Method for evaluating wheel-rail contact state based on logistic regression

Abstract

The invention discloses a logistic regression-based wheel-rail contact state evaluation method, and relates to the technical field of wheel-rail contact state evaluation. According to the logistic regression-based wheel-rail contact state evaluation method, based on a wheel-rail contact theory, the contact point distribution percentage number, the concentration ratio composite index and the 3mm equivalent taper are selected to represent the wheel-rail contact state more comprehensively and accurately through multi-dimensional parameters, so that evaluation deviation caused by one-sidedness of the parameters is avoided, and the evaluation accuracy of the wheel-rail contact state is improved. The logistic regression model is adopted as an evaluation core algorithm, the mapping relation between input and output of the model can be quantitatively explained through regression coefficients, the model is high in interpretability and adapts to engineering practice requirements, the model threshold value can be flexibly adjusted according to different line operation and maintenance requirements, and the evaluation accuracy is high. The method is advantaged in that potential safety risks are avoided, the method is adaptive to diversified railway operation scenes, intelligent grading evaluation of the wheel-rail contact state is realized, wheel-rail service life prolonging is facilitated, and operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of wheel-rail contact condition assessment technology, specifically a wheel-rail contact condition assessment method based on logistic regression. Background Technology

[0002] Wheel-rail contact condition is a core factor determining train operation safety, stability, and maintenance cycle, directly impacting the safety and economy of railway transportation. my country's high-speed rail lines have unique operating environments and technical characteristics: First, the wheel-to-rail distance is relatively short, causing the initial rail contact point to easily fall near the rail head centerline or to one side of the gauge angle, making traditional rail grinding methods unsuitable; second, the actual straight-to-curvature ratio of newly built high-speed rail lines is unreasonable, exacerbating wheel concave wear; third, mountain lines have a high proportion of bridges and tunnels, and crosswinds and sidewinds create a fluid-structure interaction effect on the car body, causing car body swaying and wheel creep wear. As the mileage increases, wheel wear becomes more severe, and wheel-rail contact matching performance declines sharply, affecting not only vehicle operation stability but also posing a potential threat to driving safety. Furthermore, my country's high-speed rail wheel-rail maintenance cycle is significantly shorter than in other countries, making wheel-rail contact condition assessment even more urgent.

[0003] To address the problem of wheel-rail contact condition assessment, scholars both domestically and internationally have conducted extensive research. Some studies have focused on single parameters, such as evaluating vehicle running stability and wheel-rail matching performance through equivalent cone values, or judging the quality of wheel-rail matching based on the lateral acceleration values ​​of the vehicle frame. However, single-parameter assessments are insufficient to comprehensively characterize the complex features of wheel-rail contact and are prone to biased results due to the one-sidedness of the parameters. Some studies have proposed wheel-rail contact matching evaluation principles or established elastic deformation body models to describe wheel-rail contact. However, these methods mostly remain at the theoretical analysis level and lack engineering-based quantitative assessment tools. Other studies have adopted intelligent methods, such as multilayer artificial neural networks, adversarial domain adaptation methods, fuzzy mathematical membership models, deep networks, and improved analytic hierarchy process (AHP). While these methods have achieved a certain degree of intelligent assessment, some intelligent algorithms are sensitive to noise, the rationality of feature selection directly determines the upper limit of algorithm performance, and some methods have high model complexity and poor interpretability, making it difficult to promote and apply them on a large scale in engineering practice.

[0004] In view of this, a wheel-rail contact state assessment method based on logistic regression is proposed to achieve accurate classification assessment of wheel-rail contact state and provide a scientific basis for train operation and maintenance. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a wheel-rail contact state assessment method based on logistic regression, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a wheel-rail contact state assessment method based on logistic regression, specifically including the following steps: S1. Contact area data collection: Based on statistical methods, the wheel profile is divided into three contact areas, denoted as LA area, LB area and LC area. The number of contact points in each contact area is counted, and the percentage of contact point distribution in each contact area is extracted. S2. Concentration Composite Index Calculation: The concentration composite index is calculated by combining the wheel-rail contact point jump characteristics; S3. Data preprocessing and dimensionality enhancement: After preprocessing the percentage of contact points in each contact area, the concentration composite index, and the 3mm equivalent taper, the data is enhanced to twenty feature vectors by combining the second power feature. S4. Wheel-rail contact condition assessment: Twenty feature vectors are input into a logistic regression model based on an improved OVO classification strategy to assess the level of wheel-rail contact condition.

[0007] This invention provides a method for evaluating wheel-rail contact state based on logistic regression. It has the following beneficial effects: This invention, based on wheel-rail contact theory, selects the percentage of contact point distribution, concentration composite index, and 3mm equivalent taper as multi-dimensional parameters to more comprehensively and accurately characterize the wheel-rail contact state. This avoids evaluation bias caused by the one-sidedness of parameters, making the grade evaluation results more consistent with the actual wheel-rail matching performance. Furthermore, it adopts a logistic regression model as the core evaluation algorithm. The mapping relationship between the model input and output can be quantitatively explained through regression coefficients, making the model highly interpretable and adaptable to engineering practice needs. Moreover, the model threshold can be flexibly adjusted according to the operation and maintenance needs of different lines, avoiding potential safety risks and adapting to diverse railway operation scenarios. At the same time, this invention realizes intelligent grade evaluation of wheel-rail contact state, improving evaluation efficiency. It is especially suitable for batch evaluation of large-scale wheel-rail samples, which helps to extend the service life of wheels and rails, reduce operation and maintenance costs, and ensure the safety and stability of train operation. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0010] Please see Figure 1 This invention provides a method for evaluating wheel-rail contact state based on logistic regression, specifically including the following steps: S1. Contact Area Data Acquisition: Establish a spatial geometric coordinate system with the X-axis parallel to the track and passing through the wheelset center, the Y-axis parallel to the wheelset axis, and the Z-axis perpendicular to the ground and upward. Based on statistical methods, divide the wheel profile into three contact areas, denoted as LA, LB, and LC. Count the number of contact points in each contact area and extract the percentage distribution of contact points in each contact area: PB, PA, PC. To obtain the percentage distribution of contact points in each contact area, it is necessary to first determine the wheel zoning area and analyze the distribution of wheel-rail contact points at different operating mileages. Statistically organize the data according to the three matching statuses: excellent, good, and poor. The wheel-rail matching contact points on the wheel profile are divided into two areas: LA and LC. Only a small number of contact points are distributed in the LB area. The lateral position coordinate range of the LA area is (712.5mm, 742.5mm); the lateral position coordinate range of the LB area is (682.5mm, 712.5mm); and the lateral position coordinate range of the LC area is (742.5mm, 801.5mm).

[0011] For detailed explanation, the evaluation criteria for the three matching statuses of Excellent, Good, and Poor are as follows: Tests were conducted on wheel-rail matching at different operating mileages. Analysis was performed on new wheels and rails, as well as wheel profiles of some wheels with less than 210,000 kilometers of service. It was found that the wheel-rail contact points are mainly distributed within the wheel tread [712.5mm, 742.5mm], and the contact points are evenly distributed. N contact =1, there are no contact points distributed at the root of the wheel flange; or, when the wheelset experiences a lateral displacement greater than 9mm, a finite number of contact points will appear in the flange contact area, and the contact points will jump around; or the contact points are not clearly divided on the wheel profile, i.e. N contact >3 and the jump between adjacent contact points is between [0.024, 0.415], i.e., 0.024 mm. J umpl The contact thickness is 0.415mm and the contact points are evenly distributed, classifying this type of contact as superior. As wear intensifies, the contact position of the wheel-rail contact point gradually shifts towards both sides of the wheel flange, or towards the outermost oblique straight side, as the degree of wear increases. The contact point jump phenomenon also intensifies, resulting in discontinuities in the wheel-rail contact point. The jump amplitude gradually increases, leading to one or two main contact areas on the tread surface and one main contact area at the wheel flange root, forming multiple relatively concentrated main contact areas within a 10mm lateral displacement. N contact At points ≥3 and where the wheel set lateral displacement ≤4mm, the contact point experiences jumping or concentration. J umpl ≤0.024mm orJ umpl ≥0.415mm, the distribution of such contact points is classified as poor; The state between excellent and poor matching performance is classified as good.

[0012] S2. Concentration Composite Index Calculation: The concentration composite index is calculated by combining the wheel-rail contact point jump characteristics. Specific methods include: A1. Calculate the contact bandwidth using the following formula: ; In the formula, For contact bandwidth, This is the lateral displacement of the wheelset. This represents the horizontal coordinate of the point of contact between the wheelset and the rail when the wheelset moves laterally to the left. The horizontal coordinates of the point of contact between the wheelset and the wheel-rail when the wheelset moves laterally to the right. A2. Calculate the contact bandwidth change rate using the following formula: ; In the formula, The contact bandwidth change rate; A3. Calculate the frequency of the wheel-rail contact point position under changes in wheelset lateral displacement. The calculation formula is: ; In the formula, The frequency of the wheel-rail contact point position under changes in wheelset lateral displacement. For the lateral displacement of the wheelset, This represents the increment of the wheelset's lateral movement. This represents the change in the wheel-rail contact point in the horizontal direction with the lateral displacement of the wheelset. This represents the change in the wheel-rail contact point in the vertical direction as a function of the wheelset's lateral displacement. A4. Calculate the concentration value using the following formula: ; In the formula, This is the concentration value. The number of times the wheelset lateral displacement is measured; A5. Define the danger threshold, and calculate it using the following formula: ; In the formula, This is the danger threshold; The number of wheel-rail contact points in the LB section. This represents the total concentration of contact points within the LB interval. The number of wheel-rail contact points in the LA section. This represents the total concentration of contact points within the LA interval. The number of wheel-rail contact points in the LC section. This represents the total concentration of contact points within the LC interval; Determine the danger zone: wheel-rail contact points with a concentration value higher than the danger threshold are classified as danger class 1 contact points, and wheel-rail contact points with a concentration value lower than the danger threshold are classified as safe class 0 contact points. When adjacent contact points form a sequence combination of 010, the area formed is called the danger zone. A6. Calculate the concentration composite index. The formula is: ; In the formula, It is a concentration composite index. The weighting coefficients for the LB interval are... This represents the total area of ​​the danger zone within the LB interval. The weighting coefficients for the LA interval are... This represents the total area of ​​the danger zone within the LA interval. These are the weighting coefficients for the LC interval. This represents the total area of ​​the hazardous area within the LC interval.

[0013] S3. Data Preprocessing and Dimensionality Upgrading: Normalize the percentage distribution of contact points in each contact area and the 3mm equivalent taper to make PB, PA, and PC... Falling within a specific range of 0-1; using Z-score standardization The values ​​were processed, and the percentage of contact points, concentration composite index, and 3mm equivalent taper of each contact area after preprocessing were combined using a power-law feature combination to increase the dimensionality to twenty feature vectors. These twenty features are shown in Table 1.

[0014] Table 1 In Table 1, It is a constant value. , , , , They represent PB, PA, PC, and 3mm equivalent taper .

[0015] S4. Wheel-rail contact condition assessment: Twenty feature vectors are input into a logistic regression model improved based on the OVO classification strategy to assess the level of wheel-rail contact condition. Specifically: In the logistic regression model improved based on the OVO classification strategy, the input feature matrix and output label value satisfy the following equation: ; In the formula, For the input feature matrix, To output the label value, , Let OVO be the probability of a certain wheel-rail contact matching level under the OVO classification strategy. The model intercept; All are model coefficients. All are predictor variables. This is the transpose of the model coefficients; Introducing the above equation Function, to obtain the probability value calculation formula: ; In the formula, This is a natural constant; after setting the model threshold, the level of wheel-rail contact condition is evaluated based on the output label value; The wheel-rail contact condition is graded as excellent, good, and poor, and the evaluation logic is as follows: .

[0016] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A wheel-rail contact state evaluation method based on a logistic regression, characterized by, Specifically comprising the following steps: S1, contact area data acquisition: based on statistical method, the wheel profile is divided into three contact areas, recorded as LA area, LB area and LC area, the number of contact points in each contact area is counted, and the contact point distribution percentage of each contact area is extracted; S2, concentration composite index calculation: the concentration composite index is calculated in combination with the jumping characteristics of wheel-rail contact points; S3, data preprocessing and dimensionality increasing: after preprocessing the contact point distribution percentage of each contact area, the concentration composite index and the 3mm equivalent taper, the secondary power feature combination is combined to increase the dimensionality to twenty feature vectors; S4, wheel-rail contact state evaluation: the twenty feature vectors are input into the improved logistic regression model based on OVO classification strategy, and the grade evaluation of wheel-rail contact state is carried out.

2. The wheel-rail contact state evaluation method based on a logistic regression according to claim 1, characterized in that, The transverse position coordinate interval of the LA area is (712.5mm, 742.5mm); The transverse position coordinate interval of the LB area is (682.5mm, 712.5mm); The transverse position coordinate interval of the LC area is (742.5mm, 801.5mm).

3. The wheel-rail contact state evaluation method based on a logistic regression according to Claim 1, characterized by, The way of calculating the concentration composite index in combination with the jumping characteristics of wheel-rail contact points includes: A1, establish a spatial geometric coordinate system: the X axis is parallel to the track and passes through the center of the wheel set, the Y axis is parallel to the wheel set axis, and the Z axis is perpendicular to the ground upward, then calculate the contact band width, the calculation formula is: ； wherein is the contact bandwidth, is the wheelset lateral displacement, is the horizontal coordinate of the contact point between the wheel and the rail when the wheelset is displaced laterally to the left, is the horizontal coordinate of the contact point between the wheel and the rail when the wheelset is displaced laterally to the right; A2, calculate the contact band width change rate, the calculation formula is: ； In the formula, is the contact bandwidth variation rate; A3, calculate the frequency of wheel-rail contact point position under the change of wheel set transverse displacement, the calculation formula is: ； wherein is the frequency of the wheel-rail contact point position for a change in the wheelset lateral displacement, is the wheelset lateral displacement variable, is the change increment of the wheelset lateral displacement, is the change in the wheel-rail contact point in the horizontal direction for a change in the wheelset lateral displacement, is the change in the wheel-rail contact point in the vertical direction for a change in the wheelset lateral displacement; A4, calculate the concentration value, the calculation formula is: ； In the formula, is the concentration value, is the number of times of the wheelset transverse displacement A5, define the dangerous threshold, and determine the dangerous area; A6, calculate the concentration composite index, the calculation formula is: ； wherein is the concentration composite index, is the LB interval weight coefficient, is the total area of dangerous regions within the LB interval, is the LA interval weight coefficient, is the total area of dangerous regions within the LA interval, is the LC interval weight coefficient, is the total area of dangerous regions within the LC interval.

4. The wheel-rail contact state evaluation method based on a logistic regression according to claim 3, characterized in that, The calculation formula of the dangerous threshold in A5 is: ； wherein is a danger threshold value; is the number of wheel-rail contact points in the LB interval, is the total value of the concentration of contact points in the LB interval, is the number of wheel-rail contact points in the LA interval, is the total value of the concentration of contact points in the LA interval, is the number of wheel-rail contact points in the LC interval, is the total value of the concentration of contact points in the LC interval.

5. The wheel-rail contact state evaluation method based on a logistic regression according to claim 4, characterized in that, The way of determining the dangerous area in A5 is: the wheel-rail contact point with a concentration value higher than the dangerous threshold is recorded as dangerous 1 type contact point, the wheel-rail contact point with a concentration value lower than the dangerous threshold is recorded as safe 0 type contact point, when the adjacent contact points appear 010 sequence combination, the area formed is called dangerous area.

6. The wheel-rail contact state evaluation method based on a logistic regression according to Claim 1, characterized by, The way of preprocessing the contact point distribution percentage of each contact area, the concentration composite index and the 3mm equivalent taper in S3 includes: The contact point distribution percentage of each contact area and the 3mm equivalent taper are normalized; The concentration composite index is standardized.

7. The wheel-rail contact state evaluation method based on a logistic regression according to Claim 1, characterized by, In the improved logistic regression model based on OVO classification strategy, the input feature matrix and output label value satisfy the following formula: ； wherein, is an input feature matrix, is an output label value, , is a probability of a certain wheel-rail contact matching level under the OVO classification strategy, is a model intercept; are model coefficients, are prediction variables, is a transpose of a model coefficient; The above formula is introduced The function, the probability value calculation formula: ； In the formula, is a natural constant; set the model threshold value according to the output label value to evaluate the level of wheel-rail contact state.

8. The wheel-rail contact state evaluation method based on a logistic regression according to claim 7, characterized in that, The grades of wheel-rail contact state include excellent, good and poor, and the evaluation logic is: 。

Images