Model data processing method and system of train energy absorption structure in collision simulation

Abstract

The invention relates to the technical field of train crashworthiness performance analysis, and discloses a model data processing method and system for a train energy absorption structure in collision simulation, so as to reduce cost, increase efficiency and ensure reliability. The method comprises the following steps: taking a force-displacement curve of each sample in a first data set as input and output of a self-supervised network model, and extracting an intermediate layer hidden variable corresponding to each force-displacement curve after dimensionality reduction in the self-supervised network model; according to the second data set, training, verifying and testing structural parameters of the energy absorption structure and a new network model of which the output is a hidden variable of the middle layer; combining the new network model with a decoder part of the self-supervised network model to obtain a target model based on a structure parameter prediction force-displacement curve; and embedding the target model into a marshalling train collision dynamic model, and predicting and outputting a force-displacement curve required by collision simulation according to input data of a dynamic simulation environment.

Description

Technical Field

[0001] The present invention relates to the technical field of train crashworthiness analysis, and in particular to a model data processing method and system for a train energy-absorbing structure in collision simulation. Background Art

[0002] Passive protection of trains involves complex mechanical properties of geometric nonlinear material contact, deformation and coupling, which are generally divided into five stages: coupler buffer, crush tube, coupler shear bolt, end energy absorption structure and car body energy absorption. Once the car body energy absorption part is generated, it will cause train deformation and passenger casualties. Therefore, the analysis and design of train crashworthiness is crucial, especially the crashworthiness design method of marshaling trains.

[0003] In the design of modern urban rail trains, impact energy management is a key link in ensuring the passive safety of the train. The multi-level energy management strategy aims to maximize the protection of the passenger area, ensure its structural integrity, and reduce injuries to the occupants. In addition to these energy management measures, a detailed marshaling crashworthiness analysis is also essential to ensure that the entire train marshaling has sufficient crashworthiness and safety in the event of a collision. Through these comprehensive measures, modern urban rail train design can better cope with potential collision risks and improve overall safety performance.

[0004] The traditional finite element method has high accuracy in calculating the crashworthiness of marshaled trains, but its computational efficiency is low. If the finite element simulation is simply embedded in the crashworthiness analysis of energy-absorbing structures and marshaled trains, it will be difficult to achieve rapid mapping and iterative optimization of structural parameters and the crashworthiness of marshaled trains. Some existing studies have introduced dynamic methods to analyze the crashworthiness of energy-absorbing structures and marshaled trains, but it is easy to produce deviations in the parameter inversion process and the efficiency improvement is limited. Summary of the invention

[0005] The present invention aims to disclose a model data processing method and system for a train energy absorbing structure in collision simulation, so as to achieve cost reduction and efficiency improvement and ensure reliability.

[0006] To achieve the above-mentioned purpose, the model data processing method of the train energy absorbing structure in the collision simulation disclosed by the present invention includes: Step S1, obtaining force-displacement curves of different energy absorbing structures based on finite element simulation to construct a first data set; Step S2: using the force-displacement curve of each sample in the first data set as the input and output of the self-supervised network model, and extracting the intermediate layer hidden variables corresponding to each force-displacement curve after dimension reduction in the self-supervised network model; Step S3, constructing a second data set according to the intermediate layer latent variables corresponding to the structural parameters of different energy absorbing structures in the self-supervised network model; Step S4, training, verifying and testing a new network model whose input is the structural parameters of the energy absorbing structure and whose output is the intermediate layer latent variables according to the second data set; Step S5, combining the new network model with the decoder part of the self-supervised network model to obtain a target model for predicting the force-displacement curve based on structural parameters, and the weight, offset value information and processing logic of the combined decoder part are consistent with the decoder weight, offset value information and processing logic obtained by training in step S2; Step S6: embedding the target model into the train-forming collision dynamics model to predict and output the force-displacement curve required for the collision simulation according to the input data of the dynamics simulation environment.

[0007] Preferably, in step S1, the finite element simulation specifically comprises constructing and experimentally verifying a finite element parametric model of the train energy absorption structure in finite element simulation software, wherein the finite element parametric model encapsulates the entire process of geometric parameter driving, material association, load application, mesh division and calculation submission in a coded manner to achieve automatic mapping of "parameter input-simulation output", and the first data set is constructed based on the finite element parametric model using a DOE (Design Of Experiments) method.

[0008] Preferably, in step S2, the force-displacement data of each sample is converted into a matrix of 2×1000 dimensions, and then four algorithms, AE, RNN-AE, LSTM-AE and GRU-AE network models, are used for model training respectively, and the prediction accuracy of different models is compared and analyzed, and the network model with the highest accuracy in predicting the output curve is used as the self-supervised network model.

[0009] Preferably, in step S4, the new network model adopts a CNN network model, the energy absorption structure is specifically an anti-climbing device in a hook buffer device, and the structural parameters include: the diameter of the parting surface between the friction tube and the friction bearing, the pre-tightening pressure of the friction bearing, the dynamic friction coefficient of the friction rubber, the pre-connection length between the friction tube and the friction bearing, the bearing thickness coefficient and the taper coefficient.

[0010] Preferably, in the marshalling train collision dynamics model, the coupler buffer device includes a coupler and the anti-climbing device, the constitutive relationship of the local plastic deformation of the coupler buffer device is defined by a generalized spring, and the mechanical characteristic curves of the energy absorption structure and the coupler are constructed by using a nonlinear hysteresis mathematical model, and the nonlinear hysteresis mathematical model is described by the following three stages: The calculation equation for the loading phase is: ;in, It is the impedance force when the coupler or anti-climber is loaded; is the initial pressure of the coupler or anti-climber; is the sliding friction; is the static friction; It represents the relative longitudinal velocity between the moving end and the stationary end of the hook. represents Stribeck critical judgment speed; The initial pressure of the coupler and the anti-climber will affect the friction factor at the beginning of the impact. In the actual calculation process, it is often calculated as a constant and rewritten in the form of stiffness: ;in, Provides linear stiffness for hook buffer loading; is the stiffness under the simultaneous action of sliding and static friction; represents the displacement of the buffer along the longitudinal direction of the train during the loading phase; In the unloading stage, the hook resistance and stroke can be considered to be linearly related, and the calculation method is: ; is the linear stiffness in the unloading stage; Among them, there is fluctuation between loading and unloading, the hook slow motion section and the static section are relatively static, and the relative speed between the two is 0, thus resulting in a zero speed stage; When the static end and the moving end of the hook-buffer system are in the stage of zero relative speed, the direction of the speed will change, resulting in the magnitude and direction of the instantaneous force in the numerical simulation not being unique; in order to accurately determine the friction damping characteristics at zero speed, so that the impedance force in the calculation process has a unique solution and avoid fluctuations in the simulation output, the mathematical model of the instantaneous impedance force at zero speed is established as follows: ; in, for The impedance of the momentary hook and buffer device; for Impedance of the momentary hook and buffer device; for The hook-and-buffer device hook-and-buffer displacement at the moment; for The hook displacement of the hook buffer device at the moment; Geometric control parameters for controlling loading and unloading boundary changes; It is the boundary force of the hook-slow characteristic curve; pass Time displacement and Time displacement The magnitude of the boundary force is determined by ,when hour, Take the impedance during the loading phase, when hour, Take the impedance during the unloading phase.

[0011] Preferably, in step S2 and step S5, before model training, the force-displacement curve and the structural parameters are normalized respectively. In the normalization process, all values ​​are directly normalized by dividing them by a maximum value. The calculation formula is: ; ; Where: and The normalized value and true value of the jth group of DOE calculation results corresponding to the i-th input structural parameter; max is the maximum true value of the i-th input structural parameter in all DOE calculation results; and are the normalized value and true value of the kth point on the impact force-displacement curve in the jth group of DOE calculation results; max is the maximum true value of the i-th crashworthiness indicator in all DOE calculation results.

[0012] Preferably, the present invention also includes: in the collision simulation, the structural parameters of the end friction energy absorption structure are used as input parameters, and the vertical displacement and acceleration of each car body of the multi-carriage train are used as output, so as to study the influence of the parameters of the train end friction energy absorption structure on the posture optimization of the multi-carriage train; and in the optimization process, the minimum vertical displacement and acceleration of the car body are used as the optimization target, and the optimal combination of energy absorption structure parameters is obtained by making the optimal decision through the Pareto solution set, so that under the condition of collision or other dynamic loads, the vertical displacement and acceleration of the train are minimized. Optionally, the specific formula of the optimization objective is as follows: ; In the formula, represents the acceleration of the stationary car in section i; represents the acceleration of the stationary car at section j; represents the vertical lift of the i-th stationary vehicle; represents the vertical lift of the jth moving vehicle; The diameter of the parting surface between the friction tube and the friction bearing for absorbing friction energy; is the bearing thickness coefficient; It is the preload pressure of the friction bearing; It is the pre-connection length between the friction tube and the friction bearing; is the dynamic friction coefficient of the friction rubber; is the taper coefficient.

[0013] To achieve the above objectives, the present invention also discloses a model data processing system for a train energy-absorbing structure in a collision simulation, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the computer program.

[0014] The present invention has the following beneficial effects: 1. In step S2, the force-displacement curve is used to predict the force-displacement curve. The latent variables in the middle layer are equivalent to the code book for sending telegrams, which are shared by the encoder and decoder. They can be regarded as the encoded values ​​after the high-dimensional time series data of the force-displacement curve is reduced in dimension.

[0015] 2. In step S4, the input and output data of the model are non-time series data, and a commonly used model can achieve reliable results.

[0016] 3. In order to analyze the energy absorption characteristics of the friction energy absorption structure under quasi-static compression, a WED-600 electro-hydraulic servo universal testing machine was used for quasi-static compression testing. During the test, the mixed friction energy absorption structure was compressed by driving the middle beam at a speed of 5 mm / min. After the test system is started, the middle beam will compress the mixed friction test energy absorption structure downward at a constant speed until the test displacement reaches 120 mm, the test stops and automatically unloads, and then the corresponding total energy absorption is calculated according to the load displacement curve obtained from the test acquisition system. In order to verify the accuracy of the combined target model of the present invention, multiple groups of force-displacement curves and measured crashworthiness indicators based on the combined target model of the present invention and finite element simulation are compared; it is concluded that the errors of the crashworthiness indicators obtained by the method of the present invention are very small, the maximum error is only 2.65%, and the overall error is less than 5%. As a result, after the combined target model of the present invention is applied to the collision dynamics model of the marshaled train, it can be used to evaluate the crashworthiness of the train without the participation of the finite element model software, and guide the parameter optimization of the energy absorption structure.

[0017] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 It is a flow chart of the model data processing method of the train energy absorbing structure in the collision simulation disclosed in Example 1 of the present invention.

[0019] Figure 2 It is a schematic diagram of the hysteresis behavior of the energy-absorbing structure of a marshaled train disclosed in Example 1 of the present invention during the continuous loading and unloading process in a collision simulation. DETAILED DESCRIPTION

[0020] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0021] Example 1 This embodiment discloses a model data processing method of a train energy absorbing structure in collision simulation.

[0022] like Figure 1 As shown, the process of the method in this embodiment includes the following steps: Step S1, obtaining force-displacement curves of different energy absorbing structures based on finite element simulation to construct a first data set.

[0023] In this step, the finite element simulation specifically includes: constructing and experimentally verifying a finite element parametric model of the train energy absorption structure in the finite element simulation software, wherein the finite element parametric model encapsulates the entire process of geometric parameter driving, material association, load application, mesh division and calculation submission in a coded manner to achieve automatic mapping of "parameter input-simulation output", and the first data set is constructed based on the finite element parametric model using the DOE method.

[0024] Preferably, the present embodiment may use Latin hypercube sampling for sampling. This method can evenly distribute sample points in multidimensional space, thereby effectively covering the entire parameter space, reducing sample bias, and improving the efficiency and accuracy of the optimization process. In order to ensure the accuracy of sample calculation, the number of DOE samples can be set to 1000 groups. This sample size can fully capture the impact of changes in input parameters on system response and improve the accuracy of model prediction.

[0025] In this step, the so-called different energy absorbing structures refer to structures with similar structures but different specific parameter values ​​so as to seek the optimal parameter combination. Optionally, the energy absorbing structure can be specifically an anti-climbing device in the hook buffer device, and the corresponding structural parameters include: the diameter of the parting surface between the friction tube and the friction bearing, the pre-tightening pressure of the friction bearing, the dynamic friction coefficient of the friction rubber, the pre-connection length of the friction tube and the friction bearing, the bearing thickness coefficient and the taper coefficient.

[0026] Step S2: using the force-displacement curve of each sample in the first data set as the input and output of the self-supervised network model, and extracting the intermediate layer hidden variables corresponding to each force-displacement curve after dimensionality reduction in the self-supervised network model.

[0027] An autoencoder is an unsupervised learning neural network model used to learn low-dimensional representations of data. It consists of two parts: an encoder and a decoder. It compresses the input data into a lower-dimensional encoded representation and then reconstructs it back to the original data through a decoder. To this end, the self-supervised network model in this step can be implemented using an autoencoder. Before training the self-supervised network model of the force-displacement curve, the force-displacement curve and structural parameter data are normalized.

[0028] Since the force-displacement curve is a high-dimensional time series curve, the biggest difference between it and the structural parameter data is that it is arranged horizontally along the last dimension. This dimension represents different types of structural parameters in the structural parameters, but in the force-displacement curve, it represents the time series data of a specific DOE calculation result. If the MinMaxScaler algorithm of the sklearn library used by the structural parameters is used for data normalization, the force-displacement curve data will be normalized to the maximum and minimum according to the column index (the second to last dimension), which will lose the time series information of the force-displacement curve, resulting in the maximum value 1 or the minimum value 0 appearing at the same time after normalization of a curve, thus turning the obviously continuous curve into discrete information. Therefore, considering that the impact force-displacement curves are stacked along the rows and they all belong to the same class, in the normalization process, all values ​​are directly normalized by dividing them by a maximum value, so that all DOE impact force-displacement curves can well retain the spatial time series continuity. The normalization method of the impact force-displacement curve is as follows: ; ; Where: and The normalized value and true value of the jth group of DOE calculation results corresponding to the i-th input structural parameter; max is the maximum true value of the i-th input structural parameter in all DOE calculation results; and are the normalized value and true value of the kth point on the impact force-displacement curve in the jth group of DOE calculation results; max is the maximum true value of the i-th crashworthiness indicator in all DOE calculation results.

[0029] In order to obtain the intermediate layer hidden variables of the force-displacement curve in the self-supervised network model, the force-displacement data of each sample is converted into a matrix of 2*1000 dimensions, and then four algorithms, AE, RNN-AE, LSTM-AE and GRU-AE network models, are used for model training. The input and output of the model are the force-displacement curve itself. Then, the prediction accuracy of different models is compared and analyzed, and the network model with the highest accuracy in predicting the output curve is used as the self-supervised network model. During the training process, optionally, the dimension of the intermediate layer hidden variables can be set to 6; the first data set can be divided into a training set, a test set and a validation set in a ratio of 7:2:1. Furthermore, in order to improve the performance of the model, the learning rate early stopping and the learning rate dynamic adjustment can be used to optimize the learning rate and training depth, and the Adam optimization algorithm can be used to optimize the model training.

[0030] In this step, after the force-displacement curve self-supervised network model and the deep learning model of the structural parameters and intermediate layer latent variables are successfully trained, the weights of the decoder part of the force-displacement curve self-supervised network model are frozen. Among them, the intermediate layer latent variables are the input of the decoder part, and the decoder is used to decode the intermediate layer latent variables into force-displacement curves; on the contrary, the encoder is used to reduce the dimension of the intermediate layer latent variables after the high-dimensional time series data of the force-displacement curve.

[0031] Step S3: construct a second data set according to the intermediate layer latent variables corresponding to the structural parameters of different energy absorbing structures in the self-supervised network model.

[0032] Step S4: training, verifying and testing a new network model whose input is the structural parameters of the energy absorbing structure and whose output is the intermediate layer latent variables according to the second data set.

[0033] In this step, since neither the structural parameters nor the intermediate layer latent variables are time series data, CNN network training can be used to obtain the interaction mechanism and correlation mechanism between the structural parameters and the intermediate layer latent variables.

[0034] Step S5, combining the new network model with the decoder part of the self-supervised network model to obtain a target model for predicting the force-displacement curve based on structural parameters, and the weights, offset value information and processing logic of the combined decoder part are consistent with the decoder weights, offset value information and processing logic obtained by training in step S2. In this step, weights are parameters connecting neurons and are used to adjust the influence of input data; offsets are bias parameters of each neuron and are used to adjust the activation threshold of neurons; both are industry terms and will not be elaborated on.

[0035] Step S6: embedding the target model into the train-forming collision dynamics model to predict and output the force-displacement curve required for the collision simulation according to the input data of the dynamics simulation environment.

[0036] In the collision simulation, the structural parameters of the end friction energy absorption structure can be used as input parameters, and the vertical displacement and acceleration of each car body of a multi-marshaling train can be used as output to study the influence of the parameters of the train end friction energy absorption structure on the posture optimization of the multi-marshaling train. In the optimization process, the minimum vertical displacement and acceleration of the car body are taken as the optimization goals, and the optimal combination of energy absorption structure parameters is obtained by making the optimal decision through the Pareto solution set, so that the vertical displacement and acceleration of the train are minimized under collision or other dynamic load conditions. For example, the specific formula for the optimization objective is as follows: ; in, represents the acceleration of the stationary car in section i; represents the acceleration of the stationary car at section j; represents the vertical lift of the i-th stationary vehicle; represents the vertical lift of the jth moving vehicle; The diameter of the parting surface between the friction tube and the friction bearing for absorbing friction energy; is the bearing thickness coefficient; It is the preload pressure of the friction bearing; It is the pre-connection length between the friction tube and the friction bearing; is the dynamic friction coefficient of the friction rubber; is the taper coefficient. Preferably, in the process of calculating the optimal decision through the Pareto solution set, the variable , , and The gain weights are set to 0.58, 0.12, 0.06 and 0.24 respectively. The aforementioned weight setting is based on an in-depth analysis and understanding of the role of each variable in the overall performance of the system. This weight allocation can better guide the optimization process of the energy-absorbing structure in order to achieve the best balance between key performance indicators such as strength and lightweight.

[0037] Furthermore, in the collision dynamics model of the marshalling train, the constitutive relationship of the local plastic deformation of the coupler and anti-climber, and the coupler structure is defined by a generalized spring, and the mechanical characteristic curves of the energy absorption structure and the coupler are constructed using a nonlinear hysteresis mathematical model.

[0038] like Figure 2As shown, the green line represents the loading curve, the red line represents the unloading curve, the abscissa X represents the displacement, and the ordinate Y represents the force; and the serial numbers ① to ⑧ in the figure correspond to the following eight parts (1) to (8) of the hysteresis behavior of the continuous loading and unloading process of the marshaling train, respectively, including: (1) Load along the loading curve until the displacement reaches .

[0039] (2) Unload along the hysteresis slope and then load again to .

[0040] (3) Load again along the loading curve until the maximum value is reached .

[0041] (4) Unload again along the hysteresis curve.

[0042] (5) Unload along the unloading curve.

[0043] (6) Load in the reverse direction along the unloading curve.

[0044] (7) Load along the hysteresis slope to the upper right until it reaches the maximum value .

[0045] (8) Further re-loading is performed along the loading curve.

[0046] Likewise, loading and unloading will occur in the opposite direction in the reverse direction of loading.

[0047] In this embodiment, the nonlinear hysteresis mathematical model is described using the following three stages: The calculation equation for the loading phase is: ;in, It is the impedance force when the coupler or anti-climber is loaded; is the initial pressure of the coupler or anti-climber; is the sliding friction; is the static friction; It represents the relative longitudinal velocity between the moving end and the stationary end of the hook. represents the Stribeck critical judgment speed.

[0048] The initial pressure of the coupler and the anti-climber will affect the friction factor at the beginning of the impact. In the actual calculation process, it is often calculated as a constant and rewritten in the form of stiffness: ;in, Provides linear stiffness for hook buffer loading; is the stiffness under the simultaneous action of sliding and static friction; Represents the displacement of the buffer along the longitudinal direction of the train during the loading phase.

[0049] In the unloading stage, the hook resistance and stroke can be considered to be linearly related, and the calculation method is: ; is the linear stiffness in the unloading stage.

[0050] There is fluctuation between loading and unloading, the hook-slow motion section and the static section are relatively static, and the relative speed between the two is 0, thus resulting in a zero speed stage.

[0051] When the static end and the moving end of the hook-buffer system are in the stage of zero relative speed, the direction of the speed will change, resulting in the magnitude and direction of the instantaneous force in the numerical simulation not being unique; in order to accurately determine the friction damping characteristics at zero speed, so that the impedance force in the calculation process has a unique solution and avoid fluctuations in the simulation output, the mathematical model of the instantaneous impedance force at zero speed is established as follows: ; in, for The impedance of the momentary hook and buffer device; for Impedance of the momentary hook and buffer device; for The hook-and-buffer device hook-and-buffer displacement at the moment; for The hook displacement of the hook buffer device at the moment; Geometric control parameters for controlling loading and unloading boundary changes; It is the boundary force of the hook-slow characteristic curve.

[0052] pass Time displacement and Time displacement The magnitude of the boundary force is determined by ,when hour, Take the impedance during the loading phase, when hour, Take the impedance during the unloading phase.

[0053] Thereby, based on the target model of the predicted force-displacement curve, the force-displacement curve can be obtained, and based on the force-displacement curve, the energy absorption characteristics of the anti-climber during the impact process can be obtained. Based on the above-mentioned nonlinear hysteresis mathematical model, it can be determined which section of the force-displacement curve and the position of the corresponding force-displacement curve are currently located according to the current number of collisions in the multiple chain collisions usually generated by the marshaled train and the displacement of adjacent trains. In the processing process of the nonlinear hysteresis mathematical model, interpolation calculations can be performed according to the force-displacement curve.

[0054] Example 2 Corresponding to the above-mentioned embodiment, this embodiment discloses a model data processing system for a train energy-absorbing structure in a collision simulation, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the method of the above-mentioned embodiment is implemented when the processor executes the computer program.

[0055] In summary, the model data processing method and system of the train energy absorbing structure in the collision simulation disclosed in the above two embodiments of the present invention respectively have at least the following beneficial effects: 1. In step S2, the force-displacement curve is used to predict the force-displacement curve. The latent variables in the middle layer are equivalent to the code book for sending telegrams, which are shared by the encoder and decoder. They can be regarded as the encoded values ​​after the high-dimensional time series data of the force-displacement curve is reduced in dimension.

[0056] 2. In step S4, the input and output data of the model are non-time series data, and a commonly used model can achieve reliable results.

[0057] 3. In order to analyze the energy absorption characteristics of the friction energy absorption structure under quasi-static compression, a WED-600 electro-hydraulic servo universal testing machine was used for quasi-static compression testing. During the test, the mixed friction energy absorption structure was compressed by driving the middle beam at a speed of 5 mm / min. After the test system is started, the middle beam will compress the mixed friction test energy absorption structure downward at a constant speed until the test displacement reaches 120 mm, the test stops and automatically unloads, and then the corresponding total energy absorption is calculated according to the load displacement curve obtained from the test acquisition system. In order to verify the accuracy of the combined target model of the present invention, multiple groups of force-displacement curves and measured crashworthiness indicators based on the combined target model of the present invention and finite element simulation are compared; it is concluded that the errors of the crashworthiness indicators obtained by the method of the present invention are very small, the maximum error is only 2.65%, and the overall error is less than 5%. As a result, after the combined target model of the present invention is applied to the collision dynamics model of the marshaled train, it can be used to evaluate the crashworthiness of the train without the participation of the finite element model software, and guide the parameter optimization of the energy absorption structure.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A model data processing method for a train energy-absorbing structure in a collision simulation, characterized in that: include: Step S1, obtaining force-displacement curves of different energy absorbing structures based on finite element simulation to construct a first data set; Step S2: using the force-displacement curve of each sample in the first data set as the input and output of the self-supervised network model, and extracting the intermediate layer hidden variables corresponding to each force-displacement curve after dimension reduction in the self-supervised network model; Step S3, constructing a second data set according to the intermediate layer latent variables corresponding to the structural parameters of different energy absorbing structures in the self-supervised network model; Step S4, training, verifying and testing a new network model whose input is the structural parameters of the energy absorbing structure and whose output is the intermediate layer latent variables according to the second data set; Step S5, combining the new network model with the decoder part of the self-supervised network model to obtain a target model for predicting the force-displacement curve based on structural parameters, and the weight, offset value information and processing logic of the combined decoder part are consistent with the decoder weight, offset value information and processing logic obtained by training in step S2; Step S6: embedding the target model into the train-forming collision dynamics model to predict and output the force-displacement curve required for the collision simulation according to the input data of the dynamics simulation environment.

2. The model data processing method of the train energy absorbing structure in collision simulation according to claim 1 is characterized in that: In the step S1, the finite element simulation specifically comprises constructing and experimentally verifying a finite element parametric model of the train energy absorption structure in the finite element simulation software, wherein the finite element parametric model encapsulates the entire process of geometric parameter driving, material association, load application, mesh division and calculation submission in a coded manner to achieve automatic mapping of "parameter input-simulation output", and the first data set is constructed based on the finite element parametric model using an experimental design method.

3. The model data processing method of the train energy absorbing structure in collision simulation according to claim 2 is characterized in that: In step S2, the force-displacement data of each sample is converted into a matrix of 2×1000 dimensions, and then four algorithms, namely AE, RNN-AE, LSTM-AE and GRU-AE network models, are used for model training. The prediction accuracy of different models is compared and analyzed, and the network model with the highest accuracy in predicting the output curve is used as the self-supervised network model.

4. The model data processing method of the train energy absorbing structure in collision simulation according to claim 3 is characterized in that: In step S4, the new network model adopts a CNN network model, and the energy absorption structure is specifically an anti-climbing device in a hook buffer device. The structural parameters include: the diameter of the parting surface between the friction tube and the friction bearing, the pre-tightening pressure of the friction bearing, the dynamic friction coefficient of the friction rubber, the pre-connection length between the friction tube and the friction bearing, the bearing thickness coefficient and the taper coefficient.

5. The model data processing method of the train energy absorbing structure in collision simulation according to claim 4 is characterized in that: In the train collision dynamics model, the coupler includes a coupler and an anti-climbing device. The constitutive relationship of the local plastic deformation of the coupler is defined by a generalized spring, and the mechanical characteristic curves of the energy absorption structure and the coupler are constructed by a nonlinear hysteresis mathematical model. The nonlinear hysteresis mathematical model is described by the following three stages: The calculation equation for the loading phase is: ;in, It is the impedance force when the coupler or anti-climber is loaded; is the initial pressure of the coupler or anti-climber; is the sliding friction; is the static friction; It represents the relative longitudinal velocity between the moving end and the stationary end of the hook. represents Stribeck critical judgment speed; The initial pressure of the coupler and the anti-climber will affect the friction factor at the beginning of the impact. In the actual calculation process, it is often calculated as a constant and rewritten in the form of stiffness: ;in, Provides linear stiffness for hook buffer loading; is the stiffness under the simultaneous action of sliding and static friction; represents the displacement of the buffer along the longitudinal direction of the train during the loading phase; During the unloading phase, the hook resistance is considered to be linearly related to the stroke and is calculated as: ; is the linear stiffness in the unloading stage; Among them, there is fluctuation between loading and unloading, the hook slow motion section and the static section are relatively static, and the relative speed between the two is 0, thus resulting in a zero speed stage; When the static end and the moving end of the hook-buffer system are in the stage of zero relative speed, the direction of the speed will change, resulting in the magnitude and direction of the instantaneous force in the numerical simulation not being unique; in order to accurately determine the friction damping characteristics at zero speed, so that the impedance force in the calculation process has a unique solution and avoid fluctuations in the simulation output, the mathematical model of the instantaneous impedance force at zero speed is established as follows: ； in, for The impedance of the momentary hook and buffer device; for Impedance of the momentary hook and buffer device; for The hook-and-buffer device hook-and-buffer displacement at the moment; for The hook displacement of the hook buffer device at the moment; Geometric control parameters used to control loading and unloading boundary changes; It is the boundary force of the hook-slow characteristic curve; pass Time displacement and Time displacement The magnitude of the boundary force is determined by ,when hour, Take the impedance during the loading phase, when hour, Take the impedance during the unloading phase.

6. The model data processing method of the train energy absorbing structure in collision simulation according to claim 5 is characterized in that: In step S2 and step S5, before model training, the force-displacement curve and structural parameters are normalized respectively. In the normalization process, all values ​​are directly normalized by dividing them by a maximum value. The calculation formula is: ； ； Where: and The normalized value and true value of the calculation result of the j-th group of experimental design for the i-th input structural parameter; max is the maximum true value of the i-th input structural parameter in all experimental design calculation results; and are the normalized value and true value of the kth point on the impact force-displacement curve in the jth group of test design calculation results; max It is the maximum true value of the i-th crashworthiness index in all experimental design calculation results.

7. The model data processing method of the train energy absorbing structure in collision simulation according to any one of claims 1 to 5, characterized in that: Also includes: In the collision simulation, the structural parameters of the end friction energy absorption structure are taken as input parameters, and the vertical displacement and acceleration of each car body of a multi-carriage train are taken as outputs. The influence of the parameters of the train end friction energy absorption structure on the posture optimization of multi-carriage trains is studied. In the optimization process, the minimum vertical displacement and acceleration of the car body are taken as the optimization targets, and the optimal combination of energy absorption structure parameters is obtained by making the optimal decision through the Pareto solution set, so that the vertical displacement and acceleration of the train are minimized under collision or other dynamic load conditions.

8. The model data processing method of the train energy absorbing structure in collision simulation according to claim 7 is characterized in that: The specific formula for the optimization objective is as follows: ； In the formula, represents the acceleration of the stationary car in section i; represents the acceleration of the stationary car at section j; represents the vertical lift of the i-th stationary vehicle; represents the vertical lift of the jth moving vehicle; The diameter of the parting surface between the friction tube and the friction bearing for absorbing friction energy; is the bearing thickness coefficient; It is the preload pressure of the friction bearing; It is the pre-connection length between the friction tube and the friction bearing; is the dynamic friction coefficient of the friction rubber; is the taper coefficient.

9. A model data processing system for a train energy-absorbing structure in a collision simulation, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Images