A rapid quantification method for mapping the vibration source intensity of a metro wheel unroundness
By acquiring multi-dimensional non-roundness attribute information of subway wheels, utilizing track wall vibration acceleration signals and deep learning models, and combining K-Means clustering and one-way ANOVA, a rapid and accurate quantitative mapping of subway wheel non-roundness and vibration source intensity was achieved. This solves the problem of real-time dynamic evaluation and prediction that is difficult to achieve in existing technologies, and improves the model's generalization ability and evaluation robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA UNIV OF GEOSCIENCES (WUHAN)
- Filing Date
- 2026-03-03
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, there is a lack of fast, accurate, and quantitative mapping methods for the relationship between the out-of-roundness of subway wheels and the intensity of vibration sources. This makes it difficult to achieve real-time dynamic assessment and prediction, and fails to meet the needs of subway operation and maintenance for rapid identification, accurate tracing, and effective control of vibration sources.
By acquiring multi-dimensional non-roundness attribute information of subway wheels, vibration source contribution is evaluated using track wall vibration acceleration signals. Combining K-Means clustering algorithm and one-way ANOVA, a non-roundness-vibration contribution evaluation channel is constructed. A deep learning model is trained, and high-frequency similarity coefficients are introduced for feature comparison to obtain vibration source contribution and calculate evaluation uncertainty coefficient.
It achieves a fast, accurate, and quantitative mapping between the out-of-roundness of subway wheels and the intensity of vibration sources, meeting the real-time dynamic assessment and prediction needs of vibration sources in subway operation and maintenance, and improving the model's generalization ability and evaluation robustness.
Smart Images
Figure CN122153370A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of vibration source strength fluctuation contribution assessment technology, specifically to a rapid quantitative method for mapping the out-of-roundness of subway wheels with vibration source strength. Background Technology
[0002] With the rapid development of urban rail transit, subways have become the backbone of urban public transportation. However, the vibration and noise problems generated during subway train operation are becoming increasingly prominent, not only affecting the normal living and working environment of residents along the line, but also potentially causing damage to surrounding buildings and precision instruments.
[0003] Wheel out-of-roundness is one of the main excitation sources causing subway vibration, and its mapping relationship with the vibration response of the track structure is complex and crucial. In existing technologies, research on the relationship between subway wheel out-of-roundness and vibration source strength mostly focuses on qualitative analysis or correlation exploration of single indicators, lacking a method that can quickly, accurately, and quantitatively map the multi-dimensional attribute information of wheel out-of-roundness to the contribution of vibration sources.
[0004] Traditional methods often rely on complex vehicle dynamics simulations or large-scale field tests, which are not only time-consuming, labor-intensive, and costly, but also make it difficult to achieve real-time and dynamic assessment and prediction of vibration source intensity caused by wheel out-of-roundness. This fails to meet the actual needs of subway operation and maintenance for rapid identification, accurate tracing, and effective control of vibration sources. Summary of the Invention
[0005] This application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources, solving the technical problem in the prior art that there is a lack of a fast, accurate, and quantitative mapping method for the relationship between out-of-roundness of subway wheels and the intensity of vibration sources, making it difficult to achieve real-time dynamic evaluation and prediction.
[0006] The technical solution to the above-mentioned technical problems in this application is as follows: This application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources, the method comprising: Obtain the wheel out-of-roundness attribute information of the subway wheel to be evaluated; Following the contribution quantification analysis strategy based on effect value, the vibration source contribution was evaluated based on the track wall vibration acceleration signal collected when the subway train passed, and the wheel roundness attribute information and vibration source contribution of K samples were obtained. Using the out-of-roundness attribute information of the K sample wheels and the vibration source contribution of the K sample wheels as training data, an out-of-roundness-vibration contribution evaluation channel is constructed. The high-frequency similarity coefficient is determined by comparing the wheel out-of-roundness attribute information with the K sample wheel out-of-roundness attribute information. The out-of-roundness-vibration contribution evaluation channel is activated based on the high-frequency similarity coefficient, and the vibration source contribution evaluation result and evaluation uncertainty coefficient are obtained by analyzing the wheel out-of-roundness attribute information.
[0007] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to vibration source intensity. First, multi-dimensional out-of-roundness attribute information of the subway wheels to be evaluated is obtained. Second, a contribution quantification analysis strategy based on effect values is adopted. Using track wall vibration acceleration signals and a K-Means clustering algorithm, the sample data is grouped, and effect values are calculated through one-way ANOVA to extract the independent share of vibration contribution from different out-of-roundness feature combinations, thus refining key sample information. Third, an out-of-roundness-vibration contribution evaluation channel is constructed. The training set is divided into P-fold cross-partitions, and multiple deep learning model units are trained, forming an integrated evaluation mechanism that effectively improves the model's generalization ability and evaluation robustness. Then, high-frequency similarity coefficients are introduced for feature comparison, which can identify the similarity between the wheel to be evaluated and high-frequency associated samples in the sample set. Finally, the number of evaluation units is adaptively selected based on the high-frequency similarity coefficients. Through parallel evaluation of multiple units and fusion of results, not only are vibration source contribution evaluation results quickly obtained, but the evaluation uncertainty coefficient is also calculated by combining the high-frequency similarity coefficients and the dispersion of the evaluation results, quantifying the reliability of the evaluation results.
[0008] Through the above technical solution, this application avoids the complexity of traditional dynamic simulation and the limitations of field testing, and realizes a fast, accurate and quantitative mapping between wheel out-of-roundness and vibration source intensity, which can meet the actual needs of real-time dynamic evaluation and prediction of vibration sources in subway operation and maintenance. Attached Figure Description
[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0010] Figure 1 This is a flowchart illustrating a rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources, provided in an embodiment of this application. Detailed Implementation
[0011] This application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources. This method addresses the technical problem that existing technologies lack a rapid, accurate, and quantifiable mapping method for the relationship between out-of-roundness of subway wheels and the intensity of vibration sources, making it difficult to achieve real-time dynamic evaluation and prediction.
[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0013] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0014] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid unnecessarily obscuring the description of this application. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0015] Examples, such as Figure 1 As shown in the figure, this application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources, including: S10: Obtain the wheel out-of-roundness attribute information of the subway wheel to be evaluated; The wheel out-of-roundness attribute information includes radial runout peak value, radial runout peak value, first-order out-of-roundness amplitude, second-order out-of-roundness amplitude, third-order out-of-roundness amplitude, higher-order out-of-roundness amplitude, out-of-roundness order energy ratio, vertical wheel-rail force peak value, and lateral wheel-rail force peak value.
[0016] In this embodiment of the application, firstly, multi-dimensional non-roundness attribute information of the subway wheels is collected.
[0017] Specifically, a high-precision laser profilometer is used to scan the wheel tread to obtain radial runout data. After data processing, the radial runout peak value and radial runout peak-to-peak value are obtained. Among them, the radial runout peak value represents the maximum deviation in the radial direction of the wheel, while the radial runout peak-to-peak value reflects the maximum range of radius variation over the entire wheel circumference.
[0018] Meanwhile, the radial runout signal is spectrally decomposed using the order analysis method to extract the first-order to higher-order out-of-roundness amplitudes. The first-order out-of-roundness amplitude corresponds to the wheel eccentricity, while the higher-order out-of-roundness amplitudes reflect the local unevenness features of the wheel surface.
[0019] Furthermore, the energy proportion of each order of non-roundness amplitude is calculated, that is, the ratio of the energy of a certain order of non-roundness to the total energy of all orders of non-roundness, so as to quantify the contribution weight of different orders of non-roundness to the overall non-roundness.
[0020] In addition, by collecting the peak values of vertical and lateral wheel-rail forces through the wheel-rail force testing system, the impact intensity of wheel out-of-roundness on the track during dynamic contact is indirectly reflected.
[0021] S20: Following the contribution quantification analysis strategy based on effect value, the vibration source contribution is evaluated based on the track wall vibration acceleration signal collected when the subway train passes, and the wheel roundness attribute information and vibration source contribution of K samples are obtained. In this embodiment of the application, following the contribution quantification analysis strategy based on effect value, firstly, the vibration acceleration signal of the track wall when the subway train passes through a specific monitoring section is collected. The collected raw vibration acceleration signal is preprocessed, including removing the DC component, using a 50Hz notch filter to eliminate power frequency interference, and suppressing environmental noise through wavelet threshold denoising to obtain the effective vibration acceleration time history curve. Then, the effective value of vibration acceleration is calculated as a quantitative index of vibration response.
[0022] K sets of sample wheel out-of-roundness attribute information for the corresponding train were collected synchronously. The K-Means clustering algorithm was used to perform cluster analysis on the K sets of sample wheel out-of-roundness attribute information to obtain the vibration source contribution of K samples corresponding to the K sample wheel out-of-roundness attribute information.
[0023] Specifically, step S20 in the method includes: At a fixed monitoring section, the track wall vibration acceleration signal and the corresponding wheel out-of-roundness attribute information are collected synchronously when each subway train passes through. Based on the track wall vibration acceleration signal, the maximum Z vibration level of the track wall for each train is calculated as a characterization quantity of vibration source strength. After standardizing the out-of-roundness attribute information of all wheels, the K-Means algorithm is used to cluster them to generate K clusters, and the centroid of each cluster is extracted as the out-of-roundness attribute information of the sample wheels, thus obtaining K sample wheels out-of-roundness attribute information, where K is an integer greater than 10. Following the contribution quantification analysis strategy based on effect value, the vibration source contribution is evaluated based on the vibration source strength characterization quantity and K clusters to assess the out-of-roundness attribute information of the K sample wheels, thereby obtaining the vibration source contribution degree of the K samples.
[0024] In this embodiment, firstly, at a fixed monitoring section, the track wall next to the subway track is selected as the vibration signal acquisition point, and a triaxial acceleration sensor is installed with a sampling frequency set to 5000Hz. Each time a subway train passes, the sensor is synchronously triggered to acquire the vibration acceleration signal of the track wall, and the acquisition time covers the entire period of the train passing through the monitoring section, typically 30 seconds to 1 minute.
[0025] Simultaneously, the out-of-roundness attribute information of all wheels on the train is acquired through an onboard wheelset detection system or a ground wheel diameter measurement device, forming a one-to-one corresponding dataset. For the collected track wall vibration acceleration signals, baseline correction is first performed to remove the DC component. Then, a 50Hz notch filter is used to eliminate power frequency interference from the urban power grid. Finally, a wavelet threshold denoising method based on the db8 wavelet basis is applied to denoise the signal, resulting in a purified vibration acceleration time history curve. According to the "Urban Area Environmental Vibration Standard" (GB10070-88), the maximum Z-level of this time history curve is calculated as a measure of the vibration source strength. The Z-level data of rounded wheels are relatively concentrated and have low dispersion, while the Z-level of non-rounded wheels exhibits greater dispersion due to varying out-of-roundness levels. Secondly, all collected wheel out-of-roundness attribute information is standardized. Specifically, for each out-of-roundness attribute index, such as radial runout peak value and out-of-roundness amplitude of each order, the Z-score standardization method is used to convert it into a standard score with a mean of 0 and a standard deviation of 1, so as to eliminate the influence of differences in the dimensions of different indices.
[0026] Then, the K-Means clustering algorithm was used to perform cluster analysis on the standardized K groups of wheel out-of-roundness attribute information. The K-means algorithm iteratively calculates and finds the optimal partition that minimizes the sum of the distances between data points within a cluster and the cluster center, thus achieving natural data aggregation.
[0027] Furthermore, during the clustering process, Euclidean distance is used as a measure of similarity between samples. By iteratively optimizing the objective function—that is, minimizing the sum of squares within each cluster—samples with similar non-roundness characteristics are clustered into one class, ultimately generating K clusters. The value of K is pre-set to an integer greater than 10 based on the total number of samples and data distribution characteristics to ensure the representativeness and diversity of the samples. The centroid vector of each cluster is extracted as a typical representative of that cluster, i.e., the non-roundness attribute information of the sample wheels, thus obtaining the non-roundness attribute information of K sample wheels.
[0028] Subsequently, following a contribution quantification analysis strategy based on effect values, the vibration source contribution was evaluated based on the vibration source intensity characterization and K clusters of wheel out-of-roundness attribute information for the K samples. The contribution quantification describes the relative importance of each vibration signal feature as a proxy variable for out-of-roundness to the road wall vibration.
[0029] Furthermore, following the contribution quantification analysis strategy based on effect value, the vibration source contribution is evaluated based on the vibration source intensity characterization and K clusters to assess the out-of-roundness attribute information of the K sample wheels, obtaining the vibration source contribution degree of the K samples, including: Using the labels of the K clusters as grouping variables and the corresponding maximum Z vibration level as the dependent variable, a one-way ANOVA was performed to calculate the effect value of the sample wheel out-of-roundness attribute information on the overall explanatory power of the vibration. Based on the completed analysis of variance and the calculated effect value, the deviation between the vibration mean of each cluster and the overall mean is calculated. Combining the sample size of each cluster, the independent contribution share of the wheel out-of-roundness attribute information represented by the centroid of each cluster to the vibration fluctuation is quantified. After normalization, the vibration source contribution of K samples corresponding one-to-one with the wheel out-of-roundness attribute information of the K samples is obtained.
[0030] In this embodiment of the application, firstly, the sample data is divided into K groups using the labels of K clusters as grouping variables, and a one-way ANOVA is performed using the maximum Z-level of each group as the dependent variable.
[0031] One-way ANOVA compares the ratio of the between-group mean square to the within-group mean square to obtain the F-value, which tests whether there are significant differences in the vibration source strength characterization corresponding to different clusters, i.e., different combinations of out-of-roundness characteristics. If the F-test result is significant, such as P < 0.05, it indicates that the wheel out-of-roundness attribute information has a significant impact on the vibration source strength.
[0032] Secondly, calculate the effect size η. 2 Its calculation formula is η 2=SS_between_groups ÷ SS_total. Where SS_between_groups is the sum of squares between groups, representing the difference in average vibration level between different groups due to wheel imperfections; SS_total is the total sum of squares, representing the overall fluctuation in vibration intensity across all trains. Effect value η 2 η represents the proportion of vibration source intensity fluctuations explained by wheel out-of-roundness. 2 The larger the value, the more significant the impact of wheel out-of-roundness properties on vibration. For example, η 2 =0.5, indicating that the non-smoothness of the wheel contributes 50% to the maximum Z-level fluctuation.
[0033] Furthermore, based on the completed analysis of variance and the calculation of the effect value, the independent contribution of the sample wheel out-of-roundness attribute information represented by the centroid of each cluster to the vibration fluctuation is further quantified.
[0034] Specifically, the vibration mean of each cluster is first calculated, which is the deviation between the average of the maximum Z vibration level of all samples in the cluster and the overall vibration mean of all samples.
[0035] Then, the bias value of each cluster is multiplied by the number of samples in that cluster to obtain the weighted contribution of that cluster to the overall vibration fluctuation. The weighted contributions of all clusters are summed to obtain the total weighted contribution.
[0036] Finally, the weighted contribution of each cluster is divided by the total weighted contribution to obtain the independent contribution share of that cluster. Then, the independent contribution shares of all clusters are normalized so that their sum equals 1. This yields the contribution degree of the vibration source of the K sample wheels corresponding one-to-one with the centroids of the K clusters, representing the K sample wheel out-of-roundness attribute information. This contribution degree reflects the proportion of the wheel with a specific combination of out-of-roundness attributes contributing to the vibration of the road wall.
[0037] For example, if K=15, clustering yields the out-of-roundness attribute information of 15 sample wheels, and a one-way ANOVA is performed to obtain the effect value η. 2 =0.72, indicating that the wheel out-of-roundness attribute information can explain 72% of the overall vibration fluctuation. Among them, the vibration mean of a certain cluster is 85dB, the overall vibration mean is 75dB, the number of samples in this cluster is 20, and the total number of samples is 150. The number of samples in other clusters is between 8 and 25. The deviation value of this cluster is calculated to be 10dB, and the weighted contribution is 10 × 20 = 200. Assuming the total weighted contribution is 1000, its independent contribution share is 200 ÷ 1000 = 0.2. After normalization, the vibration source contribution of this sample is 0.2, that is, the contribution of this out-of-roundness feature combination to the vibration accounts for 20%.
[0038] S30: Using the K sample wheel out-of-roundness attribute information and the K sample vibration source contribution as training data, construct an out-of-roundness-vibration contribution evaluation channel; In this embodiment, K sample wheel out-of-roundness attribute information and K sample vibration source contribution values are used as training data to construct an out-of-roundness-vibration contribution evaluation channel.
[0039] Specifically, the dataset is constructed by first using the out-of-roundness attribute information of K sample wheels as input feature vectors and the corresponding vibration source contribution of K sample wheels as output labels.
[0040] During model training, parameters are optimized, and root mean square error is used as the evaluation metric for model performance to ensure that the model has good fitting effect and prediction accuracy on both the training and validation sets.
[0041] Specifically, step S30 in the method includes: The K sample wheel roundness attribute information and the K sample vibration source contribution degree are used as training data, and P-fold cross-partitioning is performed to obtain P sample training sets, where P is an integer greater than or equal to 10. Using the out-of-roundness attribute information of the sample wheels as input data and the vibration source contribution of the sample as label data, the deep learning model is supervised and trained using the P sample training sets until convergence, generating P out-of-roundness-vibration contribution evaluation units, which are then combined to obtain the out-of-roundness-vibration contribution evaluation channel.
[0042] In this embodiment, the training data is first divided using P-fold cross-validation. Specifically, K sample data are randomly shuffled and then evenly divided into P parts, where P is an integer greater than or equal to 10, for example, P=10. Each time, P-1 parts of the data are selected as the training set, and the remaining 1 part is used as the validation set. This process is repeated P times to obtain P different sample training sets and corresponding validation sets.
[0043] Secondly, the out-of-roundness attribute information of the sample wheels is used as input data, including radial runout peak value, radial runout peak-to-peak value, out-of-roundness amplitude of each order, out-of-roundness order energy proportion, vertical wheel-rail force peak value, and lateral wheel-rail force peak value. The contribution of the sample vibration source is used as label data to construct the input-output relationship of the deep learning model. The deep learning model can be a multilayer perceptron, whose network structure includes an input layer, hidden layers, and an output layer.
[0044] Furthermore, the number of neurons in the input layer is consistent with the feature dimension of the non-circularity attribute information. For example, if there are 9 attribute information, then the number of neurons in the input layer is 9. The hidden layer can be set to 2-3 layers, and the number of neurons in each layer is determined through experimental optimization. For example, the first layer has 128 neurons and the second layer has 64 neurons. The ReLU activation function is used to enhance the nonlinear fitting ability. The output layer has 1 neuron, corresponding to the predicted value of the vibration source contribution. The Sigmoid activation function is used to map the output value to the range of 0-1, which is consistent with the normalization characteristics of the contribution.
[0045] Then, the deep learning model is independently trained under supervised supervision using P training samples. In each training iteration, the Adam optimizer is used with an initial learning rate of 0.001, and the root mean square error (RMSE) between the predicted and actual contributions is minimized using backpropagation. During training, the RMSE on the validation set is monitored in real time. Training is stopped when the RMSE on the validation set no longer decreases after 20 consecutive iterations to avoid overfitting. After model convergence, the model parameters corresponding to the training set are saved, generating one non-circularity-vibration contribution evaluation unit. This process is repeated P times to obtain P evaluation units with identical structures but different parameters.
[0046] Finally, the P out-of-roundness-vibration contribution evaluation units are combined to construct an out-of-roundness-vibration contribution evaluation channel.
[0047] S40: Perform feature comparison on the wheel out-of-roundness attribute information and the K sample wheel out-of-roundness attribute information to determine the high-frequency similarity coefficient; In this embodiment, firstly, high-frequency feature components are extracted from the out-of-roundness attribute information of the wheel to be evaluated. These high-frequency feature components include the out-of-roundness amplitude within a preset high-frequency order range, the proportion of high-frequency order energy, and the vertical wheel-rail force impact increment corresponding to the high-frequency out-of-roundness. Simultaneously, feature components corresponding to the high-frequency order range are extracted from the out-of-roundness attribute information of K sample wheels to form a sample high-frequency feature library.
[0048] Secondly, the cosine similarity algorithm is used to calculate the similarity value between the high-frequency feature components of the wheel to be evaluated and the high-frequency feature components of each sample. The similarity value is normalized to obtain the high-frequency similarity coefficient between the wheel to be evaluated and the out-of-roundness attribute information of K sample wheels. The coefficient ranges from 0 to 1. The closer it is to 1, the more similar the high-frequency out-of-roundness features of the two are.
[0049] Specifically, step S40 in the method includes: Randomly select the first sample of wheel out-of-roundness attribute information from the K sample wheel out-of-roundness attribute information; Based on multiple out-of-roundness attribute indicators, a multi-dimensional deviation calculation is performed on the out-of-roundness attribute information of the wheel and the out-of-roundness attribute information of the first sample wheel. Indicators with an index deviation less than a preset index deviation threshold are taken as strong correlation indicators, and a first set of strong correlation indicators is obtained. If the number of the first strongly correlated indicators is greater than the preset indicator number threshold, then the wheel roundness attribute information of the first sample is marked as high-frequency correlated information. Iterative comparison analysis is performed until the out-of-roundness attribute information of the K sample wheels has been traversed, and the proportion of the high-frequency related information is counted as the high-frequency similarity coefficient.
[0050] In this embodiment of the application, firstly, one of the K sample wheel out-of-roundness attribute information is randomly selected as the first sample wheel out-of-roundness attribute information.
[0051] Secondly, based on the out-of-roundness attribute information of the wheel to be evaluated and the first sample wheel, multiple out-of-roundness attribute indices are selected for multi-dimensional deviation calculation. These out-of-roundness attribute indices include radial runout peak value, radial runout peak-to-peak value, first-order out-of-roundness amplitude, second-order out-of-roundness amplitude, third-order out-of-roundness amplitude, higher-order out-of-roundness amplitude, out-of-roundness order energy ratio, vertical wheel-rail force peak value, and lateral wheel-rail force peak value. For each selected index, the absolute or relative deviation between the index value of the wheel to be evaluated and the index value of the first sample wheel is calculated.
[0052] Then, a preset indicator deviation threshold is set, for example, the absolute deviation of an indicator does not exceed 1 / 3 of the standard deviation of the indicator in the sample population, or the relative deviation does not exceed 5%. Indicators with deviation values less than this preset indicator deviation threshold are identified as strongly correlated indicators, and the strongly correlated indicators are aggregated to form the first strongly correlated indicator set.
[0053] Furthermore, if the number of strongly correlated indicators in the first set of strongly correlated indicators exceeds a preset threshold, then the out-of-roundness attribute information of the first sample wheel is determined to have a strong correlation with the out-of-roundness attribute information of the wheel to be evaluated in terms of high-frequency features, and is marked as high-frequency correlated information. This threshold is set according to a certain proportion of the total number of indicators; for example, if the total number of indicators is 6, the threshold is set to 3, meaning that more than half of the indicators are strongly correlated.
[0054] Then, following the same process, the next sample is randomly selected from the remaining sample wheel roundness attribute information, and the above multi-dimensional deviation calculation, strong correlation index set construction and judgment process is repeated to carry out iterative comparison analysis until all K sample wheel roundness attribute information has been traversed.
[0055] Finally, the ratio of the number of samples labeled as high-frequency related information in the K samples to K is calculated. The resulting ratio is the high-frequency similarity coefficient between the out-of-roundness attribute information of the wheel to be evaluated and the out-of-roundness attribute information of the K sample wheels. The higher this coefficient, the higher the degree of matching between the high-frequency out-of-roundness features of the wheel to be evaluated and the high-frequency features in the sample set.
[0056] S50: Activate the out-of-roundness-vibration contribution evaluation channel based on the high-frequency similarity coefficient, and analyze and obtain the vibration source contribution evaluation result and evaluation uncertainty coefficient based on the wheel out-of-roundness attribute information.
[0057] In this embodiment, a high-frequency similarity coefficient threshold is first set. When the high-frequency similarity coefficient of the wheel to be evaluated is greater than or equal to the threshold, it is determined that its high-frequency out-of-roundness feature has sufficient similarity with the high-frequency features in the sample library. At this time, the out-of-roundness-vibration contribution evaluation channel is activated.
[0058] If the high-frequency similarity coefficient is less than the threshold, it is considered that the high-frequency out-of-roundness features of the wheel to be evaluated lack representativeness in the existing sample library, and the evaluation channel is not activated for the time being. At this time, a prompt message is output, suggesting that samples of this type of feature be added or other auxiliary evaluation methods be used.
[0059] Furthermore, the evaluation channel is activated, and the out-of-roundness attribute information of the wheel to be evaluated is input into P out-of-roundness-vibration contribution evaluation units in the constructed out-of-roundness-vibration contribution evaluation channel. Each evaluation unit independently predicts a vibration source contribution value based on the input out-of-roundness attribute information. The predicted values are integrated and processed to obtain the vibration source contribution evaluation result and the evaluation uncertainty coefficient.
[0060] Specifically, step S50 in the method includes: The product of the reciprocal of the high-frequency similarity coefficient and the number of initial evaluation units is rounded down to obtain the number of adaptation units selected, Q, where the number of initial evaluation units selected is 5, and Q is greater than or equal to 3 and less than or equal to P. In the out-of-roundness-vibration contribution evaluation channel, Q evaluation units are randomly selected from the P out-of-roundness-vibration contribution evaluation units to evaluate the vibration source contribution of the wheel out-of-roundness attribute information and obtain Q initial vibration source evaluation contribution values. The average value of the contribution of the Q initial vibration sources is taken as the evaluation result of the vibration source contribution. The evaluation uncertainty coefficient is obtained based on the high-frequency similarity coefficient and the evaluation contribution analysis of the Q initial vibration sources.
[0061] In this embodiment of the application, firstly, the number Q of adapter units selected is determined.
[0062] Specifically, the reciprocal of the high-frequency similarity coefficient obtained in step S40 is multiplied by the initial number of selected evaluation units, and the product is then rounded down. Meanwhile, to ensure the stability and reliability of the evaluation, Q must satisfy the constraint condition of being greater than or equal to 3 and less than or equal to P. If the calculated result exceeds this range, the boundary value is taken as Q.
[0063] For example, the initial number of evaluation units is set to 5. If the high-frequency similarity coefficient is 0.8, its reciprocal is 1.25. 1.25×5=6.25, and after rounding down, Q=6.
[0064] Secondly, among the P evaluation units in the out-of-roundness-vibration contribution evaluation channel, Q evaluation units are selected by random sampling. The out-of-roundness attribute information of the wheel to be evaluated, including the radial runout peak value, the out-of-roundness amplitude of each order, and other features, are input into the selected Q evaluation units. Each evaluation unit independently outputs an initial vibration source evaluation contribution based on the mapping relationship it has trained, thereby obtaining Q initial evaluation results.
[0065] Then, the vibration source contribution evaluation results are calculated. The arithmetic mean of the Q initial vibration source contribution evaluations is the final vibration source contribution evaluation result.
[0066] For example, if Q=5, the five initial contribution values are 0.22, 0.18, 0.20, 0.25, and 0.19, respectively. Their average value is (0.22+0.18+0.20+0.25+0.19)÷5=0.208, which means the evaluation result is 0.208. This indicates that the out-of-roundness attribute combination of the wheel to be evaluated contributes approximately 20.8% to the vibration.
[0067] Finally, the evaluation uncertainty coefficient is obtained through analysis. The evaluation uncertainty coefficient comprehensively considers the high-frequency similarity coefficient and the dispersion of the Q initial evaluation contributions.
[0068] The evaluation uncertainty coefficient is obtained based on the high-frequency similarity coefficient and the contribution analysis of the Q initial vibration sources, including: The difference between 1 and the high-frequency similarity coefficient is used as the initial prediction uncertainty. Calculate the maximum deviation between the mean and extreme values of the evaluation contributions of the Q initial vibration sources; The uncertainty compensation coefficient is obtained by summing the maximum deviation amplitude with 1; The product of the uncertainty compensation coefficient and the initial prediction uncertainty is used as the evaluation uncertainty coefficient.
[0069] In this embodiment, the high-frequency similarity coefficient is first denoted as α, and the initial prediction uncertainty is 1-α. This value reflects the degree of difference between the high-frequency features of the wheel to be evaluated and the features of the sample library. The lower α is, the greater the initial prediction uncertainty, indicating that the evaluation result is more affected by the insufficient representativeness of the sample.
[0070] Secondly, calculate the mean μ of the evaluation contribution of the Q initial vibration sources, and find the maximum value μ. max and minimum value μ min Calculate the deviation between the mean and the extreme values, i.e., |μ-μ max |and|μ-μ min The larger of the two values is taken as the maximum deviation magnitude Δ. This value reflects the prediction dispersion of different evaluation units for the same input; the larger the Δ, the worse the stability of the model's prediction results.
[0071] Next, the maximum deviation amplitude Δ is added to 1 to obtain the uncertainty compensation coefficient (1+Δ), which is used to amplify the uncertainty caused by the dispersion of prediction.
[0072] Finally, the uncertainty compensation coefficient (1+Δ) is multiplied by the initial prediction uncertainty (1-α), and the product is the evaluation uncertainty coefficient U.
[0073] For example, if the high-frequency similarity coefficient α = 0.8, the initial prediction uncertainty is 0.2; the Q initial contributions are 0.22, 0.18, 0.20, 0.25, and 0.19, with a mean μ = 0.208 and a maximum value μ max =0.25, minimum value μ min =0.18, |0.208-0.25|=0.042, |0.208-0.18|=0.028, maximum deviation Δ=0.042; uncertainty compensation coefficient is 1+0.042=1.042; evaluation uncertainty coefficient U=0.2×1.042=0.2084.
[0074] The evaluation uncertainty coefficient U reflects the impact of sample representativeness and model prediction stability on the evaluation results, and can be used to assess the reliability of the vibration source contribution evaluation results.
[0075] In summary, compared to existing technologies, this application achieves rapid quantitative evaluation of vibration contribution from physical characteristics by deeply mapping wheel out-of-roundness attribute information with vibration source contribution. The evaluation channel constructed through P-fold cross-validation, combined with a feature comparison mechanism using high-frequency similarity coefficients, ensures the model's sufficiency in learning sample features and enhances the robustness and reliability of the evaluation results through multi-unit integrated prediction and uncertainty coefficient analysis.
[0076] In summary, the embodiments of this application have at least the following technical effects: This application provides a rapid quantification method for mapping the out-of-roundness of subway wheels to vibration source intensity. First, multi-dimensional out-of-roundness attribute information of the subway wheels to be evaluated is obtained. Second, a contribution quantification analysis strategy based on effect values is adopted. Using track wall vibration acceleration signals and a K-Means clustering algorithm, the sample data is grouped, and effect values are calculated through one-way ANOVA to extract the independent share of vibration contribution from different out-of-roundness feature combinations, thus refining key sample information. Third, an out-of-roundness-vibration contribution evaluation channel is constructed. The training set is divided into P-fold cross-partitions, and multiple deep learning model units are trained, forming an integrated evaluation mechanism that effectively improves the model's generalization ability and evaluation robustness. Then, high-frequency similarity coefficients are introduced for feature comparison, which can identify the similarity between the wheel to be evaluated and high-frequency associated samples in the sample set. Finally, the number of evaluation units is adaptively selected based on the high-frequency similarity coefficients. Through parallel evaluation of multiple units and fusion of results, not only are vibration source contribution evaluation results quickly obtained, but the evaluation uncertainty coefficient is also calculated by combining the high-frequency similarity coefficients and the dispersion of the evaluation results, quantifying the reliability of the evaluation results.
[0077] Through the above technical solution, this application avoids the complexity of traditional dynamic simulation and the limitations of field testing, and realizes a fast, accurate and quantitative mapping between wheel out-of-roundness and vibration source intensity, which can meet the actual needs of real-time dynamic evaluation and prediction of vibration sources in subway operation and maintenance.
[0078] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0079] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0080] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.
Claims
1. A rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources, characterized in that, The methods include: Obtain the wheel out-of-roundness attribute information of the subway wheel to be evaluated; Following the contribution quantification analysis strategy based on effect value, the vibration source contribution was evaluated based on the track wall vibration acceleration signal collected when the subway train passed, and the wheel roundness attribute information and vibration source contribution of K samples were obtained. Using the out-of-roundness attribute information of the K sample wheels and the vibration source contribution of the K sample wheels as training data, an out-of-roundness-vibration contribution evaluation channel is constructed. The high-frequency similarity coefficient is determined by comparing the wheel out-of-roundness attribute information with the K sample wheel out-of-roundness attribute information. The out-of-roundness-vibration contribution evaluation channel is activated based on the high-frequency similarity coefficient, and the vibration source contribution evaluation result and evaluation uncertainty coefficient are obtained by analyzing the wheel out-of-roundness attribute information.
2. The rapid quantification method for mapping subway wheel out-of-roundness to vibration source intensity according to claim 1, characterized in that, The wheel out-of-roundness attribute information includes radial runout peak value, radial runout peak value, first-order out-of-roundness amplitude, second-order out-of-roundness amplitude, third-order out-of-roundness amplitude, higher-order out-of-roundness amplitude, out-of-roundness order energy ratio, vertical wheel-rail force peak value, and lateral wheel-rail force peak value.
3. The rapid quantification method for mapping subway wheel out-of-roundness to vibration source intensity according to claim 1, characterized in that, Following an effect-value-based contribution quantification analysis strategy, vibration source contribution was evaluated based on track wall vibration acceleration signals collected during subway train passage. This yielded K sample wheel out-of-roundness attribute information and K sample vibration source contribution values, including: At a fixed monitoring section, the track wall vibration acceleration signal and the corresponding wheel out-of-roundness attribute information are collected synchronously when each subway train passes through. Based on the track wall vibration acceleration signal, the maximum Z vibration level of the track wall for each train is calculated as a characterization quantity of vibration source strength. After standardizing the out-of-roundness attribute information of all wheels, the K-Means algorithm is used to cluster them to generate K clusters, and the centroid of each cluster is extracted as the out-of-roundness attribute information of the sample wheels, thus obtaining K sample wheels out-of-roundness attribute information, where K is an integer greater than 10. Following the contribution quantification analysis strategy based on effect value, the vibration source contribution is evaluated based on the vibration source strength characterization quantity and K clusters to assess the out-of-roundness attribute information of the K sample wheels, thereby obtaining the vibration source contribution degree of the K samples.
4. The rapid quantification method for mapping subway wheel out-of-roundness to vibration source intensity according to claim 3, characterized in that, Following the contribution quantification analysis strategy based on effect values, the vibration source contribution is evaluated based on the vibration source intensity characterization and K clusters to assess the out-of-roundness attribute information of the K sample wheels, obtaining the vibration source contribution degree of the K samples, including: Using the labels of the K clusters as grouping variables and the corresponding maximum Z vibration level as the dependent variable, a one-way ANOVA was performed to calculate the effect value of the sample wheel out-of-roundness attribute information on the overall explanatory power of the vibration. Based on the completed analysis of variance and the calculated effect value, the deviation between the vibration mean of each cluster and the overall mean is calculated. Combining the sample size of each cluster, the independent contribution share of the wheel out-of-roundness attribute information represented by the centroid of each cluster to the vibration fluctuation is quantified. After normalization, the vibration source contribution of K samples corresponding one-to-one with the wheel out-of-roundness attribute information of the K samples is obtained.
5. A rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources according to claim 1, characterized in that, Using the out-of-roundness attribute information of the K sample wheels and the vibration source contribution of the K sample wheels as training data, an out-of-roundness-vibration contribution evaluation channel is constructed, including: The K sample wheel roundness attribute information and the K sample vibration source contribution degree are used as training data, and P-fold cross-partitioning is performed to obtain P sample training sets, where P is an integer greater than or equal to 10. Using the out-of-roundness attribute information of the sample wheels as input data and the vibration source contribution of the sample as label data, the deep learning model is supervised and trained using the P sample training sets until convergence, generating P out-of-roundness-vibration contribution evaluation units, which are then combined to obtain the out-of-roundness-vibration contribution evaluation channel.
6. A rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources according to claim 2, characterized in that, The high-frequency similarity coefficient is determined by feature comparison of the wheel out-of-roundness attribute information and the K sample wheel out-of-roundness attribute information, including: Randomly select the first sample of wheel out-of-roundness attribute information from the K sample wheel out-of-roundness attribute information; Based on multiple out-of-roundness attribute indicators, a multi-dimensional deviation calculation is performed on the out-of-roundness attribute information of the wheel and the out-of-roundness attribute information of the first sample wheel. Indicators with an index deviation less than a preset index deviation threshold are taken as strong correlation indicators, and a first set of strong correlation indicators is obtained. If the number of the first strongly correlated indicators is greater than the preset indicator number threshold, then the wheel roundness attribute information of the first sample is marked as high-frequency correlated information. Iterative comparison analysis is performed until the out-of-roundness attribute information of the K sample wheels has been traversed, and the proportion of the high-frequency related information is counted as the high-frequency similarity coefficient.
7. A rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources according to claim 5, characterized in that, The out-of-roundness-vibration contribution evaluation channel is activated based on the high-frequency similarity coefficient. The vibration source contribution evaluation result and evaluation uncertainty coefficient are obtained based on the wheel out-of-roundness attribute information, including: The product of the reciprocal of the high-frequency similarity coefficient and the number of initial evaluation units is rounded down to obtain the number of adaptation units selected, Q, where the number of initial evaluation units selected is 5, and Q is greater than or equal to 3 and less than or equal to P. In the out-of-roundness-vibration contribution evaluation channel, Q evaluation units are randomly selected from the P out-of-roundness-vibration contribution evaluation units to evaluate the vibration source contribution of the wheel out-of-roundness attribute information and obtain Q initial vibration source evaluation contribution values. The average value of the contribution of the Q initial vibration sources is taken as the evaluation result of the vibration source contribution. The evaluation uncertainty coefficient is obtained based on the high-frequency similarity coefficient and the evaluation contribution analysis of the Q initial vibration sources.
8. A rapid quantification method for mapping the out-of-roundness of subway wheels to the intensity of vibration sources according to claim 7, characterized in that, The evaluation uncertainty coefficient is obtained based on the high-frequency similarity coefficient and the contribution analysis of the Q initial vibration sources, including: The difference between 1 and the high-frequency similarity coefficient is used as the initial prediction uncertainty. Calculate the maximum deviation between the mean and extreme values of the evaluation contributions of the Q initial vibration sources; The uncertainty compensation coefficient is obtained by summing the maximum deviation amplitude with 1; The product of the uncertainty compensation coefficient and the initial prediction uncertainty is used as the evaluation uncertainty coefficient.