Train operation and maintenance method based on digital twinning

Abstract

The present application relates to the technical field of digital twin modeling and simulation, and particularly relates to a train operation and maintenance method based on digital twin. The method comprises the following steps: collecting vibration and acoustic sensing data from the wheel-rail interaction area, fusing train operation condition data to obtain wheel-rail interaction data; performing adaptive filtering processing on the wheel-rail interaction data to obtain filtered contact features; analyzing the spatial correlation of the vibration signals in the filtered contact features to reconstruct a contact point distribution map; performing pressure field calculation according to the contact point distribution map to obtain a contact pressure distribution map; deducing equivalent taper according to the contact pressure distribution map and the contact point distribution map, and constructing a wheel-rail contact digital twin model, and generating a contact geometric feature map; and analyzing potential wear patterns based on the contact geometric feature map. The present application realizes virtual-real mapping and closed-loop optimization of the wheel and the rail through digital twin modeling and simulation technology, which not only prolongs the train operation and maintenance cycle, but also improves the operation safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twin modeling and simulation, and particularly relates to a train operation and maintenance method based on digital twin. BACKGROUND

[0002] In the existing train operation and maintenance method, the separation management of wheel and rail detection leads to the fragmentation problem of train operation and maintenance. The traditional maintenance system detects and maintains the wheel and rail as independent objects, ignores the matching relationship of the two as a whole train operation and maintenance, cannot capture the complex dynamic characteristics of wheel-rail interaction, and leads to one-sided and inefficient maintenance decisions. The single parameter threshold judgment leads to the limitation problem of wear identification. The existing technology mainly relies on single geometric parameter thresholds such as wheel diameter and flange thickness for wear judgment, and cannot effectively identify the early characteristics of complex wear patterns such as polygonization and wavy shape, missing the best intervention opportunity. The fixed period maintenance leads to the problem of resource waste and safety hazard. The traditional maintenance method either replaces the parts too early according to the most conservative strategy, causing a large amount of resource waste, or waits until the obvious wear exceeds the standard before processing, bringing serious safety hazards and high repair costs.

[0003] In summary, the existing technology has the problems of insufficient overall analysis of wheel-rail, weak early wear feature recognition ability, and poor maintenance decision accuracy, which need to be solved. SUMMARY

[0004] Therefore, it is necessary to provide a train operation and maintenance method based on digital twin to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a train operation and maintenance method based on digital twin includes the following steps:

[0006] Step S1: Collect vibration and acoustic sensing data from the wheel and rail interaction area, and fuse train operation condition data to obtain wheel-rail interaction data;

[0007] Step S2: Perform adaptive filtering processing on the wheel-rail interaction data to obtain filtered contact features; analyze the spatial correlation of the vibration signal in the filtered contact features, and reconstruct a contact point distribution map; perform pressure field calculation according to the contact point distribution map to obtain a contact pressure distribution map; derive the equivalent taper according to the contact pressure distribution map and the contact point distribution map, and construct a wheel-rail contact digital twin model, and generate a contact geometric feature map;

[0008] Step S3: Analyze potential wear patterns based on the contact geometric feature map, and identify early wear features; calculate the flange / tread wear ratio and distribution characteristics according to the early wear features, and generate a wear pattern feature spectrum;

[0009] Step S4: different dynamic behavior analysis is performed on the wheel-rail interaction data to obtain multiple dynamic response data; a wear-dynamics correlation matrix of digital twin mapping is established according to the wear mode characteristic spectrum on the multiple dynamic response data; a performance degradation critical point of the wear-dynamics correlation matrix is identified, and a dynamic performance prediction graph is generated in the digital twin environment;

[0010] Step S5: a train operation plan is obtained to determine a train maintenance time; a maintenance technical parameter is determined according to the dynamic performance prediction graph, and a train maintenance parameter is generated in combination with the train maintenance time.

[0011] The application realizes accurate monitoring of the key area of wheel-rail by deploying sensors according to the "key point priority" principle, avoids resource waste, reduces false triggering by a double verification mechanism, improves data accuracy, ensures accurate alignment of signals by a timestamp matching algorithm, eliminates analysis errors, eliminates environmental interference by a temperature compensation mechanism, and ensures data comparability. The wheel-rail interaction data formed provides a high-quality basis for subsequent analysis.

[0012] The adaptive filtering process realizes accurate extraction of contact characteristics under different working conditions and overcomes the limitations of traditional filters; the spatial correlation analysis technology realizes "dimensional" reconstruction from one-dimensional signal to two-dimensional contact distribution without disassembling the wheel to obtain complete contact state; the pressure field calculation method establishes a quantitative mapping of contact strength and pressure value, enabling the evaluation to be improved from qualitative to quantitative; the equivalent taper derivation integrates discrete features into system-level parameters; and the contact geometric feature map realizes digital twin representation of the wheel-rail contact state.

[0013] The geometric feature change detection can identify contact abnormalities before the wear surface is visible; the typical wear mode library enables pattern recognition; the periodic analysis in the wheel circumference direction can provide early warning of polygonal wear for 3-4 weeks; the tread longitudinal wave analysis accurately quantifies wavy wear; the regional evaluation method breaks through the limitations of traditional single parameter judgment; and the wear mode characteristic spectrum comprehensively depicts the wear type, degree and distribution characteristics.

[0014] The multiple dynamic behavior analysis realizes comprehensive performance evaluation; the wear-dynamics correlation matrix breaks through the limitations of traditional fragmented analysis and realizes quantitative correlation; the speed stratification analysis considers the differences in different speed intervals; the wear combination coefficient solves the interaction effect analysis when multiple wearings coexist; the performance degradation critical point identification accurately captures the deterioration acceleration inflection point; and the dynamic performance prediction graph realizes the leap from "state monitoring" to "performance prediction".

[0015] The maintenance requirement list realizes accurate evaluation and priority ranking, avoids blindness and hysteresis; the maintenance window identification closely combines the requirement with the operation plan, maximizes the use of the stop time; the parameter optimization method customizes the maintenance parameters according to the wear mode, avoids excessive or insufficient maintenance; the collaborative benefit evaluation table promotes system-level optimization; the maintenance parameter package provides standardized instructions, realizes the transformation from 'periodic maintenance' to 'predictive maintenance', prolongs the service life of components, and improves safety and comfort.

[0016] The present application realizes accurate mapping from single-point vibration signal to overall contact state by constructing a wheel-rail contact digital twin model, breaking through the limitations of traditional detection methods. Through adaptive filtering and signal space-time conversion technology, the contact geometric features are "upgraded" and reconstructed from vibration signals, realizing collaborative analysis of the wheel-rail as a whole; secondly, the wear-dynamics correlation matrix is established to capture the complex correlation between various wear modes and dynamic performance, and the wear acceleration trend is found 3-4 weeks in advance; finally, through accurate prediction of the performance attenuation critical point, the transformation from 'periodic maintenance' to 'predictive maintenance' is realized, which not only prolongs the service life of the wheel, but also improves the operation safety. This train operation and maintenance method based on digital twin realizes virtual-real mapping and closed-loop optimization of the wheel and rail, and solves the technical problems that the traditional maintenance system cannot overcome. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 It is a step flowchart of a train operation and maintenance method based on digital twin.

[0018] The object realization, functional features and advantages of the present application will be further described with reference to the embodiments and in conjunction with the drawings. DETAILED DESCRIPTION

[0019] The technical method of the present application will be described clearly and completely below in conjunction with the drawings. Obviously, the described embodiments are 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 labor fall within the scope of protection of the present application.

[0020] In addition, the drawings are only schematic illustrations of the present application, and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated descriptions thereof will be omitted. Some block diagrams shown in the drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0021] It should be understood that, although the terms "first", "second", etc. can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the example embodiments, a first element can be called a second element, and similarly, a second element can be called a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] To achieve the above object, please refer to Figure 1 The application provides a train operation and maintenance method based on digital twinning, comprising the following steps:

[0023] Step S1: collecting vibration and acoustic sensing data from the wheel and track interaction area, fusing train operation condition data to obtain wheel-rail interaction data;

[0024] In the embodiment of the application, the "key point priority" principle is adopted to install sensors at sharp curves, before and after turnouts and large slope sections to form a sensor layout map. The vibration signal threshold is set to 5 times the average background noise, a double verification mechanism is adopted to ensure triggering accuracy, and complete signal segments before and after the wheel passes are recorded. The train operation parameters are obtained and aligned with the signal segment time, temperature compensation is applied to eliminate environmental impact, and all data are integrated into wheel-rail interaction data.

[0025] Step S2: performing adaptive filtering processing on the wheel-rail interaction data to obtain filtered contact features; analyzing the spatial correlation of the vibration signals in the filtered contact features to reconstruct a contact point distribution map; performing pressure field calculation according to the contact point distribution map to obtain a contact pressure distribution map; deriving equivalent taper according to the contact pressure distribution map and the contact point distribution map, and constructing a wheel-rail contact digital twinning model, and generating a contact geometric feature map;

[0026] In the embodiment of the application, the wavelet packet decomposition is used to decompose the vibration signal into multiple frequency bands, the relevant frequency bands are determined according to the contact feature frequency, and a frequency band feature set is formed. The mapping relationship between the features and the physical parameters is established, the working condition feature vector is extracted, the adaptive filter parameter matrix is constructed, and the filtered contact features are obtained. The time domain signal is converted into an angular domain signal to form rolling path vibration data, the local frequency spectrum entropy sequence is calculated, the entropy and contact width mapping is established, the contact distribution matrix is constructed, and the contact point distribution map is generated by using image processing technology. The single wheel load is calculated according to the load data, the effective contact area is determined, the pressure coefficient matrix is generated, the absolute pressure value is calculated, and the contact pressure distribution map is formed.

[0027] Step S3: analyzing potential wear patterns based on the contact geometric feature map, and identifying early wear features; calculating the proportion and distribution characteristics of the flange / tread wear according to the early wear features to generate a wear pattern feature spectrum;

[0028] In the embodiment of the application, the key parameters of the contact geometric characteristics are extracted, the difference with the historical data is calculated to form a geometric change vector. It is compared with the typical wear pattern library to calculate the similarity and form a pattern similarity table. The periodic changes in the wheel circumference direction are analyzed, the peak frequency is identified by Fourier transform, and the polygonization index is calculated. The longitudinal tread waves are analyzed, the scale parameters are extracted by wavelet transform, and the wave index is calculated. The wheel rim and tread areas are distinguished, the contact characteristics of each area are calculated, and the rim wear rate and tread wear rate are obtained. The various indicators are integrated to form a wear pattern characteristic spectrum.

[0029] Step S4: analyzing the wheel-rail interaction data under different dynamic behaviors to obtain multiple dynamic response data; establishing a wear-dynamics correlation matrix for digital twin mapping of the multiple dynamic response data according to the wear pattern characteristic spectrum; identifying the performance degradation critical point of the wear-dynamics correlation matrix, and generating a dynamics performance prediction map in the digital twin environment;

[0030] In the embodiment of the application, the lateral force characteristics under different curvatures are extracted to construct a guide force spectrum, the lateral displacement of the straight line segment is analyzed to identify the snaking characteristics, and the impact events are detected to extract the response characteristics. After normalizing each characteristic, the correlation between the wear pattern and the dynamic indicators is calculated, the speed stratification effect and the wear combination influence are considered, and a four-dimensional wear-dynamics correlation matrix is constructed. The wear development trend is predicted, the performance degradation curve and its derivative are calculated, the acceleration inflection point of the degradation is identified, the inflection point time is corrected according to the wear type, the performance critical threshold is set, the critical time of safety and comfort is determined, and a dynamics performance prediction map is generated.

[0031] Step S5: obtaining a train operation plan to determine a train maintenance time; determining maintenance technical parameters according to the dynamics performance prediction map, and generating train maintenance parameters in combination with the train maintenance time;

[0032] In the embodiment of the application, the maintenance type and depth are determined according to the wear type and severity, the urgency is determined according to the critical time, and a maintenance requirement list is generated. The train operation plan is obtained, the maintenance window meeting the time length requirement is identified, the feasible window set is generated considering the resource constraints. The maintenance parameters are optimized according to the wear characteristics, the initial values and constraint ranges of parameters such as cutting depth, feed speed, or polishing frequency and angle are set, and the optimal parameters are solved by multi-objective optimization. The maintenance benefits, including performance improvement, life extension, resource conservation and operation impact, are calculated to form a synergistic benefit evaluation table. The optimal scheme is selected to generate standardized maintenance instructions, which are integrated into a maintenance parameter package.

[0033] Preferably, step S1 comprises the following steps:

[0034] Step S11: obtaining and deploying vibration and acoustic sensors according to train configuration information and line geometry data to obtain a sensor layout map;

[0035] Step S12: detecting a wheel passing event section according to the sensor layout diagram;

[0036] Step S13: obtaining train operation parameters, and performing data fusion with the wheel passing event section to obtain wheel-rail interaction data.

[0037] In the embodiment of the application, train configuration information (wheel diameter, wheelbase, bogie structure and weight distribution) and line geometry data (curve radius, superelevation value, turnout position and slope) are obtained. The "key point first" principle is adopted to deploy sensors: vibration sensors (10 kHz sampling rate, 100 mV / g sensitivity) and acoustic sensors (48 kHz sampling rate, 50 mV / Pa sensitivity) are installed at sharp curves (curve radius < 300 meters), 50 meters before and after the turnout, and large slope change sections (> 3 ‰). The sensor is 15 mm away from the center line of the rail and is flush with the rail top, and is fixed with a magnetic base. The signal is transmitted to the central system through industrial Ethernet after 16-bit AD conversion, and a sensor layout diagram containing sensor information is generated.

[0038] The vibration signal threshold is set to 5 times the background noise mean value, and a double verification mechanism (vibration and acoustic signals meet the conditions at the same time and last for > 10 milliseconds) is adopted to avoid false triggering. The sampling rate is automatically increased (vibration 20 kHz, acoustic 96 kHz) to capture high-frequency characteristics when the wheel approaches. The complete signal from 0.5 seconds before the wheel passes to 1.5 seconds after the wheel passes is recorded to form an event section containing time domain waveform, time-frequency analysis and timestamp, and the root mean square value, peak factor, kurtosis and A-weighted sound pressure level are calculated as characteristic identifiers.

[0039] Train operation parameters are obtained, including speed (accuracy ±1 km / h), load (accuracy ±2%), running direction and train number information. A timestamp matching algorithm is used to align the running parameters with the event section, and temperature compensation (0.2% per degree Celsius) is applied to eliminate environmental influences. All data are integrated into wheel-rail interaction data, including original signals, operating conditions, environmental parameters and feature analysis results. A hierarchical storage structure (binary format for original signals, structured format for feature data) is used to ensure data integrity and query efficiency. Preferably, the adaptive filtering process of the wheel-rail interaction data in step S2 includes:

[0040] The wheel-rail interaction data is decomposed into multiple frequency bands, the frequency bands related to contact are identified, and a frequency band feature set is obtained;

[0041] Performing contact feature mapping on the frequency band feature set to obtain a feature mapping table;

[0042] Extracting a working condition feature vector from the frequency band feature set;

[0043] The basic filtering parameters are adaptively selected based on the feature mapping table and the operating condition feature vector;

[0044] Frequency band selective processing is performed based on the basic filter parameters and frequency band feature set to obtain the filter contact characteristics.

[0045] In this embodiment of the invention, the vibration signal is decomposed into 5 levels using the db4 wavelet basis function, resulting in 32 frequency band sub-signals. The energy distribution ratio is calculated for each frequency band sub-signal. ,in For the first Signal energy of each frequency band The total signal energy is given. The wheel-rail contact characteristic frequency is calculated based on the wheel diameter D and train speed v. The frequency band range related to contact was determined as follows. And its harmonic frequency bands. Calculate the signal-to-noise ratio for each frequency band. Frequency bands with a signal-to-noise ratio (SNR) higher than 6 dB are designated as effective frequency bands. The sub-signals, energy distribution ratio, and SNR of the effective frequency band are combined to form a frequency band feature set, which includes frequency range, energy proportion, SNR, and time-domain waveform characteristics.

[0046] Establish the correspondence between frequency band characteristics and wheel-rail contact physical parameters. First, extract the dominant frequency component from the frequency band characteristic set. and energy distribution Calculate the energy ratio matrix between frequency bands ,in Let be the signal energy in the j-th frequency band. According to wheel-rail contact theory, the low-frequency band (50-200Hz) corresponds to the contact area variation, the mid-frequency band (200-800Hz) corresponds to the contact pressure distribution, and the high-frequency band (800-2000Hz) corresponds to the surface roughness characteristics. Construct a feature-parameter mapping matrix. Matrix elements Indicates the first The frequency band characteristics for the first The contribution weights of each contact parameter are determined through regression analysis of extensive experimental data, forming a standardized feature mapping table. This table includes four dimensions: frequency band index, frequency range, corresponding physical parameter type, and contribution weight, providing a basis for subsequent filter parameter selection.

[0047] During the process of extracting operating condition feature vectors from the frequency band feature set, train speed is extracted. Axle load curve radius and orbital state index Four key operating parameters, the track condition index, represent a comprehensive evaluation index of the track's geometric and structural condition, reflecting the overall quality of track smoothness, stiffness, and stability. This index is calculated by measuring parameters such as track irregularities, track stiffness variations, and structural integrity. The calculation formula is as follows:

[0048] ;

[0049] in, The measured value of track irregularity (mm) is... The maximum allowable roughness (typically 10 mm), Given the current track stiffness (MN / m), The standard reference stiffness is typically 50 MN / m. The structural damage index is assessed based on acoustic and vibration characteristics. The maximum damage threshold (usually 1), , , The weighting coefficients are 0.5, 0.3, and 0.2, respectively, representing the relative importance of each factor; the value of T ranges from [0,2], where 0 represents the extreme orbital state, 1 represents the standard reference state, and 2 represents the ideal state.

[0050] A higher track condition index (T) indicates better track quality, more stable wheel-rail contact, and more regular vibration signal characteristics; a lower T value indicates worse track quality, less stable wheel-rail contact, and more complex vibration signal characteristics. As a component of the operating condition feature vector, this index directly influences the selection of adaptive filtering parameters, leading to different signal processing strategies for different track conditions.

[0051] These four parameters are normalized to obtain the normalized parameters. , , and The values ​​of these four normalized parameters are all in the range of [0,1]. These four normalized parameters are combined to form the working condition feature vector. To capture the trend of changing operating conditions, the Euclidean distance between the current operating condition and historical operating conditions is calculated. ,in This is the feature vector of the current operating condition. This is a historical operating condition feature vector. When the distance value is greater than 0.2, the filter parameter update mechanism is triggered. The operating condition feature vector reflects the comprehensive characteristics of the current train operating environment, directly affecting the wheel-rail contact state and signal characteristics, and is a key input for adaptive filtering.

[0052] When adaptively selecting basic filter parameters based on the feature mapping table and the operating condition feature vector, the first step is to select parameters based on the operating condition feature vector. Determine the type of contact parameter of interest. When the train speed... When the value is higher than 0.7, prioritize high-frequency characteristics; when the axle load When the value is higher than 0.8, prioritize mid-frequency characteristics; when the curve radius... When the value is below 0.3, priority is given to low-frequency features. The contribution weights of the corresponding frequency bands are extracted from the feature mapping table to construct the filter parameter matrix. , including center frequency ,bandwidth order and type Four parameters. The filter type is automatically selected from Butterworth, Chebyshev, and elliptic filters based on the signal characteristics. The filter order is determined based on the signal-to-noise ratio (SNR) distribution; lower-order filters (2nd-4th order) are used for frequency bands with high SNR, while higher-order filters (6th-8th order) are used for frequency bands with low SNR, to balance signal fidelity and noise suppression.

[0053] When performing frequency band selective processing based on basic filtering parameters and frequency band feature sets, a corresponding filter is applied to each effective frequency band. For the center frequency... ,bandwidth order and type For each given filter, calculate its frequency response. Each sub-signal in the frequency band feature set is processed through a corresponding filter to obtain the filtered sub-signal. Based on the contribution weights in the feature mapping table. The filtered sub-signals are then reconstructed using weighted summaries. ,in For the first The contribution weights of each frequency band are determined. Envelope extraction is performed on the reconstructed signal, and the signal envelope is calculated using Hilbert transform. A smooth envelope is obtained by low-pass filtering (cutoff frequency 50Hz). Finally, the filtered time-domain signal Spectral characteristics and smooth envelope These are combined to form filtered contact features, providing basic data for subsequent reconstruction of contact point distribution.

[0054] Preferably, step S2, which involves analyzing the spatial correlation of vibration signals in the filtered contact features and reconstructing the contact point distribution map, includes:

[0055] Train speed data and wheel diameter are extracted from wheel-rail interaction data, and rolling path vibration data are calculated.

[0056] extracts a local frequency spectrum entropy sequence from the rolling path vibration data by filtering contact features;

[0057] performs lateral contact width inversion according to the local frequency spectrum entropy sequence to obtain a contact width sequence;

[0058] constructs an initial contact distribution matrix from the contact width sequence and the rolling path vibration data;

[0059] generates a two-dimensional morphological pixel in a digital twin space from the initial contact distribution matrix to obtain a contact point distribution map.

[0060] In the embodiment of the application, train speed data is extracted from wheel-rail interaction data (unit: m / s) and wheel diameter (unit: m), and wheel rotation angular velocity is calculated, wherein is the wheel radius. According to the wheel rotation angular velocity and the vibration signal sampling frequency , the time-domain vibration signal is converted into an angular-domain vibration signal , wherein represents the rotation angle. By angular-domain resampling, the vibration signal is mapped from the time dimension to the space dimension to obtain the vibration distribution along the wheel circumference. The wheel circumference length is calculated, and the angular-domain vibration signal is mapped to the spatial domain and equally sampled (with a sampling interval of 1 mm) to form rolling path vibration data v(s), wherein represents the linear distance along the circumference, and the value range is . The data represents the vibration intensity distribution at each position along the circumference during one complete rotation of the wheel, and directly reflects the spatial distribution characteristics of the wheel-rail contact state.

[0061] When extracting a local frequency spectrum entropy sequence from rolling path vibration data by filtering contact features, first, the rolling path vibration data is segmented by a sliding window. The window length is set to 50 mm, and the sliding step is 5 mm. The short-time Fourier transform (STFT) is performed on the data in each window to obtain a local frequency spectrum . The normalized energy distribution is calculated for each local frequency spectrum, and then the frequency spectrum entropy is calculated, wherein represents the normalized energy at frequency . The frequency spectrum entropy reflects the complexity of the signal frequency spectrum distribution. A high entropy value indicates uniform frequency spectrum distribution, corresponding to a complex contact state; a low entropy value indicates concentrated frequency spectrum, corresponding to a simple contact state. The spectrum entropy corresponding to each window position is Composition sequence , forming a local spectrum entropy sequence. This sequence reflects the complexity change of contact state at different positions along the wheel circumference, providing the basis for lateral contact width inversion.

[0062] When performing lateral contact width inversion according to the local spectrum entropy sequence, based on wheel-rail contact theory, there is an inverse relationship between contact width and spectrum entropy. First, the mapping function of spectrum entropy and contact width is established , where represents the contact width, represents the spectrum entropy. The mapping function is obtained by fitting a large amount of experimental data, using a piecewise linear model: when , ; when , . Where is the entropy threshold, set to 0.7; is the maximum contact width, set to 15 mm; is the minimum contact width, set to 3 mm. Apply the mapping function to each entropy value H_i in the local spectrum entropy sequence H(s) to calculate the corresponding contact width , forming a contact width sequence . This sequence represents the wheel-rail contact width distribution at different positions along the wheel circumference, directly reflecting the lateral characteristics of the contact state.

[0063] When constructing the initial contact distribution matrix from the contact width sequence and rolling path vibration data, first determine the matrix size. The longitudinal dimension corresponds to the wheel circumference position , with a resolution of 1 mm and a total length of ; the lateral dimension corresponds to the contact lateral position , with a resolution of 0.5 mm and a range of . Construct an empty matrix with size . For each circumference position , according to the corresponding contact width and vibration intensity , calculate the lateral contact distribution. Assuming that the lateral contact distribution conforms to a Gaussian distribution with a center position of 0 and a standard deviation , the contact intensity at lateral position w is . Repeat this calculation for all circumference positions to fill all elements of matrix , forming the initial contact distribution matrix. This matrix represents the two-dimensional distribution characteristics of the wheel-rail contact area, with the horizontal coordinate representing the lateral position, the vertical coordinate representing the circumference position, and the matrix element value representing the contact intensity.

[0064] When generating the two-dimensional morphological pixel of the digital twin space for the initial contact distribution matrix, first, normalize the matrix , so that all element values are mapped to the range [0, 1] to obtain a normalized matrix . Apply bilateral filtering to the for smoothing processing, while retaining the edge features and reducing the noise influence, and the filtering parameters are set as the spatial domain standard deviation , the value domain standard deviation , and the window size is 5x5. Apply adaptive threshold segmentation to the filtered matrix, and the threshold value is , where represents the matrix mean, represents the standard deviation. Set the elements greater than the threshold value to 1, and the elements less than or equal to to 0 to obtain a binary matrix . Apply morphological opening operation to the to remove isolated noise points, and the structure element is a 3x3 rectangle. Apply the contour extraction algorithm to the processed binary matrix to identify the boundary of the connected region, and obtain the contour set of the contact area. Fill the contour as a color region, with the area with high contact intensity represented by red and the area with low contact intensity represented by blue, to form a color contact point distribution map. The distribution map directly displays the spatial distribution characteristics of the wheel-rail contact area, and provides a geometric basis for subsequent pressure field calculation.

[0065] Preferably, the pressure field calculation according to the contact point distribution map in step S2 comprises:

[0066] extracting train load data from the wheel-rail interaction data, and performing wheel load distribution calculation in combination with the contact point distribution map to obtain a single wheel load value;

[0067] determining an effective contact area from the contact point distribution map;

[0068] extracting a brightness value matrix from the contact point distribution map, and generating a pressure coefficient matrix in combination with the effective contact area;

[0069] performing absolute pressure value calculation according to the pressure coefficient matrix, the single wheel load value and the effective contact area to obtain a pressure numerical matrix;

[0070] performing digital twin pressure field visualization on the pressure numerical matrix to obtain a contact pressure distribution map.

[0071] In the embodiment of the application, the train load data is extracted from the wheel-rail interaction data (unit: kN), including the vehicle self-weight and the passenger and cargo weight. According to the vehicle configuration parameters, the axle load distribution coefficient is calculated, where represents the axle number, For a typical four-axle bogie, the front axle coefficient , the rear axle coefficient , and the symmetric distribution. The axle load of each axle is calculated. According to the axle load and bogie structure parameters, the left and right wheel load distribution coefficients and are calculated, considering the influence of superelevation on curves. On straight sections, ; on curved sections, the inside wheel coefficient decreases, and the outside wheel coefficient increases, with the change amount being proportional to the superelevation value and the curve radius. The single wheel load value is calculated. The single wheel load value directly determines the absolute size of the wheel-rail contact pressure and is the basic input parameter for pressure field calculation.

[0072] When determining the effective contact area from the contact point distribution map, first convert the color contact point distribution map to a grayscale map, with the grayscale value ranging from [0, 255]. Set the grayscale threshold , and consider pixels with a grayscale value greater than as the effective contact area. Apply connected component analysis to the effective contact area, identify all connected regions, and calculate the area of each region (unit: pixels). Remove small regions with an area less than the threshold pixels to reduce the impact of noise. Calculate the total area of the remaining regions. According to the spatial resolution r (unit: mm / pixel) of the contact point distribution map, convert the pixel area to the actual physical area (unit: mm²). Considering the edge blur area in the contact point distribution map, introduce an area correction coefficient to calculate the effective contact area . The effective contact area is a key parameter for calculating the average contact pressure and directly affects the absolute value of the pressure distribution.

[0073] When extracting the brightness value matrix from the contact point distribution map, normalize the grayscale contact point distribution map to map the grayscale value range from [0, 255] to [0, 1], obtaining the normalized brightness matrix . For each pixel in the effective contact area, its normalized brightness value reflects the relative contact intensity of that point. Considering the nonlinear relationship between brightness and pressure, introduce a power-law transformation to calculate the pressure coefficient , where is the pressure exponent, taking a value of 1.5. This exponent value is determined based on Hertz contact theory and reflects the nonlinear mapping relationship between contact intensity and pressure. Normalize the pressure coefficient matrix so that , to ensure the total pressure distribution is reasonable. The normalized pressure coefficient matrix represents the relative pressure distribution of each point in the contact area, providing a basis for absolute pressure value calculation.

[0074] When calculating the absolute pressure value according to the pressure coefficient matrix, single wheel load value, and effective contact area, first calculate the average contact pressure (unit: MPa). For each pixel point in the effective contact area, calculate its absolute pressure value , where is the area of the connected region where the point is located, is the total effective area. This calculation method ensures that the total pressure is balanced with the single wheel load value, while maintaining the spatial characteristics of the pressure distribution. For pixel points outside the effective contact area, set the pressure value . The pressure values of all pixel points form a pressure numerical matrix , which has the same size as the contact point distribution map, and each element value represents the absolute pressure value (unit: MPa) at the corresponding position. The pressure numerical matrix is a quantitative expression of the contact pressure distribution, providing basic data for subsequent equivalent cone angle derivation.

[0075] When visualizing the digital twin pressure field of the pressure numerical matrix, pseudo-color mapping technology is used to map the pressure value range to the color space. Set the pressure color scale, 0 MPa corresponds to blue, corresponds to red, and the intermediate values gradually change according to the rainbow color spectrum. Take the maximum value in the pressure matrix, or set a theoretical upper limit of 1000 MPa. Apply color mapping to the pressure numerical matrix to generate a color pressure field image. Add isobar lines on the image with an interval of to enhance the three-dimensional sense of the pressure distribution. Add a color scale and scale to indicate the pressure value range and unit. Label key feature points on the image, including the maximum pressure point, pressure center, and contact boundary. Save the processed image as a contact pressure distribution map, which visually displays the pressure distribution characteristics of the wheel-rail contact area, including the pressure peak position, pressure gradient, and contact boundary shape. The contact pressure distribution map is a key characterization of the wheel-rail contact state, providing important basis for wear pattern recognition and dynamic analysis.

[0076] Preferably, step S3 comprises the following steps:

[0077] Step S31: Detect the geometric feature change of the contact geometric feature map to obtain a geometric change amount;

[0078] Step S32: Compare the geometric change amount with a preset typical wear pattern library to obtain a pattern similarity table;

[0079] Step S33: Analyzing the periodic contact change in the wheel circumferential direction of the train according to the pattern similarity table and the contact geometric feature map to obtain a polygonization index;

[0080] Step S34: Analyzing the tread longitudinal corrugation of the train according to the pattern similarity table and the contact geometric feature map to obtain a waviness index;

[0081] Step S35: Evaluating the rim and tread wear according to the contact geometric feature map to obtain the rim and tread wear rates;

[0082] Step S36: Determining the wear pattern feature spectrum based on the polygonization index, the waviness index, and the rim and tread wear rates.

[0083] In the embodiment of the application, when the contact geometric feature map is subjected to geometric feature change detection, the historical data of the last 10 measurements are acquired, the phase correlation method is used for image registration, and the accuracy is 0.1 mm. Key parameters including the contact area A, the center coordinates , the major and minor axes , and the direction angle are extracted, and the differences from the average values are calculated. The Fourier transform is performed on the contact pressure distribution, the spatial frequency spectrum features are extracted, and the differences are calculated. All the differences are combined into a geometric change amount vector for comprehensively representing the contact feature change trend.

[0084] When the geometric change amount is compared with the typical wear pattern library, the pattern library containing 5 typical patterns is loaded , each pattern contains a feature vector and a weight . The cosine similarity of the geometric change amount and the feature vector of each pattern is calculated, the weighted similarity is obtained by introducing the weight coefficient. The threshold value is set to determine the early features, and the pattern similarity table is formed, reflecting the matching degree of the current features and each pattern.

[0085] When the periodic contact change in the wheel circumferential direction is analyzed, when the polygonization wear similarity , the contact feature map is unwrapped along the circumference to obtain an intensity sequence , the Fourier transform is performed to obtain the frequency spectrum . The peak frequency is identified, the wavelength and the energy proportion are calculated, wherein is the energy value at the peak frequency, representing the energy of the main frequency component of the polygonization wear, is the total signal energy, representing the sum of all frequency component energy. Calculate the polygonization index where is the wavelength representing the wavelength in the circumferential direction, is 1 / 20 of the wheel circumference, P value range [0,1], the larger the value, the more serious the polygonization.

[0086] When analyzing the longitudinal tread corrugation, when the corrugation similarity is expanded along the longitudinal direction of the tread to obtain the intensity sequence , the Morlet wavelet basis function is used for transformation. Extract the main scale parameter , calculate the wavelength , wave energy and consistency index . Calculate the corrugation index where is 10mm, value range [0,1], the larger the value, the more serious the corrugation wear.

[0087] When evaluating the rim and tread wear, the contact feature map is divided into the rim area (15mm inside the rim top to 5mm outside) and the tread area. Calculate the contact area ratio, average pressure and maximum pressure of each area to obtain the rim wear rate where is the contact area ratio of the rim area, is the average pressure of the rim area, is the maximum pressure of the rim area; the tread wear rate where is the contact area ratio of the tread area, is the average pressure of the tread area, is the maximum pressure of the tread area, and the rim / tread wear ratio .

[0088] When determining the wear pattern feature spectrum, construct the feature vector where is the polygonization value, indicating the severity of the wheel polygonization wear, the larger the value, the more serious the polygonization, the value range [0,1], is the corrugation value, indicating the severity of the tread corrugation wear, the larger the value, the more serious the corrugation wear, the value range [0,1], is the rim wear rate, indicating the wear degree of the rim area, is the tread wear rate, indicating the wear degree of the tread area, is the rim / tread wear ratio, indicating the relative relationship between the rim wear and the tread wear; construct the feature vector Normalization and setting the decision rule: for polygonal wear; for wavy wear; and for flange wear; and for tread wear; multiple results coexist as combined wear, and the decision rule is used to determine the specific wear type of the wheel according to the index values in the feature vector F. Calculate the severity index , form the wear pattern feature spectrum , provide key input for subsequent analysis.

[0089] Preferably, the different dynamic behavior analysis of the wheel-rail interaction data in step S4 includes:

[0090] Extracting the lateral force characteristics of the wheel passing through different curvature radii from the wheel-rail interaction data to obtain a curve guiding force spectrum;

[0091] Performing digital twin dynamics snake stability evaluation on the wheel-rail interaction data to obtain a snake stability index;

[0092] Identifying the impact response characteristics of the wheel passing through the track in the wheel-rail interaction data to obtain an impact response spectrum.

[0093] In the embodiment of the present application, when extracting the lateral force characteristics of the wheel passing through different curvature radii from the wheel-rail interaction data, first, according to the line geometry data, the wheel-rail interaction data is grouped by curvature radius into four categories: straight line segment > 5000m), large radius curve (1000m ≤ 5000m), medium radius curve (400m ≤ 1000m) and small radius curve ≤ 400m). Extract the lateral component of the vibration data for each type of curve segment, use three-axis acceleration sensor data, and extract the lateral acceleration perpendicular to the direction of travel. Band-pass filter the lateral acceleration, with a frequency range of 0.5-20Hz, to filter out high-frequency noise and low-frequency drift. According to Newton's second law, calculate the lateral force , where is the equivalent mass of the wheel, which is half of the axle load. Perform statistical analysis on the lateral force data to calculate the mean , standard deviation , peak factor (where is the maximum value of the lateral force) and kurtosis . Normalize the lateral force characteristic parameters under different curvature radii to obtain normalized parameters , 、 and wherein is the normalized lateral force mean value, is the normalized lateral force standard deviation, is the normalized peak factor, all in the range of [0, 1]. The curve guiding force spectrum is constructed as wherein the curve guiding force spectrum contains the curvature radius and the corresponding normalized lateral force characteristics (mean value , standard deviation , peak factor and kurtosis ), denotes the curvature category index. The curve guiding force spectrum reflects the guiding ability of the wheel under different curvature conditions, which is an important indicator for evaluating the wheel-rail matching state.

[0094] When performing digital twin dynamics snake stability evaluation on wheel-rail interaction data, first, the lateral displacement signal is extracted from the vibration data of the straight section . Wavelet denoising is performed on the lateral displacement signal, using db4 wavelet basis function, 4 layers of decomposition, and soft threshold method for threshold selection. Hilbert-Huang transform (HHT) is performed on the denoised signal to obtain intrinsic mode function (IMF) components and instantaneous frequency. The snake motion mode is identified from the IMF components, with a characteristic frequency range of 2-8 Hz. The energy proportion of the snake mode is calculated as wherein is the snake mode energy, is the total signal energy. The snake frequency is calculated as the main frequency of the snake mode. According to the vehicle speed and the snake frequency , the snake wavelength is calculated. The snake amplitude is calculated as the root mean square value of the snake mode. According to the snake theory, the critical speed is calculated, wherein is the wheelbase. The speed margin is calculated, wherein is the critical speed, indicating the speed threshold at which the vehicle snake motion begins to lose stability, exceeding this speed may lead to unstable operation, in the range of [0, 1], the larger the value, the better the snake stability. The snake stability index is constructed as wherein is the speed margin, indicating the safety margin between the current operating speed and the critical speed, which comprehensively characterizes the snake motion characteristics of the vehicle, and is a key indicator for evaluating high-speed operation stability.

[0095] ​In identifying the impact response characteristics of the wheel passing through the track in the wheel-rail interaction data, first, the peak value of the vibration signal is detected to identify the impact event. The peak value detection threshold is set to 5 times the root mean square value of the signal, and the time interval is not less than 0.1 seconds. For each detected peak value event, the time window data from 0.05 seconds before to 0.2 seconds after is extracted to form an impact response segment. Time-frequency analysis is performed on each impact response segment, and the short-time Fourier transform (STFT) is used with a window length of 256 points and an overlap rate of 75%. Impact feature parameters are extracted from the time-frequency diagram, including peak amplitude , decay time (the time required for the amplitude to drop to the peak value) , dominant frequency , and frequency bandwidth . According to the impact position information, the impact events are classified into three categories: track joint impact, turnout impact, and irregular impact. Statistical analysis is performed on the feature parameters of each type of impact event to calculate the mean and standard deviation. An impact response spectrum is constructed, where the impact response spectrum includes the impact type (such as track joint impact, turnout impact, etc.) and the corresponding feature parameters: peak amplitude , decay time , dominant frequency , and frequency bandwidth , where represents the impact type index. For track joint impact, the impact transmission ratio is additionally calculated, where is the theoretical impact input at the joint, is the measured impact response. The impact response spectrum reflects the response characteristics of the wheel-rail system to impact loads and is an important indicator for evaluating dynamic performance.

[0096] Preferably, the step S4 of establishing a wear-dynamics correlation matrix according to the wear pattern feature spectrum and multiple dynamic response data includes:

[0097] Converting the wear pattern feature spectrum, curve-guided force spectrum, hunting stability index, and impact response spectrum into a unified standard interval to obtain a normalized feature set;

[0098] Calculating the correlation degree of a single wear pattern in the normalized feature set with each dynamic indicator to obtain a single-mode influence coefficient;

[0099] Obtaining train speed data and combining the single-mode influence coefficient to perform speed interval layering to obtain a speed layering relationship;

[0100] Calculating the interaction influence of multiple wear patterns coexisting in the normalized feature set to obtain a wear combination coefficient;

[0101] The single-mode influence coefficient, the speed level relationship, and the wear combination coefficient are used to construct a wear-dynamics correlation matrix of the digital twin mapping.

[0102] In the embodiment of the application, when converting the wear mode feature spectrum, the curve guiding force spectrum, the snake stability index, and the impact response spectrum into a unified standard interval, the physical meaning and dimension of each characteristic parameter are first determined. The wear mode feature spectrum includes a polygonization index , a wave degree index , a rim wear rate , a tread wear rate , and a rim tread wear ratio ; the curve guiding force spectrum includes a lateral force mean value , a standard deviation , a peak factor , and a kurtosis ; the snake stability index includes a snake frequency , a snake wavelength , a snake amplitude , and a speed margin ; and the impact response spectrum includes a peak amplitude , a decay time , a main frequency , and a frequency bandwidth . A minimum-maximum normalization method is applied to each parameter, wherein is an original value, is a normalized value, and are the minimum value and the maximum value of the parameter, respectively. For unbounded parameters such as the rim / tread wear ratio , a sigmoid function is used for normalization, wherein is a slope parameter, and is a midpoint parameter. All normalized parameters are combined into a normalized feature set , including a wear feature vector , a guiding force feature vector , a snake stability feature vector , and an impact response feature vector . The normalized feature set makes parameters of different physical dimensions comparable and provides a unified basis for subsequent correlation analysis.

[0103] When calculating the correlation degree of a single wear mode and each dynamic index in the normalized feature set, the characteristic parameters of the single wear mode are first extracted from the wear feature vector . For polygonization wear, the polygonization index is extracted.; For wave abrasion, extract the wave degree index ; For rim wear, extract the rim wear rate ; For tread wear, extract the tread wear rate . Calculate the Pearson correlation coefficient between the characteristic parameters of each wear mode and the dynamic indicators , where is the characteristic parameter of the th wear mode, is the th dynamic indicator, denotes the covariance, denotes the standard deviation, is the standard deviation of the th wear mode characteristic parameter, is the standard deviation of the th dynamic indicator. The correlation coefficient ranges from -1 to 1, with positive values indicating positive correlation and negative values indicating negative correlation, and the absolute value indicating the correlation strength. Perform a significance test on the correlation coefficient and calculate statistic , where n is the sample size. When , the correlation is considered significant, where is the critical value of the distribution with degrees of freedom , and the significance level is set to 0.05. Group the significant correlation coefficients into a single mode influence coefficient matrix , where the matrix element represents the influence of the th wear mode on the th dynamic indicator. The single mode influence coefficient matrix reveals the independent influence of various wear modes on dynamic performance and is the basis for constructing the wear-dynamics correlation matrix.

[0104] When obtaining train speed data and performing speed interval stratification based on single mode influence coefficients, first divide the train running speed into low speed ( ≤60km / h), medium speed (60km / h ≤120km / h) and high speed ( >120km / h) three intervals. Calculate the single mode influence coefficient matrix , and for each speed interval data separately. Compare the differences in influence coefficients under different speed intervals to calculate the difference matrix , which is used to evaluate the differences between low speed , medium speed , high speed The degree of change in the influence coefficient is determined by taking the maximum difference between pairwise comparisons of the three speed ranges. When the difference matrix elements... When the value is greater than 0.3, the influence relationship is considered to have a significant velocity dependence. For influence relationships with velocity dependence, a velocity correction function is established. , where the coefficient The influence coefficients were determined through polynomial fitting. The velocity correction function was then applied to the influence coefficients to obtain the velocity-corrected influence coefficients. The correction influence coefficients for all speed ranges are combined to form a speed hierarchy relationship matrix. Matrix elements Indicates the first In the first speed range, the second... The wear mode for the first The influence of each dynamic index was determined. The velocity hierarchy matrix reveals the velocity dependence of the wear-dynamic relationship, improving the model's adaptability under different operating conditions.

[0105] When calculating the interaction effects of multiple wear modes coexisting in a normalized feature set, the coexistence of wear modes is first identified. When the polygonization index... >0.3 and wave index When the wear rate is >0.3, polygonal wear and wavy wear are considered to coexist; when the rim wear rate is... >0.3 and tread wear rate When the value is greater than 0.3, rim wear and tread wear are considered to coexist. For coexisting wear modes... Calculate its interaction coefficient ,in and Let be the single-mode influence coefficients for the i-th and j-th wear modes, respectively. , and These are the weighting coefficients, determined through regression analysis of historical data. The weighting coefficients satisfy... This reflects the ratio of linear superposition effects to nonlinear interaction effects. For cases where three or more wear modes coexist, a recursive method is used to calculate the interaction effects. ,in For pattern and The interaction coefficient, For pattern The single-mode influence coefficient, The weighting coefficient is 0.7. The interaction coefficients of all wear modes are combined to form the wear combination coefficient matrix C, where the matrix elements are... Indicates wear mode The combined influence coefficient during coexistence. The wear combination coefficient matrix captures the complex interaction effects when multiple wear modes coexist, overcoming the limitations of single-mode analysis.

[0106] When constructing the wear-kinetic correlation matrix of the digital twin mapping based on the single-mode influence coefficients, the hierarchical relationship of the partial velocity, and the wear combination coefficients, the matrix structure is first defined. The wear-kinetic correlation matrix M is a four-dimensional tensor with dimension M. ,in For the number of wear modes, For the number of dynamic indicators, The number of speed ranges, Number of wear combination types. Matrix elements. Indicates the first Within a certain speed range, in the l-th type of wear combination, the first... The wear mode for the first The degree of influence of each dynamic index. When filling the matrix elements, first, based on the single-mode influence coefficient matrix... Fill in the basic influence relationships; then, based on the velocity hierarchy relationship matrix... Speed ​​correction is performed; finally, based on the wear combination coefficient matrix... Combination effect correction is performed. The completed matrix is ​​normalized to ensure that the sum of the total influence coefficients of each dynamic index is 1. A physical mapping relationship is constructed for the matrix, establishing a correspondence between matrix elements and parameters in the digital twin model. For example, the effect of polygonal wear on serpentine stability is mapped to the equivalent taper change in the digital twin model; the effect of wavy wear on impact response is mapped to the damping ratio change in the digital twin model. The completed wear-dynamic correlation matrix achieves a quantitative mapping from wear state to dynamic performance and is the core model for performance prediction in the digital twin environment.

[0107] Preferably, step S4, which identifies the performance degradation critical point of the wear-kinetic correlation matrix and generates a kinetic performance prediction map, includes:

[0108] Calculate the performance degradation curve of the wear-kinetic correlation matrix;

[0109] Calculate the first and second rates of change of the performance degradation curve to obtain the degradation rate characteristics;

[0110] Critical inflection point identification is performed on the decay rate characteristics to obtain the decay acceleration inflection point;

[0111] Wear type correction is performed on the attenuation acceleration inflection point using the wear mode characteristic spectrum, resulting in a mode adjustment curve;

[0112] Set the performance critical threshold based on the mode adjustment curve;

[0113] A multidimensional critical evaluation is performed based on the mode adjustment curve and the performance critical threshold to obtain the performance critical time point;

[0114] A dynamic performance prediction graph is generated based on the performance critical time point and the performance decay curve.

[0115] In this embodiment of the invention, when calculating the performance degradation curve of the wear-kinetic correlation matrix, key dynamic performance indicators are first determined, including curve guiding performance. Snake-like stability and shock response performance For each performance index, the corresponding influence coefficient submatrix is ​​extracted from the wear-kinetic correlation matrix. Based on the current wear state and wear development rate, a wear state prediction model is constructed. The wear state prediction adopts an exponential growth model. ,in for The degree of wear and tear over time, Based on the current level of wear, This is the wear rate coefficient, obtained through regression from historical data. For polygonal wear, =0.015 / day; for wavy wear, =0.012 / day; for rim wear, =0.02 / day; for tread wear, =0.01 / day. Predicted wear status. Substitute the wear-kinetic correlation matrix to calculate the future. Performance metrics at any time ,in The initial performance value is set to 1. The performance degradation curve is obtained by calculating from 0 to 90 days in 1-day increments. The performance degradation curve reflects the trend of dynamic performance degradation as wear progresses, providing basic data for critical point identification.

[0116] When calculating the first and second rates of change of the performance degradation curve, the performance degradation curve... Calculating the first derivative using the central difference method ,in The time step is 1 day. First derivative. This represents the rate of performance degradation, measured in performance per day. (The first derivative is used to express this rate of degradation.) The central difference method is applied again to calculate the second derivative. Second derivative This represents the rate of change of performance degradation, expressed in performance / day². To reduce noise introduced by numerical differentiation, the first and second derivatives are smoothed using a Savitzky-Golay filter with a window length of 7 and a polynomial order of 3. The first derivative... and second derivative Combined into decay rate feature vector The decay rate feature vector comprehensively characterizes the dynamic characteristics of performance degradation, including decay rate and acceleration, providing a basis for identifying critical inflection points.

[0117] When identifying the critical inflection point of the decay rate characteristic, the first step is to use the second derivative. Find the minimum points on the curve; these points correspond to the moments when the decay acceleration is greatest, i.e., the critical inflection points in the decay process. Set the threshold condition: P''(t) < -ε and For local minimum, where This is a sensitivity parameter, with a value of 0.0001. The time point at which the condition is met. These are marked as candidate inflection points. An importance score is calculated for each candidate inflection point. A higher score indicates a more important inflection point. Candidate inflection points are ranked according to their importance scores, and the top three points with the highest scores are selected as the decay acceleration inflection points t_c1. and For each inflection point, record its time. Performance values First derivative and second derivative Forming inflection point feature vectors The inflection point of accelerated degradation marks the beginning of a rapid deterioration phase in performance, and is a critical time for maintenance decisions.

[0118] When correcting the wear type based on the attenuation acceleration inflection point using the wear mode feature spectrum, the dominant wear type is first extracted from the wear mode feature spectrum. and severity Set the correction factor according to the dominant wear type. For polygonal wear, =1.2; For wavy wear, =1.1; For rim wear, =1.3; for tread wear, =1.0; For combined wear, =1.5. The correction factor reflects the difference in the accelerating effect of different wear types on performance degradation. It varies depending on the severity of wear. Set correction index , The value range is [0.5, 1]. The time of each attenuation acceleration inflection point is corrected, , wherein is the correction coefficient corresponding to the wear type. The corrected inflection point time is usually earlier than the original inflection point time , which reflects the wear acceleration effect. According to the corrected inflection point time, the corresponding performance value, first derivative and second derivative are recalculated to form the corrected inflection point feature vector . The original performance attenuation curve is combined with the corrected inflection point to generate a mode adjustment curve by piecewise cubic spline interpolation. The mode adjustment curve takes into account the acceleration effect of the specific wear mode, improving the accuracy of performance prediction.

[0119] When setting the performance critical threshold according to the mode adjustment curve, first determine the safety limit and comfort limit of each performance index. For curve-oriented performance, the safety limit = 0.6, and the comfort limit = 0.75; for snake stability, the safety limit = 0.5, and the comfort limit = 0.7; for impact response performance, the safety limit = 0.55, and the comfort limit = 0.8. According to the dominant wear type and severity in the wear mode feature spectrum, the critical threshold is adjusted. The adjustment formula is , wherein is the performance critical threshold, the actual threshold value obtained by adjusting the basic threshold and the wear severity , which is used as the judgment standard for determining whether the performance reaches the critical state, is the adjustment coefficient, for the safety limit = 0.1, and for the comfort limit = 0.15. When <0.5, the threshold is increased, indicating tolerance for slight wear; when S>0.5, the threshold is reduced, indicating vigilance for severe wear. For combined wear, the threshold is additionally reduced by 0.05, reflecting the combined risk of multiple wear coexistence. The adjusted safety threshold and comfort threshold are finally obtained, which constitute the performance critical threshold for subsequent critical time point determination.

[0120] When performing multi-dimensional critical evaluation according to the mode adjustment curve and the performance critical threshold, first combine the mode adjustment curve with the safety threshold and comfort threshold Compare, solve equation and , get safety critical time and comfort critical time . For each performance indicator (curve orientation performance, snake stability and impact response performance), calculate its safety critical time and comfort critical time respectively. Determine the comprehensive safety critical time and the comprehensive comfort critical time . According to the urgency of critical time, set maintenance priority: when <7 days, set as high priority; when 7 days≤ <14 days, set as medium priority; when ≥14 days, set as low priority. Combine the comprehensive safety critical time , the comprehensive comfort critical time and the maintenance priority into the performance critical time point. The performance critical time point determines the time window of maintenance intervention, and provides a scientific basis for maintenance decision.

[0121] When generating the dynamic performance prediction graph according to the performance critical time point and the performance degradation curve, first construct the graph coordinate system, with the horizontal axis as time (unit: days) and the vertical axis as performance indicator P (value range [0, 1]). Draw the original performance degradation curve and the mode adjustment curve , use different colors to distinguish, the original curve is blue and the adjustment curve is red. Mark the acceleration inflection point of degradation on the graph, represented by a yellow dot. Draw the safety threshold and the comfort threshold , represented by horizontal dashed lines, the safety threshold is red and the comfort threshold is yellow. Mark the comprehensive safety critical time and the comprehensive comfort critical time , represented by vertical dashed lines, the safety critical time is red and the comfort critical time is yellow. Add performance interval bands to the graph, divide the performance value into excellent interval , warning interval and danger interval , represented by green, yellow and red undercoat respectively. Add the prediction uncertainty interval, which is centered on the mode adjustment curve and floats up and down by ±10%, represented by gray shading. Label key information on the graph, including the dominant wear type, severity, safety critical time, comfort critical time and maintenance priority. Save the processed graph as the dynamic performance prediction graph, which visually displays the performance degradation trend, critical time point and maintenance urgency, providing comprehensive visual support for maintenance decision.

[0122] Preferably, step S5 comprises the following steps:

[0123] Step S51: evaluating the maintenance urgency and type of the wheel and rail by using the wear pattern feature spectrum and the dynamic performance prediction map, and generating a maintenance requirement list;

[0124] Step S52: obtaining a train operation plan, and identifying a feasible maintenance window set in combination with the maintenance requirement list;

[0125] Step S53: optimizing maintenance parameters according to the feasible maintenance window set and the wear pattern feature spectrum, to obtain a technical parameter scheme;

[0126] Step S54: calculating the comprehensive benefits of the coordinated implementation of wheel reprofiling and rail grinding according to the technical parameter scheme, to obtain a coordinated benefit evaluation table;

[0127] Step S55: generating train maintenance parameters by using the coordinated benefit evaluation table and the technical parameter scheme.

[0128] In the embodiment of the present application, when evaluating the maintenance requirement by using the wear pattern feature spectrum and the dynamic performance prediction map, the dominant wear type and the severity are first extracted. The maintenance type is determined according to the wear type: polygonization, wheel flange and tread wear correspond to wheel reprofiling; wavy wear corresponds to rail grinding; and combined wear corresponds to coordinated implementation. The critical time and are extracted from the performance prediction map, and the urgency is determined accordingly: <7 days for "urgent"; 7-14 days for "medium"; 14-30 days for "general"; and >30 days for "routine". The combined wear additionally increases the urgency by one level. The maintenance depth is determined according to the wear severity S: S<0.3 for "mild"; 0.3-0.6 for "moderate"; and >0.6 for "deep". These information is combined into a maintenance requirement item , and a maintenance requirement list sorted by urgency is generated after evaluating all vehicles and line sections .

[0129] When identifying the maintenance window by obtaining the train operation plan, the train schedule within 30 days is obtained from the schedule. For the wheel reprofiling requirement, the window with continuous stop time exceeding 4 hours is identified; for the rail grinding requirement, the window with continuous train passing exceeding 2 hours is identified; and for the coordinated implementation requirement, the overlapping window meeting both requirements is found. Considering the resource constraints of the reprofiling equipment (2 units) and the grinding equipment (1 unit), the tasks are sorted by urgency. The window set is generated for each maintenance requirement, ensuring that the window start time is earlier than , and finally forming the feasible maintenance window set W.

[0130] According to the window set and wear characteristics, the optimization target and constraint condition are determined. The wheel rotation repair parameters include cutting depth , feed speed and profile type; the rail grinding parameters include grinding times , angle sequence and pressure . According to the wear type and severity, the initial values are set: the polygonal wear cutting depth is (mm); the wheel rim wear is (mm); the tread wear is (mm). The wave wear grinding times is times. The total maintenance time is ensured not to exceed the window length, the multi-objective optimization algorithm is used to solve the optimal parameters, and the technical parameter scheme is formed.

[0131] When calculating the synergistic implementation benefits, an evaluation index system is established: performance improvement benefit , wherein is the weight coefficient of performance improvement, is the performance difference before and after maintenance, representing the performance improvement value brought by maintenance; life extension benefit , wherein is the weight coefficient of life extension, is the expected life difference before and after maintenance, representing the use life extension value brought by maintenance; resource saving benefit , wherein is the weight coefficient of resource saving, is the resource consumption difference before and after maintenance, representing the resource (such as energy, material) saving value brought by maintenance; operation impact cost , wherein is the weight coefficient of operation impact, is the operation interruption degree caused by maintenance, representing the negative impact cost of maintenance activities on normal operation. The comprehensive benefit and the return on investment are calculated, wherein is the maintenance cost, and the synergistic benefit evaluation table is formed.

[0132] When generating maintenance parameters, the top 20% of the schemes with the highest comprehensive benefits are selected as the candidate set, the parameters and window information thereof are extracted, the standardized maintenance instructions are generated, including maintenance objects, types, times, locations, parameters, quality requirements and acceptance standards. The operation guide for the synergistic task is generated, the instructions are converted into a device executable format, the effect prediction information is added, and the maintenance parameter package is integrated.

[0133] Therefore, the embodiments should be regarded, at any point, as being exemplary and not limiting, the scope of the application being defined by the appended claims and not by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

[0134] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A train operation and maintenance method based on digital twins, characterized in that, Includes the following steps: Step S1: Collect vibration and acoustic sensor data from the wheel-rail interaction area, and fuse them with train operation data to obtain wheel-rail interaction data; Step S2: Perform adaptive filtering on the wheel-rail interaction data to obtain the filtered contact characteristics; Train speed data and wheel diameter are extracted from wheel-rail interaction data, and rolling path vibration data are calculated. Local spectral entropy sequence is extracted from rolling path vibration data using filtered contact features. Lateral contact width is inverted based on local spectral entropy sequence to obtain contact width sequence. Initial contact distribution matrix is ​​constructed from contact width sequence and rolling path vibration data. Two-dimensional morphological pixel generation is performed on initial contact distribution matrix in digital twin space to obtain contact point distribution map. The pressure field is calculated based on the contact point distribution map to obtain the contact pressure distribution map; the equivalent taper is derived based on the contact pressure distribution map and the contact point distribution map, and a wheel-rail contact digital twin model is constructed to generate a contact geometric feature map; Step S3: Analyze potential wear modes based on the contact geometry feature map and identify early wear characteristics; calculate the wear ratio and distribution characteristics of the wheel flange / tread based on the early wear characteristics, and generate a wear mode feature spectrum; Step S4: Analyze different dynamic behaviors of the wheel-rail interaction data to obtain various dynamic response data; extract the lateral force characteristics of the wheel when passing through different radii of curvature from the wheel-rail interaction data to obtain the curve guiding force spectrum; perform digital twin dynamics hunting stability assessment on the wheel-rail interaction data to obtain the hunting stability index; identify the impact response characteristics of the wheel passing through the track in the wheel-rail interaction data to obtain the impact response spectrum; convert the wear mode feature spectrum, curve guiding force spectrum, hunting stability index, and impact response spectrum into a unified standard interval to obtain a normalized feature set; calculate the correlation between a single wear mode and each dynamic index in the normalized feature set to obtain the single-mode influence coefficient; acquire train speed data, and perform speed interval layering based on the single-mode influence coefficient to obtain the sub-speed hierarchy relationship; calculate the interaction effect when multiple wear modes coexist in the normalized feature set to obtain the wear combination coefficient; construct a wear-dynamic correlation matrix of digital twin mapping based on the single-mode influence coefficient, sub-speed hierarchy relationship, and wear combination coefficient; identify the performance degradation critical point of the wear-dynamic correlation matrix and generate a dynamic performance prediction map in the digital twin environment; Step S5: Obtain the train operation plan and determine the timing of train maintenance; determine the maintenance technical parameters based on the dynamic performance prediction diagram, and generate train maintenance parameters in combination with the timing of train maintenance.

2. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain and deploy vibration and acoustic sensors based on train configuration information and track geometry data to obtain a sensor point layout diagram; Step S12: Detect wheel passage events based on the sensor point layout diagram to obtain wheel passage event segments; Step S13: Obtain train operation parameters and fuse them with wheel-rail interaction data through the event segment to obtain wheel-rail interaction data.

3. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S2, which involves adaptive filtering of the wheel-rail interaction data, includes: The wheel-rail interaction data is decomposed into multiple frequency bands, and the frequency bands related to contact are identified to obtain the frequency band feature set; Contact feature mapping is performed on the frequency band feature set to obtain a feature mapping table; Extract operating condition feature vectors from the frequency band feature set; The basic filtering parameters are adaptively selected based on the feature mapping table and the operating condition feature vector; Frequency band selective processing is performed based on the basic filter parameters and frequency band feature set to obtain the filter contact characteristics.

4. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S2, which involves calculating the pressure field based on the contact point distribution diagram, includes: Train load data is extracted from wheel-rail interaction data, and wheel load distribution is calculated by combining the contact point distribution map to obtain the single wheel load value; Determine the effective contact area from the contact point distribution diagram; Extract the brightness value matrix from the contact point distribution map and combine it with the effective contact area to generate the pressure coefficient matrix; The absolute pressure value is calculated based on the pressure coefficient matrix, single wheel load value, and effective contact area to obtain the pressure numerical matrix. A digital twin pressure field visualization of the pressure numerical matrix is ​​performed to obtain a contact pressure distribution map.

5. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Detect geometric feature changes in the contact geometric feature map to obtain the amount of geometric change; Step S32: Compare the geometric changes with a preset library of typical wear patterns to obtain a pattern similarity table; Step S33: Analyze the periodic contact changes in the wheel circumference direction of the train based on the pattern similarity table and contact geometry feature map to obtain the polygonization index; Step S34: Analyze the longitudinal ripples of the train tread based on the pattern similarity table and contact geometry feature map to obtain the waviness index; Step S35: Evaluate the wear of the wheel flange and tread based on the contact geometry feature diagram to obtain the wear rate of the wheel flange and tread; Step S36: Determine the wear mode characteristic spectrum based on the polygonization index, waviness index, and rim and tread wear rates.

6. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S4 involves identifying the performance degradation critical point of the wear-kinetic correlation matrix and generating a kinetic performance prediction map, including: Calculate the performance degradation curve of the wear-kinetic correlation matrix; Calculate the first and second rates of change of the performance degradation curve to obtain the degradation rate characteristics; Critical inflection point identification is performed on the decay rate characteristics to obtain the decay acceleration inflection point; Wear type correction is performed on the attenuation acceleration inflection point using the wear mode characteristic spectrum, resulting in a mode adjustment curve; Set the performance critical threshold based on the mode adjustment curve; A multidimensional critical evaluation is performed based on the mode adjustment curve and the performance critical threshold to obtain the performance critical time point; A dynamic performance prediction graph is generated based on the performance critical time point and the performance decay curve.

7. The train operation and maintenance method based on digital twins according to claim 1, characterized in that, Step S5 includes the following steps: Step S51: Use wear mode characteristic spectrum and dynamic performance prediction map to assess the urgency and type of maintenance for wheels and rails, and generate a maintenance requirement list; Step S52: Obtain the train operation plan and identify a set of feasible maintenance windows based on the maintenance requirements list; Step S53: Optimize maintenance parameters based on the feasible maintenance window set and wear mode feature spectrum to obtain a technical parameter scheme; Step S54: Calculate the comprehensive benefits of the coordinated implementation of wheel overhaul and rail grinding based on the technical parameter scheme, and obtain the coordinated benefit evaluation table; Step S55: Generate train maintenance parameters using the synergy benefit assessment table and technical parameter scheme.

Images