation eigenvector related to the curvature of the wheel-rail contact surface in the geometric deformation gradient tensor, and generate a feature decomposition sequence including the curvature degradation gradient; Execute a noise masking mechanism on the feature decomposition sequence at the local computing node, and generate an anti-interference degradation gradient encryption chain by superimposing random masking noise through a dynamic attenuation window; Construct a joint parameter space with the stress distribution time series signal in the dynamic contact patch deformation characteristics for the degradation gradient encryption chain, and generate a global parameter topology chain of degradation and deformation coupling through cross-node parameter interpolation operation in the collaborative learning framework; Dynamically iterate and encrypt the degradation gradient in the global parameter topology chain at the local computing node, fuse the stress distribution in the wheel-rail contact area to generate an encrypted weight gradient chain and distribute it to the global aggregation node; Aggregate the encrypted gradient chains of multiple nodes in the global aggregation node, eliminate the random mask noise, and extract the degradation and deformation coupling phase offset across nodes to generate an offset correction amount for the preliminary positioning position of the wheel center.

[0009] Optionally, constructing a joint parameter space for the degraded gradient encryption chain and the stress distribution time series signal in the dynamic contact patch deformation characteristics, and generating a global parameter topology chain of degradation and deformation coupling through cross-node parameter interpolation operations in the collaborative learning framework, including: Align the sampling timestamps of the spatio-temporal distribution and the stress distribution time series signal in the degraded gradient encryption chain, match the geometric deformation phase of the wheel-rail contact area and the dynamic stress fluctuation period, and generate a parameter matrix of degradation and stress with spatio-temporal alignment; Perform cross-node parameter interpolation operations on the parameter matrix to compensate for the dynamic contact patch deformation gradient differences in adjacent track sections, and generate an interpolation compensation parameter chain with continuous cross-node distribution; Fuse the degradation gradient and the stress distribution amplitude in the interpolation compensation parameter chain, adjust the fusion weight of the gradient and the stress based on the instantaneous deformation rate, and generate a dynamic weight fusion parameter set; Perform spatio-temporal convolution superposition of the degradation gradient and the stress distribution on the dynamic weight fusion parameter set, extract the parameter aggregation characteristics in the wheel-rail contact center area, and generate a feature topology network of degradation and deformation coupling; Analyze the co-oscillation mode of the degradation gradient and the stress distribution in the feature topology network, and generate a global parameter topology chain of degradation and deformation coupling.

[0010] Optionally, the millimeter-wave radar network distributed longitudinally along the track captures the dynamic contact patch spatial coordinates in the wheel-rail contact area in real time, and combines the spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface to analyze the deformation characteristics of the dynamic contact patch, including: Deploy a millimeter-wave radar array longitudinally along the track, synchronously collect the reflected signal pulse sequence in the wheel-rail contact area, capture the multi-path echo time delay and amplitude fluctuation related to the deformation of the track contact surface in the reflected signal, and generate an original signal grid of the dynamic contact patch spatial coordinates; Perform multi-path interference suppression processing on the original signal grid to compensate for the attenuation distortion of the reflected signal during high-speed operation, and generate an anti-interference dynamic contact patch spatial coordinate grid; Extract the pulse response amplitude and the time series phase fluctuation of the dynamic contact patch spatial coordinate grid, and construct a pulse response topology structure of the dynamic contact patch deformation characteristics; Perform spatio-temporal correlation matching between the pulse response topology structure and the spatial geometric mapping relationship of the wheel flange center line, and generate a coupling parameter set of the dynamic contact patch deformation characteristics and the geometric mapping; Analyze the amplitude fluctuation of the impulse response and the spatial offset of the geometric mapping in the set of coupling parameters to generate the deformation characteristics of the dynamic contact patch.

[0011] Optionally, the spatio-temporal correlation matching of the impulse response topological structure and the spatial geometric mapping relationship of the wheel rim center line to generate a set of coupling parameters of the dynamic contact patch deformation characteristics and geometric mapping includes: Calibrate the spatio-temporal coordinate origin of the impulse response topological structure and the geometric mapping spatial reference point of the wheel rim center line, eliminate the coordinate offset of the propagation path of the deformation in the wheel-rail contact area and the geometric degradation parameters, and generate a coordinate set of deformation and geometry correlation with spatio-temporal reference synchronization; Track the diffusion rate of the deformation propagation path in the coordinate set, map the attenuation trend of the spatial geometric degradation gradient of the wheel rim center line with the wear of the track contact surface, and generate a dynamic correlation map of deformation diffusion and geometric degradation; Quantify the phase offset angle between the deformation propagation path and the spatial geometric degradation gradient in the dynamic correlation map, and fuse the instantaneous load intensity of the wheel-rail contact area to calculate the spatial coupling weight, and generate a coupling weight chain of deformation and geometry; Reconstruct the dynamic correlation map in the coupling weight chain, superimpose the multi-dimensional influence factors of the instantaneous load distribution on the wheel-rail contact surface, and generate a multi-dimensional coupling parameter field of deformation propagation and geometric degradation; Analyze the oscillation frequency of the deformation propagation path and the attenuation amplitude of the geometric degradation gradient in the multi-dimensional coupling parameter field to generate a set of coupling parameters of the dynamic contact patch deformation characteristics and geometric mapping.

[0012] Optionally, based on the offset correction amount, combined with the spatial distribution law of the deformation characteristics of the dynamic contact patch, drive the actuator to adjust the preliminary positioning position, and maintain the positioning stability of the wheel-rail contact center during high-speed operation through the real-time matching of the wheel rim geometric characteristics and the contact patch coordinates, including: Generate the dynamic baseline parameters corresponding to the offset correction amount, decompose the high-frequency oscillation component and the low-frequency attenuation trend in the spatial distribution law of the deformation characteristics of the dynamic contact patch, construct a baseline offset field for the positioning of the wheel-rail contact center and send it to the actuator drive nodes of each track section; Map the phase difference between the wheel rim geometric characteristics and the contact patch coordinates in the baseline offset field at the actuator drive node, compensate the time delay effect in the spatial distribution law, and generate a driving signal for adjusting the wheel center coordinate offset; Analyze the instantaneous amplitude fluctuation of the high-frequency oscillation component in the driving signal for adjusting the wheel center coordinate offset, combine the propagation path of the contact patch deformation characteristics, and distribute dynamic weights to the actuator drive nodes to generate multi-node dynamic weight drive parameters; Convert the multi-node dynamic weight driving parameters into a pulse width modulation signal of the actuator, match the phase synchronization of the rim geometric features and the contact patch coordinates, and generate a dynamic control instruction set for the wheel-rail contact center; Execute the dynamic control instruction set, and through the real-time closed-loop feedback of the pressure gradient and contact patch coordinates in the wheel rim contact area by the actuator, adjust the wheel center position and maintain the positioning stability of the wheel-rail contact center.

[0013] Optionally, the determining the chord length of the wheel according to the detection process of the first detection unit for the wheel includes: When the first detection unit detects that the wheel enters the detection area, start the encoder to record the first displacement signal of the wheel movement; When the first detection unit does not detect the wheel, stop the encoder recording and read the second displacement signal recorded by the encoder; Determine the initial displacement value when the wheel edge starts to enter the detection area according to the first displacement signal recorded by the encoder; Determine the final displacement value when the wheel edge completely passes through the first detection unit according to the second displacement signal recorded by the encoder; Determine the chord length of the wheel according to the final displacement value and the initial displacement value.

[0014] Optionally, when the second detection unit detects the wheel, calculate half of the chord length and control the second detection unit to move half of the chord length in the same direction, including: When the second detection unit detects the wheel, use the microprocessor to calculate half of the chord length and drive the second detection unit to move half of the chord length in the same direction through the servo motor.

[0015] In a second aspect, the present application provides a wheel center positioning system for a rail vehicle, including: An installation module for pre-installing a first detection unit and a second detection unit on both sides of the wheel; A measurement module for determining the chord length of the wheel according to the detection process of the first detection unit for the wheel; A calculation module for calculating half of the chord length when the second detection unit detects the wheel and controlling the second detection unit to move half of the chord length in the same direction; A positioning module for determining the current position of the second detection unit as the preliminary positioning position of the wheel center when controlling the second detection unit to move half of the chord length in the same direction; A correction module is used to generate an offset correction amount for the preliminary positioning position of the wheel center based on a collaborative learning framework of dynamic contact patch deformation characteristics and wheel flange geometric degradation, and drive an actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center.

[0016] In the embodiment of the present application, a first detection unit and a second detection unit are pre-installed on both sides of the wheel; according to the detection process of the first detection unit for the wheel, the chord length of the wheel is determined; when the second detection unit detects the wheel, half of the chord length is calculated, and the second detection unit is controlled to move a distance of half of the chord length in the same direction; when the second detection unit is controlled to move a distance of half of the chord length in the same direction, the current position of the second detection unit is determined as the preliminary positioning position of the wheel center; based on a collaborative learning framework of dynamic contact patch deformation characteristics and wheel flange geometric degradation, an offset correction amount for the preliminary positioning position of the wheel center is generated, and an actuator is driven according to the offset correction amount to dynamically correct the preliminary positioning position to maintain the positioning stability of the wheel-rail contact center.

[0017] The technical solution of the present application has the following beneficial effects: Through the collaborative positioning and dynamic correction mechanism of the dual detection units, the present application significantly improves the positioning accuracy and stability of the wheel-rail contact center. Based on chord length measurement and half-distance movement, rapid preliminary positioning of the wheel center is achieved. Combining the collaborative analysis of dynamic contact patch deformation characteristics and wheel flange degradation data, an offset correction amount is generated in real time to drive the actuator for multi-dimensional dynamic compensation, effectively suppressing the accumulation of positioning deviations caused by geometric wear and uneven stress distribution in wheel-rail dynamic contact, reducing the risk of abnormal wear caused by contact stress concentration, enhancing the adaptability of the system to wheel-rail material deformation and dynamic load changes, extending the service life of wheels and tracks, and ensuring the smoothness and safety of train operation.

[0018] Furthermore, by integrating static wheel flange geometric degradation characteristics and dynamic contact patch deformation data, a full-dimensional dynamic compensation mechanism for the wheel-rail contact center is constructed, significantly improving the positioning stability under high-speed operating conditions. Based on three-dimensional laser scanning, accurate geometric mapping of wheel flange thickness and tread wear is generated. Combining with a millimeter-wave radar network to capture the deformation law of the dynamic contact patch in real time, the correlation characteristics between wheel flange degradation and contact stress distribution are analyzed through a collaborative learning framework, driving a multi-axis actuator to achieve dynamic correction compensation in the lateral, vertical, and rotational dimensions, effectively suppressing the offset of the contact center caused by wheel-rail geometric deformation and dynamic load fluctuations, reducing the risk of abnormal wear, enhancing the adaptive matching ability of the wheel-rail contact interface, extending the service life of the track system, and providing technical support for the smoothness and safety of train operation at high speed.

[0019] These aspects or other aspects of the present application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0021] Figure 1 The flowchart of a method for positioning the wheel center of a rail vehicle provided by the present application is shown; Figure 2 The structural schematic diagram of a wheel center positioning system of a rail vehicle provided by the present application is shown. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to enable those skilled in the art of the present technology to better understand the solutions of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application.

[0023] In some processes described in the specification, claims and the above drawings of the present application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0024] The present application aims to break through the defects of traditional wheel set positioning technology that relies on static geometric calibration and cannot adapt to the interference of dynamic wear of the wheel flange and deformation of the contact patch, and solve the problems of low efficiency of manual calibration and the risk of wheel-rail side wear and derailment caused by the drift of the positioning reference. The present application constructs a wheel center dynamic correction and contact mechanics collaborative control system by integrating multi-source detection data and wheel-rail coupling dynamic characteristics, aiming to achieve the autonomous perception, real-time correction and stability maintenance of the wheel contact center under complex working conditions, and improve the monitoring accuracy of the wheel-rail system throughout the life cycle and the train operation safety redundancy.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0026] Figure 1 The following is a flowchart of a method for positioning the wheel center of a rail vehicle provided by an embodiment of the present application. As Figure 1 shown, the method includes: 101. Pre-install a first detection unit and a second detection unit on both sides of the wheel; In this step, the first detection unit refers to a laser range finder installed on the outer side of the wheel for capturing the geometric characteristics of the wheel flange.

[0027] The second detection unit refers to a capacitive displacement sensor arranged below the tread of the wheel for assisting in positioning.

[0028] In the embodiments of the present application, the deployment of the detection unit is realized through the combined configuration of the laser range finder and the capacitive displacement sensor. The first detection unit is installed on the outer side of the wheel flange (with a spacing of 15 mm), and the second detection unit is arranged below the tread of the wheel (with an elevation angle of 30°). The CAN bus is used to synchronize the sampling timings of the two units (frequency 1 kHz). Based on the geometric symmetry of the wheel set, the wheel contour point cloud data is extracted through adaptive threshold segmentation (Otsu algorithm). The highest point of the wheel flange and the lowest point of the tread are detected by combining the Hough transform, and the installation tilt compensation parameters of the two sensors are dynamically calibrated (compensation accuracy ±0.02 mm) to ensure that the double detection units are in the wheel lathe workshop of the depot of the head-end car. First, the wheel lathe operator starts the Hegenscheidt CNC non-drop wheel lathe and installs the standard template wheel set (diameter 860 mm, inside distance 1353 mm) to the processing position. The template wheel is repeatedly measured three times with a laser wheel diameter gauge to ensure that the wheel diameter error ≤ 0.05 mm, and the fluctuation value of the inside distance is controlled within the range of ±0.2 mm. During the calibration process, the operator synchronously calibrates the axial positioning accuracy of the machine tool, detects the straightness of the guide rail through a micrometer (error < 0.01 mm / m), and verifies the feed resolution of the cutting tool holder (0.001 mm / pulse). After the calibration is completed, the system generates a "Device Status Confirmation Form" to provide a reference coordinate system for subsequent wheel turning.

[0029] 102. Determine the chord length of the wheel according to the detection process of the first detection unit for the wheel; In this step, the chord length refers to the maximum straight-line distance of the local arc segment of the wheel tread measured by the first detection unit.

[0030] The detection process refers to the process of fitting the geometric features of the wheel flange based on the sliding window method and the least squares method.

[0031] In the embodiment of the present application, based on the wheel flange contact point sequence (sampling interval 0.1 mm) collected by the first detection unit, the sliding window method (window width 50 mm) is used to extract the continuous curvature feature of the wheel tread, and the least squares method is used to fit the local arc segment to generate the initial value of the chord length (fitting residual < 0.05 mm). Combining the prior parameter of the wheel diameter (standard value 840 - 920 mm) to constrain the effective range of the chord length, and using the Kalman filter to eliminate the lateral vibration noise of the wheel set (vibration amplitude ±3 mm), finally the dynamic chord length measurement value (update frequency 200 Hz) is output, and its accuracy is optimized to ±0.1 mm through cubic polynomial interpolation, providing a geometric reference for the subsequent half-chord movement.

[0032] After receiving the wheel set of the CRH5A type multiple unit (cumulative operating mileage 128,000 km), the wheel lathe operator uses a portable wheel flange profiler to scan the tread, and measures the wheel flange thickness of 28.5 mm (close to the scrapping threshold of 28 mm) and the equivalent taper of 0.18 mm (exceeding the standard by 0.05 mm). By comparing the wheel diameter attenuation curve recorded by the TCMS system, it is found that the diameter of the No. 3 wheel set has worn from 860 mm to 854.3 mm. The operator inputs the "progressive cutting" mode into the numerical control system, and automatically generates the turning parameters: the target diameter is 853 mm (leaving a safety margin of 0.3 mm), the single-sided cutting amount is 0.35 mm, and the Q value is set to 0.12. The system synchronously calculates the tool path compensation amount to eliminate the radial error of 0.07 mm caused by the wheel set installation eccentricity.

[0033] 103. When the second detection unit detects the wheel, calculate half of the chord length, and control the second detection unit to move half of the chord length distance in the same direction; In this step, half of the chord length refers to the displacement amount calculated according to the chord length measurement value for adjusting the position of the second detection unit.

[0034] The same direction refers to the longitudinal movement path of the track that is consistent with the movement direction of the wheel.

[0035] In the embodiment of the present application, when the second detection unit triggers the wheel flange contact signal, a timestamp alignment mechanism is adopted to synchronize the data streams of the two units, and the target displacement of the half chord length is calculated by dividing the instantaneous value of the chord length by two (resolution 0.01 mm). The linear motor (repeated positioning accuracy ±5 μm) is driven to drive the second detection unit to move longitudinally along the track. PID closed-loop control (proportional coefficient Kp = 0.8) is used to compensate the trajectory deviation caused by track unevenness in real time. At the same time, the data of the inertial measurement unit (IMU) is fused to eliminate mechanical vibration interference (acceleration threshold 0.3 g), ensuring that the position error during the movement of the half chord length is stable within ±0.15 mm, and the target pose adjustment of the detection unit is completed.

[0036] After starting the turning program, the cemented carbide tool cuts into the wheel flange at a speed of 650 r / min, with a feed per revolution of 0.02 mm, and the cutting depth is fine-tuned in real time through the hydraulic servo system. The operator observes the temperature of the cutting area through the infrared monitoring screen (the peak value reaches 320 °C). When the thickness of the blue iron chips exceeds 0.1 mm, the high-pressure air blowing device (pressure 0.6 MPa) is immediately triggered to break the continuous strip-shaped iron chips into chips with a length <5 cm. Wear cut-resistant gloves and goggles throughout the operation, and clean the chip collection tank every 15 minutes to prevent overheating alarms caused by the accumulation of high-temperature iron chips. After 2 hours and 36 minutes of processing, the wheel pair diameter is restored to 853.1 mm, and the surface roughness Ra ≤ 1.6 μm.

[0037] 104. When controlling the second detection unit to move half of the chord length in the same direction, determine the current position of the second detection unit as the preliminary positioning position of the wheel center; In this step, the current position refers to the absolute space coordinates after the second detection unit completes the movement of the half chord length.

[0038] The preliminary positioning position refers to the spatial reference point of the wheel center generated based on the data fusion of the dual detection units.

[0039] In the embodiment of the present application, after the second detection unit completes the displacement of the half chord length, its absolute position coordinates are read through a high-precision grating scale (resolution 0.001 mm), and spatial coordinate conversion is performed in combination with the initial installation matrix (4×4 homogeneous transformation matrix) of the first detection unit. The position feedback data of the two units is fused by the weighted average method (weight ratio 6:4) to eliminate the non-linear error of the sensor. At the same time, the thermal expansion compensation parameters of the wheel pair (temperature coefficient × / °C) are loaded to correct the influence of environmental temperature drift, and finally the preliminary positioning coordinates of the wheel center are generated (the radius of the three-dimensional error ellipse ≤ 0.25 mm), and the initial spatial reference system of the wheel-rail contact center is established.

[0040] After the repair is completed, the wheel operator uses a three-point contact wheel diameter gauge to remeasure the wheelset diameter (the average of three measurements is 853.09mm, and the standard deviation is 0.03mm), and uses a rim template ruler to detect the contour fit to ensure that the gap with the standard R100 arc surface is less than 0.05mm. The wheel-rail contact simulation software verifies that the equivalent taper is reduced to 0.03mm, and the lateral force fluctuation value is optimized from 14.6kN before correction to 8.2kN. All data is entered into the "Wheelset Repair Record Sheet" and cross-checked with the stored values of the on-board TCMS system to confirm that the wheel diameter parameter error rate is less than 0.05%. After discovering that the inner distance deviation of wheelset No. 4 was +0.3mm, the secondary calibration procedure was immediately started.

[0041] 105. Based on the collaborative learning framework of dynamic contact spot deformation characteristics and wheel rim geometric degradation, an offset correction value of the initial positioning position of the wheel center is generated, and the actuator is driven to dynamically correct the initial positioning position according to the offset correction value to maintain the positioning stability of the wheel-rail contact center.

[0042] In this step, the dynamic contact patch deformation characteristics refer to the geometric changes in the wheel-rail contact area caused by the force during operation.

[0043] Wheel rim geometry degradation refers to the reduction in geometric dimensions of the wheel rim due to wear during long-term operation.

[0044] The offset correction refers to the compensation value calculated by the collaborative learning framework for adjusting the wheel center positioning.

[0045] The actuator refers to the linear servo motor used to achieve dynamic adjustment of the wheel center position.

[0046] Positioning stability refers to the ability of the wheel-rail contact center to maintain position error within the allowable range during dynamic correction.

[0047] In the embodiment of the present application, a time series correlation model of contact spot deformation (sampling rate 500Hz) and wheel rim wear (3D scanning accuracy 0.02mm) is constructed based on LSTM neural network, and dynamic contact force (six-dimensional force sensor data) and wheel diameter degradation parameters (monthly average change) are input, and the synergistic features of contact spot area fluctuation (standard deviation threshold 0.8mm²) and wheel rim thickness loss are extracted through the attention mechanism. Random forest regression is used to generate position correction (output resolution 0.01mm), drive the linear servo motor (response time 10ms) to perform sub-micron position compensation, and at the same time constrain the correction amplitude (maximum correction amount ±1.5mm) through the Lyapunov stability criterion, so as to achieve the stability of the wheel-rail contact center positioning error within ±0.3mm, and ensure the dynamic balance of the train running posture.

[0048] Upload the corrected data to the vehicle health management system. Combining with the wheel pair material fatigue model (based on the Archard wear equation), predict that the remaining service life of this wheel pair is increased from 70,000 km to 110,000 km. The system automatically generates an "Evaluation Report on the Re-profiling Effect", showing that the wheel-rail contact stress is reduced by 18% and the vibration energy is reduced by 23 dB. At the same time, optimize the next re-profiling cycle to 90,000 km, and it is recommended to adjust the running route of this train number to reduce the frequency of passing through small-radius curves with a radius R < 600 m. Finally, install an RFID electronic tag on the wheel pair to achieve full-life-cycle data traceability.

[0049] In summary, steps 101 to 105 achieve high-precision dynamic self-compensation positioning of the wheel-rail contact center. Through the collaborative measurement of the dual detection units and the chord length adaptive movement mechanism, combined with the real-time fusion analysis of the contact patch deformation characteristics and the wheel flange degradation parameters, a full-closed-loop correction model for the wheel center position is constructed. Based on the dynamic contact mechanics feedback and geometric feature decoupling algorithm, the system automatically eliminates the positioning deviations caused by wheel flange wear, tread peeling, and wheel-rail lateral vibration, forming a stable tracking ability of the wheel-rail contact center against mechanical deformation interference, improving the robustness and positioning repeatability of the dynamic detection of the wheel pair of rail transit vehicles, and providing a real-time dynamic benchmark for optimizing the wheel-rail relationship.

[0050] In order to break through the defects of traditional wheel-rail contact center positioning technology that relies on off-line calibration and cannot adapt to the dynamic wear of the wheel flange and the random deformation of the contact patch, and to solve the positioning drift and derailment risks caused by the lag of static parameter update and inaccurate dynamic deformation modeling, this technology constructs a wheel-rail geometry degradation compensation system based on collaborative learning by integrating multi-dimensional wear characteristics and real-time contact mechanics data, aiming to achieve adaptive perception, dynamic correction, and stability closed-loop control of the wheel center position, and provide high-precision wheel-rail relationship optimization support for the rail transit system under all working conditions and the full cycle.

[0051] In some embodiments, as described in step 105, based on the collaborative learning framework of the dynamic contact patch deformation characteristics and the wheel flange geometry degradation, generate the offset correction amount of the preliminary positioning position of the wheel center, including: 201. Obtain the geometric morphology of the wheel surface with different wear degrees through three-dimensional laser scanning, generate a set of static parameters including the thickness of the wheel flange and the tread wear depth, and establish a spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface based on the thickness of the wheel flange and the tread wear depth; In step 201, three-dimensional laser scanning refers to the technology of using a laser beam to collect high-precision three-dimensional point clouds on the wheel surface.

[0052] The thickness of the wheel flange refers to the vertical distance from the highest point of the wheel flange part to the base surface.

[0053] The tread wear depth refers to the amount of surface material loss of the wheel tread due to wear.

[0054] The static parameter set refers to a fixed data set of the wheel geometric characteristics obtained through measurement.

[0055] The spatial geometric mapping relationship refers to the spatial position and angle correlation model between the center line of the wheel flange and the track contact surface.

[0056] In the embodiments of the present application, a high-precision three-dimensional laser scanner (resolution 0.02 mm) is used to collect the full circumferential point cloud of the wheel surface. The wheel geometric data in different wear stages is fused through a multi-view registration algorithm (such as ICP iterative closest point) to generate a static parameter set of the wheel flange thickness (measurement error ±0.1 mm) and the tread wear depth (quantization accuracy ±0.05 mm). Based on the non-uniform rational B-spline (NURBS) surface reconstruction technology, a three-dimensional wheel model is established. Combining the inclination angle of the track contact surface (in the range of 5° - 15°) and the standard parameters of the wheel-rail clearance, the center line of the wheel flange is mapped to the track contact surface coordinate system through a spatial geometric projection algorithm to generate a spatial mapping relationship matrix including the geometric degradation of the wheel flange and the track adaptability, providing a static benchmark for dynamic contact analysis.

[0057] 202. A millimeter-wave radar network distributed longitudinally along the track, which captures the spatial coordinates of the dynamic contact patch in the wheel-rail contact area in real time, and analyzes the deformation characteristics of the dynamic contact patch in combination with the spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface; In step 202, the millimeter-wave radar network refers to an array of high-frequency radar devices arranged along the track for capturing dynamic contact patches.

[0058] The spatial coordinates of the dynamic contact patch refer to the spatial position data of the wheel-rail contact area that changes with time during operation.

[0059] The deformation characteristics refer to the geometric shape change law of the contact patch after being stressed.

[0060] In the embodiments of the present application, millimeter-wave radar nodes (operating frequency 77 GHz) are deployed every 10 meters along the track, and the three-dimensional coordinates of the wheel-rail contact patch are captured in real time through MIMO array technology (refresh rate 100 Hz). A multi-target tracking algorithm (such as JPDA joint probabilistic data association) is used to separate the wheel-rail contact patch from other interference signals. Combining the wheel flange center line mapping relationship in step 201, the deformation amount of the contact patch (such as the change in ellipticity ±0.3 mm) is calculated through a curvature matching algorithm. Time series analysis (ARIMA model) is introduced to extract the area fluctuation (standard deviation threshold 1.2 mm²) and position drift trend (speed 0.5 mm / s) of the contact patch, and a deformation characteristic spectrum of the dynamic contact patch is constructed to realize the real-time state perception of wheel-rail contact mechanics.

[0061] 203. Input the rim geometry degradation characteristics in the static parameter set and the dynamic contact patch deformation characteristics into the collaborative learning framework, and generate the offset correction amount of the preliminary positioning position of the wheel center through the interaction of local model parameter encryption iteration and global parameter aggregation; In step 203, the collaborative learning framework refers to a distributed learning architecture that realizes model training through multi-node data interaction.

[0062] Local model parameter encryption iteration refers to the process of encrypting and updating model parameters and optimizing them at the local node.

[0063] Global parameter aggregation refers to the operation of fusing and unifying the model parameters of multiple local nodes.

[0064] The offset correction amount refers to the compensation value used to adjust the position of the wheel center.

[0065] In the embodiment of the present application, based on the federated learning framework, the local node uses a lightweight convolutional network (MobileNet-V3) to extract features of the rim thickness degradation rate (0.05 - 0.2 mm per month on average) and the contact patch deformation rate (0.1 - 0.8 mm / s), and uses the homomorphic encryption technology to securely transmit the model gradient parameters (dimension 256). The global server fuses the features of multiple nodes through the adaptive weighted aggregation algorithm (weights are assigned according to the confidence of node data), constructs a time series prediction model using the gated recurrent unit (GRU), and outputs the lateral offset correction amount of the wheel center (resolution 0.01 mm) and the longitudinal compensation amount (step size 0.05 mm), forming a distributed collaborative correction strategy against the data island effect.

[0066] 204. Dynamically correct the preliminary positioning position by driving the actuator according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center, including: In step 204, the actuator refers to a mechanical device used to realize the dynamic adjustment of the wheel center position.

[0067] Positioning stability refers to the ability of the wheel-rail contact center to keep the position error within the allowable range during the dynamic correction process.

[0068] In the embodiment of the present application, the actuator uses a linear voice coil motor (thrust 120N, response time 5ms) to receive the offset correction instruction, restricts the correction amplitude (maximum ±2mm) through the Lyapunov stability criterion, combines the thermogram of the spatial distribution of the contact patch deformation (Gaussian kernel radius 50mm), and dynamically adjusts the PID control parameters (proportional coefficient 0.6 - 1.2). In the wheel-rail contact force feedback loop, the data of the six-dimensional force sensor (sampling rate 1kHz) and the wheel flange geometry scanning result are integrated, and the real-time matching of the actuator displacement (accuracy ±0.03mm) and the contact patch coordinates is achieved through Kalman filtering to ensure that the correction process is synchronized with the wheel-rail dynamics state.

[0069] 205. Based on the offset correction amount, combined with the spatial distribution law of the deformation characteristics of the dynamic contact patch, drive the actuator to adjust the preliminary positioning position, and maintain the positioning stability of the wheel-rail contact center during high-speed operation through the real-time matching of the wheel flange geometric features and the contact patch coordinates.

[0070] In step 205, the spatial distribution law refers to the change trend and pattern of the deformation characteristics of the dynamic contact patch in space.

[0071] The wheel flange geometric features refer to the geometric shape and dimensional parameters of the wheel flange.

[0072] The real-time matching of the contact patch coordinates refers to the process of dynamically aligning the spatial position of the contact patch with the wheel flange geometric features.

[0073] In the embodiment of the present application, based on the principal component analysis of the spatial distribution of the contact patch deformation (cumulative variance contribution rate > 85%), extract the deformation direction eigenvector (such as the axial proportion is 60%), and drive the actuator to perform dynamic compensation along the track lateral-vertical composite motion trajectory (interpolation period 0.1ms). Through the real-time registration of the wheel flange geometric features (such as the wheel flange angle 35° - 42°) and the contact patch coordinates (matching error < 0.1mm), combined with the wheel pair hunting frequency (1 - 3Hz), adaptively adjust the correction frequency, and control the offset of the wheel-rail contact center within ±0.25mm at a running speed of 380km / h to achieve closed-loop control of the positioning stability under high-speed conditions.

[0074] The following is a specific example: In the scenario of wheel-rail contact center positioning for heavy-haul freight trains, the system uses a high-frame-rate 3D laser scanner (resolution 0.05 mm) to perform a full circumferential scan of wheels in long-term service. The phase unwrapping algorithm is used to reconstruct the flange thickness (measurement range 30 - 40 mm) and the tread wear gradient (accuracy ±0.08 mm), and a wheel-rail spatial geometric mapping matrix is established in combination with the curvature radius of the track contact surface (standard value 300 - 350 mm). 77 GHz millimeter-wave radar nodes are arranged every 15 meters along the freight dedicated line, and the Doppler frequency shift compensation technology is used to eliminate the influence of the low-speed swaying of the train (amplitude ±5 mm) on the contact patch detection. The contact patch area (reference value 200 - 300 mm²) and the position offset (detection error ±0.3 mm) of the time series are fused through the Bayesian filtering algorithm. The federated learning framework based on blockchain technology encrypts and trains the flange degradation rates (average daily 0.02 - 0.05 mm) and the contact patch fluctuation characteristics (frequency 0.5 - 2 Hz) of 20 groups of freight formation vehicles, and the gated attention mechanism is used to screen key features to generate offset correction parameters (compensation step 0.03 - 0.1 mm). The six-degree-of-freedom parallel mechanism (repeated positioning accuracy ±0.02 mm) is driven to perform dynamic compensation. Combining the real-time matching of the contact patch heat map (grid density 5 mm × 5 mm) and the flange angle (35° - 45°), the offset of the wheel-rail contact center is stabilized within the threshold of ±0.4 mm under the heavy-haul condition of 80 km / h, effectively solving the hunting instability problem caused by the flange wear of freight trains.

[0075] To sum up, steps 201 to 205 achieve the full-life-cycle dynamic self-healing positioning of the wheel-rail contact center. By fusing the static wear parameters of 3D laser scanning and the dynamic contact patch deformation characteristics captured by millimeter-wave radar, a wheel flange geometric degradation compensation model with multi-modal data collaboration is constructed. The system realizes the encrypted training of local wear data and the distributed aggregation of global dynamic parameters based on the federated learning framework, generates a positioning correction strategy resistant to mechanical deformation interference, and combines the spatial distribution law of the contact patch and the real-time matching mechanism of the wheel flange geometric characteristics to break through the limitation of the lag of traditional static detection, realizes sub-millimeter-level dynamic compensation of the wheel center position under high-speed conditions, effectively inhibits the phenomena of wheel-rail wear and hunting instability, and improves the train operation safety and the wheel set operation and maintenance efficiency.

[0076] In order to overcome the difficulties in decoupling geometric degradation characteristics and dynamic stress deformation in the wheel-rail contact center positioning technology and the problem of correction lag caused by multi-node data noise interference, and to solve the positioning drift risk caused by the traditional method relying on single-dimensional features and the lack of cross-node data security, this technology constructs a global topology analysis model of degradation-deformation coupling by fusing gradient tensor decomposition and anti-noise encryption mechanism, aiming to achieve high-robustness dynamic correction of the wheel center position and cross-node collaborative optimization, and improve the collaborative control accuracy and system stability of wheel-rail contact mechanics in complex scenarios such as heavy-haul and high-speed.

[0077] In some embodiments, as described in step 203, the rim geometry degradation features in the static parameter set and the dynamic contact patch deformation features are input into a collaborative learning framework, and an offset correction amount of the preliminary positioning position of the wheel center is generated through an interactive manner of local model parameter encryption iteration and global parameter aggregation, including: 301. Decompose the geometric deformation gradient tensor of the rim geometry degradation features, extract the degradation feature vectors related to the curvature of the wheel-rail contact surface in the geometric deformation gradient tensor, and generate a feature decomposition sequence including the curvature degradation gradient; In step 301, the geometric deformation gradient tensor refers to a mathematical structure that describes the change rate and direction of the rim geometry degradation features in space.

[0078] The curvature degradation gradient refers to the geometric deformation change rate related to the curvature of the wheel-rail contact surface.

[0079] The feature decomposition sequence refers to a set of feature vectors including the curvature degradation gradient obtained by decomposing the geometric deformation gradient tensor.

[0080] In the embodiments of the present application, a tensor decomposition algorithm is used to perform dimensionality reduction processing on the rim geometry degradation features, and the feature vectors related to the curvature of the wheel-rail contact surface are extracted through principal component analysis to generate a feature decomposition sequence including the curvature degradation gradient. Combining the rim thickness degradation rate (0.02 - 0.05 mm per day on average) and the tread wear depth (quantization accuracy ±0.08 mm), the singular value decomposition technique is used to separate the curvature-sensitive components in the geometric deformation gradient tensor, and a correlation matrix between the degradation features and the curvature of the track contact surface (standard value 300 - 350 mm) is constructed to provide core feature data for subsequent noise masking and cross-node fusion.

[0081] 302. Execute a noise masking mechanism on the feature decomposition sequence at the local computing node, and generate an anti-interference degradation gradient encryption chain by superimposing random masking noise through a dynamic attenuation window; In step 302, the noise masking mechanism refers to a technique for enhancing data security and anti-interference ability by superimposing random noise.

[0082] The dynamic attenuation window refers to a data processing window that adjusts the noise amplitude over time.

[0083] The degradation gradient encryption chain refers to an anti-interference degradation gradient data sequence after noise masking processing.

[0084] In the embodiment of the present application, at the local computing node, a dynamic decay window (window width 50 ms) is used to perform noise masking processing on the feature decomposition sequence. Masking noise with a mean of zero (standard deviation 0.1) is generated through a Gaussian random function and superimposed on the degraded gradient sequence. The adaptive threshold segmentation algorithm (OTSU) is used to screen out effective feature components, and the homomorphic encryption technology is combined to encrypt the gradient sequence after noise masking, generating a degradation-resistant gradient encryption chain to ensure the security and noise resistance during data transmission.

[0085] 303. Construct a joint parameter space with the stress distribution time series signal in the dynamic contact patch deformation feature for the degradation gradient encryption chain, and generate a global parameter topology chain coupling degradation and deformation through cross-node parameter interpolation operations in the collaborative learning framework. In step 303, the joint parameter space refers to a multi-dimensional data space constructed by fusing the degradation gradient encryption chain and the dynamic contact patch stress distribution signal.

[0086] Cross-node parameter interpolation operation refers to the operation of data interpolation and fusion between different computing nodes.

[0087] The global parameter topology chain refers to a global data chain coupling degradation and deformation generated through cross-node fusion.

[0088] In the embodiment of the present application, based on the degradation gradient encryption chain and the dynamic contact patch stress distribution time series signal (sampling rate 1 kHz), the kernel function mapping technology is used to construct a joint parameter space. Through cross-node parameter interpolation operations (such as bicubic spline interpolation) in the collaborative learning framework, the local degradation characteristics and the global stress distribution law are fused to generate a global parameter topology chain coupling degradation and deformation. Combining the wheel-rail contact force feedback data (six-dimensional force sensor), the phase alignment accuracy of the degradation gradient and the stress distribution in the topology chain is optimized, providing a basis for subsequent dynamic iterative encryption.

[0089] 304. Dynamically iteratively encrypt the degradation gradient in the global parameter topology chain at the local computing node, fuse the stress distribution in the wheel-rail contact area to generate an encrypted weight gradient chain, and distribute it to the global aggregation node. In step 304, dynamic iterative encryption refers to the process of performing multiple encryption processes on the degradation gradient data to enhance security.

[0090] The encrypted weight gradient chain refers to a gradient data sequence with encrypted weights generated after fusing the stress distribution data.

[0091] In the embodiments of the present application, at the local computing node, a dynamic iterative encryption algorithm is used to encrypt the degraded gradients in the global parameter topology chain, and an encrypted weight gradient chain is generated through the stress distribution data (quantization accuracy ±0.05 MPa) in the wheel-rail contact area. The distributed hash table (DHT) technology is used to distribute the encrypted gradient chain to the global aggregation node. Combining the wheel flange geometric degradation rate (0.2 - 0.5 mm per month) and the contact patch deformation frequency (0.5 - 2 Hz), the encrypted weight allocation strategy is optimized to ensure the accuracy and efficiency of cross-node data fusion.

[0092] 305. Aggregate the encrypted gradient chains of multiple nodes in the global aggregation node, eliminate the random mask noise, and extract the cross-node degradation and deformation coupling phase offset to generate an offset correction amount for the preliminary positioning position of the wheel center.

[0093] In step 305, the cross-node degradation and deformation coupling phase offset refers to the phase difference between the degradation and deformation characteristics extracted by aggregating the data of multiple nodes.

[0094] The offset correction amount refers to the compensation value used to adjust the position of the wheel center.

[0095] In the embodiments of the present application, at the global aggregation node, a weighted average algorithm is used to aggregate the encrypted gradient chains of multiple nodes, and the cross-node degradation and deformation coupling phase offset (resolution 0.01 mm) is extracted through Fourier transform. A low-pass filter is used to eliminate the random mask noise (cutoff frequency 10 Hz), and an offset correction amount is generated in combination with the wheel-rail contact center positioning error (threshold ±0.3 mm). Based on the Lyapunov stability criterion, the correction amplitude is constrained (maximum ±1.5 mm) to achieve high-precision dynamic compensation and stability control of the wheel center position.

[0096] The following is a specific example: In the scenario of high-speed rail wheel center positioning, after a certain train set has been in long-term operation, non-uniform wear appears on the wheel flange. First, the geometric data of the wheel flange cross-section is obtained through a three-dimensional laser scanner, and a deformation gradient tensor matrix is constructed based on the curvature differential algorithm to extract the contact surface curvature change rate exceeding The feature vector of the wheel-rail dynamic contact patch is used to generate a degraded gradient sequence containing different wear phases (0°, 120°, 240°). The local edge computing node uses an adaptive Hamming window to dynamically suppress the noise of the gradient sequence, and superimposes quantum random noise (amplitude controlled at ±5μm) that meets the NIST standard to form an encryption chain. In the cross-node collaboration stage, the encryption chain is spatially and temporally aligned with the finite element stress cloud map (sampling rate 1kHz) of the wheel-rail dynamic contact patch, and a joint feature space containing 12-dimensional deformation parameters is constructed through bicubic spline interpolation. After three rounds of federated learning iterations, the global aggregation node uses an improved K-SVD algorithm to eliminate multi-node noise interference, and finally outputs the wheel center positioning correction based on the phase offset analysis of the contact patch stress extreme point (maximum stress 380MPa). In actual tests on a CR400BF vehicle model, this method reduced the lateral positioning error of the wheelset from 2.3mm to 0.7mm, effectively eliminating the harmonic component of the snaking motion caused by wheel rim wear (amplitude reduced by 63%), and meeting the sub-millimeter positioning accuracy requirements under conditions of 350km / h.

[0097] In summary, steps 301 to 305 realize the multi-dimensional anti-noise collaborative positioning compensation of the wheel-rail contact center, and construct a coupled analysis system of degradation characteristics and contact spot deformation through joint modeling of geometric deformation gradient tensor decomposition and dynamic stress distribution signal. The system generates an anti-interference encrypted gradient chain based on the noise mask mechanism, and uses cross-node parameter interpolation operation to fuse local degradation characteristics and global deformation laws, breaking through the characteristic deviation limitations of traditional single-node learning, and realizing phase synchronization analysis of wheel rim degradation gradient and contact stress, forming a high-precision correction amount generation capability that is resistant to data noise and node heterogeneity, ensuring the dynamic stability and anti-interference of wheel center positioning under complex working conditions, and providing a holographic compensation benchmark for wheel-rail dynamic adaptation.

[0098] In order to overcome the problems of compensation lag and deviation accumulation caused by asynchrony of spatiotemporal data and isolation of local parameters in wheel-rail contact center compensation technology, and to solve the problem of positioning inaccuracy caused by the inability of traditional methods to coordinate degradation gradients and dynamic stress fluctuations and cross-regional data faults, this technology integrates spatiotemporal alignment and multi-node collaborative optimization mechanism to construct a degradation-deformation global topological analysis framework, aiming to achieve real-time phase matching and dynamic weight adaptive compensation of the wheel-rail contact center, and provide a highly robust wheel-rail collaborative control solution for extreme working conditions such as high speed and heavy load.

[0099] In some embodiments, as described in step 303, the degraded gradient encryption chain and the stress distribution time series signal in the dynamic contact spot deformation feature are used to construct a joint parameter space, and a global parameter topology chain of degradation and deformation coupling is generated through the cross-node parameter interpolation operation in the collaborative learning framework, including: 401. Align the sampling timestamps of the spatio-temporal distribution and the stress distribution time series signals in the degraded gradient encryption chain, match the geometric deformation phase of the wheel-rail contact area with the dynamic stress fluctuation period, and generate a parameter matrix of the spatio-temporal alignment of degradation and stress; In step 401, the degraded gradient encryption chain refers to an anti-interference degraded gradient data sequence after noise masking processing.

[0100] The stress distribution time series signal refers to the stress fluctuation data of the wheel-rail contact area changing with time.

[0101] The geometric deformation phase refers to the periodic change characteristics of the geometric deformation of the wheel-rail contact area in time.

[0102] The dynamic stress fluctuation period refers to the periodic change law of the stress distribution signal in time.

[0103] The parameter matrix refers to a data matrix generated after spatio-temporal alignment of the degraded gradient and the stress distribution.

[0104] In the embodiment of the present application, the sampling time series of the degraded gradient encryption chain and the stress distribution signal are matched through a time series alignment algorithm (such as dynamic time warping), and the geometric deformation phase (accuracy ±0.5°) and the dynamic stress fluctuation period (frequency 0.1 - 5 Hz) of the wheel-rail contact area are synchronized by using a phase-locked loop technology. Based on the wheel flange degradation rate (0.02 mm per day on average) and the stress amplitude change rate (gradient 0.3 MPa / s), a spatio-temporal synchronization compensation matrix (dimension 128×128) is constructed, and the null data caused by the timestamp deviation is filled by cubic spline interpolation (interpolation error <0.1%). Finally, a parameter matrix containing the spatio-temporal correlation of the degraded gradient and the stress distribution is generated, providing a reference data source for cross-node interpolation.

[0105] 402. Perform parameter interpolation operations between multiple nodes on the parameter matrix, compensate for the differences in the dynamic contact patch deformation gradients of adjacent track sections, and generate an interpolation compensation parameter chain with continuous distribution across nodes; In step 402, the parameter interpolation operation between multiple nodes refers to the operation of data interpolation and fusion between different computing nodes.

[0106] The dynamic contact patch deformation gradient difference refers to the degree of difference in the contact patch deformation characteristics of adjacent track sections.

[0107] The interpolation compensation parameter chain refers to a continuously distributed parameter sequence generated by cross-node interpolation.

[0108] In the embodiments of the present application, the spatial Kriging interpolation algorithm is adopted among multiple nodes. According to the track section length (standard 25m) and the dynamic contact patch deformation gradient difference (threshold ±0.5mm / m), the adjacent node parameter faults are compensated. The spatial continuity of the contact patch deformation gradient (range 1 - 8mm / m) is adjusted through the adaptive interpolation step size (0.5 - 2m), and the interpolation parameter distribution is optimized by combining the node confidence weights (0.6 - 0.9), generating a continuous interpolation compensation chain covering the entire track (resolution 0.1mm), eliminating the compensation lag problem caused by local data islands, and forming a global spatial correlation model of the degradation gradient and deformation characteristics.

[0109] 403. Integrate the degradation gradient and the stress distribution amplitude in the interpolation compensation parameter chain, adjust the fusion weight of the gradient and the stress based on the instantaneous deformation rate, and generate a dynamic weight fusion parameter set; In step 403, the degradation gradient refers to the change rate of the geometric degradation characteristics of the wheel flange.

[0110] The stress distribution amplitude refers to the intensity value of the stress fluctuation of the contact patch.

[0111] The instantaneous deformation rate refers to the instantaneous change speed of the geometric deformation in the wheel-rail contact area.

[0112] The dynamic weight fusion parameter set refers to the fusion parameter set generated after adjusting the weight according to the instantaneous deformation rate.

[0113] In the embodiments of the present application, based on the instantaneous deformation rate (quantization accuracy ±0.05mm / s), the fusion weight of the degradation gradient and the stress amplitude is dynamically adjusted, and the exponential decay function (time constant 50ms) is used to allocate the gradient weight (0.3 - 0.7) and the stress weight (0.7 - 0.3). The contact patch stress peak value (threshold 2MPa) and the degradation gradient extreme value (threshold 0.1mm / m) are extracted through a sliding window (width 100ms), and the fuzzy logic rule base (32 rules) is used to optimize the weight allocation strategy, generating a dynamic weight fusion parameter set (update frequency 50Hz) to achieve real-time adaptive fusion of the degradation and deformation characteristics.

[0114] 404. Perform spatio-temporal convolution superposition of the degradation gradient and the stress distribution on the dynamic weight fusion parameter set, extract the parameter aggregation characteristics in the wheel-rail contact center area, and generate a feature topology network coupling degradation and deformation; In step 404, the spatio-temporal convolution superposition refers to the process of performing convolution operations on the degradation gradient and the stress distribution in the time and space dimensions.

[0115] The parameter aggregation characteristics refer to the key characteristics in the wheel-rail contact center area extracted through convolution superposition.

[0116] The feature topology network refers to the relationship network among features coupling degradation and deformation.

[0117] In the embodiment of the present application, a three-dimensional spatio-temporal convolution kernel (size 3×3×5) is used to extract features from the dynamic weight fusion parameter set, and the dilated convolution (dilation rate 2) is used to enhance the long-range dependence relationship between the degradation gradient and the stress distribution. Combining with the max pooling layer (stride 2) to compress the redundant feature dimension, and using the residual connection to retain the original parameter distribution characteristics, finally a feature topology network coupling degradation and deformation (the number of nodes is 512) is generated, and its topological structure strengthens the parameter aggregation ability of the contact center region through the graph attention mechanism (the number of heads is 8), forming a global feature expression for wheel-rail dynamic adaptation.

[0118] 405. Analyze the cooperative oscillation mode between the degradation gradient and the stress distribution in the feature topology network, and generate a global parameter topology chain coupling degradation and deformation.

[0119] In step 405, the cooperative oscillation mode refers to the law of synchronous change between the degradation gradient and the stress distribution.

[0120] The global parameter topology chain refers to the global parameter correlation chain generated by analyzing the cooperative oscillation mode.

[0121] In the embodiment of the present application, based on the Fourier transform, the fundamental frequency oscillation mode (main frequency 0.5 - 2 Hz) between the degradation gradient and the stress distribution in the feature topology network is extracted, and the cooperative oscillation components are screened through the coherence analysis (threshold 0.85). The spectral clustering algorithm (the number of clusters is 5) is used to divide the phase synchronization region, and combined with the Lyapunov exponent (threshold 0.2) to evaluate the oscillation stability, finally a global parameter topology chain coupling degradation and deformation (dimension 256) is generated, and its topological relationship is optimized by the node betweenness centrality (threshold 0.4), providing a full-dimensional parameter correlation benchmark for the dynamic compensation of the wheel-rail contact center.

[0122] To sum up, steps 401 to 405 achieve the global spatio-temporal synchronous dynamic compensation of the wheel-rail contact center. Through the spatio-temporal phase alignment of the degradation gradient and the stress distribution and multi-node interpolation compensation, a parameter fusion system with cross-region continuous distribution is constructed. The system is based on the dynamic weight adjustment mechanism and the spatio-temporal convolution superposition algorithm, analyzes the cooperative oscillation law of degradation and deformation, breaks through the spatio-temporal fragmentation limitation of traditional local compensation, forms a global topological correlation model of the degradation gradient and the stress fluctuation, realizes the high-precision anti-disturbance correction of the contact center position, effectively suppresses the phase mismatch and stress concentration problems of the wheel-rail dynamic contact, and improves the dynamic adaptability and service life of the wheel-rail system under complex working conditions.

[0123] The following is a specific example: In the dynamic compensation scenario of the suspension gap of an urban maglev train, the system uses a submillimeter-wave three-dimensional profiler (wavelength 0.3 mm) to scan the surface topography of the electromagnet pole shoe in real time. By using the dynamic time warping algorithm to align the suspension force fluctuation signal (frequency band 50 - 500 Hz) with the pole shoe wear gradient data (daily change 0.005 - 0.015 mm), a spatio-temporal correlation matrix (dimension 256×256) is constructed. 60 GHz phased array radar nodes (spacing 8 m) are deployed along the suspension track, and the adaptive Kriging interpolation algorithm is used to compensate for the suspension gap gradient difference between adjacent sections (threshold ±0.08 mm / m), generating a continuous gap compensation chain across nodes. Based on the suspension electromagnetic field intensity (0.8 - 1.5 T) and the current ripple characteristics (harmonic distortion rate < 5%), the fusion weight (0.4 - 0.6) of the wear gradient and electromagnetic stress is dynamically adjusted by a fuzzy logic controller. A three-dimensional spatio-temporal convolution kernel (size 5×5×7) is used to extract the magnetic-force coupling characteristics of the suspension center region, constructing a feature topology network containing 512 nodes. Wavelet coherence analysis is used to analyze the phase synchronization mode of the suspension force pulsation (main frequency 120 Hz) and the pole shoe wear gradient (weekly change 0.1 mm). The dynamic compensation domain is divided by spectral clustering, driving a multi-stage voice coil motor array (response time 2 ms) to implement nanoscale suspension gap compensation (accuracy ±0.02 mm). Under the operating condition of 430 km / h, the suspension center offset is controlled within ±0.15 mm, effectively suppressing the suspension force oscillation instability problem caused by pole shoe uneven wear.

[0124] In some embodiments, as described in step 202, the millimeter-wave radar network longitudinally distributed along the track captures the dynamic contact patch spatial coordinates in the wheel-rail contact area in real time. Combining the spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface, the deformation characteristics of the dynamic contact patch are analyzed, including: 501. Deploy a millimeter-wave radar array longitudinally along the track, synchronously collect the reflected signal pulse sequence in the wheel-rail contact area, capture the multipath echo time delay and amplitude fluctuation related to the deformation of the track contact surface in the reflected signal, and generate the original signal grid of the dynamic contact patch spatial coordinates; In step 501, the millimeter-wave radar array refers to a combination of high-frequency radar devices arranged longitudinally along the track for capturing the reflected signals in the wheel-rail contact area.

[0125] The reflected signal pulse sequence refers to the data stream of the reflected signals in the wheel-rail contact area collected by the millimeter-wave radar.

[0126] The multipath echo time delay refers to the time delay generated by the reflected signal due to different paths.

[0127] The amplitude fluctuation refers to the change in the intensity of the reflected signal during propagation.

[0128] The original signal grid refers to the initial data matrix of the spatial coordinates of the dynamic contact patches generated by the reflected signals.

[0129] In the embodiments of the present application, a 77 GHz millimeter-wave radar array (spacing 2 m) is longitudinally deployed along the track, and the reflected pulse signals (bandwidth 4 GHz) in the wheel-rail contact area are synchronously collected through MIMO technology. Based on the pulse compression algorithm, the multi-path echo time delay (resolution 0.1 ns) is extracted, and the effective echoes related to the deformation of the contact surface are separated by combining adaptive threshold segmentation (Otsu algorithm). The original spatial coordinate grid (grid density 5 mm × 5 mm) is constructed by using cubic spline interpolation. Through the joint analysis of the echo amplitude fluctuation (dynamic range 60 dB) and the time delay difference (maximum ±3 ns), the original signal grid containing the three-dimensional coordinates of the contact patch, the reflection intensity, and the deformation correlation degree is generated, providing a data basis for multi-path suppression.

[0130] 502. Perform multi-path interference suppression processing on the original signal grid, compensate for the attenuation distortion of the reflected signals during high-speed operation, and generate an anti-interference spatial coordinate grid of the dynamic contact patches; In step 502, the multi-path interference suppression processing refers to the process of eliminating the signal interference caused by the multi-path effect through an algorithm.

[0131] The attenuation distortion refers to the signal distortion caused by attenuation during the propagation of the reflected signals.

[0132] The anti-interference spatial coordinate grid of the dynamic contact patches refers to the spatial coordinate data matrix after multi-path suppression processing.

[0133] In the embodiments of the present application, an adaptive filtering algorithm (LMS filter order 32) is adopted for the original signal grid, and the multi-path interference suppression coefficient is calculated in real time in combination with the train running speed (200 - 350 km / h). The signal attenuation distortion is compensated by Kalman filtering (attenuation coefficient 0.05 - 0.2 dB / m), and the real spatial distribution of the dynamic contact patches is predicted by using a backpropagation neural network (64 hidden layer nodes), generating an anti-interference coordinate grid (positioning error ±0.3 mm). The confidence weights of the grid nodes are optimized by fusing the data of the pulse arrival angle (accuracy ±0.5°) and the Doppler frequency shift (±5 kHz), and finally a dynamic contact patch spatial grid with a time synchronization error <1 ms is output.

[0134] 503. Extract the pulse response amplitude and the time sequence phase fluctuation of the spatial coordinate grid of the dynamic contact patches, and construct a pulse response topological structure of the dynamic contact patch deformation characteristics; In step 503, the pulse response amplitude refers to the intensity value of the reflected signal pulse.

[0135] The time sequence phase fluctuation refers to the phase change of the reflected signal pulse in time.

[0136] The impulse response topology refers to a deformation feature relationship network constructed based on the amplitude and phase fluctuations of the impulse response.

[0137] In the embodiments of the present application, the impulse response envelope is extracted based on the Hilbert transform, the range of amplitude fluctuations (threshold 0.5 - 2V) is statistically calculated using a sliding window (width 50ms), and the temporal phase jump is eliminated by combining the phase unwrapping algorithm (jump threshold π / 4). The impulse response topology is constructed through a graph convolutional network (convolution kernel size 3×3), where the node represents the deformation intensity of the contact patch (0 - 100%), and the edge weight is associated with the phase synchronization of adjacent regions (coherence > 0.8). The impulse repetition frequency (10kHz) and the moving speed of the contact patch (0.1 - 5m / s) are fused to generate a dynamically updated topological network, realizing the spatial correlation modeling of deformation features.

[0138] 504. Perform spatio-temporal correlation matching on the spatial geometric mapping relationship between the impulse response topology and the wheel rim center line to generate a set of coupling parameters for the dynamic contact patch deformation features and geometric mapping; In step 504, the spatial geometric mapping relationship of the wheel rim center line refers to the spatial position and angle correlation model between the wheel rim center line and the track contact surface.

[0139] Spatio-temporal correlation matching refers to the process of aligning the impulse response topology and the geometric mapping relationship in time and space.

[0140] The set of coupling parameters refers to the data set containing degradation and deformation features generated through spatio-temporal correlation matching.

[0141] In the embodiments of the present application, the spatio-temporal registration algorithm (ICP iterative closest point) is used to map the coordinates of the impulse response topology nodes to the wheel rim center line coordinate system (registration error ±0.2mm), and the geometric mapping weight (0.3 - 0.7) is calculated in combination with the wheel-rail contact force data (six-dimensional force sensor sampling rate 2kHz). The deformation rate of the topology nodes (0.05 - 0.5mm / s) and the wheel rim geometric degradation gradient (average daily 0.01 - 0.03mm) are aligned through the dynamic time warping (DTW) algorithm to generate a parameter set including spatio-temporal correlation intensity, phase offset (±0.8mm), and coupling confidence (0.6 - 0.95), establishing a quantitative correlation model between the contact patch deformation and the wheel rim degradation.

[0142] 505. Analyze the spatial offset between the impulse response amplitude fluctuation and the geometric mapping in the set of coupling parameters to generate the deformation features of the dynamic contact patch.

[0143] In step 505, the impulse response amplitude fluctuation refers to the change law of the reflection signal pulse intensity.

[0144] The spatial offset of the geometric mapping refers to the spatial position deviation between the center line of the wheel rim and the track contact surface.

[0145] The deformation characteristics of the dynamic contact patch refer to the deformation law and characteristics of the contact patch generated by analyzing the coupled parameter set.

[0146] In the embodiments of the present application, key fluctuation modes in the coupled parameter set are extracted based on principal component analysis (cumulative variance contribution rate > 85%), and a random forest regression model is used to analyze the non-linear relationship between the amplitude fluctuation (variance threshold 0.5V²) and the spatial offset (0.1 - 1.2 mm). Stable oscillation modes are screened through the Lyapunov exponent (threshold 0.25), and the curvature of the contact patch movement trajectory ( ) is fused to generate a deformation characteristic spectrum, and finally a contact patch deformation characteristic set including deformation intensity grading (grades I - V), offset direction (8 azimuth partitions), and dynamic risk index is output, providing a quantitative decision-making basis for wheel-rail maintenance.

[0147] The following is a specific example: In the scenario of monitoring the wheel-rail contact state in a high-speed railway tunnel, the system longitudinally deploys a 24GHz millimeter-wave radar array (spacing 10m) along the tunnel wall, synchronously collects the reflected pulse sequence (bandwidth 2GHz) in the wheel-rail contact area through beamforming technology, and uses time-frequency analysis to extract the multi-path echo delay (resolution 0.2ns) and amplitude fluctuation (dynamic range 50dB), generating an original signal grid containing the three-dimensional coordinates of the contact patch (grid density 10mm × 10mm). An adaptive filtering algorithm (LMS filter order 64) is used to suppress the reflection interference of the tunnel wall, and the signal attenuation distortion (attenuation coefficient 0.1dB / m) is compensated in real time in combination with the train running speed (250 - 350km / h), generating an anti-interference dynamic contact patch spatial coordinate grid (positioning error ±0.5mm). Based on the Hilbert transform, the pulse response envelope is extracted, and a graph convolutional network (convolution kernel size 5×5) is used to construct the pulse response topological structure, and the nodes represent the deformation intensity of the contact patch (0 - 100%). The topological nodes are mapped to the wheel rim center line coordinate system through a spatio-temporal registration algorithm (ICP iterative closest point) (registration error ±0.3mm), and a coupled parameter set is generated in combination with the wheel-rail contact force data (sampling rate 1kHz). Principal component analysis is used to extract key fluctuation modes, and the amplitude fluctuation (variance threshold 0.8V²) and spatial offset (0.2 - 1.5mm) are analyzed through random forest regression. Finally, a contact patch deformation characteristic set including deformation intensity grading (grades I - V) and dynamic risk index is output, providing a quantitative decision-making basis for wheel-rail maintenance in the tunnel.

[0148] In summary, steps 501 to 505 achieve millimeter-wave multimodal dynamic perception and compensation control of the deformation characteristics of the wheel-rail contact patch. By analyzing the multipath echoes of the millimeter-wave radar array and modeling the pulse response topology, the spatio-temporal resolution limitations of traditional single-point detection are overcome. Based on the multipath interference suppression and pulse sequence spatio-temporal correlation algorithm, the system couples the deformation characteristics of the dynamic contact patch with the geometric mapping of the wheel flange, forms a deformation compensation model that resists attenuation and distortion, realizes sub-millimeter real-time analysis of the spatial offset of the contact patch, overcomes the problems of signal distortion and positioning drift caused by multipath effects, and significantly improves the dynamic monitoring accuracy and stability of the high-speed wheel-rail contact state.

[0149] In some embodiments, as described in step 504, the spatio-temporal correlation matching is performed between the pulse response topology structure and the spatial geometric mapping relationship of the wheel flange center line to generate a coupling parameter set of the deformation characteristics of the dynamic contact patch and the geometric mapping, including: 601. Calibrate the spatio-temporal coordinate origin of the pulse response topology structure and the geometric mapping spatial reference point of the wheel flange center line, eliminate the coordinate system offset of the propagation path and geometric degradation parameters of the deformation in the wheel-rail contact area, and generate a coordinate set of deformation and geometry correlation with spatio-temporal reference synchronization; In step 601, the pulse response topology structure refers to a deformation characteristic relationship network constructed based on the amplitude and phase fluctuations of the pulse response.

[0150] The spatio-temporal coordinate origin refers to the time and space reference points of the pulse response topology structure.

[0151] The geometric mapping spatial reference point of the wheel flange center line refers to the spatial position reference point between the wheel flange center line and the track contact surface.

[0152] The deformation propagation path refers to the diffusion trajectory of the deformation in the wheel-rail contact area in time and space.

[0153] The coordinate set of deformation and geometry correlation with spatio-temporal reference synchronization refers to the data set generated after aligning the deformation propagation path and geometric degradation parameters.

[0154] In the embodiments of the present application, the iterative closest point (ICP) algorithm is used to calibrate the spatio-temporal coordinate origin of the pulse response topology structure, and the geometric mapping reference point of the wheel flange center line is calibrated by a laser tracker (accuracy ±0.02 mm). The coordinate system offset of the deformation propagation path is compensated by combining the wheel-rail contact force distribution data (sampling rate 1 kHz), and the spatio-temporal reference of the contact patch deformation and the wheel flange degradation parameters is aligned by using spatio-temporal registration technology (registration error <0.1 mm). Based on the thermal expansion coefficient of the wheel-rail material ( × Compensate for the influence of environmental temperature drift at (°C), and finally generate a spatio-temporal synchronous coordinate set including the deformation propagation rate (0.05 - 0.5 mm / s) and the geometric degradation gradient (0.01 - 0.03 mm per day on average), providing a benchmark data source for dynamic correlation analysis.

[0155] 602. Track the diffusion rate of the deformation propagation path in the coordinate set, map the attenuation trend of the spatial geometric degradation gradient of the wheel rim center line with the wear of the track contact surface, and generate a dynamic correlation map of deformation diffusion and geometric degradation; In step 602, the deformation diffusion rate refers to the diffusion speed of the deformation in the wheel-rail contact area in time and space.

[0156] The spatial geometric degradation gradient refers to the geometric deformation change rate of the wheel rim center line in space.

[0157] The dynamic correlation map refers to the dynamic relationship network between deformation diffusion and geometric degradation.

[0158] In the embodiment of the present application, based on the optical flow method, track the diffusion rate of the deformation propagation path in the coordinate set (quantization accuracy ±0.1 mm / s), combine the time-series data of the wear depth of the track contact surface (0.1 - 0.4 mm per month on average), and fit the degradation gradient attenuation curve by polynomial regression (fitting error < 2%). Use the graph convolutional network (GCN) to construct a dynamic correlation map of deformation diffusion and geometric degradation, where the node represents the deformation intensity of the contact patch (0 - 100%), and the edge weight is associated with the synchronization of the degradation rates in adjacent regions (correlation coefficient > 0.85). Integrate the train running speed (200 - 350 km / h) and the wheel-rail clearance data (standard value 10 - 14 mm) to generate a dynamic correlation map covering the entire track, revealing the quantitative correlation law between deformation propagation and geometric degradation.

[0159] 603. Quantify the phase offset angle between the deformation propagation path and the spatial geometric degradation gradient in the dynamic correlation map, integrate the instantaneous load intensity in the wheel-rail contact area to calculate the spatial coupling weight, and generate a coupling weight chain of deformation and geometry; In step 603, the phase offset angle refers to the phase difference between the deformation propagation path and the geometric degradation gradient.

[0160] The instantaneous load intensity refers to the load magnitude in the wheel-rail contact area at a certain moment.

[0161] The spatial coupling weight refers to the fusion weight of deformation and geometric features calculated according to the instantaneous load intensity.

[0162] The coupling weight chain refers to the data sequence of deformation and geometric features including the spatial coupling weight.

[0163] In the embodiments of the present application, wavelet transform is used to extract the phase offset angle (resolution ±0.5°) between the deformation propagation path and the geometric degradation gradient in the dynamic correlation map, and the spatial coupling weight (0.3 - 0.7) is calculated through the instantaneous load intensity (peak value 20 - 50 kN). Based on the fuzzy logic control rules (48 rules), the weight distribution strategy is dynamically adjusted, and combined with the real-time data of the contact patch stress distribution (mesh density 5 mm × 5 mm), a coupling weight chain of deformation and geometry (update frequency 50 Hz) is generated. Kalman filtering is used to eliminate sensor noise (signal-to-noise ratio > 30 dB) to ensure the temporal continuity and spatial consistency of the weight chain, providing input conditions for the reconstruction of the multi-dimensional parameter field.

[0164] 604. Reconstruct the dynamic correlation map in the coupling weight chain, and superimpose the multi-dimensional influence factors of the instantaneous load distribution on the wheel-rail contact surface to generate a multi-dimensional coupling parameter field of deformation propagation and geometric degradation; In step 604, the multi-dimensional influence factors refer to the multi-dimensional influencing factors of the instantaneous load distribution on the wheel-rail contact surface.

[0165] The multi-dimensional coupling parameter field refers to a data field containing multi-dimensional correlation characteristics of deformation propagation and geometric degradation.

[0166] In the embodiments of the present application, principal component analysis (cumulative variance contribution rate > 90%) is used to reconstruct the dynamic correlation map in the coupling weight chain, and multi-dimensional factors such as the temperature gradient (0 - 50 °C), humidity influence (20 - 95% RH), and vibration spectrum (5 - 200 Hz) of the instantaneous load distribution on the wheel-rail contact surface are superimposed. Tensor decomposition technology (Tucker decomposition rank 3) is used to compress the parameter dimension, and a multi-dimensional coupling parameter field (mesh resolution 1 mm × 1 mm × 0.1 s) of deformation propagation and geometric degradation is generated through a spatio-temporal interpolation algorithm (cubic spline). Combining the Lyapunov exponent (threshold 0.25) to screen the stable parameter region, a global coupling field model resistant to environmental interference is formed.

[0167] 605. Analyze the oscillation frequency of the deformation propagation path and the attenuation amplitude of the geometric degradation gradient in the multi-dimensional coupling parameter field to generate a coupling parameter set of dynamic contact patch deformation characteristics and geometric mapping.

[0168] In step 605, the oscillation frequency refers to the periodic change frequency of the deformation propagation path in time.

[0169] The attenuation amplitude refers to the change amplitude of the geometric degradation gradient in space.

[0170] The coupling parameter set refers to a set of deformation and geometric mapping characteristic data generated by analyzing the multi-dimensional coupling parameter field.

[0171] In the embodiments of the present application, based on Fourier transform, the oscillation frequency (main frequency 0.5 - 3 Hz) of the deformation propagation path in the multi-dimensional coupled parameter field is analyzed, and a random forest regression model is used to quantify the attenuation amplitude (range 0.1 - 1.2 mm) of the geometric degradation gradient. The phase synchronization region of deformation and degradation is extracted through coherence analysis (threshold 0.8), and combined with the real-time matching data of the flange angle (35° - 45°) and the contact patch curvature radius (200 - 300 mm), a coupled parameter set including deformation intensity grading (Level I - V), offset direction (8 azimuth partitions), and risk index is generated. Finally, through adaptive PID control (response time 10 ms), closed-loop compensation of the dynamic contact patch deformation characteristics and geometric mapping is achieved, and the error is stabilized within the threshold of ±0.2 mm.

[0172] The following is a specific example: In the scenario of wheel-rail dynamic monitoring in the throat area of urban rail transit stations, the system longitudinally deploys an 80 GHz millimeter-wave radar array (spacing 5 m) along the turnout section. Combining with the transient three-dimensional structure reconstruction technology of Hefei Zhongke Junda Vision Co., Ltd., a sub-millimeter three-dimensional point cloud (resolution 0.05 mm) of the wheel-rail contact area is generated through high-speed multi-view laser scanning, and the reflection pulse sequence (bandwidth 5 GHz) is synchronously collected to capture the microscopic deformation of the contact patch of the switch point rail. The iterative closest point (ICP) algorithm is used to calibrate the spatial reference of the pulse response topology and the flange center line (registration error ±0.1 mm), and the 360° high-definition image data of the trackside panoramic intelligent detection system is fused to generate a spatio-temporal synchronous coordinate set including the deformation propagation path (speed 0.1 - 0.8 mm / s) and the flange wear gradient (0.02 - 0.05 mm per day) in the turnout area. The deformation diffusion trajectory is tracked by the optical flow method, and the lateral force data (peak value 15 - 30 kN) when the train passes through the turnout is superimposed to construct a deformation-wear dynamic correlation map (update frequency 50 Hz), and the wavelet packet decomposition is used to quantify the phase offset angle (accuracy ±0.3°). Based on tensor decomposition, multi-dimensional factors such as the temperature rise gradient (ΔT 5 - 15 °C) of the contact patch and the humidity penetration coefficient (0.1 - 0.5) are fused to reconstruct the coupled parameter field of deformation propagation and geometric degradation (grid density 2 mm × 2 mm × 0.2 s). Finally, the main oscillation frequency band (0.8 - 2.5 Hz) and the wear attenuation threshold (0.3 - 1.0 mm) are extracted through coherent spectrum analysis, and a coupled parameter set including the safety level (Level A - E) and maintenance priority of the turnout contact patch is generated, providing a decision-making basis for the optimization of the wheel-rail dynamic adaptability in the throat area of the station yard.

[0173] In summary, steps 601 to 605 achieve full-dimensional dynamic perception and collaborative compensation of wheel-rail contact deformation and geometric degradation. Through spatio-temporal reference synchronization calibration and deformation propagation path tracking technology, a dynamic correlation model of deformation diffusion and geometric degradation is constructed. Based on phase shift quantization and multi-dimensional coupled parameter field reconstruction, the system breaks through the accuracy limitation of traditional single-dimensional compensation, forms the oscillation frequency-attenuation amplitude collaborative analysis ability of the deformation propagation path and the wheel flange degradation gradient, realizes the high-precision dynamic matching of the contact patch deformation characteristics and geometric mapping, effectively suppresses the stress concentration and geometric degradation mismatch problems in the wheel-rail contact area, and improves the dynamic adaptability and service reliability of the wheel-rail system.

[0174] In some embodiments, as described in step 105, based on the offset correction amount and in combination with the spatial distribution law of the deformation characteristics of the dynamic contact patch, the actuator is driven to adjust the preliminary positioning position. By real-time matching of the wheel flange geometric characteristics and the contact patch coordinates, the positioning stability of the wheel-rail contact center during high-speed operation is maintained, including: 701. Generate dynamic baseline parameters corresponding to the offset correction amount, decompose the high-frequency oscillation component and the low-frequency attenuation trend in the spatial distribution law of the deformation characteristics of the dynamic contact patch, and construct a baseline offset field for wheel-rail contact center positioning and send it to the actuator drive nodes in each track section; In step 701, the dynamic baseline parameter refers to the reference parameter for dynamic compensation generated according to the offset correction amount.

[0175] The high-frequency oscillation component refers to the rapidly changing periodic component in the deformation characteristics of the dynamic contact patch.

[0176] The low-frequency attenuation trend refers to the slowly changing trend component in the deformation characteristics of the dynamic contact patch.

[0177] The baseline offset field refers to the wheel-rail contact center positioning reference data field covering the entire track.

[0178] In the embodiments of the present application, the wavelet packet decomposition algorithm is used to separate the high-frequency oscillation component (frequency band 50 - 200 Hz) and the low-frequency attenuation trend (frequency band 0 - 5 Hz) in the deformation characteristics of the dynamic contact patch, and the spatial interpolation algorithm is used to construct the baseline offset field for wheel-rail contact center positioning (grid accuracy 1 mm × 1 mm). Based on the train running speed (200 - 350 km / h) and the dynamic data of the wheel-rail clearance (range 8 - 14 mm), a baseline parameter field covering the entire track is generated, and the parameter field is distributed to the linear motor drive nodes in each track section using the timestamp synchronization protocol (such as IEEE 1588) (communication delay < 1 ms) to provide a reference spatial distribution model for subsequent dynamic compensation.

[0179] 702. The actuator drive node maps the phase difference between the rim geometric features and the contact patch coordinates in the baseline offset field, compensates for the time delay effect in the spatial distribution law, and generates a driving signal for adjusting the wheel center coordinate offset; In step 702, the rim geometric features refer to the geometric shape and dimensional parameters of the wheel rim.

[0180] The contact patch coordinates refer to the spatial position data of the wheel-rail contact area.

[0181] The phase difference refers to the phase deviation between the rim geometric features and the contact patch coordinates.

[0182] The time delay effect refers to the time delay phenomenon during the deformation propagation of the wheel-rail contact area.

[0183] The driving signal for adjusting the wheel center coordinate offset refers to the compensation control signal for adjusting the wheel center position.

[0184] In the embodiment of the present application, a phase detection module (accuracy ±0.5°) is deployed at the actuator drive node. The phase difference (range 0 - 30°) between the rim geometric features and the contact patch coordinates is calculated through the cross-correlation algorithm. Combining the deformation propagation speed of the contact patch (0.1 - 0.8 mm / s), the time delay effect compensation amount (compensation accuracy ±0.02 ms) is predicted by the Kalman filter. The driving signal for adjusting the wheel center coordinate offset (resolution 0.01 mm) is generated by fusing the six-dimensional force sensor data (sampling rate 2 kHz). The signal amplitude overshoot is corrected in real time through the PID control algorithm (proportional coefficient Kp = 0.6) to ensure that the driving signal strictly matches the dynamic state of the wheel-rail.

[0185] 703. Analyze the instantaneous amplitude fluctuation of the high-frequency oscillation component in the driving signal for adjusting the wheel center coordinate offset. Combining the propagation path of the contact patch deformation characteristics, dynamic weights are assigned to the actuator drive nodes to generate multi-node dynamic weight drive parameters; In step 703, the instantaneous amplitude fluctuation refers to the intensity change of the high-frequency oscillation component at a certain moment.

[0186] The propagation path refers to the diffusion trajectory of the contact patch deformation in time and space.

[0187] The dynamic weight refers to the compensation weight of the actuator assigned according to the contact patch deformation characteristics.

[0188] The multi-node dynamic weight drive parameters refer to the data set containing multiple node weight compensation parameters.

[0189] In the embodiment of the present application, the Hilbert-Huang transform is used to analyze the instantaneous amplitude of the high-frequency oscillation component in the driving signal (fluctuation range 0.5 - 3 V). Combining the curvature change rate of the contact patch deformation propagation path ( ), dynamically allocate the weights (0.3 - 0.7) of the execution nodes through a fuzzy logic controller (with 64 rules in the rule base). Optimize the weight allocation strategy based on the thermo-mechanical stress distribution heat map of the contact patch (mesh density 5mm × 5mm), and generate a multi-node dynamic weight driving parameter set including node priorities (levels 1 - 5) and response speeds (10 - 50ms) to achieve the adaptive optimal configuration of the compensated resources.

[0190] 704. Convert the multi-node dynamic weight driving parameters into pulse width modulation signals of the actuator, match the phase synchronization of the rim geometric features and the contact patch coordinates, and generate a dynamic control instruction set for the wheel-rail contact center; In step 704, the pulse width modulation signal refers to the signal that controls the actuator by adjusting the pulse width.

[0191] The phase synchronization refers to the consistency in phase between the rim geometric features and the contact patch coordinates.

[0192] The dynamic control instruction set refers to the set of control instructions used to dynamically adjust the wheel-rail contact center.

[0193] In the embodiment of the present application, convert the dynamic weight parameters into a PWM signal with adjustable duty cycle (adjustment accuracy ±0.5%) through a pulse width modulator (carrier frequency 20kHz), and synchronize the phase of the rim geometric features and the contact patch coordinates using the phase-locked loop technology (synchronization error < 0.1°). Combine the Lyapunov stability criterion to constrain the amplitude of the PWM signal (threshold ±5V), generate a dynamic control instruction set including lateral compensation amount (±2mm) and vertical compensation amount (±1mm), and send it to the actuator through the EtherCAT bus (cycle 125μs) to ensure that the instruction timing is strictly synchronized with the wheel-rail contact state.

[0194] 705. Execute the dynamic control instruction set, and adjust the wheel center position and maintain the positioning stability of the wheel-rail contact center through the real-time closed-loop feedback of the pressure gradient in the rim contact area and the contact patch coordinates by the actuator.

[0195] In step 705, the pressure gradient refers to the spatial change rate of the pressure in the rim contact area.

[0196] The real-time closed-loop feedback refers to the control process of real-time monitoring and adjusting the wheel center position through sensors.

[0197] The positioning stability refers to the ability of the wheel-rail contact center to maintain position accuracy during the dynamic adjustment process.

[0198] In the embodiments of the present application, the actuator uses a voice coil motor array (thrust 120N, repeat positioning accuracy ±0.01mm) to receive dynamic control instructions, and the pressure sensor array (range 0 - 50MPa) is used to real-time feedback the pressure gradient in the wheel rim contact area (gradient resolution ±0.05MPa / mm). Combining the laser tracking data of the contact patch coordinates (sampling rate 1kHz), a closed-loop control loop is constructed. Model predictive control (prediction step 10ms) is used to dynamically adjust the wheel center position, so that the contact center offset is stabilized within the range of ±0.15mm. Finally, the stress equilibrium algorithm (equilibrium error <5%) is used to maintain the dynamic stability of the wheel-rail contact.

[0199] The following is a specific example: In the wheel-rail dynamic adaptation control scenario of the high-speed railway curved track section, the system is based on the contact patch laser Doppler vibrometer (LDV) to real-time collect the sub-micron vibration spectrum (frequency band 0.5 - 800Hz) of the wheel-rail contact surface, and combines the wheel set three-dimensional profile scanner (accuracy ±5μm) to construct a wheel rim geometric feature database. The high-frequency vibration component (main frequency 50 - 200Hz) and the low-frequency creep trend (period 0.2 - 2s) of the contact patch deformation are separated by the improved HHT transform. The dynamic detection data of the track inspection vehicle (sampling interval 10cm) is fused by using the Kriging spatial interpolation algorithm to generate a dynamic baseline parameter field including the lateral offset baseline (±0.3 - 1.2mm) and the longitudinal creep gradient (0.05 - 0.15mm / m) of the curve section. The millimeter-wave radar phase interference array (resolution 0.1°) is deployed at the actuator node, and the improved cross-correlation algorithm is used to real-time match the spatio-temporal difference (time delay compensation amount ±0.8ms) between the wheel rim profile phase and the contact patch coordinates. Combining the six-dimensional sensor data of the wheel-rail contact force (range ±50kN), an adaptive PID compensation signal (adjustment period 5ms) is generated. Based on the propagation vector field of the contact patch stress nephogram (density 1mm×1mm), the compensation weights (0.4 - 0.9) of 16 groups of actuators are dynamically allocated by using the fuzzy neural network, and the weight parameters are converted into pulse width modulation instructions (carrier frequency 25kHz) through the carrier reconstruction technology. Finally, relying on the distributed piezoelectric actuator array (response time 0.2ms), the contact patch stress equilibrium control is implemented, and the pressure gradient (accuracy ±0.01MPa / mm) is real-time feedback by combining the fiber Bragg grating sensor, so that the wheel-rail contact center offset in the curve section is stabilized within the range of ±0.1mm, effectively suppressing the wheel rim wear and snake instability.

[0200] In summary, steps 701 to 705 achieve the global dynamic closed-loop control and self-adaptive compensation of the wheel-rail contact center. Through the construction of a dynamic baseline parameter field and the separation technology of high-frequency oscillation components, the phase lag bottleneck of traditional static compensation is broken through. Based on the phase difference mapping between the flange geometric characteristics and the contact patch coordinates, combined with a multi-node dynamic weight allocation mechanism, the system forms a collaborative control ability to decouple oscillation suppression and attenuation trends, realizing millisecond-level dynamic adjustment of the wheel center position. By the phase synchronization matching of the pressure gradient feedback and the pulse modulation signal, the uneven stress distribution and the accumulation of dynamic offsets in the wheel-rail contact area are eliminated, and finally the cross-scale collaborative optimization of the contact center positioning accuracy and stability is achieved.

[0201] In order to overcome the problem that it is difficult to accurately detect the wheel geometric parameters of moving vehicles such as trains and cars in dynamic working conditions, and to solve problems such as low efficiency of traditional manual measurement and susceptibility to environmental interference in contact detection, this technology constructs a closed-loop detection system for the wheel movement trajectory and geometric dimensions by integrating encoder displacement signals and edge trigger logic, aiming to achieve non-contact dynamic measurement of the wheel chord length and self-adaptive compensation for abnormal working conditions, providing reliable data support for wheel set turning and safety warning, reducing operation and maintenance costs and improving the running safety of vehicles.

[0202] In some embodiments, as described in step 102, determining the chord length of the wheel according to the detection process of the first detection unit for the wheel includes: 801. When the first detection unit detects that the wheel enters the detection area, start the encoder to record the first displacement signal of the wheel movement; In step 801, the first detection unit refers to an infrared or photoelectric sensor used to detect the wheel entering the detection area.

[0203] The encoder refers to a measuring device that converts the linear displacement of the wheel into a pulse signal.

[0204] The first displacement signal refers to the displacement pulse signal recorded by the encoder when the wheel enters the detection area.

[0205] In the embodiment of the present application, a photoelectric sensor or an infrared array is used to detect the entry of a wheel into the detection area, triggering the rotary encoder to start pulse counting. The encoder adopts an incremental design, and converts the linear displacement of the wheel into an angular displacement pulse signal through a rack and pinion or a magnetic roller. The first detection unit uses an infrared transmissive sensor (such as the combination of SE303A and PH302) to monitor the wheel flange occlusion state in real time. When the wheel edge first blocks the optical path, a trigger signal is sent to the encoder. The encoder is built-in with a high-speed counter (such as the TIM module of STM32) to start recording the pulse count with a microsecond-level response. The initial pulse value is obtained through the input capture function of a PLC or a single-chip microcomputer (such as S7-200 SMART). Finally, through photoelectric signal synchronization and debounce algorithms, the initial displacement value at the moment when the wheel enters is determined, with an accuracy of ±0.1 mm.

[0206] 802. When the first detection unit does not detect the wheel, stop the encoder from recording, and read the second displacement signal recorded by the encoder; In step 802, the second displacement signal refers to the displacement pulse signal recorded by the encoder when the wheel completely passes through the detection area.

[0207] In the embodiment of the present application, when the wheel completely passes through the detection area, the infrared sensor optical path resumes conduction, triggering the encoder to stop counting. At this time, the encoder transmits the accumulated pulse count to the controller (such as the SST89E564RD single-chip microcomputer) through the SPI or EtherCAT communication protocol. The system uses anti-interference technology (such as digital filtering) to eliminate false triggers caused by mechanical vibration, and ensures the continuity of the displacement signal through timestamp calibration (accuracy ±0.5 ms). The final displacement value is calculated through a pulse-displacement conversion coefficient (such as 0.02 mm per pulse), and the sliding error is corrected by combining the wheel diameter dynamic compensation algorithm (based on laser ranging data) to generate a second displacement signal dataset.

[0208] 803. Determine the initial displacement value when the wheel edge starts to enter the detection area according to the first displacement signal recorded by the encoder; In step 803, the initial displacement value refers to the displacement reference value when the wheel edge starts to enter the detection area.

[0209] In the embodiment of the present application, the calculation of the initial displacement value is based on the relationship between the encoder pulse count and the mechanical transmission ratio. For example, for an encoder with a rack and pinion drive, each revolution corresponds to a linear displacement of 10 mm, then the initial pulse count is divided by the pulse count per revolution (such as 2000 PPR) to obtain the initial displacement. The system removes signal noise through wavelet transform and uses an edge detection algorithm (such as the Canny operator) to accurately calibrate the moment when the wheel enters. Combining the three-dimensional scanning data of the wheel set (such as the tread profile library), the influence of the geometric deviation of the wheel flange on the initial position is dynamically corrected, and finally an initial displacement value with a sub-millimeter-level accuracy is output.

[0210] 804. Determine the final displacement value when the wheel edge completely passes through the first detection unit according to the second displacement signal recorded by the encoder; In step 804, the final displacement value refers to the displacement termination value when the wheel edge completely passes through the detection area.

[0211] In the embodiment of the present application, the determination of the final displacement value needs to compensate for the encoder return error and temperature drift. The system uses a dual-channel calibration technology (A / B phase pulse phase difference analysis) to detect the rotation direction of the encoder to prevent reverse counting. The Kalman filter is used to fuse multi-sensor data (such as ultrasonic ranging) to calibrate the pulse accumulation value in real time. For example, when the wheel completely passes through, the number of pulses recorded by the encoder is converted into displacement through a linear interpolation algorithm, and the geometric correction amount of the laser sensor (such as L2 + 70mm reference point) is superimposed, and finally a displacement termination value with an accuracy of ±0.05mm is generated.

[0212] 805. Determine the chord length of the wheel according to the final displacement value and the initial displacement value.

[0213] In step 805, the chord length refers to the straight-line distance of the wheel tread calculated according to the final displacement value and the initial displacement value.

[0214] In the embodiment of the present application, the chord length calculation dynamically compensates for the wheel-rail contact deformation through the displacement difference. The system subtracts the final displacement value from the initial displacement value, combines the geometric parameters of the wheel rolling circle reference point (such as at the tread L2 + 70mm), and uses the least squares method to fit the true value of the chord length. For the curve condition, a curvature radius compensation coefficient is introduced (such as compensating 0.2% when R > 300m), and parallel operation is realized through FPGA to ensure real-time performance at a 1kHz update rate. The final output result is verified by the stress balance algorithm, so that the chord length measurement error is stabilized within ±0.3mm, meeting the wheel set turning repair accuracy requirements.

[0215] The following is a specific example: In the scenario of detecting the geometric parameters of the wheels of rail vehicles, a certain subway depot uses a chord length measurement system based on an incremental encoder to monitor the wear of the wheel set. Specifically, when the wheel (with a diameter of 840mm) enters the detection area composed of two laser pair sensors, the 2000-line incremental encoder is triggered to start recording the displacement pulse signal. The system collects the initial displacement value of the wheel edge contacting the first laser beam in real time with a detection accuracy of 0.4mm = 125.6mm. When the wheel completely passes through the second laser sensor with a spacing of 500mm, the encoder automatically stops recording and reads the final displacement value = 681.2 mm. The displacement difference ΔX = 555.6 mm is obtained through differential operation. Combining with the fixed spacing L = 500 mm of the double laser sensors, the actual chord length value is calculated as 832.7 mm using the geometric relation chord length C = √(4ΔX² - L²). In the actual measurement of the vehicle depot on Shenzhen Metro Line 6, by comparing with the reference data of the 3D laser scanner, it is found that the chord length of a certain train wheel is shortened by 3.5 mm due to long-term operation (exceeding the 2 mm maintenance threshold), triggering an automatic alarm and generating a wheel flange turning work order. This solution uses an anti-slip frame with a counterweight of 100 kg to ensure a constant contact pressure between the coding wheel and the rail surface. Combining with the chord length reference calibration once a week (using a calibration wheel with a standard diameter of 845 mm), the measurement error is controlled within the design index of ±0.8 mm, effectively solving the problem of cumulative error caused by wheel slip in traditional wheel diameter measurement.

[0216] In summary, steps 801 to 805 achieve the full-process automation and dynamic error compensation of wheel chord length detection. Through the timing control and edge trigger mechanism of the encoder displacement signal, the technical bottlenecks of fuzzy positioning and response lag in traditional manual measurement are broken through. The system calculates the chord length based on the displacement difference of the wheel entering and leaving the detection area, and combines with the self-calibration algorithm of the encoder signal to eliminate the cumulative error caused by mechanical vibration and speed fluctuation, realizing the sub-millimeter-level real-time dynamic detection of wheel geometric parameters. Through the precise calibration and compensation of the start / end points of the displacement signal, the measurement deviation caused by wheel tilt or deflection is synchronously solved, providing a stable data source for wheel diameter calculation and tread wear analysis.

[0217] In order to overcome the problem of positioning inaccuracy in wheel-rail contact center detection caused by wheel geometric deviation and dynamic stress fluctuation, and solve the technical bottleneck that traditional methods rely on manual calibration and cannot adapt to complex working conditions, this technology constructs an adaptive dynamic positioning system for the wheel-rail contact center by integrating chord length half-value calculation and servo drive control, aiming to achieve real-time precise detection and dynamic compensation of the wheel center position, and provide an efficient and reliable solution for wheel-rail maintenance and safe operation.

[0218] In some embodiments, as described in step 103, when the second detection unit detects the wheel, calculate half of the chord length, and control the second detection unit to move half of the chord length distance in the same direction, including: 901. When the second detection unit detects the wheel, use a microprocessor to calculate half of the chord length, and drive the second detection unit to move half of the chord length distance in the same direction through a servo motor.

[0219] In step 901, the second detection unit refers to an infrared or photoelectric sensor used to detect the wheel entering the target area.

[0220] The microprocessor refers to an embedded computing chip used for performing the calculation of the half value of the chord length.

[0221] Half of the chord length refers to one - half of the straight - line distance of the wheel tread.

[0222] The servo motor refers to a closed - loop control motor used to drive the precise movement of the second detection unit.

[0223] In the embodiments of this application, the photoelectric sensor or laser array is used to detect the entry of the wheel into the second detection area in real - time, triggering the microprocessor (such as ARM Cortex - M4) to start the calculation of the half value of the chord length. The microprocessor loads the pre - stored chord length data (such as 600 mm), uses the floating - point arithmetic unit (FPU) to quickly calculate the half value (300 mm), and transmits the calculation result to the servo - motor controller (such as Delta ASDA - A2 series) through the SPI or CAN bus. The servo motor adopts closed - loop feedback control (encoder resolution 10000 PPR), combines the PID adjustment algorithm (proportional coefficient Kp = 0.8) to dynamically adjust the moving speed and acceleration, ensuring that the second detection unit accurately moves to the target position (error ±0.1 mm). The system eliminates the displacement deviation caused by mechanical vibration and track unevenness through real - time position monitoring and dynamic compensation mechanism, and finally realizes the precise positioning of the wheel center position.

[0224] The following is a specific example: In the dynamic detection scenario of the wheel set of a mine heavy - duty truck, the system deploys a laser array along the truck running track as the second detection unit. When the wheel enters the detection area, it triggers the STM32F407 microprocessor to start the calculation of the half value of the chord length. The microprocessor loads the pre - stored chord length data (such as 850 mm), quickly calculates the half value (425 mm) through the floating - point arithmetic unit, and transmits the result to the Yaskawa servo - motor controller (SGD7S series) through the CAN bus. The servo motor adopts a 20 - bit high - resolution encoder (accuracy ±0.01 mm) and the PID closed - loop control algorithm (proportional coefficient Kp = 0.7), driving the second detection unit to accurately move a half - value distance (425 mm) along the track direction. The system uses a laser ranging sensor (accuracy ±0.05 mm) to monitor the moving position in real - time, combines the Kalman filter to eliminate the displacement deviation caused by track vibration and mechanical clearance (compensation amount ±0.2 mm). Finally, the second detection unit is positioned at the wheel center position (error ±0.1 mm), providing high - precision reference data for the wheel set turning repair and safety monitoring of the mine truck, and significantly improving the dynamic adaptability and running stability of the wheel - rail system.

[0225] In summary, step 901 realizes the automatic dynamic positioning and compensation of the wheel-rail contact center. By calculating the half value of the chord length in real time with a microprocessor and controlling the servo motor to move precisely, it breaks through the limitations of response lag and positioning deviation in traditional static detection. The system dynamically adjusts the position of the second detection unit based on the half value of the chord length, combined with the closed-loop feedback control of the servo motor, to eliminate the uneven stress distribution and geometric offset in the wheel-rail contact area and achieve sub-millimeter precise positioning of the wheel center position. Through the coordinated control of chord length half-value compensation and servo drive, the stability and reliability of wheel-rail contact center detection are significantly improved, providing high-precision data support for wheel-rail dynamic adaptation.

[0226] Figure 2 The following is a schematic structural diagram of a wheel center positioning system for a rail vehicle provided by an embodiment of the present application, as Figure 2 shown. The system includes: An installation module 21 for pre-installing a first detection unit and a second detection unit on both sides of the wheel; A measurement module 22 for determining the chord length of the wheel according to the detection process of the wheel by the first detection unit; A calculation module 23 for calculating half of the chord length when the second detection unit detects the wheel and controlling the second detection unit to move half of the chord length in the same direction; A positioning module 24 for determining the current position of the second detection unit as the preliminary positioning position of the wheel center when controlling the second detection unit to move half of the chord length in the same direction; A correction module 25 for generating an offset correction amount for the preliminary positioning position of the wheel center based on a collaborative learning framework of dynamic contact patch deformation characteristics and wheel flange geometric degradation, and driving an actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center.

[0227] Figure 2 The wheel center positioning system for a rail vehicle described above can execute Figure 1 the wheel center positioning method for a rail vehicle described in the embodiment shown. Its implementation principle and technical effects will not be elaborated again. For the wheel center positioning system for a rail vehicle in the above embodiment, the specific ways for each module and unit to perform operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0228] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A method for positioning the wheel center of a rail vehicle, characterized in that, Including: Pre - install a first detection unit and a second detection unit on both sides of the wheel; Determine the chord length of the wheel according to the detection process of the first detection unit for the wheel; When the second detection unit detects the wheel, calculate half of the chord length, and control the second detection unit to move half of the chord length in the same direction; When controlling the second detection unit to move half of the chord length in the same direction, determine the current position of the second detection unit as the preliminary positioning position of the wheel center; Based on the collaborative learning framework of dynamic contact patch deformation characteristics and wheel - rim geometric degradation, generate an offset correction amount for the preliminary positioning position of the wheel center, and drive an actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel - rail contact center.

2. The method according to claim 1, wherein The generating of the offset correction amount for the preliminary positioning position of the wheel center based on the collaborative learning framework of dynamic contact patch deformation characteristics and wheel - rim geometric degradation includes: Obtain the geometric morphology of the wheel surface with different wear degrees through three - dimensional laser scanning, generate a set of static parameters including the thickness of the wheel - rim and the wear depth of the tread surface, and establish a spatial geometric mapping relationship between the center line of the wheel - rim and the track contact surface based on the thickness of the wheel - rim and the wear depth of the tread surface; Distribute a millimeter - wave radar network along the longitudinal direction of the track, capture the spatial coordinates of the dynamic contact patch in the wheel - rail contact area in real time, and analyze the deformation characteristics of the dynamic contact patch in combination with the spatial geometric mapping relationship between the center line of the wheel - rim and the track contact surface; Input the wheel - rim geometric degradation characteristics and the dynamic contact patch deformation characteristics in the set of static parameters into the collaborative learning framework, and generate an offset correction amount for the preliminary positioning position of the wheel center through the interaction of local model parameter encryption iteration and global parameter aggregation; The driving of the actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel - rail contact center includes: Based on the offset correction amount, combined with the spatial distribution law of the deformation characteristics of the dynamic contact patch, drive the actuator to adjust the preliminary positioning position, and maintain the positioning stability of the wheel - rail contact center during high - speed operation through the real - time matching of the wheel - rim geometric characteristics and the contact patch coordinates.

3. The method according to claim 2, characterized in that, The inputting of the wheel - rim geometric degradation characteristics and the dynamic contact patch deformation characteristics in the set of static parameters into the collaborative learning framework, and generating an offset correction amount for the preliminary positioning position of the wheel center through the interaction of local model parameter encryption iteration and global parameter aggregation includes: Decompose the geometric deformation gradient tensor of the wheel - rim geometric degradation characteristics, extract the degradation feature vector related to the curvature of the wheel - rail contact surface in the geometric deformation gradient tensor, and generate a feature decomposition sequence including the curvature degradation gradient; Execute a noise masking mechanism on the feature decomposition sequence at the local computing node, and generate an anti - interference degradation gradient encryption chain by superimposing random masking noise through a dynamic attenuation window; Construct a joint parameter space with the degenerate gradient encryption chain and the stress distribution time-series signal in the dynamic contact patch deformation characteristics. Through the cross-node parameter interpolation operation in the collaborative learning framework, generate a global parameter topology chain coupling degradation and deformation; Dynamically iteratively encrypt the degenerate gradient in the global parameter topology chain at the local computing node, fuse the stress distribution in the wheel-rail contact area to generate an encrypted weight gradient chain and distribute it to the global aggregation node; Aggregate the encrypted gradient chains of multiple nodes in the global aggregation node, eliminate the random mask noise and extract the cross-node degradation and deformation coupling phase offset to generate an offset correction amount for the preliminary positioning position of the wheel center.

4. The method according to claim 3, characterized in that, The construction of the joint parameter space with the degenerate gradient encryption chain and the stress distribution time-series signal in the dynamic contact patch deformation characteristics, and the generation of the global parameter topology chain coupling degradation and deformation through the cross-node parameter interpolation operation in the collaborative learning framework includes: Align the sampling timestamps of the spatio-temporal distribution in the degenerate gradient encryption chain and the stress distribution time-series signal in the dynamic contact patch deformation characteristics, match the geometric deformation phase of the wheel-rail contact area and the dynamic stress fluctuation period, and generate a parameter matrix of degradation and stress with spatio-temporal alignment; Perform a parameter interpolation operation between multiple nodes on the parameter matrix, compensate for the dynamic contact patch deformation gradient difference in adjacent track sections, and generate an interpolation compensation parameter chain with continuous cross-node distribution; Fuse the degenerate gradient and the stress distribution amplitude in the interpolation compensation parameter chain, adjust the fusion weight of the gradient and the stress based on the instantaneous deformation rate, and generate a dynamic weight fusion parameter set; Perform spatio-temporal convolution superposition of the degenerate gradient and the stress distribution on the dynamic weight fusion parameter set, extract the parameter aggregation characteristics in the wheel-rail contact center area, and generate a feature topology network coupling degradation and deformation; Analyze the co-oscillation mode of the degenerate gradient and the stress distribution in the feature topology network, and generate a global parameter topology chain coupling degradation and deformation.

5. The method according to claim 2, characterized in that, The millimeter-wave radar network longitudinally distributed along the track captures the dynamic contact patch spatial coordinates in the wheel-rail contact area in real time. Combining the spatial geometric mapping relationship between the center line of the wheel flange and the track contact surface, analyze the deformation characteristics of the dynamic contact patch, including: Deploy a millimeter-wave radar array longitudinally along the track, synchronously collect the reflected signal pulse sequence in the wheel-rail contact area, capture the multi-path echo time delay and amplitude fluctuation related to the deformation of the track contact surface in the reflected signal, and generate an original signal grid of the dynamic contact patch spatial coordinates; Perform multi-path interference suppression processing on the original signal grid, compensate for the attenuation distortion of the reflected signal during high-speed operation, and generate an anti-interference dynamic contact patch spatial coordinate grid; Extract the pulse response amplitude and time-series phase fluctuation of the dynamic contact patch spatial coordinate grid, and construct a pulse response topology structure of the dynamic contact patch deformation characteristics; Perform spatio-temporal correlation matching between the pulse response topology structure and the spatial geometric mapping relationship of the wheel flange center line to generate a coupling parameter set of the dynamic contact patch deformation characteristics and the geometric mapping; Analyze the spatial offset between the pulse response amplitude fluctuation and the geometric mapping in the coupling parameter set to generate the deformation characteristics of the dynamic contact patch.

6. The method according to claim 5, characterized in that, Performing spatio-temporal correlation matching on the spatial geometric mapping relationship between the pulse response topology and the rim centerline to generate a coupling parameter set of dynamic contact patch deformation characteristics and geometric mapping, including: Calibrating the spatio-temporal coordinate origin of the pulse response topology with the geometric mapping spatial reference point of the rim centerline, eliminating the coordinate offset of the propagation path of the deformation in the wheel-rail contact area and the geometric degradation parameters, and generating a coordinate set of deformation and geometry correlation with spatio-temporal reference synchronization; Tracking the diffusion rate of the deformation propagation path in the coordinate set, mapping the attenuation trend of the spatial geometric degradation gradient of the rim centerline with the wear of the track contact surface, and generating a dynamic correlation map of deformation diffusion and geometric degradation; Quantifying the phase offset angle between the deformation propagation path and the spatial geometric degradation gradient in the dynamic correlation map, fusing the instantaneous load intensity in the wheel-rail contact area to calculate the spatial coupling weight, and generating a coupling weight chain of deformation and geometry; Reconstructing the dynamic correlation map in the coupling weight chain, superimposing the multi-dimensional influence factors of the instantaneous load distribution on the wheel-rail contact surface, and generating a multi-dimensional coupling parameter field of deformation propagation and geometric degradation; Analyzing the oscillation frequency of the deformation propagation path and the attenuation amplitude of the geometric degradation gradient in the multi-dimensional coupling parameter field, and generating a coupling parameter set of dynamic contact patch deformation characteristics and geometric mapping.

7. The method according to claim 1, wherein Based on the offset correction amount, combined with the spatial distribution law of the deformation characteristics of the dynamic contact patch, driving the actuator to adjust the preliminary positioning position, and maintaining the positioning stability of the wheel-rail contact center during high-speed operation through the real-time matching of the rim geometric characteristics and the contact patch coordinates, including: Generating dynamic baseline parameters corresponding to the offset correction amount, decomposing the high-frequency oscillation component and the low-frequency attenuation trend in the spatial distribution law of the dynamic contact patch deformation characteristics, constructing a baseline offset amount field for wheel-rail contact center positioning and sending it to the actuator drive nodes in each track section; Mapping the phase difference between the rim geometric characteristics and the contact patch coordinates in the baseline offset amount field at the actuator drive node, compensating for the time delay effect in the spatial distribution law, and generating a wheel center coordinate offset amount adjustment drive signal; Analyzing the instantaneous amplitude fluctuation of the high-frequency oscillation component in the wheel center coordinate offset amount adjustment drive signal, combining the propagation path of the contact patch deformation characteristics, and distributing dynamic weights to the actuator drive nodes to generate multi-node dynamic weight drive parameters; Converting the multi-node dynamic weight drive parameters into a pulse width modulation signal of the actuator, matching the phase synchronization of the rim geometric characteristics and the contact patch coordinates, and generating a dynamic control instruction set for the wheel-rail contact center; Executing the dynamic control instruction set, adjusting the wheel center position and maintaining the positioning stability of the wheel-rail contact center through the real-time closed-loop feedback of the pressure gradient and the contact patch coordinates in the rim contact area by the actuator.

8. The method according to claim 1, characterized in that, Determining the chord length of the wheel according to the detection process of the wheel by the first detection unit, including: When the first detection unit detects that the wheel enters the detection area, starting the encoder to record the first displacement signal of the wheel movement; When the first detection unit fails to detect the wheel, stop the encoder recording and read the second displacement signal recorded by the encoder; Determine the initial displacement value when the wheel edge starts to enter the detection area according to the first displacement signal recorded by the encoder; Determine the final displacement value when the wheel edge completely passes the first detection unit according to the second displacement signal recorded by the encoder; Determine the chord length of the wheel according to the final displacement value and the initial displacement value.

9. The method according to claim 1, characterized in that When the second detection unit detects the wheel, calculate half of the chord length and control the second detection unit to move half of the chord length in the same direction, including: When the second detection unit detects the wheel, use a microprocessor to calculate half of the chord length and drive the second detection unit to move half of the chord length in the same direction through a servo motor.

10. A wheel center positioning system for a rail vehicle, characterized in that, Include: An installation module for pre-installing a first detection unit and a second detection unit on both sides of the wheel; A measurement module for determining the chord length of the wheel according to the detection process of the first detection unit for the wheel; A calculation module for calculating half of the chord length and controlling the second detection unit to move half of the chord length in the same direction when the second detection unit detects the wheel; A positioning module for determining the current position of the second detection unit as the preliminary positioning position of the wheel center when controlling the second detection unit to move half of the chord length in the same direction; A correction module for generating an offset correction amount for the preliminary positioning position of the wheel center based on the collaborative learning framework of dynamic contact patch deformation characteristics and wheel rim geometry degradation, and driving an actuator to dynamically correct the preliminary positioning position according to the offset correction amount to maintain the positioning stability of the wheel-rail contact center.

Images