Motor train unit wheel set turning repair management method and system

Abstract

The invention relates to a motor train unit wheel set turning repair management method and system. The method comprises the following steps that the actually-measured profile, the wheel diameter value, the transverse movement amount information and the running mileage data of a motor train unit wheel set are collected; based on the collected data, establishing a nonlinear mapping model of the wheel flange abrasion area and the running mileage by utilizing a Gaussian process regression model, and outputting a wheel flange abrasion trend curve and a confidence interval; a multi-target optimization model with the longest total service life mileage of wheels, the maximum unit cost mileage and the most uniform turning repair interval as targets is constructed; the multi-objective optimization model is solved, and an optimized turning repair strategy containing the coarse scale template number is obtained; performing fine scale mapping on the coarse scale template number based on a template recommendation engine to obtain a final turning repair strategy; and the final turning repair strategy is pushed to a lathe numerical control system and used for turning repair of the wheel with the abnormal rim abrasion rate. Compared with the prior art, the method has the advantages of realizing balance between economical efficiency and safety and the like.

Description

Technical Field

[0001] This invention relates to the field of high-speed train operation and maintenance technology, and in particular to a method and system for managing the turning and repair of train wheelsets. Background Technology

[0002] Wheelsets are the only components of a high-speed train in contact with the track, and their wear condition directly determines driving safety, ride quality, and maintenance costs. Existing wheel-turning strategies suffer from the following three major drawbacks:

[0003] 1) Static threshold: Fixed flange thickness or fixed running mileage is commonly used as the turning trigger condition, ignoring the actual measured profile difference, resulting in the coexistence of "over-turning" or "under-turning";

[0004] 2) Coarse template: Only 2mm or 4mm template steps are provided, which cannot accurately match the actual wear, resulting in a waste of 6-10mm of wheel diameter and a reduction of more than 15% in wheel life.

[0005] 3) Lack of economic viability and no quantitative optimization

[0006] Existing revetting strategies rely solely on fixed mileage or thickness thresholds for decision-making, without establishing a multi-dimensional objective function of "lifetime-cost-uniformity". This makes it impossible to quantitatively evaluate the comprehensive economics of different revetting nodes, resulting in the remaining mileage after the last revetting often being too long or too short, which disrupts maintenance scheduling and increases driving safety hazards.

[0007] A search revealed that Chinese invention patent application publication number CN112115581A discloses an analysis and prediction algorithm for wheel life, comprising four parts: first, modeling and solving wheel diameter wear based on wheel turning data; second, modeling and solving wheel flange wear based on wheel turning data; third, modeling and solving wheel turning based on turning data; and finally, constructing a wheel life limit prediction calculation model based on the above wheel wear turning data analysis results. In actual execution, constrained by the flange thickness limit, the flange wears continuously, and when its thickness degrades to a certain limit, subsequent routine or advanced repairs will cause additional wheel diameter loss due to flange thickness compensation. This wheel diameter loss is usually much greater than the relatively fixed amount of daily turning, which also has a significant impact on the wheel life limit. In this invention, this wheel diameter loss is also included in the wheel life prediction model. In addition, the actual turning execution error is also considered among the relevant influencing factors for wheel life prediction. The existing patent application fails to predict wheel turning from multiple perspectives, thus limiting the economic feasibility of the prediction.

[0008] How to achieve the most economical wheel turning and repair management has become a technical problem that needs to be solved. Summary of the Invention

[0009] The purpose of this invention is to overcome the defects of the prior art and provide a method and system for the maintenance and management of wheelsets of high-speed trains.

[0010] The objective of this invention can be achieved through the following technical solutions:

[0011] According to one aspect of the present invention, a method for managing the wheelset turning and repair of high-speed trains is provided, the method comprising the following steps:

[0012] Collect measured profiles, wheel diameters, lateral displacement information, and mileage data of EMU wheelsets;

[0013] Based on the collected data, a nonlinear mapping model between the wear area of ​​the wheel flange and the running mileage is established using a Gaussian process regression model, and the wear trend curve and confidence interval of the wheel flange are output.

[0014] A multi-objective optimization model is constructed with the objectives of maximizing the total wheel life mileage, maximizing the unit cost mileage, and achieving the most uniform refining interval. By combining the wheel flange wear trend and confidence interval, the multi-objective optimization model is solved to obtain an optimized refining strategy that includes coarse scale template numbers.

[0015] Based on the template recommendation engine, the coarse scale template number is mapped to a fine scale, and the corresponding fine scale template number is matched to obtain the final refining strategy containing the fine scale template number.

[0016] The final turning strategy is pushed to the turning machine CNC system for turning wheels with abnormal flange wear rates.

[0017] Preferably, the process of establishing the nonlinear mapping model includes: inputting the train's running mileage, line curve radius, axle position, and train type to a Gaussian process regression model, and outputting the wheel flange wear trend curve and confidence interval of the train's wheelset;

[0018] The hyperparameters of the Gaussian process regression model include the covariance function, signal variance, noise variance, and length scale. The correlation strength between the wear area of ​​the wheel flange at two running mileage positions of the EMU wheelset is used as the covariance function, the dynamic range of the wear area changing with mileage is used as the signal variance, the comprehensive uncertainty introduced by the trackside laser measurement, signal transmission and digital filtering is used as the noise variance, and the running mileage is used as the length scale.

[0019] More preferably, the hyperparameters are automatically estimated by maximizing the marginal log-likelihood function, so that the error between the rate of change of the rim wear area and the true value is minimized.

[0020] Preferably, the process of solving the multi-objective optimization model includes: inferring the mileage corresponding to the limit value of the wheel flange wear area and the lower limit of wheel life in the worst case based on the wheel flange wear trend curve and confidence interval, and solving the turning mileage node and wheel diameter cutting amount that satisfy the constraints of the multi-objective optimization model.

[0021] More preferably, the constraints of the multi-objective optimization model include:

[0022] The wheel diameter is not less than the wheel diameter threshold.

[0023] The difference in wheel diameter on the same axis does not exceed the wheel diameter difference threshold.

[0024] The envelope ratio of the profile is not lower than the envelope ratio threshold;

[0025] The coarse scale template number must be one or more combinations of 28mm, 30mm, and 32mm;

[0026] The standard deviation of the mileage interval between each overhaul is the smallest.

[0027] Preferably, the coarse scale template number is converted into a fine scale template number from the 0.1mm subdivision template library that meets the mapping conditions, wherein the mapping conditions must simultaneously meet the following conditions:

[0028] 1) The closest coarse scale template number;

[0029] 2) The proportion of the measured envelope profile is not lower than the envelope rate threshold;

[0030] 3) Minimum wheel diameter cutting amount.

[0031] Preferably, the final turning strategy includes: turning mileage node, coarse scale template number, and wheel diameter cutting amount.

[0032] Preferably, the process of identifying wheels with abnormal flange wear rates by voting includes: detecting the flange wear rate of the wheel using the K-sigma criterion, HBOS, and local outlier factor respectively; if at least two of these methods determine that the flange wear rate of the wheel is abnormal, then the flange wear rate of the wheel is determined to be abnormal.

[0033] According to another aspect of the present invention, a trainset wheelset turning and maintenance management system is provided, the system comprising:

[0034] The sensing layer collects data on the measured profile, wheel diameter, lateral displacement, and mileage of the train wheelsets.

[0035] The computational layer, based on the collected data, as well as the total train set information and track line information, establishes a nonlinear mapping model between wheel flange wear area and running mileage, and outputs the wheel flange wear trend curve and confidence interval.

[0036] The control layer is used to establish a multi-objective optimization model and solve the multi-objective optimization model by combining the rim wear trend curve and confidence interval to obtain the final refinishing strategy;

[0037] The evaluation layer is used to identify wheels with abnormal rim wear rates, visualize the results of the calculation layer, and display the refinishing plan and list of abnormal wheels.

[0038] Preferably, the sensing layer includes a data acquisition module for collecting data on the wheelsets of the high-speed train, and a preprocessing module for processing the collected data;

[0039] The computational layer includes a wear law modeling module, which uses a Gaussian process regression model to establish a nonlinear mapping model between the wear area of ​​the wheel flange and the running mileage, and outputs the wear trend curve and confidence interval of the wheel flange.

[0040] The control layer includes an economic optimization decision module and a template recommendation engine. The economic optimization decision module is used to obtain an optimization repair strategy containing coarse-scale template numbers for the constructed multi-objective optimization model. The template recommendation engine is used to map coarse-scale template numbers to fine-scale template numbers and generate a final repair strategy containing fine-scale template numbers.

[0041] The evaluation layer includes an anomaly detection module, a visual management and control platform, and a digital twin verification module.

[0042] Anomaly detection module: It integrates three algorithms: K-sigma criterion, HBOS and local outlier factor. If at least two of these methods determine that the wheel flange wear rate is abnormal, then the wheel flange wear rate is determined to be abnormal.

[0043] Visualized management and control platform: Real-time display of the entire train's wear heat map, life prediction curve, Gantt chart of the refining plan, and list of abnormal wheels;

[0044] Digital twin verification module: Establish a virtual wheelset, import actual profile data to replay the life evolution process of the wheelset under different refinishing strategies, and intuitively verify whether the life extension rate meets the standard.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1) This invention utilizes a Gaussian process regression model to establish the rim wear trend curve and confidence interval; solves a multi-objective optimization model to quickly obtain an optimized turning strategy; matches the coarse scale template number of the optimized turning strategy with the corresponding fine scale template number; and uses the final turning strategy to perform fine turning on wheels with abnormal rim wear rates. The multi-objective optimization model aims to maximize the total wheel life mileage, the unit cost mileage, and the turning interval, achieving a balance between economy and safety, and solving the problem that previous turning schemes only considered a single objective and did not consider economy.

[0047] 2) This invention uses a Gaussian process regression model to output the wheel flange wear trend curve and confidence interval of the EMU wheelset, which serves as the basis for subsequent optimization of the refinishing strategy and also provides a baseline for identifying wheels with abnormal wheel flange wear rates. Compared with baselines set based on experience, the identification accuracy is higher. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the structure and method flow of the EMU wheelset turning and repair management system in this invention;

[0049] Figure 2 This is a schematic diagram showing the relationship between the wear area of ​​the wheel rim and the operating mileage in this invention;

[0050] Figure 3 This is a schematic diagram of the wear prediction confidence interval in this invention;

[0051] Figure 4 This is a schematic diagram illustrating the relationship between the repair nodes and the unit cost mileage in the digital twin lifetime verification of this invention.

[0052] Figure 5 This is a schematic diagram showing the relationship between the repair nodes and the total lifespan mileage in the digital twin lifespan verification of this invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0054] Example 1

[0055] This embodiment relates to a wheelset turning and maintenance management system for high-speed trains, such as... Figure 1 The system adopts a four-layer architecture consisting of a perception layer, a computing layer, a control layer, and an evaluation layer to achieve intelligent decision-making for wheelset maintenance throughout its entire lifecycle.

[0056] 1. The perception layer performs data acquisition and preprocessing, including a data acquisition module and a preprocessing module.

[0057] Data acquisition module: By deploying a high-precision laser profile measuring device by the trackside, the laser profiler can acquire the measured profile, wheel diameter and lateral displacement data of the entire wheelset in real time each time the train passes by. The repeatability of the device is better than 0.05mm, providing high-quality raw data for subsequent analysis.

[0058] Preprocessing module: Performs profile denoising, alignment, missing value interpolation and standardization on the collected data to ensure that the profile data quality meets the requirements of subsequent modeling and optimization analysis.

[0059] The standardized processing includes the following, which is used for subsequent wheel anomaly identification:

[0060] Calculate the rate of change (wear / mileage) of the wheel flange wear area within the current sliding window;

[0061] Calculate the mean and standard deviation of the wear rate samples of all wheels on the entire train within the same window.

[0062] Calculate the HBOS anomaly score of the wheel;

[0063] Calculate the observed value of the i-th wheel on the j-th index;

[0064] Calculate the frequency percentage of the j-th indicator;

[0065] Calculate the local density of the sample.

[0066] 2. Computational layer, used for modeling profile wear patterns and predicting wheel life, including:

[0067] The wear pattern modeling module uses a Gaussian process regression (GPR) composite kernel function to establish a nonlinear mapping model between the wear area of ​​the wheel flange and the mileage. It can output the wear trend curve and confidence interval of the wheel flange, and realize the early prediction of wheel life.

[0068] 3. Control layer, used for economic optimization and template recommendation, including:

[0069] Economic optimization decision module: The NSGA-Ⅲ three-objective optimization algorithm is adopted to establish a multi-objective optimization model with the objectives of maximizing the total wheel life mileage, minimizing unit cost (i.e. maximizing unit cost mileage) and maximizing the uniformity of the turning interval, and solves the optimal turning mileage node and the optimal coarse scale template number.

[0070] Template Recommendation Engine: From a 0.1mm-level subdivision template library of 28mm, 30mm, and 32mm templates, the system intelligently matches the optimal precision template number based on the principles of maximizing profile envelope and minimizing wheel diameter loss, ensuring minimal wheel diameter loss and a profile envelope ≥95%. Lateral displacement determines the lateral position of the wheel-rail contact point, affecting the matching degree between the template and the measured profile. Through lateral displacement range scanning, the system can determine whether the candidate template can "enclose" the measured profile within the entire lateral displacement range, avoiding local "exposed areas" or steps.

[0071] 4. Evaluation layer, used for anomaly detection, visualization, and digital twins, including:

[0072] Anomaly detection module: It integrates three algorithms: K-sigma criterion, HBOS (Histogram-based OutlierScore) histogram scoring and Local Outlier Factor (LOF) to identify wheels with abnormal flange wear rate and trigger early warning.

[0073] Visualized management and control platform: Based on WebGL, it displays the wear heat map, life prediction curve, Gantt chart of the turning and repair plan and list of abnormal wheels of the whole train in real time, and supports SMS / WeChat push of abnormal information.

[0074] Digital twin verification module: Create a virtual wheelset in Unity3D, import actual profile data to replay the life evolution process of the wheelset under different refinishing strategies, and intuitively verify whether the life extension rate meets the standard.

[0075] Through the above four-layer collaboration, the system realizes intelligent, economical and visualized wheelset turning repair, significantly improving wheel life and reducing maintenance costs.

[0076] In the computational layer, the wear pattern modeling module uses Gaussian process regression (GPR), taking "mileage + curve radius + axle position + vehicle type" as joint input, and outputting the probability distribution of flange wear area and wheel diameter loss at any future mileage (i.e., flange wear trend curve and confidence interval, such as...). Figure 2 and Figure 3 The curve radius is obtained from the track information, while the axle position and train type are obtained from the EMU information.

[0077] The Gaussian process regression model used in the wear pattern modeling module has a composite kernel structure, including a radial basis function (RBF) kernel and a white noise kernel. When building a model (e.g., Gaussian process regression), the model itself has some parameters that need to be set in advance; these parameters are called "hyperparameters" (e.g., the signal variance in the kernel function). Length scale L, noise variance These hyperparameters cannot be directly calculated from the training data, but they significantly affect the model's predictive performance. We employ Maximum Likelihood Estimation (MLE) to automatically estimate the optimal values ​​of these hyperparameters from the observed data, ensuring the model best fits the data without requiring manual trial and error adjustments. The hyperparameters are automatically optimized using MLE.

[0078] Choose a composite kernel function:

[0079]

[0080] Wherein, k(l,l′) is the covariance function (kernel function), which describes the correlation strength between the wear areas of the wheel rim at two running mileage positions l and l′. The larger the value, the more "similar" the wear evolution is between the two locations.

[0081] (signal variance): A hyperparameter that controls the amplitude of the kernel function. The larger the value, the greater the dynamic range (fluctuation amplitude) of the wear area as a function of mileage is allowed. Physically, it reflects the "overall severity of the wear rate".

[0082] L (length-scale): A hyperparameter that determines the relevant range of influence, referring to the operating mileage in this case. The smaller L is, the faster the covariance decays with the mileage difference, meaning that the wear curve can show a significant inflection point over a short distance; the larger L is, the smoother the curve; the unit is the same as mileage (km).

[0083] (noise variance): Characterizes measurement noise or unexplained variance in the model. The larger the value, the greater the deviation of the observed value from the regression curve, which can prevent overfitting; When the GPR approaches 0, it degenerates into an exact interpolation. In the scenario of monitoring the wheelset profile of high-speed trains, This is used to quantify the comprehensive uncertainties introduced by trackside laser measurement, signal transmission, and digital filtering. The larger the value, the higher the tolerance of the Gaussian process regression model to deviations from the fitted curve at the observation point, effectively suppressing the impact of measurement random errors on the estimation of wheel flange wear trends and reducing the risk of overfitting; when... As the value approaches zero, the Gaussian process regression degenerates into strict interpolation, forcing the curve to pass through every observation point, thus misinterpreting random fluctuations as true wear characteristics. Therefore, a reasonable setting... The key is to balance fitting accuracy and generalization ability. On-site, maximum marginal likelihood estimation is used for automatic optimization, so that the model can retain the main features of wear evolution while having appropriate robustness to measurement noise.

[0084] δ l,l′ (Kroneckerdelta): 1 when l = l′, 0 otherwise. A noise term is added only at the "autocovariance" position to ensure that the observed values ​​are fully correlated with themselves and contain noise.

[0085] MLE (Maximum Likelihood Estimation): Automatically estimates the probability of an observed vector (the wear area rate of the training set's wheel rims) by maximizing the marginal log-likelihood function. This log-probability is maximized by adjusting the model's hyperparameter θ.

[0086]

[0087] θ={σ f , L,σ n No manual settings are required.

[0088] In the formula: y is the observation vector (the rate of change of the wear area of ​​the wheel rim in the training set); X is the input matrix (corresponding to mileage or other features); θ is the set of hyperparameters to be optimized, including the signal variance σ. f Length scale L and noise standard deviation σ n K: Covariance matrix (kernel function value), K ij =k(x i ,x j |θ); n: Number of training samples (number of wheels × number of mileage points); is a normalization constant, independent of the parameters, and can be ignored during optimization; I is the identity matrix.

[0089] In the control layer, the economic optimization decision module uses the NSGA-III algorithm to solve the Pareto front and automatically recommends the most balanced remediation strategy from the set of non-dominated solutions using the entropy-weighted TOPSIS method. Its objective function is:

[0090]

[0091] The constraints are:

[0092]

[0093] Among them, L total : The total mileage that the entire train wheelset can operate over its lifespan (including the remaining mileage before multiple overhauls and the final scrapping); C total Total cost over the life cycle, including all refinishing costs and the final wheel replacement cost; The unit cost per mileage reflects an economic indicator; the lower the value, the more economical it is. σ interval : Standard deviation of the mileage interval between each turning operation, used to measure the uniformity of the interval; the smaller the value, the more stable the production scheduling; D is the wheel diameter; ΔD 同轴 This represents the difference in diameter of the coaxial wheels.

[0094] The rim wear trend curve given by the Gaussian process regression model can be directly extrapolated to the mileage corresponding to the rim wear area reaching the limit, thus quantifying the remaining wheel life and the next mileage that can be operated. The confidence interval is used to calculate the lower limit of life under the worst case, providing a probabilistic basis for the cost-life trade-off. When solving the Pareto front, the NSGA-Ⅲ algorithm needs to repeatedly call the objective function for candidate re-repair mileage nodes. For each input of a turning mileage node, the rim wear trend curve immediately returns the corresponding wear area, wheel diameter cutting amount, and remaining life. The algorithm uses this information to determine the turning mileage node. In other words, the turning mileage node is essentially the result of a point-by-point scan and comparison on the rim wear trend curve.

[0095] Construct a "profile-template" mapping library, for a given wear profile P wear The optimal template number is determined to satisfy the constraints and minimize the wheel diameter ΔD to be removed during the wheel turning. The template recommendation engine supports a 0.1mm-level template subdivision library and sorts and recommends templates based on the dual criteria of "minimum cutting amount + maximum envelope rate". The recommended template order is 28mm, 30mm and 32mm to achieve minimum cutting.

[0096]

[0097] In the formula, T i Candidate turning template number, corresponding to rim thickness grades such as 28mm, 30mm, and 32mm; P wear : Actual wear profile data of the vehicle (two-dimensional coordinate point set); ΔD cut (P wear ,T i ): Turn the worn profile into a template T. i The minimum wheel diameter cutting amount (diameter loss) required at that time.

[0098] Envelope ratio greater than 95%: When recommending a refinishing template, it must be ensured that the selected template profile (standard new wheel flange curve) can be completely "enclosed" by the current measured wear profile for an area ≥95%. The envelope ratio is calculated using the upper confidence limit profile of the confidence interval in a nonlinear mapping model. That is, the measured worn wheel cross-section is considered as the outer contour; the candidate template is considered as the inner contour. Only when the inner contour falls within the outer contour for more than 95% of its arc length is the template considered to be able to enclose the wheel and can be used for refinishing. Otherwise, a thinner template (with a smaller flange thickness) is used to continue trying until the 95% envelope ratio is met. This prevents the template from being partially "exposed," which could lead to missing material, steps, or abnormal contact on the wheel after refinishing, ensuring driving safety and a smooth transition of the profile.

[0099] In the evaluation layer, the anomaly detection module includes:

[0100] a) Time series anomaly detection submodule based on K-sigma criterion;

[0101] like Marking an anomaly;

[0102] In the formula, The rate of change of the flange wear area of ​​the i-th wheel within the current sliding window (wear / mileage); μ: the mean of the wear rate samples of all wheels in the train within the same window; σ: the standard deviation of the wear rate samples of all wheels in the train within the same window; 3σ: three times the standard deviation threshold, the outlier boundary in the classic 3σ criterion.

[0103] b) An unsupervised scoring submodule based on HBOS;

[0104]

[0105] Among them, S i : HBOS anomaly score for the i-th wheel, the larger the value, the more abnormal; m is the number of indicators involved in the scoring (e.g., the rate of change of rim wear area, the rate of change of rolling circle radius difference (RRD)); x ij h: The observed value of the i-th wheel on the j-th index; j (x ij ) refers to the frequency percentage of the j-th indicator.

[0106] c) Multidimensional outlier identification submodule based on relative density (LOF);

[0107]

[0108] In the formula, LOF k (i): The local outlier factor of sample i, used to measure the density of sample i relative to its k nearest neighbors. A value greater than 1 indicates an outlier, and the larger the value, the more abnormal the sample i is. The larger the value, the more likely sample i is to be an outlier. ρ k (i,o): Local density of sample (i,o); ρ k (o): The local density of sample o, i.e., the density around sample o in the sample space, where o represents any sample in the k nearest neighbor of sample i; ρ k (i): The local density of sample i, i.e., the density around sample i in the sample space; N k (i) is the k-nearest neighbors set of sample i, where k is the algorithm hyperparameter, which is fixed and written into the configuration; i is the index of the wheel being evaluated.

[0109] All three anomaly detection methods must use the normal wear rate as a reference. The expected trend of the rim wear trend curve provides a baseline: when the measured rim wear area change rate deviates from the expectation by more than 3σ, and the local density is significantly lower than the neighborhood, it is judged as an anomaly. Without a baseline, the threshold can only be set empirically, making it difficult to control the false alarm rate and false negative rate. If at least two of the three anomaly detection methods indicate anomalies, then the rim wear rate of the wheel is judged to be abnormal.

[0110] The visual management and control platform supports the following functions:

[0111] 1) Batch import and automatic cleaning of multi-format profile data, supporting trackside measurements, JSON / CSV / TXT and database interfaces, with built-in coordinate flipping, baseline alignment and missing value imputation;

[0112] 2) Batch analysis of the profile geometry parameters of single wheels or entire vehicle convoys, which can output the rim thickness, height, qR value and lateral / vertical / normal wear area at one time, and the results can be exported with one click;

[0113] 3) Real-time calculation of wheel-rail contact geometry, providing the rolling circle radius difference, equivalent taper and contact point trajectory according to the set lateral step length, and plotting the RRD curve for diagnostic reference;

[0114] 4) Wheel flange wear trend monitoring and abnormal early warning, integrating K-sigma, HBOS and LOF algorithms, automatically marking wheels exceeding the limit in red, and supporting SMS and WeChat push notifications;

[0115] 5) Intelligent generation of economical repair plans, calling the "life-cost-uniformity" three-objective optimization kernel, and returning recommended repair mileage nodes, template numbers, and unit cost mileage;

[0116] 6) Batch evaluation of Hertzian contact stress: calculate the maximum contact stress based on the lateral displacement range and generate stress-displacement curves;

[0117] 7) Reverse optimization of turning cutting amount: Based on the minimum envelope principle, the optimal feed amount, cutting area and turning wheel diameter of each wheel of the same bogie are solved, and it is compatible with 0.1mm level template library;

[0118] 8) Automatic output of charts and reports, covering profile images before and after turning and three-dimensional wear evolution, all exported in high-definition 300dpi, which can be directly used for on-site decision-making and patent document preparation.

[0119] The digital twin verification module is built on the Unity3D engine. The Unity3D virtual wheelset drives wear evolution based on actual profile data, replaying life curves under different strategies. It supports the following functions: importing actual profile data to drive virtual wheelset wear evolution; simulating lifespan extension rates under different refinishing strategies; and outputting lifespan-cost curves and strategy scoring reports. The output is as follows:

[0120]

[0121] In the formula, η 延长 Total wheel life extension rate (percentage); L opt The total mileage of a wheelset from its first turning to its final operational status after the optimization of the turning strategy using this invention is L. fixThe total lifespan of the same vehicle model when using a traditional fixed-mileage (approximately 312,000 km) maintenance strategy.

[0122] The recommended turning strategy was fed back into the Unity3D digital twin and simulated 1 million km of operation, verifying an 8.5–12.4% lifespan extension and a 22% reduction in wheel diameter waste.

[0123] Example 2

[0124] This embodiment relates to a method for managing the wheelset turning and repair of high-speed trains, such as... Figure 1 This includes the following steps:

[0125] S1 collects the measured profile, wheel diameter, lateral displacement, and running mileage information of the EMU wheelsets through a trackside laser profile measurement device;

[0126] S2, preprocesses the collected data, including alignment, interpolation and standardization;

[0127] S3, a nonlinear mapping model between wheel rim wear area and running mileage is established based on Gaussian process regression;

[0128] S4. Construct a multi-objective optimization model with the objectives of total wheel life mileage, unit cost mileage, and uniformity of refinishing intervals.

[0129] S5 uses the NSGA-III algorithm to solve the multi-objective optimization model and obtain the optimized turning strategy. The optimized turning strategy includes: the optimal turning mileage node, i.e., the mileage node corresponding to the next (and subsequent) turning; the optimal coarse scale template number at this node (e.g., 28.3mm); and the minimum wheel diameter that must be removed in this turning (wheel diameter cutting amount). The optimal turning mileage node and template number; here the template number is the coarse scale template number, only selected from 28, 30, and 32mm, the purpose of which is to allow the multi-objective optimization to converge quickly.

[0130] S6, based on the template recommendation engine, matches the optimal turning template: Building upon S5, a fine-scale mapping is performed on the coarse-scale template number. The coarse-scale template is replaced with the fine-scale template number from the 0.1mm sub-library that is closest to the measured profile while still achieving 95% envelopment and minimizing wheel diameter cutting amount. This transforms the optimization results into data that can be immediately processed on-site. First, the thinnest (minimum cutting) template that meets the requirements of "envelopment rate ≥ 95%, wheel diameter ≥ 770mm, coaxial difference ≤ 0.3mm" is found. The exact wheel diameter cutting amount ΔD is calculated, generating a machining program that can be directly executed by the CNC turning machine. If no sub-template meeting the constraints is found in the 30mm range, it automatically downgrades to 29.9mm, 29.8mm... up to the 28mm series; conversely, if all 28mm series are overcut, the 30mm series is borrowed. The final output includes the optimized turning strategy, including the fine-scale template number, wheel diameter cutting amount, and turning mileage nodes.

[0131] S7 uses an anomaly detection module to identify wheels with abnormal wear rates;

[0132] S8 pushes the optimized turning strategy, after S6 precision mapping, to the turning machine CNC system and transmits back measured data for model closed-loop update. The optimized turning strategy includes: the mileage node corresponding to the next (and subsequent) turning; the precision scale template number (e.g., 28.3mm) after 0.1mm-level precision turning at that node; and the minimum wheel diameter that must be removed in this turning cycle.

[0133] The constraints of the multi-objective optimization model include:

[0134] 41) The wheel diameter is not less than the wheel diameter threshold (770mm);

[0135] 42) The diameter difference of the coaxial wheels does not exceed the wheel diameter difference threshold (0.3mm);

[0136] 43) The profile envelope ratio is not lower than the envelope ratio threshold (e.g., 95%);

[0137] 44) The coarse scale template number must be one or more combinations of 28mm, 30mm, and 32mm.

[0138] The closed-loop update mechanism includes:

[0139] The measured profile data after turning is fed back to the wear pattern modeling module;

[0140] Automatically update the hyperparameters of the Gaussian process regression model: re-estimate the hyperparameters in an incremental manner, and dynamically correct the flange wear trend curve as the actual wheel condition drifts; subsequent optimization and early warning strategies remain consistent with the field, realizing a closed-loop update of data, model, and decision;

[0141] And dynamically adjust the anomaly detection threshold and optimize the weight coefficients.

[0142] Example 3

[0143] This embodiment also relates to the application of a trainset wheelset turning and maintenance management system. To verify the applicability and economy of the system on actual lines, a CRH380B trainset was selected as the test object. Combined with a trackside laser profile measurement device, the measured profile data of the train was collected. Based on the four-layer architecture system of perception layer, calculation layer, control layer and evaluation layer proposed by the system, a complete closed-loop verification was carried out.

[0144] Step 1, Data Acquisition and Preprocessing

[0145] From January 2023 to February 2024, the system collected 1,128 measured profile data points from a CRH380B high-speed train, covering the wear status of the left and right wheelsets, different axle positions, and different operating mileages. The data preprocessing module performed denoising, alignment, interpolation, and standardization of the profile data to ensure that the data quality met the requirements for subsequent modeling and optimization analysis.

[0146] Step 2, Modeling and Predicting Wear Patterns

[0147] A Gaussian process regression (GPR) model is used, combined with multi-dimensional features such as mileage, curve radius, axle position, and vehicle type, to establish a nonlinear mapping model between wheel flange wear area and mileage, such as... Figure 2 As shown. This model employs a composite kernel function (RBF kernel + white noise kernel) and automatically optimizes hyperparameters through maximum marginal likelihood estimation, achieving a prediction accuracy of R0. 2 =0.93, with an average prediction error of less than 3.5%.

[0148] Step 3, Multi-objective optimization decision

[0149] Based on the NSGA-III algorithm, a three-objective optimization model is constructed with the goals of maximizing total wheel life, minimizing unit cost, and achieving the most uniform turning interval. The model inputs include:

[0150] Measured rim wear area curve;

[0151] Template turning wheel diameter loss function;

[0152] Constraints such as coaxial wheel diameter difference and profile envelope ratio;

[0153] Parameters for turning cost (200 yuan / wheel) and wheel replacement cost (8000 yuan / wheel).

[0154] The Pareto front is obtained by solving the system, and the optimal equilibrium strategy is recommended by the entropy-weighted TOPSIS method, as shown in Table 1.

[0155] Recommended turning nodes are 217,000 km, 250,000 km, and 293,000 km, corresponding to coarse scale template numbers 30→28→28→28→28→28. This means the entire lifespan is divided into four operating segments using three turning nodes.

[0156] First segment: 0km → 217,000km

[0157] Second segment: 217,000 km → 250,000 km

[0158] Third segment: 250,000 km → 293,000 km

[0159] Fourth paragraph: 293,000 km → Wheels worn out.

[0160] Only one resurfacing operation is scheduled at the end of each operating section, so the actual number of resurfacing operations on site is 4.

[0161] During economic optimization calculations, to incorporate both the wheel diameter remaining after the previous turning and the wheel diameter to be cut to in the next turning into the objective function, the program adds an extra "virtual" section at both the beginning and end, resulting in six coarse scale template numbers: 30→28→28→28→28→28. In other words, the six coarse scale template numbers represent the complete sequence used for calculation, while the four turning operations represent the sequence executed on-site. The two are automatically mapped through the interval start-end point.

[0162] Step 4: Template Recommendation and Profile Envelope Validation

[0163] The template recommendation engine, based on the principle of "minimum cutting amount + maximum envelope rate," dynamically matches precise-scale template numbers from the 28mm, 30mm, and 32mm template libraries to form the final turning strategy. The system sets the profile envelope rate to ≥95% to ensure a smooth profile transition after turning, without any steps or missing material. Verification was conducted using the Unity3D digital twin module to simulate the wheel life evolution under different strategies, confirming that the recommended turning strategy effectively extends wheelset life.

[0164] Step 5, Anomaly Detection and Early Warning

[0165] The system integrates three algorithms—K-sigma criterion, HBOS scoring, and LOF outlier factor—to monitor the rate of change of rim wear area in real time. During the test, six abnormally worn wheels were identified, one of which was marked in advance and scheduled for refinishing due to its rim wear area change rate exceeding the limit, thus avoiding potential safety risks in subsequent operation.

[0166] Step 6, Visualization and Strategy Push

[0167] Through the visual management and control platform, maintenance personnel can view the following in real time:

[0168] Wear thermogram;

[0169] Lifetime prediction curve;

[0170] Gantt chart of the repair plan;

[0171] List of abnormal wheels and warning information.

[0172] The system automatically generates a laminating schedule for the next 12 months and supports manual fine-tuning. After the strategy is confirmed, the system pushes the laminating nodes and template numbers to the laminating machine CNC system via an interface to achieve closed-loop execution.

[0173] The implementation results are compared in Table 1, and the template selection scheme for the resurfacing is shown in Table 2.

[0174] Table 1

[0175]

[0176] Table 2

[0177]

[0178] Each row in Table 1 represents a candidate strategy. The first column is the node (in ten thousand km) where the first resurfacing occurred, and the numbers 1 to 7 afterward indicate the template thickness (mm) selected for each resurfacing operation in sequence.

[0179] 197,000 km node strategy: A total of 6 turning operations were carried out, with 30mm used for the first operation and 28mm used for the subsequent 5 operations;

[0180] 224,000 km node strategy: A total of 5 turning repairs were carried out, the first one was 30 mm, and the subsequent 4 were 28 mm.

[0181] 258,000 km node strategy: A total of 4 turning repairs were carried out, the first one was 30 mm, and the subsequent 3 were 28 mm.

[0182] 300,000 km node strategy: Three turning operations were carried out in total, the first one was 30 mm, and the subsequent two were 28 mm.

[0183] The " / " indicates that under this strategy, no further turning will be arranged, and the wheel will be used directly until its life limit is reached and scrapped. After the system sends these parameters to the template recommendation engine, it will further refine the 28mm to a 0.1mm level (e.g., 28.3mm) and give the minimum wheel diameter cutting amount. Therefore, Table 1 is a "coarse template ladder" for rapid comparison in multi-objective optimization.

[0184] Table 2 quantifies the measured results of the four typical strategies in Table 1 into three core indicators:

[0185] 1. Total lifespan: The cumulative mileage a wheel can operate from its first turning to its final disposal.

[0186] 2. Cost efficiency: How many kilometers can be traveled per yuan (km / yuan);

[0187] 3. Lifespan extension rate: the percentage increase compared to the original fixed 312,000 km strategy.

[0188] The data comparison in Table 2 shows that: the earlier the first resurfacing is started (moving the node forward), the longer the total lifespan, but the number of resurfacing operations increases, and the cost efficiency decreases slightly; the 197,000 km strategy achieves the longest lifespan of 1,582,000 km, which is 8.5% longer than the original strategy, and the cost efficiency is still 3.5% higher than the original strategy; the 258,000 km strategy has the highest cost efficiency (3.59 km / yuan) while extending the lifespan by 6.2%, making it suitable for the "lifespan and cost trade-off" scenario; although the original strategy has the fewest resurfacing operations, it does not make full use of the remaining wear space, resulting in the shortest total lifespan and the lowest economic efficiency.

[0189] Therefore, the system can select any row in the Pareto frontier according to the operator's preference (prioritizing lifespan or cost), and automatically generate a turning calendar for the next 12 months after confirmation, and send it to the on-site CNC turning machine for execution.

[0190] This system has been validated on the CRH380B model. The relationships between maintenance milestones and unit cost mileage and total lifespan mileage are as follows: Figure 4 and Figure 5 As shown, from Figure 4 As can be seen, the horizontal axis represents the mileage at the first turning node (i.e., the number of kilometers driven corresponding to the first turning), and the vertical axis represents the unit cost mileage (km / yuan). The curve reaches a peak of 3.51km / yuan near 197,000 km, and then slowly decreases as the node moves forward. At the 300,000 km node, the cost efficiency is still higher than the original fixed strategy of 3.45km / yuan, indicating that appropriately advancing the first turning can significantly improve economic efficiency; turning too early (<190,000 km) or too late (>300,000 km) will reduce cost efficiency due to increased turning frequency or wasted wheel diameter.

[0191] from Figure 5 As can be seen, the horizontal axis represents the "re-turning milestone mileage," while the vertical axis represents the "total service life mileage." The total service life increases monotonically as the milestone is moved forward, reaching a maximum of 1,582,000 km at 197,000 km, which is about 8.5% higher than the original strategy. When the milestone is delayed to 300,000 km, the service life still remains at 1,549,000 km. This indicates that using early precision-calibrated templates can effectively extend the total service mileage of wheels and provides a degree of tolerance for the timing of the first re-turning, offering operating units a flexible range that balances service life and cost.

[0192] This system has good adaptability to various train models and supports flexible configuration of various flange thickness templates (28 / 30 / 32mm) and different tread types (such as DIN5573, LMA, XP55, etc.). It is suitable for various rail transit train models such as high-speed trains, urban subway trains, and trams.

[0193] This embodiment verifies the system's comprehensive advantages in terms of measured data-driven approach, economic optimization, and strategy executability. Compared to traditional fixed-mileage strategies, the system is significantly more effective in extending wheelset life, reducing wheel diameter waste, and improving operational efficiency, demonstrating good engineering application value.

[0194] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for managing the turning and maintenance of wheelsets on high-speed trains, characterized in that, The method includes the following steps: Collect measured profiles, wheel diameters, lateral displacement information, and mileage data of EMU wheelsets; Based on the collected data, a nonlinear mapping model between the wear area of ​​the wheel flange and the running mileage is established using a Gaussian process regression model, and the wear trend curve and confidence interval of the wheel flange are output. A multi-objective optimization model is constructed with the objectives of maximizing the total wheel life mileage, maximizing the unit cost mileage, and achieving the most uniform refining interval. By combining the wheel flange wear trend and confidence interval, the multi-objective optimization model is solved to obtain an optimized refining strategy that includes coarse scale template numbers. Based on the template recommendation engine, the coarse scale template number is mapped to a fine scale, and the corresponding fine scale template number is matched to obtain the final refining strategy containing the fine scale template number. The final turning strategy is pushed to the turning machine CNC system for turning wheels with abnormal flange wear rates.

2. The method for managing the turning and maintenance of wheelsets in high-speed trains according to claim 1, characterized in that, The process of establishing the nonlinear mapping model includes: inputting the train's running mileage, line curve radius, axle position, model type, and collected data into a Gaussian process regression model, and outputting the wheel flange wear trend curve and confidence interval of the train wheelset; The hyperparameters of the Gaussian process regression model include the covariance function, signal variance, noise variance, and length scale. The correlation strength between the wear area of ​​the wheel flange at two running mileage positions of the EMU wheelset is used as the covariance function, the dynamic range of the wear area changing with mileage is used as the signal variance, the comprehensive uncertainty introduced by the trackside laser measurement, signal transmission and digital filtering is used as the noise variance, and the running mileage is used as the length scale.

3. The method for managing the turning and maintenance of wheelsets in a high-speed train according to claim 2, characterized in that, The hyperparameters are automatically estimated by maximizing the marginal log-likelihood function, which minimizes the error between the rate of change of the rim wear area and the true value.

4. The method for managing the turning and maintenance of wheelsets in a high-speed train according to claim 1, characterized in that, The process of solving the multi-objective optimization model includes: inferring the mileage corresponding to the limit of the rim wear area and the lower limit of wheel life in the worst case based on the rim wear trend curve and confidence interval; and solving the turning mileage node and wheel diameter cutting amount that satisfy the constraints of the multi-objective optimization model.

5. The method for managing the turning and maintenance of wheelsets in a high-speed train according to claim 4, characterized in that, The constraints of the multi-objective optimization model include: The wheel diameter is not less than the wheel diameter threshold. The difference in wheel diameter on the same axis does not exceed the wheel diameter difference threshold. The envelope ratio of the profile is not lower than the envelope ratio threshold; The coarse scale template number must be one or more combinations of 28mm, 30mm, and 32mm; The standard deviation of the mileage interval between each overhaul is the smallest.

6. The method for managing the turning and maintenance of wheelsets in a high-speed train according to claim 1, characterized in that, Convert the coarse scale template number to a fine scale template number from the 0.1mm subdivision template library that meets the mapping conditions. The mapping conditions must simultaneously meet the following conditions: 1) The closest coarse scale template number; 2) The proportion of the measured envelope profile is not lower than the envelope rate threshold; 3) Minimum wheel diameter cutting amount.

7. The method for managing the turning and maintenance of wheelsets in a high-speed train according to claim 1, characterized in that, The final turning strategy includes: turning mileage nodes, precision scale template number, and wheel diameter cutting amount.

8. A method for managing the turning and maintenance of wheelsets in high-speed trains according to claim 1, characterized in that, The process of identifying wheels with abnormal flange wear rates by voting includes: detecting the flange wear rate of the wheel using the K-sigma criterion, HBOS, and local outlier factor respectively; if at least two of these methods determine that the flange wear rate of the wheel is abnormal, then the flange wear rate of the wheel is determined to be abnormal.

9. A system utilizing the wheelset turning and maintenance management method for high-speed trains according to any one of claims 1 to 8, characterized in that, The system includes: The sensing layer collects data on the measured profile, wheel diameter, lateral displacement, and mileage of the train wheelsets. The computational layer, based on the collected data, as well as the total train set information and track line information, establishes a nonlinear mapping model between wheel flange wear area and running mileage, and outputs the wheel flange wear trend curve and confidence interval. The control layer is used to establish a multi-objective optimization model and solve the multi-objective optimization model by combining the rim wear trend curve and confidence interval to obtain the final refinishing strategy; The evaluation layer is used to identify wheels with abnormal rim wear rates, visualize the results of the calculation layer, and display the refinishing plan and list of abnormal wheels.

10. The system according to claim 9, characterized in that, The perception layer includes a data acquisition module for collecting data on the wheelsets of the high-speed train, and a preprocessing module for processing the collected data. The computational layer includes a wear law modeling module, which uses a Gaussian process regression model to establish a nonlinear mapping model between the wear area of ​​the wheel flange and the running mileage, and outputs the wear trend curve and confidence interval of the wheel flange. The control layer includes an economic optimization decision module and a template recommendation engine. The economic optimization decision module is used to obtain an optimization repair strategy containing coarse-scale template numbers for the constructed multi-objective optimization model. The template recommendation engine is used to map coarse-scale template numbers to fine-scale template numbers and generate a final repair strategy containing fine-scale template numbers. The evaluation layer includes an anomaly detection module, a visual management and control platform, and a digital twin verification module. Anomaly detection module: It integrates three algorithms: K-sigma criterion, HBOS and local outlier factor. If at least two of these methods determine that the wheel flange wear rate is abnormal, then the wheel flange wear rate is determined to be abnormal. Visualized management and control platform: Real-time display of the entire train's wear heat map, life prediction curve, Gantt chart of the refining plan, and list of abnormal wheels; Digital twin verification module: Establish a virtual wheelset, import actual profile data to replay the life evolution process of the wheelset under different refinishing strategies, and intuitively verify whether the life extension rate meets the standard.

Images