Glue joint insulation joint fatigue damage evaluation method and system considering random parameters

By combining the finite element method with the proxy model, the systematic problems of fatigue life assessment of adhesively bonded insulating joints were solved, high-precision fatigue damage assessment under random factor loads was achieved, and the reliability and efficiency of the assessment were improved.

CN120781632AInactive Publication Date: 2025-10-14SOUTHWEST JIAOTONG UNIV

Patent Information

Application Number
CN202511285365.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-09-10
Publication Date
2025-10-14
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the existing technology, the fatigue life reliability assessment method of adhesively bonded insulated joints lacks systematicity and cannot consider the complex nonlinear dynamic impact effects under random factor loads, resulting in the inability to effectively assess the dynamic impact and derailment risk of wheel-rail under high-speed transient rolling vibration conditions.

Method used

The finite element method is used to establish a dynamic contact model between the wheel and the bonded insulating joint. By obtaining random parameter data for simulation, the dynamic response parameters are calculated. The fatigue life is predicted by combining the surrogate model, and a sensitivity analysis is performed to evaluate the failure probability and reliability of the bonded insulating joint.

Benefits of technology

The accuracy and reliability of fatigue damage assessment of adhesively bonded insulated joints have been significantly improved, the contribution of random parameters has been quantified, and the authenticity of dynamic response modeling and the efficiency of fatigue life prediction have been optimized.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120781632A_ABST
    Figure CN120781632A_ABST
Patent Text Reader

Abstract

The invention provides a glue joint insulation joint fatigue damage evaluation method and system considering random parameters, and relates to the technical field of rail vehicle wheel-rail contact analysis, and the method comprises the steps: obtaining first information; establishing a dynamic contact model of the wheel and the cementing insulation joint according to the first information, and performing simulation to obtain a dynamic response parameter in the contact process of the wheel and the track; performing fatigue damage analysis according to the dynamic response parameters to obtain fatigue damage parameters; performing fatigue life prediction according to the fatigue damage parameter to obtain a fatigue life prediction result; performing sensitivity analysis according to the fatigue life prediction result to obtain a sensitivity analysis result; and calculating the failure probability of the glue joint insulation joint according to the sensitivity analysis result, and evaluating the reliability of the glue joint insulation joint. According to the method, the dynamic contact model of the wheel and the cementing insulation joint is established by using the finite element method and simulation is carried out, so that high-precision wheel-rail contact dynamic response parameters are directly obtained.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rail vehicle wheel-rail contact analysis, in particular to a fatigue damage evaluation method and system of a bonded insulating joint considering random parameters. BACKGROUND

[0002] As a key isolation component of the seamless track signal system, the bonded insulating joint is widely used in the turnout area, but its irregular structure causes damage such as rail head indentation and crack, significantly aggravating the wheel-rail dynamic impact and derailment risk. In recent years, deep learning and machine learning technologies have improved the fatigue life prediction accuracy, but the research on the strong coupling nonlinear dynamic effect of wheel-rail and the influence of random mechanical parameters under high-speed transient rolling vibration contact impact conditions is still insufficient. As a weak link that bears severe load in high-speed turnouts, it is of great practical significance to study its fatigue damage under impact load.

[0003] In the prior art, there is a blank in the fatigue life reliability evaluation method of the bonded insulating joint, there is no corresponding evaluation means, and the complex nonlinear dynamic impact under random factor load cannot be considered. For example, the patent with publication number CN202211517312 can only evaluate geometric irregularities and cannot analyze fatigue damage; the agent model in the patent with publication number CN202410483106 does not consider the nonlinear dynamic effect of joint wheel-rail contact impact and does not quantify the contribution of wheel-rail dynamic parameters; the acceleration adaptability evaluation method in the patent with publication number CN202310051843 lacks systematicness in analyzing the fatigue damage of the bonded insulating joint in the turnout area, and does not consider the influence of multi-source wheel-rail random factors such as axle load and contact position.

[0004] Based on the above-mentioned shortcomings of the prior art, there is an urgent need for a fatigue damage evaluation method and system of a bonded insulating joint considering random parameters. SUMMARY

[0005] The purpose of the present application is to provide a fatigue damage evaluation method and system of a bonded insulating joint considering random parameters to improve the above-mentioned problems. In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows: In a first aspect, the present application provides a fatigue damage evaluation method of a bonded insulating joint considering random parameters, comprising: obtaining first information, the first information including random parameter data of vehicle parameters, rail parameters and wheel-rail contact parameters; establishing a dynamic contact model of the wheel and the bonded insulating joint and performing simulation by using the finite element method according to the first information, and obtaining dynamic response parameters in the wheel-rail contact process; performing fatigue damage analysis according to the dynamic response parameters, obtaining fatigue damage parameters by calculating the stress and strain state at the cross section of the bonded insulating joint; According to the fatigue damage parameter, fatigue life prediction is performed, fatigue life of each section is predicted based on a preset proxy model, and fatigue life prediction results are obtained; According to the fatigue life prediction results, sensitivity analysis is performed, and sensitivity analysis results are obtained; According to the sensitivity analysis results, the failure probability of the glued insulation joint is calculated, and the reliability of the glued insulation joint is evaluated according to the failure probability curve under different design life.

[0006] In a second aspect, the application also provides a glued insulation joint fatigue damage evaluation system considering random parameters, comprising: An acquisition module is configured to acquire first information, wherein the first information comprises random parameter data of vehicle parameters, rail parameters and wheel-rail contact parameters; A simulation module is configured to establish a dynamic contact model of a wheel and a glued insulation joint and perform simulation by using a finite element method according to the first information, and obtain dynamic response parameters in a wheel-rail contact process; A calculation module is configured to perform fatigue damage analysis according to the dynamic response parameters, obtain fatigue damage parameters by calculating stress and strain states at a section of the glued insulation joint; A prediction module is configured to perform fatigue life prediction according to the fatigue damage parameters, predict fatigue life of each section based on a preset proxy model, and obtain fatigue life prediction results; An analysis module is configured to perform sensitivity analysis according to the fatigue life prediction results, and obtain sensitivity analysis results; An evaluation module is configured to calculate the failure probability of the glued insulation joint according to the sensitivity analysis results, and evaluate the reliability of the glued insulation joint according to the failure probability curve under different design life.

[0007] The application has the following advantages: The application directly obtains high-precision wheel-rail contact dynamic response parameters by establishing a dynamic contact model of a wheel and a glued insulation joint and performing simulation by using a finite element method, predicts fatigue life of each section based on a proxy model to generate fatigue life distribution data, quantifies random parameter contribution and outputs a sensitivity index by performing sensitivity analysis according to the fatigue life prediction results, and significantly improves the modeling authenticity of dynamic response, fatigue life prediction efficiency and reliability optimization. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. Other related drawings can also be obtained by those of ordinary skill in the art without creative labor on the basis of these drawings.

[0009] Figure 1 A flow chart of a method for evaluating fatigue damage of a bonded insulation joint considering random parameters; Figure 2 A structure diagram of a system for evaluating fatigue damage of a bonded insulation joint considering random parameters; Figure 3 An explicit finite element wheel-rail impact contact model of a bonded insulation rail joint; Figure 4 A diagram showing the relationship between the angles of the critical plane and the global coordinate system; Figure 5 A diagram showing the relationship between the normal vector of the rail and the critical plane; Figure 6 A time-domain curve of wheel-rail force of the interaction between the wheel and the bonded insulation joint; Figure 7 A diagram showing the training error of the most dangerous section of the ANN model under the unoptimized condition; Figure 8 A diagram showing the training error of the most dangerous section of the ANN model under the WOA optimized condition; Figure 9 A diagram showing the probability density function distribution of the fatigue life of the most dangerous section; Figure 10 A cloud chart of the probability density function of the fatigue life of the 70 key sections along the longitudinal direction; Figure 11 A diagram showing the vertical distribution of the probability density function of the fatigue life of the most dangerous section; Figure 12 A longitudinal waterfall chart of the probability density function distribution of the fatigue life; Figure 13 A diagram showing the fatigue failure probability of the most dangerous section; Figure 14 A diagram showing the vertical fatigue failure probability distribution of the most dangerous section along the rail head; Figure 15 A diagram showing the fatigue life of the most dangerous section under the fatigue failure probability of 5%; Figure 16 A diagram showing the fatigue parameter sensitivity analysis result of the 12# section; Figure 17A schematic diagram of fatigue parameter sensitivity analysis results of the rail head five-layer unit.

[0010] In the figure, 901 is an acquisition module, 902 is a simulation module, 903 is a calculation module, 904 is a prediction module, 905 is an analysis module, and 906 is an evaluation module. DETAILED DESCRIPTION

[0011] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments but not all of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but merely represents selected embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art without creative work based on the embodiments in the present application belong to the scope of protection of the present application.

[0012] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used for distinguishing description, and cannot be understood as indicating or implying relative importance.

[0013] Embodiment 1 The embodiment provides a fatigue damage evaluation method for a bonded insulation joint considering random parameters.

[0014] Referring to Figure 1 , the method includes steps S100 to S600.

[0015] In step S100, first information is acquired, and the first information includes random parameter data in vehicle parameters, steel rail parameters and wheel-rail contact parameters. The present application first constructs a multi-source uncertainty input framework, extracts three types of random parameter data of a vehicle, a steel rail and wheel-rail contact, establishes a probability distribution characteristic library of key variables, and provides a random load spectrum basis conforming to actual service conditions for subsequent dynamic modeling.

[0016] Specifically, the present application considers the influence of multiple uncertainties on wheel-rail contact dynamics, and selects the following dynamic random parameters: axle load , running speed , lateral contact position , rail bottom slope , traction coefficient coefficient of friction primary suspension vertical stiffness primary suspension vertical damping primary suspension longitudinal stiffness primary suspension longitudinal damping fastener vertical stiffness fastener vertical damping fastener longitudinal stiffness fastener longitudinal damping The following table is a distribution table of random parameters.

[0017] Table 1 Probability distribution characteristic parameters of the wheel-glued insulation joint impact transient dynamics model

[0018] Step S200, according to the first information, a dynamic contact model of the wheel and the glued insulation joint is established by using the finite element method and simulation is performed, and dynamic response parameters in the wheel and track contact process are obtained; It can be understood that this step is based on a set of random parameters, and an explicit finite element method is innovatively used to construct a high-frequency impact contact model. Through fine meshing and face-face penalty function contact algorithm, the transient nonlinear dynamic behavior of the wheel rolling over the glued insulation joint is directly captured, and dynamic response parameters such as wheel-rail force and stress wave propagation, which cannot be characterized by traditional quasi-static models, are output.

[0019] Step S300, fatigue damage analysis is performed according to the dynamic response parameters, stress and strain states at the cross section of the glued insulation joint are calculated, and fatigue damage parameters are obtained; It should be noted that this step uses dynamic response data to implement a critical plane search strategy for 70 key cross sections. Through coordinate system rotation, the stress and strain components of each plane are quantified, and the fatigue parameters are calculated by combining a multi-axial fatigue model to reveal the damage evolution mechanism of weak points in the structure.

[0020] Step S400, fatigue life prediction is performed according to the fatigue damage parameters, the fatigue life of each cross section is predicted based on a predetermined proxy model, and fatigue life prediction results are obtained; It can be understood that this step realizes data mapping of “random input-life output” by constructing a proxy model, and uses a whale optimization algorithm to adaptively search for optimal network parameters. A multiple error control strategy (fatigue life + stress response double constraint) guarantees the prediction accuracy.

[0021] Step S500, sensitivity analysis is performed according to the fatigue life prediction results, and sensitivity analysis results are obtained; Preferably, this step quantifies the contribution entropy of random parameters to the failure probability by Sobol variance decomposition, condenses the interaction effect of variables into first-order sensitivity index, and outputs visualized conclusions.

[0022] Step S600, calculate the failure probability of the bonded insulation joint according to the sensitivity analysis result, and evaluate the reliability of the bonded insulation joint according to the failure probability curve under different design life.

[0023] It should be noted that this step combines the failure probability curve and the sensitivity conclusion, calculates the cumulative failure probability by traversing the design life threshold, combines the contribution degree sorting of high sensitivity parameters, and outputs the service reliability grade and fatigue resistance design optimization path of the bonded insulation joint under the random load spectrum.

[0024] Further, step S200 includes steps S210 to S240.

[0025] Step S210, according to the first information, the random distribution characteristics and the numerical range are defined, and the model input parameter set is obtained; Step S220, according to the model input parameter set, a three-dimensional explicit finite element model is constructed, a geometric model including a wheel, a rail, a bonded insulation joint and a suspension system is established, and meshing and material property assignment are performed, to obtain a dynamic contact finite element model; Step S230, according to the dynamic contact finite element model, wheel-rail impact dynamic simulation is carried out, the process of wheel rolling through the bonded insulation joint is simulated, and wheel-rail contact dynamic response data is obtained; Step S240, according to the wheel-rail contact dynamic response data, by analyzing the stress and strain components of the cross-section position of the bonded insulation joint, the multi-axial stress-strain state is calculated, and the dynamic response parameters are obtained.

[0026] Specifically, in this embodiment, a three-dimensional wheel-bonded insulation joint impact transient dynamics model for the bonded insulation joint region is established. As shown in FIG. 2, the model includes a wheel, a rail, a bonded insulation joint and a suspension system. Figure 3The model consists of a 15.0 m long half track and half wheelset, including the suspension mass of the car body and bogie. The wheel tread is modeled with a standard LMA worn tread profile. The track is modeled based on the standard profile geometry of a 60 N rail. The wheel, rail and bonded joint are modeled using hexahedral solid elements. Through the adaptive refinement technique, the element mesh size in the 1 m length of the joint influence zone is 2 mm as the finest mesh. For the parts away from the evaluation zone, the element mesh size can reach 40 mm. The lumped mass of the car body and bogie is modeled as mass elements connected to the wheel through the primary suspension system, which consists of parallel linear spring and viscous damper elements. Since the sleepers and ballast have a small influence on the high frequency dynamic behavior studied, the foundation is simplified as vertical spring and damper elements with displacement limits in the lateral and longitudinal directions. The model parameters are listed in Table 2. The joint is connected using fishplates based on the face-to-face penalty function contact algorithm, and the contact pair between the wheel and the rail is formed.

[0027] Table 2 Deterministic parameters of the wheel-rail transient rolling contact finite element model

[0028] The explicit finite element method is used to realize the wheel-rail dynamic interaction process and frictional contact behavior when the wheel passes through the bonded joint. Through the implicit-explicit sequential solution method, ANSYS / LS-DYNA software is used to simulate the above dynamic process. In the current model, the initial position of the wheel is set as the origin of the coordinate system, and the length of the bonded joint is 1 meter.

[0029] In the wheel-rail rolling contact problem, the rail is subjected to multi-axial and non-proportional loading, which leads to changes in the stress direction and strain range during wheel passage. Therefore, in order to predict the service life of the rail, a multi-axial criterion including non-proportional loading is needed. A combined method based on energy density and critical plane is used to predict the fatigue life at the cross point, which is proposed by Jiang and Sehitoglu (J-S). In this criterion, it is believed that both the normal stress and shear stress components on the critical plane contribute to the damage of the material. The fatigue parameter is defined as: (1); where is the maximum fatigue parameter, , is the maximum normal stress, is the range of normal stress (i.e. maximum normal strain), is the range of shear stress, is the range of shear strain.

[0030] wherein the McLaughlin bracket is denoted as <>, <x>equal to 0.5 * |x|; is the maximum normal stress; is the normal strain range; is the shear stress range; is the shear strain range. The maximum fatigue parameters are determined by investigating all possible planes at a material point through tensor rotation of stress and strain and critical planes. The energy density is calculated as the product of the stress and strain range components on each material plane and each loading increment. The critical plane is defined as the plane with the maximum .

[0031] Here represents the product of the stress and strain range components. The model is physically relevant to both tensile and shear crack initiation and propagation for two loading modes, i.e. tensile and shear. The crack behavior is material and loading size dependent. Some materials exhibit shear cracks, while some exhibit tensile cracks, and some exhibit mixed crack behavior. Therefore, the fatigue crack initiation life is calculated on the critical crack plane.

[0032] (2). where and are the elastic / shear modulus; and are the tensile / shear fatigue strength coefficients; and are the tensile / shear fatigue ductility coefficients.

[0033] Further, the step S300 includes steps S310 to S340.

[0034] The step S310, according to the dynamic response parameters, carries out identification processing, by evaluating the norm equivalent stress distribution of each key section of the bonded insulation joint, selects the element with the maximum norm equivalent stress value in each section as the most dangerous element, and obtains a dangerous element set; The step S320, according to the dangerous element set, carries out element-level stress-strain matrix extraction processing, and obtains a stress-strain matrix dataset; The step S330, according to the stress-strain matrix dataset, carries out critical plane searching, calculates the stress and strain range of all planes through coordinate system rotation, and obtains a fatigue parameter dataset of all planes based on a preset fatigue model; The step S340, according to the fatigue parameter dataset, by screening the maximum fatigue parameter value of all planes of each dangerous element, and recording the spatial angle of the corresponding critical plane, obtains a fatigue damage parameter.

[0035] In this example, fatigue parameters of bonded insulating rail joint (BIRJ) and fatigue life are analyzed in detail as follows. Dynamic responses such as wheel-rail contact force, displacement and acceleration, rail stress and strain can be obtained by the above finite element modeling, which can have a preliminary understanding of the dynamic interaction of wheel-BIRJ, and then some key section positions are selected for key analysis, a total of 70 key sections , ,…, ).

[0036] First, the stress and strain state of the BIRJ solution zone is solved. Von Mises stress is selected as an index to determine the most dangerous element at the same section ), is the element most likely to occur fatigue cracks at section , the principle is to select the element with the maximum Von Mises stress value for fatigue analysis. The fatigue parameters in this element are calculated using the J-S fatigue model. The value of of each plane in this element is calculated according to equation (2), and the plane with the maximum value is considered as the critical plane, where fatigue cracks are most likely to occur. At the same time, the angle of this critical plane is obtained to describe the crack direction. After calculating of all sections, further analysis is needed for the section with the maximum value. Generally, it occurs near the endpost of BIRJ because of the wheel-rail impact. That is, the stress and strain relationship of the elements from the rail head to the rail jaw, i.e., from top to bottom, needs to be analyzed, and the distribution of fatigue parameters along the vertical direction is selected. Five elements from top to bottom are , ,…, .

[0037] The critical plane angle calculation process for fatigue crack initiation of a unit is shown in Figure 4 and Figure 5 : Figure 5 , is the normal vector of the crack plane in the rail section and the angle with the horizontal axis, is its angle with the horizontal axis in the horizontal plane.

[0038] First, the stress matrix and strain matrix : (3); (4); By rotating matrix, The stress and strain matrix of coordinate system are and . Figure 4 There are two auxiliary angles and in the coordinate system, where is the projection of vector on the plane and the angle with the positive direction of axis, is the angle between vector and its projection on the plane. The transformation from the original coordinate system to coordinate system is through two auxiliary angles rotation twice. The first rotation is coordinate system rotating around axis by angle , and the second rotation is rotating around axis by space angle , which satisfies the right-hand inertial system. Assuming the rotation matrix of the first rotation is , and the rotation matrix of the second rotation is , as shown in equations (5) and (6).

[0039] (5); (6); (7); Then, The stress matrix and strain matrix of coordinate system are respectively: (8); (9); The verification work and the simple summary of the dynamics law are as follows. Wheel-rail force is the key to measure whether the instantaneous rolling wheel-rail contact relationship is correct or not. Figure 6 ​The time domain response of wheel-rail force throughout the process is shown, i.e. the wheel starts from the initial position, passes through the dynamic relaxation zone, then the wheel rolling contact impacts the glued insulation joint and passes through. The results show that the wheel-rail force quickly reaches equilibrium, and before passing through the influence zone of the glued insulation joint, i.e. due to the influence of the stiffness of the insulating sheet material, the wheel-rail force decreases slightly with the increase of the vertical wheel-rail force, and then impacts the other end of the rail. The red line is the longitudinal wheel-rail force, which decreases slightly with the increase of the vertical wheel-rail force before passing through the glued insulation joint, and becomes a short reverse when the wheel impacts the glued insulation joint. The lateral wheel-rail force is basically not affected by the glued insulation joint. The amplitude of the vertical wheel-rail force curve reaches a maximum at a position slightly behind the joint. The oscillation of the wheel-rail force in the time domain curve when the wheel impacts the joint appears obvious P1, P2 forces. Therefore, subsequent evaluation will focus on the distribution of stress and strain in the rail section and along the longitudinal direction of the rail before and after the joint, which is the cause of the fatigue damage of the glued insulation joint.

[0040] The oscillation of the wheel-rail force in the time domain curve when the wheel impacts the joint appears obvious P1, P2 forces. Therefore, subsequent evaluation will focus on the distribution of stress and strain in the rail section and along the longitudinal direction of the rail before and after the joint, which is the cause of the fatigue damage of the glued insulation joint.

[0041] Further, the step S400 includes steps S410 to S430.

[0042] Step S410, input feature construction according to fatigue damage parameters, by extracting the fatigue parameters, critical plane angle and corresponding random parameter samples of each dangerous unit, the input feature set is obtained; Step S420, proxy model training according to the proxy model input feature set, the optimal weight and bias parameters of the artificial neural network are searched by the whale optimization algorithm, and the model accuracy is optimized by adopting a multiple error control strategy, to obtain the trained proxy model; Step S430, fatigue life prediction according to the trained proxy model, by inputting the random parameter sample set into the proxy model and outputting the fatigue life value of each section, the fatigue life prediction result is obtained.

[0043] Further, the step S500 includes steps S510 to S530.

[0044] Step S510, fatigue failure probability calculation according to the fatigue life prediction result, by setting the design life threshold of the glued insulation joint, the sample proportion whose predicted life is lower than the threshold is counted, to obtain the failure probability value; Step S520: Calculate sensitivity indices based on the failure probability values, and obtain a set of sensitivity indices by decomposing the total variance of the failure probability into independent contributions and interaction contributions of the input random variables. Step S530: sort the sensitivity indices according to the sensitivity index set and calculate the contribution ratio to obtain the sensitivity analysis result.

[0045] Specifically, in order to efficiently map multiple related responses, the present invention proposes an ANN agent model with multiple error control strategies, including a basic modeling architecture and an error strategy control algorithm.

[0046] The traditional fatigue life proxy model uses ANN neural network to build multiple input features and single output features. However, when facing the complex wheel-rail contact fatigue life prediction, it is not possible to capture the complex nonlinear mapping relationship more accurately through limited data. In this embodiment, the fatigue life model formula is used to calculate , stress and strain amplitude, etc., as output features. In this case, by to multiple related responses The basic idea is: first, a collaborative sampling technique is used to extract a small amount of multi-dimensional input features at the same time. and multiple related responses , in order to establish a training sample data set containing the correlation between each output response, specifically referring to the JS-model , , , The variables are captured as output quantities and the fatigue life is predicted Also as the output; In addition, considering the rectified linear unit function (RELU) with the ability to prevent gradient disappearance and the identity activation (IA) function with the ability to protect the mapping result as the transfer function, the basic architecture of the ANN model is constructed, and the multi-dimensional features are layered. mapped to multiple related responses; finally, in order to avoid falling into local optimization and convergence instability in solving the weight and bias parameters of ANN modeling, a multi-error control algorithm introducing the White Whale Optimization Algorithm (WOA) is proposed to search for the optimal parameters and complete ANN modeling.

[0047] ANN model with a set of input features After multiple hidden layers, each layer is nonlinearly weighted mapped by the RELU function, and the value of the final output layer is obtained by the identity activation function IA function at the final output layer. Assume that the nonlinear action function RELU function of each hidden layer is , the input layer random variable and the response variable of the output layer wheel-rail contact fatigue life can be expressed as: (10); wherein, represents the hidden layer The second nested mapping has: (11); For the parameters of the output layer , it can be extended as follows: (12); The weight vector and the bias vector in formulas (10)-(12) are model training parameters that need to be modified by the ANN neural network through the calculation of the neural network proxy model error. The accurate modeling problem of the ANN model can be converted into the optimization problem of and .

[0048] The fitness function is constructed as follows: Once the optimal uncertain parameters are searched by the training algorithm, the establishment of a high-precision ANN proxy model can be guaranteed.

[0049] (13); wherein represents the optimal solution of the neural network model weight and bias parameters after training; is the fitness function of the model. However, for the calculation of the relatively complex wheel-rail contact problem and the fatigue life and other parameters involved in this paper, it is difficult for the traditional training to obtain a relatively ideal effect.

[0050] In order to improve the accuracy of the complex ANN model proxy, the present application proposes a new error control algorithm: it is assumed that the output of the model contains two parts, one part is the error function of the fatigue life, and the other part is the error function of the stress-strain response value related to the wheel-rail contact calculation related to the fatigue life calculation. The total fitness error can be expressed by the weighted sum of the two error parts: (14); wherein is the weight factor of the error function related to the stress-strain response value, which is used to adjust the size of the fitness function and the dynamic relationship between the two error parts.

[0051] To improve the prediction accuracy and efficiency of the model, the whale optimization algorithm (WOA) is introduced in the training process of the ANN model. WOA is an intelligent optimization algorithm based on the hunting behavior of humpback whales, which is suitable for solving nonlinear optimization problems. In this invention, by introducing it into the training process of ANN, the hyperparameters of neural network model such as learning rate and number of hidden layers and weight factor , so as to further improve the prediction accuracy and efficiency of complex ANN model.

[0052] The core idea of whale optimization algorithm is to simulate the spiral hunting behavior of humpback whales around the prey and dynamic search behavior. The algorithm process includes three key mechanisms.

[0053] Spiral mechanism: assuming that the current optimal solution is the position of the prey, the rest of the individuals gradually approach the position of the current optimal solution. This behavior is realized by the following formula: (15); (16); Where, represents the current iteration, where the position of the current optimal solution (prey), is the position vector, and control the movement behavior of the whale, defined as: (17); (18); With the number of iterations from 2 linearly decreasing to 0, is a random number, taking values in the range [0, 1].

[0054] Shrinking enclosure mechanism: when the whale surrounds the prey, it will approach the prey along the spiral trajectory, and its behavior is simulated as follows: (19); Where, denotes the distance between the whale and the prey in the th iteration, is a spiral shape parameter, is a random number, taking values in the range [-1, 1].

[0055] It should be noted that the humpback whale swims around the prey in a shrinking circle, and at the same time it swims along a spiral path. In order to simulate this simultaneous behavior, it is assumed that there is a 50% probability of choosing the shrinking enclosure mechanism or the spiral model to update the position of the whale during the optimization process.

[0056] (20); Search mechanism: When ≥ 1, the whale will randomly search for a new position to enhance the global search ability: (21); (22); where is a randomly selected individual in the current population.

[0057] The WOA algorithm starts from a set of random solutions. In each iteration, the search agent updates its position according to a randomly selected search agent or the best solution obtained so far. The parameter a is gradually reduced from 2 to 0 to achieve exploration and exploitation, respectively. When > 1, a random search agent is selected; while < 1, the current optimal solution is selected to update the position of the search agent. According to the value of , WOA can switch between spiral motion and circular motion. Finally, the WOA algorithm ends when the termination condition is met. According to the set number of whale individuals N and the number of iterations T, the loop calculation is performed, and the fitness function of each whale individual is sorted in each iteration. After the loop is completed, the optimal hyperparameter distribution When the fatigue field knowledge is embedded in the loss function, the training process of the machine learning model is not only constrained by the fatigue test data, but also by the fatigue field knowledge, thereby realizing the fusion of the fatigue field knowledge and the machine learning model. According to the characteristics of the research problem, researchers can flexibly select the fatigue knowledge and adjust its contribution in the loss function to achieve the best learning effect.

[0058] According to the distribution of input random variables in Table 1, Latin hypercube method is used to sample 400 samples. The 400 samples obtained are divided into two groups, containing 320 samples selected as training samples for ANN modeling, and the remaining 200 samples as test samples, 80 groups for testing the constructed ANN model. In order to be reasonable and accurate for modeling, mean absolute error ( ), mean square error ( ) and R2(R-squared) are used to reflect the effect of ANN modeling. , and the expression of R2 is: (23); (24) (25); where, is the predicted value of the i-th sample; is the true value of the i-th sample; is the true value of the i-th sample; denotes the number of samples.

[0059] Figure 7 and Figure 8 The training accuracy of the un-optimized and optimized ANN models is shown. The ANN model without optimization uses 3 layers with 30 neurons in each layer and is trained using the Scaled Conjugate Gradient algorithm with a learning rate of 0.01. The WOA-ANN algorithm optimizes the learning rate, the number of hidden layers, and the weight factor The fitness function constructed is optimized The trained ANN uses 3 layers with 13 neurons in each layer, a learning rate of 0.00055, and a weight factor of 0.29. Figure 7 and Figure 8 It can be seen from the comparison that the ANN model without optimized parameters has an error in the predicted value of some samples that can exceed 5% for complex nonlinear optimization problems, and the test data set is significantly less than the training data set. The overall error of the optimized network model is very small, with a maximum error of 1.76%. Although the generalization error is slightly higher than the training error, the ANN model can still maintain high generalization accuracy outside the sample points.

[0060] Based on this, 100,000 samples are sampled by the LHS normal sampling method, and a probability statistical analysis is performed on the fatigue life results under the condition of large sample data. The fatigue life Nf in different dimensions is calculated, as shown in Figures 9-11 The life value is logarithmically processed in the figure. Figure 9 is the Nf distribution diagram of the most dangerous section of the entire bonded insulation joint. It can be seen that it obeys the logarithmic normal distribution. Figure 10 is the probability density distribution cloud diagram of the logarithmic life of the 70 key sections along the longitudinal direction. Figure 11 is the logarithmic life distribution of the 5-layer elements from the outer surface to the subsurface at the most dangerous section. The dashed line is the life distribution calculated when the random parameters take the mean value. It can be seen that the mean parameter is conservative in calculating the outermost life of the impact section, and is aggressive in calculating the subsurface.

[0061] Figure 12 The probability density distribution of the fatigue life at some sections before and after the impact is shown. The fatigue life distribution before the impact is narrow and high, and the fatigue life distribution at the rail end where the impact does not occur is wide and low, and appears more discrete.

[0062] Failure probability is one of the core indicators in reliability analysis, which measures the possibility of a structural component failing to meet the design requirements under certain conditions. In the study of fatigue life, when the actual fatigue life is lower than the design life, the structure will enter a failure state. To quantify this possibility of failure, the fatigue life distribution under a large sample is statistically analyzed, and the failure probability is calculated. The failure probability can be calculated by the following formula: (26); (27); where, is the design life, is the actual life. When , it represents the failure of the structure, and the failure probability is . The failure probability under each design life can be obtained by statistically analyzing the calculated fatigue life values of 100,000 samples, and a failure probability curve is obtained. As shown in Figure 13 , it is the failure probability curve of the fatigue life of the most dangerous section of BIRJ. The red dashed line in the figure marks the fatigue failure life of 341,871 times under a 5% failure probability. Figure 14 The fatigue failure probability curves of the five positions along the outer surface to the inner of the most dangerous section are shown, and the 95% reliable lives of the five positions are 341,871 times, 831,763 times, 5,495,408 times, 13,489,628 times, and 74,131,024 times, respectively. Figure 15 The fatigue failure lives of different sections under a 5% failure probability are shown. It can be found that Section #11, Section #12, and Section #13 are more dangerous than other sections. Among them, Section #11 and Section #12 are the sections on both sides of the rail end. Although Section #11 is not subjected to impact, it still bears a large stress-strain change process due to the structural irregularity of the glued insulation joint, resulting in a fatigue life significantly lower than other areas.

[0063] Sensitivity analysis quantifies the influence of uncertain factors, enabling the determination of the relative importance of each random variable on the rolling contact fatigue damage state of BIRJ. In random parameter analysis, Sobol sensitivity analysis is widely used to quantify the uncertainty contribution of input random variables to output responses. Sobol method comprehensively evaluates the importance of each random variable by decomposing the total variance of the system into independent effects of each input variable and their interaction effects. Specifically, for the objective function , the Sobol index is calculated by the following formula: (28); (29); in, is the input variable The first-order sensitivity index of , which indicates the independent contribution of the variable to the output response, The variable is measured and In practical applications, Sobol sensitivity analysis numerically approximates the sensitivity indices of each order by evenly dividing the high-dimensional random input space and combining it with the Monte Carlo sampling method. This method is efficient and robust, capable of capturing the complex interaction effects in high-dimensional nonlinear systems. This analysis provides theoretical support for understanding the impact of uncertainties on system reliability and provides a decision-making basis for optimized design. By identifying key parameters, engineers can prioritize the variables that have the greatest impact on fatigue life during the design phase, thereby achieving more efficient and reliable BIRJ structural design.

[0064] First, Sobol sensitivity analysis was used to quantify the effect of each random parameter on the fatigue parameter. contribution. Figure 16 and Figure 17 The results show that the lateral contact position and the primary suspension vertical stiffness have an impact on fatigue parameters. It shows high sensitivity and certain sensitivity to speed. At the impact position, from the rail surface to the inside, the sensitivity of the lateral position gradually decreases, while the sensitivity of the vertical stiffness of the primary suspension gradually increases.

[0065] Furthermore, step S600 includes steps S610 to S630.

[0066] Step S610: Calculate the fixed life failure probability based on the sensitivity analysis results, and obtain the failure probability value by statistically predicting the proportion of samples whose life is lower than the design life; Step S620: quantify the sensitivity coefficient according to the failure probability value, and obtain a parameter sensitivity coefficient set by calculating the sensitivity coefficient percentage of each parameter; Step S630: Perform visualization and optimization suggestion generation processing based on the parameter sensitivity coefficient set, and output the reliability optimization direction conclusion of the adhesively bonded insulating joint by making a sensitivity pie chart and marking the key parameter contribution values.

[0067] In a specific embodiment, based on the above results, 341871 times is set as the design life, and the failure probability is calculated. The sensitivity coefficients for each parameter were calculated. Further sensitivity analysis quantified the contribution of 14 random parameters to the failure probability. The sensitivity coefficients for lateral contact position were 35.7%, operating speed was 21.9%, and axle load, fastener vertical damping, and primary suspension longitudinal damping were 10.4%, 9.8%, and 6.2%, respectively. The total contribution of the remaining parameters was 15.0%. The distribution of sensitivity coefficients for each parameter indicated that lateral contact position and operating speed were the dominant factors affecting reliability.

[0068] Example 2: like Figure 2 As shown, this embodiment provides a fatigue damage assessment system for adhesively bonded insulating joints considering random parameters, the system comprising: An acquisition module 901 is configured to acquire first information, where the first information includes random parameter data among vehicle parameters, rail parameters, and wheel-rail contact parameters; A simulation module 902 is configured to establish a dynamic contact model between the wheel and the adhesively bonded insulating joint using a finite element method based on the first information and perform simulation to obtain dynamic response parameters during the wheel-rail contact process; The calculation module 903 is used to perform fatigue damage analysis based on the dynamic response parameters and obtain fatigue damage parameters by calculating the stress and strain state at the cross section of the adhesively bonded insulating joint; The prediction module 904 is used to predict fatigue life according to fatigue damage parameters, and predict the fatigue life of each section based on a preset proxy model to obtain fatigue life prediction results; An analysis module 905 is used to perform sensitivity analysis based on the fatigue life prediction result to obtain a sensitivity analysis result; The evaluation module 906 is used to calculate the failure probability of the adhesively bonded insulating joint according to the sensitivity analysis results, and evaluate the reliability of the adhesively bonded insulating joint according to the failure probability curves under different design lifespans.

[0069] In a specific embodiment of the present invention, the simulation module 902 includes: A first simulation unit is configured to obtain a model input parameter set by defining random distribution characteristics and a value range based on the first information; The second simulation unit is used to construct a three-dimensional explicit finite element model based on the model input parameter set. By establishing a geometric model including wheels, rails, adhesively bonded insulated joints and suspension systems, and performing meshing and material property assignment, a dynamic contact finite element model is obtained; The third simulation unit is used to perform wheel-rail impact dynamic simulation based on the dynamic contact finite element model, and obtain the wheel-rail contact dynamic response data by simulating the process of the wheel rolling through the adhesively bonded insulating joint; The fourth simulation unit is configured to calculate a multi-axial stress-strain state according to the wheel-rail contact dynamic response data by analyzing stress and strain components of the section position of the glued insulation joint, and obtain the dynamic response parameter.

[0070] In an embodiment of the present application, the calculation module 903 includes: The first calculation unit is configured to perform identification processing according to the dynamic response parameter, select a cell with the maximum norm equivalent stress value in each section as the most dangerous cell by evaluating the norm equivalent stress distribution of each key section of the glued insulation joint, and obtain a dangerous cell set; The second calculation unit is configured to perform unit-level stress-strain matrix extraction processing according to the dangerous cell set, and obtain a stress-strain matrix data set; The third calculation unit is configured to perform critical plane search according to the stress-strain matrix data set, calculate the stress and strain range of all planes by coordinate system rotation, and obtain a fatigue parameter data set of all planes based on a preset fatigue model; The fourth calculation unit is configured to obtain a fatigue damage parameter by screening the maximum fatigue parameter value of all planes of each dangerous cell and recording the spatial angle of the corresponding critical plane according to the fatigue parameter data set.

[0071] In an embodiment of the present application, the prediction module 904 includes: The first prediction unit is configured to perform input feature construction according to the fatigue damage parameter, obtain an input feature set by extracting the fatigue parameter, critical plane angle and corresponding random parameter sample of each dangerous cell; The second prediction unit is configured to perform proxy model training according to the proxy model input feature set, search for optimal weight and bias parameters of an artificial neural network by a whale optimization algorithm, and optimize model accuracy by a multiple error control strategy, and obtain a trained proxy model; The third prediction unit is configured to perform fatigue life prediction according to the trained proxy model, input the random parameter sample set into the proxy model and output the fatigue life value of each section, and obtain a fatigue life prediction result.

[0072] In an embodiment of the present application, the analysis module 905 includes: The first analysis unit is configured to perform fatigue failure probability calculation according to the fatigue life prediction result, set a design life threshold of the glued insulation joint, count the sample proportion whose predicted life is lower than the threshold, and obtain a failure probability value; The second analysis unit is configured to perform sensitivity index calculation according to the failure probability value, decompose the total variance of the failure probability into independent contributions and interaction effect contributions of each input random variable, and obtain a sensitivity index set; The third analysis unit is configured to sort the sensitivity indexes according to the sensitivity index set and calculate the contribution ratio to obtain a sensitivity analysis result.

[0073] In one specific embodiment of the present application, the evaluation module 906 comprises: The first evaluation unit is configured to calculate the fixed life failure probability according to the sensitivity analysis result, calculate the sample proportion with the predicted life lower than the designed life, and obtain a failure probability value. The second evaluation unit is configured to quantify the sensitivity coefficients according to the failure probability value, calculate the sensitivity coefficient percentage of each parameter, and obtain a parameter sensitivity coefficient set. The third evaluation unit is configured to perform visualization and optimization suggestion generation processing according to the parameter sensitivity coefficient set, make a sensitivity pie chart, mark the contribution value of the key parameter, and output a reliability optimization direction conclusion of the adhesive insulation joint.

[0074] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.< / x>

Claims

1. A fatigue damage assessment method for adhesively bonded insulated joints considering random parameters, characterized in that: include: Acquiring first information, wherein the first information includes random parameter data among vehicle parameters, rail parameters, and wheel-rail contact parameters; Based on the first information, a dynamic contact model of the wheel and the adhesively bonded insulating joint is established using the finite element method and simulated to obtain dynamic response parameters during the wheel-rail contact process; Performing fatigue damage analysis based on the dynamic response parameters, and obtaining fatigue damage parameters by calculating stress and strain states at a cross section of the adhesively bonded insulating joint; Fatigue life prediction is performed according to the fatigue damage parameters, and fatigue life of each section is predicted based on a preset proxy model to obtain fatigue life prediction results; Perform sensitivity analysis based on the fatigue life prediction result to obtain a sensitivity analysis result; The failure probability of the adhesively bonded insulating joint is calculated based on the sensitivity analysis results, and the reliability of the adhesively bonded insulating joint is evaluated based on the failure probability curves under different design lifespans.

2. The fatigue damage assessment method for adhesively bonded insulated joints considering random parameters according to claim 1 is characterized in that: Based on the first information, a dynamic contact model between the wheel and the adhesively bonded insulating joint is established using the finite element method and simulated, including: According to the first information, a model input parameter set is obtained by defining random distribution characteristics and value ranges; A three-dimensional explicit finite element model is constructed according to the model input parameter set, and a dynamic contact finite element model is obtained by establishing a geometric model including wheels, rails, adhesively bonded insulated joints and a suspension system, and performing meshing and material property assignment; Performing a wheel-rail impact dynamic simulation based on the dynamic contact finite element model, and obtaining wheel-rail contact dynamic response data by simulating a process in which a wheel rolls through a bonded insulating joint; According to the wheel-rail contact dynamic response data, the stress and strain components at the cross-section of the adhesively bonded insulating joint are analyzed, the multiaxial stress-strain state is calculated, and the dynamic response parameters are obtained.

3. The fatigue damage assessment method for adhesively bonded insulated joints considering random parameters according to claim 1, characterized in that: Performing fatigue damage analysis based on the dynamic response parameters includes: Identification processing is performed based on the dynamic response parameters, and by evaluating the normalized equivalent stress distribution of each key section of the adhesively bonded insulating joint, the unit with the largest normalized equivalent stress value in each section is selected as the most dangerous unit to obtain a dangerous unit set; Performing unit-level stress-strain matrix extraction processing according to the dangerous unit set to obtain a stress-strain matrix data set; Performing a critical plane search based on the stress-strain matrix data set, calculating the stress and strain ranges of all planes by rotating the coordinate system, and obtaining fatigue parameter data sets of all planes based on a preset fatigue model; According to the fatigue parameter data set, the fatigue damage parameter is obtained by screening the maximum fatigue parameter values ​​of all planes of each dangerous unit and recording the spatial angle of the corresponding critical plane.

4. The fatigue damage assessment method for adhesively bonded insulated joints considering random parameters according to claim 1, characterized in that: Fatigue life prediction is performed based on the fatigue damage parameters, including: constructing input features according to the fatigue damage parameters, and obtaining an input feature set by extracting fatigue parameters, critical plane angles, and corresponding random parameter samples of each dangerous unit; Performing proxy model training based on the proxy model input feature set, searching for optimal weight and bias parameters of the artificial neural network using a whale optimization algorithm, and optimizing the model accuracy using a multiple error control strategy to obtain a trained proxy model; Fatigue life prediction is performed based on the trained proxy model, and a fatigue life prediction result is obtained by inputting a random parameter sample set into the proxy model and outputting a fatigue life value of each section.

5. The fatigue damage assessment method for adhesively bonded insulated joints considering random parameters according to claim 1, characterized in that: A sensitivity analysis is performed based on the fatigue life prediction results, including: Calculating the fatigue failure probability based on the fatigue life prediction results, setting a design life threshold for the adhesively bonded insulating joint, and calculating the proportion of samples with a predicted life below the threshold to obtain a failure probability value; Calculating sensitivity indices based on the failure probability values, and obtaining a set of sensitivity indices by decomposing the total variance of the failure probability into independent contributions and interaction effect contributions of the input random variables; The sensitivity index is sorted according to the sensitivity index set and the contribution ratio is calculated to obtain the sensitivity analysis result.

6. A fatigue damage assessment system for adhesively bonded insulated joints considering random parameters, characterized in that: include: an acquisition module, configured to acquire first information, wherein the first information includes random parameter data among vehicle parameters, rail parameters, and wheel-rail contact parameters; a simulation module, configured to establish a dynamic contact model between the wheel and the adhesively bonded insulating joint using a finite element method based on the first information and perform simulation to obtain dynamic response parameters during the wheel-rail contact process; a calculation module, configured to perform fatigue damage analysis based on the dynamic response parameters and obtain fatigue damage parameters by calculating stress and strain states at a cross section of the adhesively bonded insulating joint; A prediction module, configured to predict fatigue life according to the fatigue damage parameters, and to predict the fatigue life of each section based on a preset proxy model to obtain a fatigue life prediction result; An analysis module, configured to perform sensitivity analysis based on the fatigue life prediction result to obtain a sensitivity analysis result; An evaluation module is used to calculate the failure probability of the adhesively bonded insulating joint according to the sensitivity analysis result, and to evaluate the reliability of the adhesively bonded insulating joint according to the failure probability curves under different design lifespans.

7. The fatigue damage assessment system for adhesively bonded insulated joints considering random parameters according to claim 6, characterized in that: The simulation module includes: A first simulation unit is configured to obtain a model input parameter set by defining random distribution characteristics and a value range based on the first information; a second simulation unit, configured to construct a three-dimensional explicit finite element model according to the model input parameter set, and obtain a dynamic contact finite element model by establishing a geometric model including wheels, rails, adhesively bonded insulated joints, and a suspension system, and performing meshing and material property assignment; a third simulation unit, configured to perform a wheel-rail impact dynamic simulation based on the dynamic contact finite element model, and obtain wheel-rail contact dynamic response data by simulating a process in which a wheel rolls through a bonded insulating joint; The fourth simulation unit is used to calculate the multi-axial stress-strain state based on the wheel-rail contact dynamic response data by analyzing the stress and strain components at the cross-section position of the adhesively bonded insulating joint to obtain dynamic response parameters.

8. The fatigue damage assessment system for adhesively bonded insulating joints considering random parameters according to claim 6, characterized in that: The calculation module includes: A first calculation unit is configured to perform identification processing based on the dynamic response parameters, evaluate the normalized equivalent stress distribution of each key section of the adhesively bonded insulating joint, select the unit with the largest normalized equivalent stress value in each section as the most dangerous unit, and obtain a dangerous unit set; A second calculation unit is used to perform unit-level stress-strain matrix extraction processing based on the dangerous unit set to obtain a stress-strain matrix data set; a third calculation unit, configured to search for a critical plane according to the stress-strain matrix data set, calculate the stress and strain ranges of all planes by rotating the coordinate system, and obtain fatigue parameter data sets of all planes based on a preset fatigue model; The fourth calculation unit is used to obtain fatigue damage parameters by screening the maximum fatigue parameter values ​​of all planes of each dangerous unit according to the fatigue parameter data set and recording the spatial angle of the corresponding critical plane.

9. The fatigue damage assessment system for adhesively bonded insulated joints considering random parameters according to claim 6, characterized in that: The prediction module includes: A first prediction unit is configured to construct input features based on the fatigue damage parameters, and obtain an input feature set by extracting fatigue parameters, critical plane angles, and corresponding random parameter samples of each dangerous unit; a second prediction unit, configured to train the proxy model based on the proxy model input feature set, search for optimal weight and bias parameters of the artificial neural network using a whale optimization algorithm, and optimize the model accuracy using a multiple error control strategy to obtain a trained proxy model; The third prediction unit is used to perform fatigue life prediction based on the trained proxy model, and obtain fatigue life prediction results by inputting a random parameter sample set into the proxy model and outputting the fatigue life value of each section.

10. The fatigue damage assessment system for adhesively bonded insulated joints considering random parameters according to claim 6, characterized in that: The analysis module includes: A first analysis unit is configured to calculate the fatigue failure probability based on the fatigue life prediction result, and obtain a failure probability value by setting a design life threshold of the adhesively bonded insulating joint and calculating a proportion of samples with a predicted life lower than the threshold; A second analysis unit is configured to calculate a sensitivity index based on the failure probability value, and obtain a sensitivity index set by decomposing the total variance of the failure probability into independent contributions and interaction contributions of the input random variables; The third analysis unit is used to sort the sensitivity index according to the sensitivity indicator set and calculate the contribution ratio to obtain a sensitivity analysis result.

Citation Information

Patent Citations

  • Method for evaluating short-wave geometric irregularity of steel rail weld zone of high-speed railway

    CN115730493A

  • Method for evaluating acceleration adaptability of high-speed turnout

    CN116029138A

  • Structure fatigue reliability evaluation method based on WOA-XGBoost proxy model

    CN118313203A

Cited By

  • Cutting process parameter design method, device and equipment and storage medium

    CN121365572A

  • Cutting process parameter design method, device, equipment and storage medium

    CN121365572B