A hierarchical calibration method and system for a railway vehicle dynamics simulation model

Abstract

This invention discloses a hierarchical calibration method and system for railway vehicle dynamics simulation models, belonging to the technical field of railway vehicle dynamics simulation parameter calibration. This invention divides the parameters to be calibrated into multiple calibration levels from bottom to top according to the physical dependency relationship and coupling strength of the parameters. Each level is matched with dedicated experimental data and error indicators. A layer-by-layer calibration method is adopted, locking the parameters after the previous layer is calibrated. When trigger conditions are met, the coupled parameters are selectively released for joint fine-tuning. Calibration is completed using a multi-convergence evaluation mechanism. This invention effectively reduces the impact of parameter coupling on convergence, improves the utilization rate and identification accuracy of experimental data, ensures traceability of the calibration process, avoids the amplification of errors in basic parameters, and is applicable to the calibration and credibility assessment of various railway vehicle dynamics models.

Description

Technical Field

[0001] This invention relates to the field of parameter calibration technology for railway vehicle dynamics simulation models, and in particular to a hierarchical calibration method and system for railway vehicle dynamics simulation models that can be applied to the calibration, simulation verification, and credibility assessment of dynamics models for high-speed trains, railway locomotives, railway passenger cars, railway freight cars, and urban rail vehicles. Background Technology

[0002] With the deepening of railway vehicle structural optimization design, operation safety performance evaluation, ride stability analysis, curve passing performance verification and serpentine motion stability analysis, the engineering field's reliance on dynamic simulation models continues to increase. The simulation accuracy and reliability of the model directly determine the engineering reference value of the simulation analysis results.

[0003] Railway vehicle multibody dynamics simulation models include various subsystems such as car body, bogie, wheelsets, primary suspension, secondary suspension, and wheel-rail contact. The number and types of parameters to be calibrated are enormous, and there are strong coupling relationships between different parameters. Currently, the mainstream model calibration methods in the industry fall into two categories:

[0004] Manual parameter tuning method: Based on the results of real vehicle tests and bench tests, technicians repeatedly and manually modify the model parameters to complete the rough calibration of the model.

[0005] The whole vehicle one-time overall optimization calibration method: all vehicle mass parameters, suspension stiffness and damping parameters, wheel-rail contact parameters, and vehicle dynamic response parameters are included in the same optimization range. With the goal of minimizing the global comprehensive error, the whole set of calibration parameters is solved in one go.

[0006] While the two calibration methods described above can achieve preliminary model correction, they have significant technical drawbacks:

[0007] First, the parameters are highly coupled, resulting in poor overall calibration convergence. The dynamic model has high-dimensionality and a large number of parameters, and different parameters can simultaneously affect the same dynamic response index. A one-time overall optimization will cause parameters to compensate for each other and the algorithm's search space to be too large, which can easily lead to slow iterative convergence or even failure to converge.

[0008] Second, there is no precise correspondence between the test data and the parameters. Vehicle mass, inertia, and center of gravity parameters dominate static load distribution and basic modal response; primary and secondary suspension parameters correspond to dynamic vibration responses at different frequencies and positions; wheel-rail contact and adhesion parameters directly affect curve passing performance, wheel-rail forces, and hunting stability. The existing overall calibration does not match and divide parameters with data, resulting in low utilization of test data and poor parameter identification accuracy.

[0009] Third, the calibration process lacks interpretability and traceability. Traditional overall calibration only outputs the final set of parameter results, making it impossible to pinpoint which subsystem or parameter the simulation error originates from. The calibration process lacks traceability, making it difficult to assess the credibility of the simulation model.

[0010] Fourth, errors in basic parameters are amplified step by step. If there are initial deviations in the mass, inertia, and center of gravity parameters of basic rigid bodies such as the car body, bogie, and wheelsets, subsequent suspension parameters and wheel-rail parameters will passively produce compensatory offsets. Although they can barely match some test indicators, the physical mechanism of the model is distorted, which seriously reduces the engineering application value of the simulation model.

[0011] Therefore, there is an urgent need to propose a calibration scheme for railway vehicle dynamics simulation models that can reduce the influence of parameter coupling, improve identification accuracy, and have traceability, so as to solve the above-mentioned problems existing in the current technology. Summary of the Invention

[0012] This application provides a hierarchical calibration method and system for railway vehicle dynamics simulation models, aiming to solve one of the problems existing in the current manual experience-based parameter tuning method or the one-time overall optimization calibration method for the whole vehicle, such as strong parameter coupling, poor overall calibration convergence, no precise correspondence between test data and parameters, lack of interpretability and traceability of the calibration process, or the stepwise amplification of errors in basic parameters.

[0013] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0014] This invention provides a hierarchical calibration method for a railway vehicle dynamics simulation model, comprising the following steps:

[0015] Step S1: Construct an initial simulation model of the multibody dynamics of the railway vehicle to be calibrated, obtain the original parameters and corresponding experimental data of the initial simulation model, and preprocess the experimental data to establish a one-to-one correspondence between the experimental data and the simulation output;

[0016] Step S2: Based on the physical dependency relationship and coupling strength of the parameters, divide the parameters to be calibrated into multiple calibration levels from the bottom to the top.

[0017] Step S3: Match a dedicated test dataset and quantification error index for each calibration level, and establish a mapping relationship between level, test data and error index;

[0018] Step S4: For each calibration level, construct the hierarchical calibration objective function for that level by combining the corresponding operating conditions and evaluation indicators;

[0019] Step S5: Perform layer-by-layer parameter calibration in order from bottom to top. During the calibration process of the current level, the parameters of the previous level that have been calibrated are locked to constant values ​​and do not participate in the optimization solution of the current level.

[0020] Step S6: When the subsequent level calibration meets the preset triggering conditions, interrupt the regular calibration process of the current level, perform sensitivity analysis on the previous locked parameters, identify and selectively release the coupling parameters that have a significant impact on the residual of the current level, fine-tune the released coupling parameters together with the current level parameters, re-verify and lock the released parameters after fine-tuning, and then continue the regular calibration process of the current level.

[0021] Step S7: Use a multiple convergence evaluation mechanism to determine the calibration convergence status of each level. If the convergence criterion is met, proceed to the next level of calibration.

[0022] Step S8: After all levels of calibration are completed, the calibrated model is comprehensively verified using multi-condition vehicle operation test data. After the verification is passed, the calibration results are output.

[0023] Furthermore, the calibration hierarchy is divided into five layers, from bottom to top as follows:

[0024] The first layer is the basic rigid body parameter layer, which includes the mass, moment of inertia and center of gravity position parameters of the vehicle body, bogie frame and wheelsets;

[0025] The second layer is the primary suspension parameter layer, including the three-dimensional stiffness of the primary axle box springs, the damping of the primary vertical damper, and the three-dimensional stiffness parameters of the axle box positioning.

[0026] The third layer is the secondary suspension parameter layer, which includes the three-dimensional stiffness of the secondary springs, the damping of the secondary vertical damper, the damping of the secondary lateral damper, the stiffness of the anti-roll torsion bar, and the damping parameters of the anti-hunting damper.

[0027] The fourth layer is the wheel-rail contact parameter layer, which includes wheel-rail profile parameters, wheel equivalent taper, and wheel-rail friction coefficient parameters;

[0028] The fifth layer is the vehicle operation response calibration layer, which is used to release and fine-tune at least one parameter that has been locked in the first four layers and is strongly coupled with the vehicle dynamic response.

[0029] Furthermore, the specific experimental data and error indicators for each level are as follows:

[0030] The basic rigid body parameter layer corresponds to the vehicle weighing data, wheel weight data, center of gravity position test data, and moment of inertia test data. The error indicators include total mass error, wheel load distribution error, center of gravity position error, and moment of inertia error.

[0031] The primary suspension parameter layer corresponds to the primary suspension characteristic test data. The error indicators include primary static deflection error, primary axle box spring stiffness error, primary vertical damper damping error, and axle box positioning stiffness error.

[0032] The secondary suspension parameter layer corresponds to the secondary suspension characteristic test data. The error indicators include secondary static deflection error, secondary spring stiffness error, secondary damper damping error, anti-roll torsion bar stiffness error, and anti-hunting damper damping error.

[0033] The wheel-rail contact parameter layer corresponds to the measured data of wheel-rail profile and the wheel-rail force test data. The error indicators include wheel-rail contact geometric error and wheel-rail friction coefficient error.

[0034] The vehicle operation response calibration layer corresponds to the dynamic response data of the vehicle's line operation or bench test. The error indicators include vehicle body vibration acceleration error, bogie frame vibration acceleration error, wheel-rail force error, suspension component relative displacement error, hunting critical speed error, and hunting frequency error.

[0035] Furthermore, the construction process of the hierarchical calibration objective function is as follows:

[0036] Define the parameter vector to be identified in the k-th layer and its physical feasible region, and determine the set of working conditions and indexes for calibration in this layer;

[0037] Calculate the individual residuals between the simulation output and the experimental observations, and convert the residuals into normalized errors by using a normalized scaling factor;

[0038] The Huber loss function or pseudo-Huber loss function is used as a robust loss function to process the normalization error. Combined with the weight coefficients of multiple working conditions and multiple indicators, a hierarchical basic objective function is constructed.

[0039] A core indicator threshold constraint is introduced, which sets an allowable upper limit on the error of a pre-selected core indicator and adds it to the objective function in the form of a soft constraint penalty term.

[0040] A parameter deviation penalty term is introduced, which is calculated using L2 norm regularization based on the deviation between the parameter vector to be identified and the initial nominal value of the parameter.

[0041] By combining the basic objective function of the comprehensive hierarchy, the core indicator threshold constraint penalty term, and the parameter deviation penalty term, a complete calibration and optimization objective function for the k-th layer is formed.

[0042] Furthermore, the specific execution order of the layer-by-layer parameter calibration is as follows:

[0043] First, perform the calibration of the first layer of basic rigid body parameters to obtain and lock the calibrated rigid body structure parameters;

[0044] Under the condition of locking the first layer of parameters, perform calibration of the second layer of primary suspension parameters, and lock the second layer of parameters after calibration is completed;

[0045] Under the condition that the parameters of the first two layers are locked, the third layer of secondary suspension parameter calibration is performed, and the third layer of parameters is locked after the calibration is completed.

[0046] Under the condition that the first three layers of parameters are locked, the fourth layer of wheel-rail contact parameter calibration is performed, and the fourth layer of parameters is locked after calibration is completed.

[0047] Based on the parameter locking of the first four layers, the fifth layer of vehicle operation response calibration is performed for fine-tuning.

[0048] Furthermore, the inter-layer parameter locking and selective release strategy specifically includes:

[0049] Once the current level calibration is complete and the preset error threshold is met, the parameters of that level are marked as locked.

[0050] During subsequent hierarchical calibration, if the core error decreases by less than the preset proportion in multiple consecutive iterations and the current hierarchical index does not meet the threshold, then the locked parameter review will be triggered.

[0051] Sensitivity analysis was performed on the parameters of the preceding locking layer to identify coupling parameters that have a significant impact on the residuals of the current layer.

[0052] Only release the coupling parameters, and then fine-tune them together with the current level parameters;

[0053] After the joint fine-tuning is completed, the core error index of the level to which the released parameter belongs is checked. If it still meets the corresponding threshold, it is locked again; if it does not meet the threshold, the release range of the coupled parameter is narrowed first and then optimized again; if the corresponding threshold is still not met after narrowing the release range, a rollback operation is performed to restore the released parameter to its original locked value.

[0054] Furthermore, the multiple convergence evaluation mechanism includes error attainment determination, change stabilization determination, and verification condition determination, specifically as follows:

[0055] Error compliance determination includes objective function threshold determination and core indicator worst value determination. If both are satisfied, the error is deemed to be compliant. Core indicator worst value determination refers to selecting at least one core indicator from the error indicators at this level and determining whether its maximum error value under all calibration conditions meets the preset error upper limit.

[0056] The determination of stabilization of change includes the determination of the relative decrease of the objective function and the determination of the relative change of parameters. If the magnitude threshold is met in multiple consecutive iterations, it is determined that the change is stabilizing.

[0057] The verification condition determination is to substitute the calibration result into a set of verification conditions that have been pre-separated from the test dataset of that level. If the objective function satisfies the preset upper limit, the verification is deemed to have passed. The set of verification conditions consists of independent test conditions that do not participate in the calibration optimization process and are used to verify the generalization ability of the calibration result.

[0058] When the error meets the standard, the change stabilizes, and the verification condition is met simultaneously, the calibration at that level is considered to have converged.

[0059] Furthermore, the preprocessing of the test data includes time synchronization, coordinate unification, outlier removal, filtering, feature extraction, and working condition segmentation, so as to establish a one-to-one correspondence between the test data and the simulation output.

[0060] Correspondingly, the present invention also provides a hierarchical calibration system for a railway vehicle dynamics simulation model, comprising:

[0061] The model and data preprocessing module is used to build the initial simulation model, obtain the raw parameters and experimental data, and complete the experimental data preprocessing.

[0062] The parameter hierarchical module is used to divide the parameters to be calibrated into multiple calibration levels from the bottom to the top according to the physical subordination and coupling strength of the parameters;

[0063] The data mapping module is used to establish the corresponding mapping relationship between each calibration level and test data and error indicators;

[0064] The hierarchical calibration module is used to construct the calibration objective function for each level and perform layer-by-layer parameter optimization in order from the bottom to the top.

[0065] The parameter locking and releasing module is used to lock the inter-layer parameters, and when the subsequent layer calibration meets the preset trigger conditions, perform sensitivity analysis on the previously locked parameters, selectively release the coupling parameters and fine-tune them together with the current layer parameters, and re-verify and lock them after fine-tuning.

[0066] The convergence evaluation module is used to determine the calibration convergence status of each level through multiple criteria.

[0067] The integrated verification module is used to perform multi-condition integrated verification of the model after all levels of calibration are completed and output the calibration results.

[0068] Furthermore, the system can be deployed on industrial computers, simulation workstations, or servers, or it can be integrated into the background calibration module of railway dynamics simulation software.

[0069] Beneficial effects:

[0070] 1. This invention divides the high-dimensional parameters to be calibrated into multiple independent levels according to their physical hierarchical relationship and coupling strength, and performs optimization and solution layer by layer. This greatly reduces the parameter search space of single-step optimization, avoids the mutual compensation problem caused by simultaneous optimization of all parameters, and significantly improves the convergence speed and stability of the calibration algorithm.

[0071] 2. This invention configures exclusive test data and error indicators for each level, so that different types of test data can be used to calibrate the most physically related subset of parameters, avoiding interference from irrelevant data on parameter identification, and improving the utilization rate of test data and the physical consistency of parameter identification results.

[0072] 3. This invention adopts a layered calibration and layered error recording mode. Each level outputs independent error results and parameter change records, which can directly locate the source level and parameter category of simulation error, provide a quantitative basis for model credibility assessment, and solve the problem of poor interpretability of traditional overall calibration.

[0073] 4. This invention adopts a calibration sequence from the bottom basic rigid body to the top vehicle response. The basic rigid body parameters are accurately calibrated first, and then the upper subsystem parameters are calibrated. Combined with the inter-layer parameter locking mechanism, it avoids the compensatory shift of the upper layer parameters to the basic parameter error, ensures the authenticity of the physical mechanism of the calibrated model, and improves the engineering application value.

[0074] 5. This invention employs a parameter locking and selective release strategy to perform cross-layer joint fine-tuning on only a small number of strongly coupled parameters while ensuring the efficiency of hierarchical calibration. This avoids the computational burden of full rollback and compensates for the neglect of cross-layer coupling characteristics in pure hierarchical calibration, thus balancing calibration efficiency and accuracy. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 This is a schematic diagram of the overall process of the layered calibration method of the present invention.

[0077] Figure 2 This is a schematic diagram of the hierarchical structure of parameters for a railway vehicle dynamics simulation model.

[0078] Figure 3 This diagram illustrates the correspondence between parameters at each level and experimental data and error indicators.

[0079] Figure 4This is a flowchart illustrating the interlayer parameter locking and selective release strategy.

[0080] Figure 5 This is a structural block diagram of the hierarchical calibration system of the present invention.

[0081] Figure 6 This is an example of basic rigid body parameter layer parameter calibration in Example 3, wherein, Figure 6 a is the normalized residual of the basic rigid body layer. Figure 6 b is the normalized residual of the first-order suspension layer. Figure 6 c represents the normalized residual of the bipolar suspended layer. Figure 6 d represents the normalized residual of the wheel-rail contact layer. Figure 6 e represents the normalized residual of the vehicle response layer.

[0082] Figure 7 For error compliance judgment ( Figure 7 a), the relative error between parameter identification results and design values ​​( Figure 7 b).

[0083] Figure 8 This is a schematic diagram of the calibrated vehicle parameters, where, Figure 8 'a' represents the calibrated time-domain signal of the vehicle's operating response – vehicle acceleration. Figure 8 b represents the calibrated vehicle operating response error - vehicle acceleration. Figure 8 c represents the calibrated time-domain signal of the vehicle's operating response – wheel-rail force. Figure 8 d represents the calibrated time-domain signal of the vehicle's operating response - critical speed / frequency of the serpentine pattern. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0085] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0086] Example 1:

[0087] This embodiment addresses the technical problem of the progressive amplification of basic parameter errors in existing technologies. It details the specific implementation process of this invention, which uses a layer-by-layer calibration and locking mechanism from the bottom to the top layer to block the upward propagation of basic parameter errors. The overall process of this embodiment is as follows: Figure 1 As shown, the parameter hierarchical structure is as follows: Figure 2 As shown.

[0088] Step 1: Initial Model Building and Data Preprocessing

[0089] First, an initial simulation model of the multibody dynamics of a certain type of EMU to be calibrated is constructed. The model includes the car body subsystem, bogie frame subsystem, wheelset subsystem, primary suspension subsystem, secondary suspension subsystem, and wheel-rail contact subsystem. It is constructed based on the SIMPACK dynamics simulation software, and the initial parameters are the design values ​​in the vehicle design drawings.

[0090] Simultaneously collect test data corresponding to the actual vehicle, including: static weighing data (total mass of empty vehicle, vertical load of each wheel), center of gravity and moment of inertia test data, primary / secondary suspension characteristic test data, wheel-rail profile measured data, and vehicle dynamic response data under straight running conditions.

[0091] The test data were preprocessed as follows: time synchronization of multi-sensor data was achieved by aligning timestamps; all measurement data were converted to a unified vehicle coordinate system; abnormal measurement values ​​were eliminated by the 3σ criterion; noise reduction of vibration acceleration signals was achieved by Butterworth low-pass filtering; characteristic indicators such as peak value and root mean square value were extracted from the time history data; and data was segmented according to the operating conditions so that each test data could be established in a one-to-one correspondence with the simulation output.

[0092] Step 2: Parameter hierarchical partitioning

[0093] Based on the physical hierarchy and coupling strength of the parameters, the parameters to be calibrated are divided into five calibration levels from the bottom to the top, such as... Figure 2 As shown:

[0094] L1 Basic Rigid Body Parameter Layer: Basic structural parameters such as vehicle body mass, vehicle body moment of inertia, vehicle center of gravity position, bogie frame mass, and wheelset mass;

[0095] L2 primary suspension parameter layer: primary axle box spring vertical / lateral / longitudinal stiffness, primary vertical damper damping, axle box positioning three-dimensional stiffness;

[0096] L3 secondary suspension parameters: triaxial stiffness of secondary air springs, damping of secondary vertical / lateral dampers, stiffness of anti-roll torsion bar, damping of anti-hunting dampers;

[0097] L4 wheel-rail contact parameter layer: wheel equivalent taper, wheel-rail friction coefficient;

[0098] L5 Vehicle Operation Response Calibration Layer: Used for fine-tuning of the strongly coupled parameters of the first four layers.

[0099] Step 3: Establishing the Hierarchical-Data-Indicator Mapping

[0100] Each level is matched with specific experimental data and error indicators, and the mapping relationship is as follows: Figure 3 As shown:

[0101] L1 layer matches static weighing, center of gravity, and inertia test data, with error indices including total mass error, wheel weight distribution error, center of gravity position error, and rotational inertia error;

[0102] The L2 layer is matched with the primary suspension characteristic test data, and the error indexes are primary static deflection error, spring stiffness error, damper damping error, and positioning stiffness error.

[0103] The L3 layer matched the secondary suspension characteristic test data, and the error indexes were secondary static deflection error, spring stiffness error, damper damping error, anti-roll torsion bar stiffness error, and anti-hunting damper damping error.

[0104] L4 layer matches wheel-rail profile and wheel-rail force test data, and the error indicators are contact geometry error and friction coefficient error;

[0105] The L5 layer matches the test data of the whole vehicle line, and the error indicators are the vehicle body acceleration error, the frame acceleration error, the wheel-rail force error, and the critical speed error of the snake.

[0106] Step 4: Layer-by-layer calibration and locking execution

[0107] Calibration is performed in order from the bottom layer to the top layer. Once the parameters of the preceding layer are calibrated, they are locked and used as constants in subsequent layer calculations.

[0108] Perform L1 basic rigid body parameter layer calibration: Based on static and inertia test data, optimize rigid body mass, inertia, and center of gravity parameters to ensure that the total mass error is less than 1%, the wheel weight distribution error is less than 2%, and the center of gravity and inertia error is less than 3%. After calibration, mark all rigid body parameters as locked and treat them as constant values ​​in subsequent layer calibrations, and they will not participate in optimization.

[0109] Perform L2 primary suspension parameter layer calibration: Based on the locked L1 parameters, optimize the primary suspension stiffness and damping parameters using the primary suspension characteristic test data as a benchmark. Since the basic rigid body parameters have been accurately calibrated, the suspension parameters do not need to compensate for load errors caused by mass deviations, thus avoiding the upward transmission of basic errors.

[0110] Perform L3, L4, and L5 layer calibrations sequentially. After each layer calibration is completed, lock the parameters before proceeding with the next layer optimization.

[0111] Through the progressive and locking mechanism in this embodiment, the error of the basic rigid body parameters will not be transmitted to the upper subsystems such as suspension and wheel-rail. The calibration results of the upper parameters only reflect their own characteristic deviations, ensuring the authenticity of the physical mechanism of the model and solving the problem of passive compensation of the basic error by the upper parameters and model distortion in traditional overall calibration.

[0112] Example 2:

[0113] This embodiment addresses the technical problems of strong parameter coupling and poor overall calibration convergence in the prior art, and details the specific implementation process of the interlayer parameter locking and selective release strategy of the present invention. The strategy flow is as follows: Figure 4 As shown.

[0114] Based on the layer-by-layer calibration in Example 1, when performing the L4 wheel-rail contact parameter layer calibration, it was found that the error of the wheel-rail lateral force index decreased by less than 1% after 5 iterations and could not meet the preset 5% error threshold, thus triggering the parameter lock verification mechanism.

[0115] Step 1: Sensitivity Analysis of Coupling Parameters

[0116] A global sensitivity analysis was performed on the previously locked parameters of layers L1-L3, and the sensitivity coefficients of each parameter to the wheel-rail lateral force error were calculated.

[0117] L1 rigid body parameters: sensitivity coefficients are all less than 0.05, having minimal impact on the current error;

[0118] L2 series parameters: The sensitivity coefficient of the longitudinal positioning stiffness of the axle box is 0.12, which has a weak influence;

[0119] L3 secondary parameters: The sensitivity coefficient of the anti-hunting damper is 0.38, which has a significant impact on the wheel-rail lateral force and hunting characteristics, and belongs to the strongly coupled parameters.

[0120] Ultimately, the damping of the anti-hunting damper was identified as a coupling parameter that significantly affects the residual of the current layer.

[0121] Step 2: Parameter Release and Joint Fine-tuning

[0122] Only the damping parameters of the anti-hunting shock absorber in layer L3 are released, and these parameters, along with the wheel-rail friction coefficient and equivalent taper parameter in layer L4, are used to form a joint optimization variable. Joint fine-tuning is carried out with the goal of minimizing the wheel-rail lateral force error. The remaining pre-locked parameters are kept constant and do not participate in this optimization, thereby reducing the number of optimization variables from dozens of parameters to 3, significantly narrowing the search space.

[0123] Step 3: Verification and Relocking

[0124] After the joint fine-tuning was completed, the core error indicators of the L3 layer (static deflection error of the secondary suspension and damping error of the anti-hunting damper) were re-verified and confirmed that they still met the preset 5% error threshold requirement. Therefore, the damping parameter of the anti-hunting damper was re-marked as locked and the subsequent L5 layer calibration was continued.

[0125] If the core L3 index exceeds the threshold after joint fine-tuning, the release range is narrowed first, and only the parameter with the highest sensitivity is retained for re-optimization; if the narrowing range still does not meet the requirements, a rollback operation is performed, the release parameter is restored to the original locked value, and optimization continues only through the current layer parameter.

[0126] The selective release strategy in this embodiment avoids the dimensionality curse and convergence difficulties caused by simultaneous optimization of all parameters, and also makes up for the neglect of cross-layer coupling characteristics by pure hierarchical calibration. While ensuring convergence efficiency, it also takes into account the correction capability of coupled parameters, effectively solving the problem of poor convergence of strongly coupled parameter calibration.

[0127] Example 3:

[0128] This embodiment addresses the technical problems of inaccurate correspondence between experimental data and parameters and low identification accuracy in the prior art. It details the construction and calculation process of the hierarchical calibration objective function of the present invention and supplements the complete algorithm formula.

[0129] Taking the k-th layer calibration as an example, the complete objective function construction and calculation process is as follows:

[0130] 1. Basic variable definition: Let the parameter vector to be identified in the k-th layer be: Its physical feasible region satisfies: Where k is the calibration layer number, in this embodiment the vehicle dynamics parameters are divided into k layers according to the coupling strength, the larger k is, the stronger the parameter coupling. Let be the parameter vector to be identified in the k-th layer. The number of parameters in the k-th layer; Let be the lower bound vector of the parameters of the k-th layer. The upper bound vector of the parameters of the k-th layer is given by engineering physics constraints.

[0131] Define the set of operating conditions participating in calibration at level k. Indicator set ,in, For the set of test / simulation conditions participating in the calibration at layer k, For the m-th working condition of the k-th layer (such as the linear uniform speed working condition, the curve passing working condition, etc.); The total number of working conditions at the k-th level. This is the set of index numbers extracted from the k-th layer under each working condition; denoted as the number of indicators at the k-th level.

[0132] Under any working condition With index i, simulation output Compared with experimental observations ;in, During the calibration process for the k-th layer, the parameters of that layer are given. At that time, working conditions The simulation output value of the i-th index; For working conditions The experimental observation value of the i-th index (obtained from bench tests or line tests); For dynamic simulation mapping, the parameters locked in the previous layer and the parameters of the current layer are input into the dynamic model, and the corresponding working conditions and index values ​​are output as functions. (j=1,…,k-1) represents the parameter values ​​after the j-th layer calibration is completed and locked.

[0133] 2. Normalized Residual Calculation: To account for the differences in the dimensions of different indicators and ensure numerical stability, we define individual residuals and normalized residuals: To eliminate dimensional differences, a normalization scaling factor is defined: in, To prevent the denominator from being too small a positive number, the normalization error is set according to the magnitude of the index; the normalization error is: in, The original residual (difference between simulation and experiment) of index i for the k-th working condition m. It is a normalization scaling factor used to make indicators with different dimensions dimensionless; For the normalized scalar error, see details. Figure 6 As shown.

[0134] 3. Robust Loss Function Processing: The Huber loss function is used as the robust loss function to process the normalized residuals and suppress the influence of outliers.

[0135] For normalized residuals Huber loss is defined as:

[0136]

[0137] in, The threshold parameter for Huber loss is δ=1.345 in this embodiment, which balances robustness with estimation efficiency. The error after robust loss processing is: ,in, The normalized error is processed by the robust loss function; This is the robust loss function.

[0138] 4. Hierarchical basic objective function: The hierarchical basic objective function is a weighted sum of multiple working conditions and multiple indicators.

[0139]

[0140] in, The scalar objective function value for the k-th layer; The weight coefficient of index i for the working condition m at the k-th level reflects the importance of the index in the calibration of this level. The weight of the core index is set to 0.15~0.2, and the weight of the secondary index is set to 0.05~0.1.

[0141] 5. Core indicator threshold constraints: Select the set of core indicators for this level. Set an upper limit for the allowable error for each core indicator. The hard constraint of the core indicator threshold can be expressed as:

[0142]

[0143] The soft constraint of the core indicator threshold can be expressed as:

[0144]

[0145] in, This is the set of core indicator numbers for the k-th layer.

[0146] This represents the upper limit of the allowable error for the i-th core indicator under operating condition m.

[0147] , which is the penalty factor at the k-th level, applies a second penalty to the objective function when the core indicator exceeds the limit.

[0148] 6. Parameter deviation penalty term: To prevent parameters from deviating too much from the reasonable initial values ​​for the project, a parameter prior deviation penalty term is introduced: ,in, The penalty function for parameter deviation in the k-th layer; For the nominal value / design value / empirical initial value of the parameter at level k; Let be the regularization weight matrix for the k-th layer.

[0149] 7. Finally, there is the complete optimization problem. For the comprehensive objective function, regularization term, and constraint interval of the k-th level complete optimization, the k-th level optimization problem is written as: st ,in, For about variables Find the minimum value; is the regularization coefficient for the k-th layer, used to balance the fitting accuracy and the degree of parameter deviation.

[0150] Through the hierarchical objective function design in this embodiment, each level uses only the corresponding physically related experimental data for parameter identification, avoiding interference from irrelevant data. Simultaneously, robust loss, core index constraints, and regularization terms ensure the accuracy and physical rationality of parameter identification, solving the problems of chaotic data matching and low identification accuracy in traditional overall calibration. The relative errors of all parameters are controlled within ±5%.

[0151] Example 4: This example addresses the technical problem of lack of interpretability and traceability in the calibration process in the prior art, and details the multiple convergence evaluation mechanism and error traceability method of the present invention.

[0152] This invention employs a triple convergence criterion of "error reaching the standard - change stabilizing - verification passed" to avoid false convergence and overfitting. At the same time, each level independently records the error, enabling error traceability.

[0153] To ensure that the calibration of the k-th layer meets the error index requirements while avoiding numerical oscillations, overfitting, or spurious convergence, this invention employs a multi-judgment convergence evaluation method. Let the parameters of the k-th layer in the t-th iteration be... The corresponding objective function value and normalization error are:

[0154]

[0155] Where t is the iteration number of the k-th layer, t=0, 1, 2, ...;

[0156] This is the parameter vector obtained in the k-th layer and t-th iteration;

[0157] This represents the objective function value at the t-th iteration.

[0158] Let be the normalized error of the operating condition index i at the t-th iteration.

[0159] 1. Error compliance judgment (dual channel)

[0160] Criterion 1: Threshold determination of objective function:

[0161]

[0162] Criterion 2: Determination of the worst value of the core indicator:

[0163]

[0164] in, A threshold is preset for the objective function of the k-th layer; see details below. Figure 7 As shown.

[0165] 2. Determination of trend stabilization (stability index):

[0166] To avoid spurious convergence where the objective function is small but oscillating, a criterion of the magnitude of the descent and the change in parameters is introduced:

[0167] Criterion 3: Relative decrease in the objective function:

[0168]

[0169] Criterion 4: Relative change in parameters:

[0170]

[0171] in, The relative decrease in the objective function at the t-th iteration;

[0172] This represents the relative change in the parameter vector at the t-th iteration;

[0173] , To place small positive numbers with a denominator of zero;

[0174] 3. Verify the operating condition determination (to prevent overfitting)

[0175] Substitute the calibration result of the k-th layer into the set of verification conditions and define the verification objective function:

[0176]

[0177] Criterion 5: Verification Error Judgment

[0178]

[0179] in, To verify the objective function corresponding to the working condition;

[0180] To verify the weighting coefficients corresponding to the working conditions;

[0181] To verify the upper limit of the objective function.

[0182] 4. Unified convergence criterion:

[0183] Define the unified convergence criterion for the k-th layer in the t-th iteration as:

[0184]

[0185] in, This represents the Boolean result for the convergence determination of the k-th layer in the t-th iteration. It is the logical operator "AND".

[0186] When all the above criteria are met, the calibration for that level is considered converged. After each level of calibration is completed, an error report and parameter change record for that level are generated, including the error values ​​of each indicator, a comparison of parameters before and after adjustment, and the number of convergence iterations.

[0187] After all levels of calibration are completed, comprehensive vehicle verification is performed. Multi-condition track test data are used to verify the vehicle's dynamic response, and the errors in indicators such as vehicle acceleration, wheel-rail force, and critical hunting speed are all controlled within 5%. See details... Figure 8 As shown, if the comprehensive verification does not meet the requirements, the error source level can be located in reverse according to the error records of each level, and the corresponding level can be returned for recalibration, so as to achieve full-link traceability of the calibration process.

[0188] Through the multiple convergence evaluation and hierarchical error recording mechanism in this embodiment, the calibration process has clear quantitative basis and traceability path, which solves the problems of traditional overall calibration that only outputs the final result, cannot locate the source of error, and has poor interpretability, and provides complete data support for model credibility assessment.

[0189] Example 5:

[0190] This embodiment corresponds to the system scheme of the present invention. The structure of the hierarchical calibration system includes the following functional modules:

[0191] Model and data preprocessing module: responsible for importing the initial dynamic simulation model, reading the original design parameters, nominal parameters and experimental data, performing preprocessing operations such as time synchronization, coordinate unification, filtering, feature extraction, etc., and outputting standardized calibration input data.

[0192] Parameter hierarchical module: Based on the physical subordination and coupling strength of the parameters, it automatically divides the parameters to be calibrated into multiple calibration levels and generates a hierarchical parameter list and physical constraint interval.

[0193] Data mapping module: Establishes a one-to-one mapping relationship between each calibration level and the test dataset and error indicators, and assigns exclusive calibration data and evaluation indicators to each level.

[0194] Hierarchical calibration module: Constructs a hierarchical calibration objective function for each level, calls an optimization algorithm to perform parameter identification, and completes the calibration layer by layer in order from the bottom to the top.

[0195] Parameter locking and releasing module: manages the locking / releasing status of parameters at each level, triggers parameter verification based on the convergence status, and performs sensitivity analysis, selective release, joint fine-tuning, and relocking operations.

[0196] Convergence evaluation module: The convergence status of each calibration layer is determined by three criteria: error compliance, change stabilization, and verification of operating conditions. The module outputs the error report and convergence record for each layer.

[0197] Comprehensive Validation Module: Performs multi-condition comprehensive validation on the model after all levels of calibration are completed, and generates the final calibration parameter set, error analysis report and model credibility evaluation results.

[0198] This system can be deployed on industrial computers, simulation workstations, or servers, and can also be integrated into the background calibration module of railway dynamics simulation software such as SIMPACK and ADAMS / Rail for automatic operation.

[0199] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A hierarchical calibration method for a railway vehicle dynamics simulation model, characterized in that, Includes the following steps: Step S1: Construct an initial simulation model of the multibody dynamics of the railway vehicle to be calibrated, obtain the original parameters and corresponding experimental data of the initial simulation model, and preprocess the experimental data to establish a one-to-one correspondence between the experimental data and the simulation output; Step S2: Based on the physical dependency relationship and coupling strength of the parameters, divide the parameters to be calibrated into multiple calibration levels from the bottom to the top. Step S3: Match a dedicated test dataset and quantification error index for each calibration level, and establish a mapping relationship between level, test data and error index; Step S4: For each calibration level, construct the hierarchical calibration objective function for that level by combining the corresponding operating conditions and evaluation indicators; Step S5: Perform layer-by-layer parameter calibration in order from bottom to top. During the calibration process of the current level, the parameters of the previous level that have been calibrated are locked to constant values ​​and do not participate in the optimization solution of the current level. Step S6: When the subsequent level calibration meets the preset triggering conditions, interrupt the regular calibration process of the current level, perform sensitivity analysis on the previous locked parameters, identify and selectively release the coupling parameters that have a significant impact on the residual of the current level, fine-tune the released coupling parameters together with the current level parameters, re-verify and lock the released parameters after fine-tuning, and then continue the regular calibration process of the current level. Step S7: Use a multiple convergence evaluation mechanism to determine the calibration convergence status of each level. If the convergence criterion is met, proceed to the next level of calibration. Step S8: After all levels of calibration are completed, the calibrated model is comprehensively verified using multi-condition vehicle operation test data. After the verification is passed, the calibration results are output.

2. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 1, characterized in that, The calibration hierarchy consists of five layers, from bottom to top: The first layer is the basic rigid body parameter layer, which includes the mass, moment of inertia and center of gravity position parameters of the vehicle body, bogie frame and wheelsets; The second layer is the primary suspension parameter layer, including the three-dimensional stiffness of the primary axle box springs, the damping of the primary vertical damper, and the three-dimensional stiffness parameters of the axle box positioning. The third layer is the secondary suspension parameter layer, which includes the three-dimensional stiffness of the secondary springs, the damping of the secondary vertical damper, the damping of the secondary lateral damper, the stiffness of the anti-roll torsion bar, and the damping parameters of the anti-hunting damper. The fourth layer is the wheel-rail contact parameter layer, which includes wheel-rail profile parameters, wheel equivalent taper, and wheel-rail friction coefficient parameters; The fifth layer is the vehicle operation response calibration layer, which is used to release and fine-tune at least one parameter that has been locked in the first four layers and is strongly coupled with the vehicle dynamic response.

3. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 2, characterized in that, The specific test data and error indicators for each level are as follows: The basic rigid body parameter layer corresponds to the vehicle weighing data, wheel weight data, center of gravity position test data, and moment of inertia test data. The error indicators include total mass error, wheel load distribution error, center of gravity position error, and moment of inertia error. The primary suspension parameter layer corresponds to the primary suspension characteristic test data. The error indicators include primary static deflection error, primary axle box spring stiffness error, primary vertical damper damping error, and axle box positioning stiffness error. The secondary suspension parameter layer corresponds to the secondary suspension characteristic test data. The error indicators include secondary static deflection error, secondary spring stiffness error, secondary damper damping error, anti-roll torsion bar stiffness error, and anti-hunting damper damping error. The wheel-rail contact parameter layer corresponds to the measured data of wheel-rail profile and the wheel-rail force test data. The error indicators include wheel-rail contact geometric error and wheel-rail friction coefficient error. The vehicle operation response calibration layer corresponds to the dynamic response data of the vehicle's line operation or bench test. The error indicators include vehicle body vibration acceleration error, bogie frame vibration acceleration error, wheel-rail force error, suspension component relative displacement error, hunting critical speed error, and hunting frequency error.

4. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 1, characterized in that, The construction process of the hierarchical calibration objective function is as follows: Define the parameter vector to be identified in the k-th layer and its physical feasible region, and determine the set of working conditions and indexes for calibration in this layer; Calculate the individual residuals between the simulation output and the experimental observations, and convert the residuals into normalized errors by using a normalized scaling factor; The Huber loss function or pseudo-Huber loss function is used as a robust loss function to process the normalization error. Combined with the weight coefficients of multiple working conditions and multiple indicators, a hierarchical basic objective function is constructed. A core indicator threshold constraint is introduced, which sets an allowable upper limit on the error of a pre-selected core indicator and adds it to the objective function in the form of a soft constraint penalty term. A parameter deviation penalty term is introduced, which is calculated using L2 norm regularization based on the deviation between the parameter vector to be identified and the initial nominal value of the parameter. By combining the basic objective function of the comprehensive hierarchy, the core indicator threshold constraint penalty term, and the parameter deviation penalty term, a complete calibration and optimization objective function for the k-th layer is formed.

5. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 2, characterized in that, The specific execution order of the layer-by-layer parameter calibration is as follows: First, perform the calibration of the first layer of basic rigid body parameters to obtain and lock the calibrated rigid body structure parameters; Under the condition of locking the first layer of parameters, perform calibration of the second layer of primary suspension parameters, and lock the second layer of parameters after calibration is completed; Under the condition that the parameters of the first two layers are locked, the third layer of secondary suspension parameter calibration is performed, and the third layer of parameters is locked after the calibration is completed. Under the condition that the first three layers of parameters are locked, the fourth layer of wheel-rail contact parameter calibration is performed, and the fourth layer of parameters is locked after calibration is completed. Based on the parameter locking of the first four layers, the fifth layer of vehicle operation response calibration is performed for fine-tuning.

6. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 1, characterized in that, The interlayer parameter locking and selective release strategy specifically includes: Once the current level calibration is complete and the preset error threshold is met, the parameters of that level are marked as locked. During subsequent hierarchical calibration, if the core error decreases by less than the preset proportion in multiple consecutive iterations and the current hierarchical index does not meet the threshold, then the locked parameter review will be triggered. Sensitivity analysis was performed on the parameters of the preceding locking layer to identify coupling parameters that have a significant impact on the residuals of the current layer. Only release the coupling parameters, and then fine-tune them together with the current level parameters; After the joint fine-tuning is completed, the core error index of the level to which the released parameter belongs is checked. If it still meets the corresponding threshold, it is locked again; if it does not meet the threshold, the release range of the coupled parameter is narrowed first and then optimized again; if the corresponding threshold is still not met after narrowing the release range, a rollback operation is performed to restore the released parameter to its original locked value.

7. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 1, characterized in that, The multiple convergence evaluation mechanism includes error compliance determination, change stabilization determination, and verification condition determination, specifically: Error compliance determination includes objective function threshold determination and core indicator worst value determination. If both are satisfied, the error is deemed to be compliant. Core indicator worst value determination refers to selecting at least one core indicator from the error indicators at this level and determining whether its maximum error value under all calibration conditions meets the preset error upper limit. The determination of stabilization of change includes the determination of the relative decrease of the objective function and the determination of the relative change of parameters. If the magnitude threshold is met in multiple consecutive iterations, it is determined that the change is stabilizing. The verification condition determination is to substitute the calibration result into a set of verification conditions that have been pre-separated from the test dataset of that level. If the objective function satisfies the preset upper limit, the verification is deemed to have passed. The set of verification conditions consists of independent test conditions that do not participate in the calibration optimization process and are used to verify the generalization ability of the calibration result. When the error meets the standard, the change stabilizes, and the verification condition is met simultaneously, the calibration at that level is considered to have converged.

8. The hierarchical calibration method for the railway vehicle dynamics simulation model according to claim 1, characterized in that, The preprocessing of the test data includes time synchronization, coordinate unification, outlier removal, filtering, feature extraction, and working condition segmentation, so as to establish a one-to-one correspondence between the test data and the simulation output.

9. A hierarchical calibration system for a railway vehicle dynamics simulation model, which, after operation, implements the hierarchical calibration method described in any one of claims 1-8, characterized in that, include: The model and data preprocessing module is used to build the initial simulation model, obtain the raw parameters and experimental data, and complete the experimental data preprocessing. The parameter hierarchical module is used to divide the parameters to be calibrated into multiple calibration levels from the bottom to the top according to the physical subordination and coupling strength of the parameters; The data mapping module is used to establish the corresponding mapping relationship between each calibration level and test data and error indicators; The hierarchical calibration module is used to construct the calibration objective function for each level and perform layer-by-layer parameter optimization in order from the bottom to the top. The parameter locking and releasing module is used to lock the inter-layer parameters, and when the subsequent layer calibration meets the preset trigger conditions, perform sensitivity analysis on the previously locked parameters, selectively release the coupling parameters and fine-tune them together with the current layer parameters, and re-verify and lock them after fine-tuning. The convergence evaluation module is used to determine the calibration convergence status of each level through multiple criteria. The integrated verification module is used to perform multi-condition integrated verification of the model after all levels of calibration are completed and output the calibration results.

10. The hierarchical calibration system for the railway vehicle dynamics simulation model according to claim 9, characterized in that, The system can be deployed on industrial computers, simulation workstations, or servers, or it can be integrated into the background calibration module of railway dynamics simulation software.

Images