Railway passenger car whole vehicle assembling method based on explicit dynamic integral method

Abstract

The application discloses a railway passenger car whole vehicle completion method based on an explicit dynamic integral method, and comprises the following steps: obtaining the parameters of materials to be matched with a railway passenger car whole vehicle to be completed; constructing a statically indeterminate balance model of the railway passenger car whole vehicle to be completed, and obtaining the displacement response of the statically indeterminate balance model; calculating first theoretical pad parameters according to the displacement response and the completion requirements of the railway passenger car whole vehicle; inputting the parameters of the materials to be matched into a full connection feedforward neural network, and obtaining second theoretical pad parameters; constructing a linear regression model, combining the first theoretical pad parameters and the second theoretical pad parameters, and outputting the combined parameters as pad parameters to be added; and guiding the railway passenger car whole vehicle to be completed according to the pad parameters to be added. The application can more accurately calculate the required pad parameters, improve the completion precision, reduce repeated operations, and improve the production and maintenance efficiency; through data analysis and machine learning technology, the application can adapt to various physical changes that may occur in the long-term operation of the railway passenger car.

Description

Technical Field

[0001] This invention relates to the field of railway passenger car technology, and in particular to a railway passenger car assembly method based on explicit dynamic integration. Background Technology

[0002] As core technical equipment for passenger transport, railway passenger cars play a crucial role in ensuring operational performance, safety, and reliability. To maintain their optimal performance, railway passenger cars must undergo regular inspections and maintenance. During this process, ensuring the accurate assembly of the passenger car body is paramount, as the car body's balance directly affects driving safety and passenger comfort. According to the "Railway Passenger Car Depot Maintenance Regulations" issued by China Railway Corporation, strict regulations govern the primary and secondary train springs, coupler height, and other aspects during the assembly of passenger cars. Due to factors such as numerous bogie manufacturers, different production batches, and performance degradation of components (such as springs and wheels) during long-term service due to material fatigue, stress relaxation, or wear, variations exist in the passenger car body's mass distribution, spring height, coupler height, and side bearing clearance, thus affecting the efficiency and quality of the depot maintenance assembly process.

[0003] In the traditional railway passenger car assembly process, multiple parts need to be manually measured after unloading to adjust the thickness of the shims. This process often requires repeated lifting and unloading multiple times (usually 3 to 5 times) to meet the standard requirements. Given the size and weight of railway passenger cars, repeatedly lifting and unloading operations are not only time-consuming and labor-intensive, but also significantly reduce production efficiency and increase safety hazards.

[0004] To address the aforementioned issues, the published paper "Research on Key Technologies for One-Time Completion of Bus Body in Bus Depot Repair" proposes a numerical calculation method based on a statically indeterminate equilibrium model of the entire vehicle. This method aims to determine the required thickness and quantity of padding during the unloading process in advance, enabling one-time completion of the bus body. This technical solution addresses the statically indeterminate problem by constructing and solving the dynamic equations. However, this method has certain limitations in practical applications, particularly when applied to bus bodies and frames that have experienced fatigue and minor deformation after long-term operation. Because the model assumes an ideal physical state for the body and frame, significant deviations may occur in the numerical calculation of the unloading parameters. While this deviation can be reduced by constructing a more accurate equation model, this improvement process can be quite time-consuming and require substantial computational resources. Summary of the Invention

[0005] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a simple and effective method for the assembly of railway passenger cars, which has error elimination and iterative update capabilities.

[0006] The technical solutions disclosed in this invention include:

[0007] A method for the overall assembly of railway passenger cars based on explicit dynamic integration includes the following steps:

[0008] Obtain the associated material parameters of the railway passenger car to be completed; the associated material parameters include: the mass of each part of the car body, the distributed load of the car body, the spring damping parameters, and the wheel parameters;

[0009] A statically indeterminate equilibrium model of the railway passenger car to be completed is constructed, and the displacement response of the statically indeterminate equilibrium model is obtained by solving the model based on the explicit dynamic integration method and the associated material parameters; the first theoretical padding parameters are calculated based on the displacement response and the completion requirements of the railway passenger car.

[0010] A fully connected feedforward neural network is constructed, and historical whole-vehicle commissioning and debugging data of the same model of railway passenger car as the railway passenger car to be commissioned are obtained. The data is divided into training set and test set to complete the training of the fully connected feedforward neural network. The parameters of the associated materials are input into the fully connected feedforward neural network to obtain the second theoretical padding parameters.

[0011] Construct a linear regression model θ = w1·θ1 + w2·θ2 + b, and output the parameter to be added θ by combining the first theoretical padding parameter θ1 and the second theoretical padding parameter θ2. Here, w1 and w2 are the weights of the first theoretical padding parameter and the second theoretical padding parameter, respectively, and w1 + w2 = 1. b is the bias term.

[0012] The installation of the railway passenger car is guided by the parameters to be added.

[0013] In some preferred embodiments, solving the statically indeterminate equilibrium model of the railway passenger car to be completed based on the explicit dynamic integration method and the associated material parameters includes the following steps:

[0014] Construct the dynamic equations for the entire railway passenger car to be completed:

[0015] Where [M], [C], and [K] are the mass matrix, spring damping matrix, and stiffness matrix of each part of the railway passenger car body to be completed, respectively. These are acceleration, velocity, and displacement vectors, respectively, and {P} is the vehicle body distributed load vector;

[0016] Construct velocity profiles based on linear integral and integral control parameters α and β, respectively. and displacement X t+Δt The difference expression is as follows:

[0017]

[0018]

[0019] Substituting the difference expression into the dynamic equation of the railway passenger car to be completed, the displacement response of the statically indeterminate equilibrium model is obtained, that is, the velocity of the railway passenger car to be completed at time t+Δt. and displacement X t+Δt .

[0020] In some preferred embodiments, to ensure that the calculated dynamic equations of the completed railway passenger car are unconditionally stable, integral control parameters α≥1 / 2 and β≥(α+1 / 2) are set. 2 / 4.

[0021] In some preferred embodiments, after obtaining historical whole-vehicle commissioning and testing data of the same model of railway passenger car as the one to be commissioned, a preprocessing step is also included:

[0022] The historical completion and commissioning data includes the material parameters, number of trial completions, and actual padding parameters for each trial completion of several models of railway passenger cars during their historical completion and commissioning.

[0023] Clean historical commissioning data, retain the actual padding parameters of the last trial commissioning for each commissioning, associate the cleaned historical commissioning data with the corresponding assigned material parameters and use it as the first sample;

[0024] Pre-set parameter perturbation is applied to the material parameters of samples with the same actual padding parameters to form a new second sample;

[0025] The first and second samples are combined to form the sample set for training the fully connected feedforward neural network.

[0026] In some preferred embodiments, the training method for the linear regression model θ = w1·θ1 + w2·θ2 + b includes:

[0027] Initialize w1 = 1, w2 = b = 0;

[0028] Constructing a loss function using mean squared error: Where N is the number of samples in the training sample set. The first theoretical padding parameter for the training samples, θ i The actual padding parameters for the training samples;

[0029] The linear regression model is iteratively optimized using gradient descent until the value of the loss function no longer decreases or the preset number of iterations is reached.

[0030] In some preferred embodiments, the padding parameters include: the thickness of the core pad, the thickness of the spring pad, and the thickness of the side bearing pad.

[0031] In some preferred embodiments, the fully connected feedforward neural network further includes a mapping layer after the output layer;

[0032] The mapping layer contains a thickness list of the actual thicknesses of the disc pad, spring pad, and side bearing pad in the field. The mapping layer matches the disc pad, spring pad, and side bearing pad thickness with the smallest absolute value of the difference between the predicted value and the predicted value in the thickness list according to the predicted value output by the output layer, and outputs the matching result as the second theoretical padding parameter.

[0033] Beneficial effects

[0034] 1. Completion accuracy: By constructing a statically indeterminate equilibrium model and using explicit dynamic integration methods combined with machine learning algorithms, this invention can more accurately calculate the required padding parameters. This accurate calculation result helps to complete the completion of the entire railway passenger car in one go, significantly improving the completion accuracy, significantly reducing repetitive operations, thereby optimizing the completion process and improving production and maintenance efficiency.

[0035] 2. Enhanced Operational Safety: Repetitive lifting and lowering operations are not only time-consuming and labor-intensive, but may also pose safety hazards. This invention helps improve operational safety by reducing these operations, creating a safer working environment for workers.

[0036] 3. High adaptability: Considering the various physical changes that may occur in railway passenger cars during long-term operation (such as spring fatigue, wheel wear, etc.), this invention can adapt to these changes through data analysis and machine learning technology, and provide customized installation solutions for railway passenger cars in different states. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating a preferred embodiment of the railway passenger car assembly method based on an explicit dynamic integration method according to the present invention.

[0038] Figure 2 This is a front view schematic diagram of the statically indeterminate equilibrium model of a railway passenger car to be completed in a preferred embodiment of the present invention.

[0039] Figure 3 This is a rear view schematic diagram of the statically indeterminate equilibrium model of a railway passenger car to be completed, constructed in a preferred embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings. In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention.

[0041] Example

[0042] like Figure 1 As shown, this embodiment provides a method for the assembly of a railway passenger car based on an explicit dynamic integration method, including the following steps:

[0043] S1. Obtain the associated material parameters of the railway passenger car to be completed; the associated material parameters include: the mass of each part of the car body, the distributed load of the car body, the spring damping parameters, and the wheel parameters.

[0044] Traditional passenger car unloading and commissioning operations do not consider the impact of material parameters on unloading quality, which is a major reason why unloading often fails to pass the test on the first attempt. However, railway passenger cars involve a large number of auxiliary materials. This invention mainly involves the mass of various parts of the car body, the distributed load of the car body, spring damping parameters, and wheel parameters. The spring damping parameters mainly include the initial height parameters of the primary / secondary springs and the compression stiffness parameters of the primary / secondary springs. The workshops involved are the car body disassembly workshop, the spring testing workshop, and the wheel refinishing workshop. The above-mentioned auxiliary material parameters can be obtained using conventional methods in the field. In the car body disassembly workshop, force sensors and a lifting synchronization control system are added to the existing car-lifting machines used for unloading to achieve synchronous lifting of four car-lifting machines used for the same car body. During the process, the lifting error is controlled to ensure that the height difference of the four machines does not exceed 2mm after the car is lifted. At this time, the distributed load of the car body obtained by the four force sensors is transmitted to the computer terminal at the unloading station. The spring parameters have already been obtained in the test bench of the existing spring testing workshop. Similarly, the wheel diameter information is obtained by the computer of the wheel turning machine and uploaded to the computer terminal of the unloading station.

[0045] S2. Construct a statically indeterminate equilibrium model of the entire railway passenger car to be completed, such as... Figures 2-3 As shown, the displacement response of the statically indeterminate equilibrium model is obtained by solving the problem based on the explicit dynamic integration method and the parameters of the associated materials; the first theoretical padding parameters are calculated based on the displacement response and the requirements for the completion of the railway passenger car.

[0046] It should be noted that the padding parameters described in this invention are based on the actual structure of railway passenger cars and the process requirements of the maintenance workshop, and actually include: the thickness of the center plate pad, the thickness of the spring pad, and the thickness of the side bearing pad.

[0047] A vehicle with spring suspension is a multi-degree-of-freedom vibration system. A vehicle generally consists of a car body, secondary suspension springs and dampers, a bogie frame, primary suspension springs and dampers, and wheelset axle boxes. In existing technologies, when establishing mathematical models of the dynamic characteristics of a vehicle or the entire train, components of the vehicle system such as the car body, bogie frame, and axle boxes are considered as rigid bodies. Each rigid body has six degrees of freedom, namely three displacements and three rotations. In the vehicle coordinate system, the three rotations are generally expressed as head-shaking, roll, and pitching motions. This paper constructs a statically indeterminate equilibrium model for the current vehicle and performs precise equilibrium solutions on the model. In some preferred embodiments, due to the complexity of the vehicle structure, the statically indeterminate equations have multiple solutions and do not meet the requirements for vehicle landing. This invention proposes a method to solve this statically indeterminate problem using dynamic equations. The vehicle system is considered as having six degrees of freedom: car body buoyancy, car body pitching motion, car body roll motion, bogie frame buoyancy, bogie frame roll motion, and bogie frame pitching motion. Their equations of motion are as follows:

[0048] (1) Equations of motion of the vehicle body

[0049] 1. Buoyancy

[0050]

[0051] Among them, M c For vehicle body mass; C sz For the two-stage suspension damping; [K] is a diagonal matrix, and {X} is a column vector:

[0052]

[0053] In the formula, k sz1 k represents the parallel stiffness of the springs in positions 1 and 3 of the bolster. sz2 k represents the parallel stiffness of the 2nd and 4th position bolster springs. sz3 The parallel stiffness of the 5th and 7th position bolster springs; k sz4 The parallel stiffness of the 6th and 8th position bolster springs; d s It is half the lateral distance of the second-stage suspension; c It is half the vehicle's fixed distance; Z c , β c These represent the vehicle's rising, rolling, and pitching movements, respectively; Z t1 Z t2 The rising and falling motions of the first and second position bogie frames are respectively; The roll motions are for the 1st and 2nd position bogie frames, respectively.

[0054] 2. Side Rolling Motion

[0055]

[0056] Where [K] is a diagonal matrix, and {X} is a column vector:

[0057]

[0058] 3. Head nodding exercise

[0059]

[0060] Where [K] is a diagonal matrix, and {X} is a column vector:

[0061]

[0062] (2) Equations of motion of the frame

[0063] 1. Buoyancy

[0064]

[0065] Among them, M t For vehicle body mass; C pz For the two-stage suspension damping; [K] is a diagonal matrix, and {X} is a column vector:

[0066]

[0067] Where, k pz1 k represents the parallel stiffness of the springs in positions 1 and 3 of the axle box. pz2 k represents the parallel stiffness of the springs in positions 2 and 4 of the axle box. pz3 The parallel stiffness of the springs in positions 5 and 7; k pz4 The parallel stiffness of the springs in positions 6 and 8; d w It is half the lateral distance of the suspension system; t It is half the fixed distance of the bogie wheelset.

[0068] (2) Side Rolling Motion

[0069]

[0070] Where [K] is a diagonal matrix, and {X} is a column vector:

[0071]

[0072] (3) Nodding movement

[0073]

[0074] Where [K] is a diagonal matrix, and {X} is a column vector:

[0075]

[0076] The dynamic equations of the railway passenger car to be completed are constructed. The above formulas are then substituted into the dynamic equations of the railway passenger car to be completed, and the displacement response of the above dynamic equations is solved using numerical methods. Specifically, this includes:

[0077] Where [M], [C], and [K] are the mass matrix, spring damping matrix, and stiffness matrix of each part of the railway passenger car body to be completed, respectively. These are acceleration, velocity, and displacement vectors, respectively, and {P} is the vehicle body distributed load vector.

[0078] Construct velocity profiles based on linear integral and integral control parameters α and β, respectively. and displacement X t+Δt The difference expression is as follows:

[0079]

[0080]

[0081] Substituting the difference expression into the dynamic equation of the railway passenger car to be completed, we get:

[0082]

[0083] Among them, effective quality and effective force vector They are respectively:

[0084]

[0085]

[0086] Solving the above equations yields the speed of the entire railway passenger car to be completed at time t+Δt. and displacement X t+Δt .

[0087] It should be understood that while the first theoretical shim-addition parameters calculated through this step already possess a high accuracy rate, the model may exhibit significant deviations in the numerical calculation of completion parameters when dealing with passenger car bodies and frames that have experienced fatigue and minor deformations after long-term operation. This is because the model assumes an ideal physical state for the car body and frame. Therefore, this invention also provides the following steps to eliminate such deviations. Through data analysis and machine learning techniques, it can adapt to these changes and provide customized completion solutions for railway passenger cars in different states.

[0088] S3. Construct a fully connected feedforward neural network, obtain historical whole-vehicle commissioning and debugging data of the same model of railway passenger car as the railway passenger car to be commissioned, and complete the training of the fully connected feedforward neural network by dividing it into training set and test set; input the associated material parameters into the fully connected feedforward neural network to obtain the second theoretical padding parameters.

[0089] A fully connected feedforward neural network, also known as a multilayer perceptron (MLP), consists of multiple fully connected layers, each fully connected to the previous layer, with no feedback connections (i.e., the network does not form loops). It has a strong ability to capture non-linear relationships when processing non-image, non-temporal (i.e., tabular or structured data) data, making it particularly suitable for the technical problems addressed in this application.

[0090] However, it also has some limitations, such as a tendency to overfit (especially with limited data). Therefore, some preferred embodiments also provide a superior training method to avoid overfitting, specifically including:

[0091] After obtaining the historical completion and commissioning data of railway passenger cars of the same model as the one to be completed, a preprocessing step is also included:

[0092] The historical commissioning data includes the material parameters, number of trial commissionings, and actual padding parameters for each trial commissioning of several models of railway passenger cars. This historical data can be obtained from the railway depot's maintenance workshop.

[0093] The historical installation and commissioning data is cleaned, and the actual padding parameters from the last trial installation are retained for each installation and commissioning. The cleaned historical installation and commissioning data is associated with the corresponding material parameters and used as the first sample. Considering that the number of the first sample is limited, it is not sufficient to train the neural network effectively. Therefore, this invention considers using parameter perturbation to increase the number of samples.

[0094] For samples with the same actual padding parameters, the parameters of the associated materials are subject to preset parameter perturbations to form new second samples. These parameter perturbations create new samples by making small changes to the existing dataset, thereby increasing not only the amount of data but also its diversity, which is very helpful in improving the model's generalization ability. During operation, a reasonable perturbation range needs to be determined based on an understanding of the influence of each parameter. For parameters with a large influence, the perturbation range may need to be relatively conservative to prevent the generation of unrealistic data points. For parameters with a smaller influence, a larger perturbation range can be considered. The specific settings can be configured by those skilled in the art according to the actual needs of the site; this invention does not impose further limitations.

[0095] The first and second samples are combined to form the sample set for training the fully connected feedforward neural network.

[0096] It should be noted that the thickness of the venturi pads and spring washers added on-site during the repair process is subject to specifications. For example, spring washers are only available in 5mm sizes, and the venturi pads must be installed within 50mm. The prediction results directly output by the fully connected feedforward neural network may not meet these specifications (e.g., they may contain decimal values ​​or exceed the specifications). Therefore, in some preferred embodiments, a mapping between the actual specifications and the prediction output is established to address this issue. Specifically, this includes:

[0097] The fully connected feedforward neural network further includes a mapping layer after the output layer;

[0098] The mapping layer contains a thickness list of the actual thicknesses of the disc pad, spring pad, and side bearing pad in the field. The mapping layer matches the disc pad, spring pad, and side bearing pad thickness with the smallest absolute value of the difference between the predicted value and the predicted value in the thickness list according to the predicted value output by the output layer, and outputs the matching result as the second theoretical padding parameter.

[0099] S4. Construct a linear regression model θ = w1·θ1 + w2·θ2 + b. Combine the first theoretical padding parameter θ1 and the second theoretical padding parameter θ2 to output the padding parameter θ. Here, w1 and w2 are the weights of the first theoretical padding parameter and the second theoretical padding parameter, respectively, and w1 + w2 = 1. b is the bias term.

[0100] The technical considerations for this step are as follows:

[0101] As mentioned earlier, the first theoretical padding parameters obtained through theoretical equations do not consider the errors caused by fatigue of structural components after long-term operation, while the second theoretical padding parameters obtained through a fully connected feedforward neural network are obtained in a black-box state, without considering the actual theoretical basis of the acquisition process. Therefore, both are bound to have certain errors. To eliminate these errors, this invention introduces a model fusion-based method, fully considering the advantages and limitations of each parameter, and combining them through a weighted approach. Furthermore, when using the weighted method, the allocation of weights usually depends on the reliability, historical performance, expert experience, or statistical evidence derived from data analysis of each parameter. No fixed method can be applied to all situations; therefore, those skilled in the art can determine the specific weight settings based on the specific circumstances and available data. In some preferred embodiments, a training method for a linear regression model is provided to obtain a more objective and accurate weight configuration, specifically including:

[0102] Initialize w1 = 1 and w2 = b = 0. For simplicity, w2 can be represented as 1 - w1, so only w1 and b need to be optimized.

[0103] Constructing a loss function using mean squared error (MSE): Where N is the number of samples in the training sample set. The first theoretical padding parameter for the training samples, θ i The actual padding parameters for the training samples.

[0104] The linear regression model is iteratively optimized using gradient descent until the loss function no longer decreases or a preset number of iterations is reached. Gradient descent is essentially the optimizer for the linear regression model, and its purpose is to minimize the loss function. The specific steps of gradient descent are as follows:

[0105] Calculate the gradient of the loss function L with respect to w1 and b:

[0106]

[0107]

[0108] Then use these gradients to update w1 and b:

[0109]

[0110]

[0111] Here, α is the learning rate, which is set by those skilled in the art according to actual needs. Too large a value may lead to convergence too quickly or divergence, while too small a value may lead to training too slowly.

[0112] S5. Guide the installation of the railway passenger car to be completed according to the parameters to be added.

[0113] It should be understood that since the specific completion method is not the focus of this invention, and there are standard completion methods available for reference in the bus depot repair process, this invention does not impose specific limitations on this step.

[0114] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for the overall assembly of a railway passenger car based on an explicit dynamic integration method, characterized in that, Including the following steps: Obtain the associated material parameters of the railway passenger car to be completed; the associated material parameters include: the mass of each part of the car body, the distributed load of the car body, the spring damping parameters, and the wheel parameters; A statically indeterminate equilibrium model of the railway passenger car to be completed is constructed, and the displacement response of the statically indeterminate equilibrium model is obtained by solving the model based on the explicit dynamic integration method and the associated material parameters; the first theoretical padding parameters are calculated based on the displacement response and the completion requirements of the railway passenger car. A fully connected feedforward neural network is constructed, and historical whole-vehicle commissioning and debugging data of the same model of railway passenger car as the railway passenger car to be commissioned are obtained. The data is divided into training set and test set to complete the training of the fully connected feedforward neural network. The parameters of the associated materials are input into the fully connected feedforward neural network to obtain the second theoretical padding parameters. Construct a linear regression model θ = w1·θ1 + w2·θ2 + b, and output the parameter to be added θ by combining the first theoretical padding parameter θ1 and the second theoretical padding parameter θ2. Here, w1 and w2 are the weights of the first theoretical padding parameter and the second theoretical padding parameter, respectively, and w1 + w2 = 1. b is the bias term. The installation of the railway passenger car is guided by the parameters to be added.

2. The railway passenger car assembly method based on explicit dynamic integration as described in claim 1, characterized in that, The steps for solving the statically indeterminate equilibrium model of the railway passenger car to be completed based on the explicit dynamic integration method and the associated material parameters are as follows: Construct the dynamic equations for the entire railway passenger car to be completed: Where [M], [C], and [K] are the mass matrix, spring damping matrix, and stiffness matrix of each part of the railway passenger car body to be completed, respectively. {X} represents the acceleration, velocity, and displacement vectors, respectively, and {P} represents the vehicle body distributed load vector; Construct velocity profiles based on linear integral and integral control parameters α and β, respectively. and displacement X r+Δt The difference expression is as follows: Substituting the difference expression into the dynamic equation of the railway passenger car to be completed, the displacement response of the statically indeterminate equilibrium model is obtained, that is, the velocity of the railway passenger car to be completed at time t+Δt. and displacement X t+Δt ; in, Let X be the acceleration of the railway passenger car to be completed at time t+Δt. t , These represent the displacement, velocity, and acceleration of the railway passenger car to be completed at time t.

3. The railway passenger car assembly method based on explicit dynamic integration as described in claim 2, characterized in that: To ensure that the calculated dynamic equations of the completed railway passenger car are unconditionally stable, the integral control parameters are set as α≥1 / 2 and β≥(α+1 / 2). 2 / 4.

4. The railway passenger car assembly method based on explicit dynamic integration as described in claim 1, characterized in that, After obtaining the historical completion and commissioning data of railway passenger cars of the same model as the one to be completed, a preprocessing step is also included: The historical completion and commissioning data of railway passenger cars includes the material parameters, number of trial completions, and actual padding parameters for each trial completion of several models of railway passenger cars during the historical completion and commissioning process. Clean historical commissioning data, retain the actual padding parameters of the last trial commissioning for each commissioning, associate the cleaned historical commissioning data with the corresponding material parameters and use it as the first sample; Pre-set parameter perturbation is applied to the material parameters of samples with the same actual padding parameters to form a new second sample; The first and second samples are combined to form the sample set for training the fully connected feedforward neural network.

5. The railway passenger car assembly method based on explicit dynamic integration as described in claim 4, characterized in that, The training methods for the linear regression model θ = w1·θ1 + w2·θ2 + b include: Initialize w1 = 1, w2 = b = 0; Constructing a loss function using mean squared error: Where N is the number of samples in the training sample set. The first theoretical padding parameter for the training samples, θ i The actual padding parameters for the training samples; The linear regression model is iteratively optimized using gradient descent until the value of the loss function no longer decreases or the preset number of iterations is reached.

6. The railway passenger car assembly method based on explicit dynamic integration as described in claim 1, characterized in that, The padding parameters include: the thickness of the core pad, the thickness of the spring pad, and the thickness of the side bearing pad.

7. The railway passenger car assembly method based on explicit dynamic integration as described in claim 6, characterized in that: The fully connected feedforward neural network further includes a mapping layer after the output layer; The mapping layer contains a thickness list of the actual thicknesses of the disc pad, spring pad, and side bearing pad in the field. The mapping layer matches the disc pad, spring pad, and side bearing pad thickness with the smallest absolute value of the difference between the predicted value and the predicted value in the thickness list according to the predicted value output by the output layer, and outputs the matching result as the second theoretical padding parameter.

Images