Starting vibration analysis method and device based on reduced order model
Through the starting vibration analysis method based on the down-order model, the problem of lack of technical guidance in suspended NVH performance adjustment is solved, and the suspension stiffness and position are quickly predicted, the adjustment cost is reduced, and the development efficiency is improved.
Patent Information
- Application Number
- CN202411356655.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-02-18
AI Technical Summary
The prior art lacks clear technical guidance in the performance adjustment of suspended NVH, resulting in large deviations in starting vibration, poor robustness, and high adjustment costs.
The starting vibration analysis method based on the down-order model is adopted. By obtaining the model parameters under the starting conditions, a sample database is constructed, a down-order model is established, and the output parameters are estimated through model training and verification analysis to evaluate the model quality.
This method can quickly predict suspension stiffness and position, reduce adjustment costs, improve development efficiency, shorten R&D time, and provide a cost-reducing solution.
Smart Images
Figure CN119475958A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of vehicle testing and analysis, and in particular to a starting vibration analysis method and device based on a reduced-order model. Background Art
[0002] The existing suspension NVH performance tuning is mostly based on experience. The approximate suspension layout range is given, and the suspension is subsequently matched according to the actual vehicle. For the same powertrain to match different body positions, it is necessary to re-simulate the suspension system and analyze the system, and the whole vehicle tuning work has a long cycle. At the same time, the actual tuning work is heavily dependent on the actual vehicle. The cost of tuning a set of assemblies is 20W. Generally, the platform prototype needs to be tuned once, and the OTS prototype needs to be tuned once. The cost of the trial prototype ranges from 40W to 70W, which is costly. For the starting vibration condition, the suspension position is reasonably arranged, and the difference in the vibration size of multiple starts is small. However, when the arrangement is unreasonable, the starting vibration deviation is large and the robustness is poor. At this stage, there is no clear technical guidance for the early stage of the suspension layout position, and the above problems need to be solved urgently. Summary of the invention
[0003] The object of the present invention is to provide a starting vibration analysis method and device based on a reduced-order model to solve the problems raised in the above-mentioned background technology.
[0004] To achieve the above object, the present invention provides the following technical solution: a starting vibration analysis method based on a reduced-order model, comprising:
[0005] Based on the starting condition, model parameters of the reduced-order model are obtained, and a sample database of the model parameters is constructed, wherein the model parameters include input parameters and output parameters;
[0006] Dividing the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio;
[0007] Establish a reduced-order model, train the reduced-order model through modeling parameters, and verify and analyze the trained reduced-order model through verification parameters to obtain estimated output parameters;
[0008] The quality of the reduced-order model was assessed by comparing the estimated output parameters with the output parameters in the modeling parameters.
[0009] Furthermore, based on the starting condition, the model parameters of the reduced-order model are obtained, and a sample database of the model parameters is constructed, which also includes:
[0010] Design tests based on starting conditions and process test data;
[0011] The model parameters related to the starting vibration in the processed test data are obtained, and a sample database of the model parameters is constructed.
[0012] Furthermore, based on the starting conditions, the test is designed and the test data is processed, including:
[0013] The test includes simulation analysis and calculation of the whole vehicle under starting conditions and actual measurement of the whole vehicle;
[0014] The model parameters obtained include simulation analysis data of the vehicle simulation analysis calculation model under starting conditions and vehicle measured data, a ratio of the simulation analysis data to the vehicle measured data is preset, and a sample database of model parameters is constructed according to the ratio.
[0015] Furthermore, the model parameters include input parameters and output parameters, wherein the input parameters include starter torque peak, starter torque slope, engine cylinder pressure, powertrain suspension stiffness, powertrain suspension position and shock absorber stiffness damping, and the output parameters include seat rail vibration acceleration.
[0016] Further, under the vehicle starting condition, the seat rail vibration acceleration value is obtained;
[0017] Filter the seat rail vibration acceleration value;
[0018] The seat rail vibration acceleration value obtained is the average of the square of the peak and trough differences in the X, Y, and Z axis directions in the vehicle coordinate system.
[0019] Further, a reduced-order model is established, the reduced-order model is trained by modeling parameters, and the reduced-order model after model training is verified and analyzed by verification parameters to obtain estimated output parameters, which also includes:
[0020] Using the modeling parameters, the reduced-order model is trained by calling a Kriging algorithm and / or a neural network regression model;
[0021] The verification parameters are analyzed in the trained reduced-order model to obtain the estimated seat rail vibration acceleration values. The estimated vibration values are compared with the actual starting vibration values in the sample database to evaluate the quality of the reduced-order model.
[0022] Furthermore, the estimated vibration value is compared with the actual starting vibration value to evaluate the quality of the reduced-order model, which also includes:
[0023] The average normalized error calculation and Pearson coefficient are used to evaluate whether the quality of the reduced-order model meets the qualification standard;
[0024] The calculation formula for the average normalized error is:
[0025]
[0026] Where: MNE is the mean normalized error, f(x i ) is the estimated value of the reduced-order model, yi is the original actual value, y max and min are the maximum and minimum values of the model output parameters respectively;
[0027] The calculation formula of Pearson coefficient is:
[0028]
[0029] Where: R 2 is the Pearson coefficient, RSS is the sum of squares of deviations between the true value and the estimated value caused by the error, and TSS is the total variance.
[0030] Furthermore, a qualified standard for the quality of the reduced-order model is set, and a first threshold of MNE is set according to the qualified standard, and R 2 The second threshold value of
[0031] The MNE value is not greater than the first threshold, and R 2 When it is not less than the second threshold, the quality of the reduced-order model meets the qualification standard;
[0032] Select the lower MNE among the reduced-order models that meet the quality criteria while maintaining a high R 2 The reduced-order model of the value is used as a large data reduced-order model.
[0033] Further, a third threshold value of the seat rail vibration acceleration value is set and recorded as A;
[0034] Export the big data reduced-order model into an m file recognizable by MATLAB, and set the vibration acceleration value of the seat rail to ≤A;
[0035] The lower deviation of the suspension and shock absorber stiffness is taken as the lower boundary of stiffness, and the upper deviation of the suspension and shock absorber stiffness is taken as the upper boundary of stiffness;
[0036] According to the actual project powertrain suspension position parameters and suspension stiffness boundaries, the suspension position coordinates, suspension stiffness, shock absorber stiffness and damping are obtained through the big data reduction model.
[0037] On the other hand, a starting vibration analysis device based on a reduced-order model is provided, comprising:
[0038] An acquisition module, used for acquiring model parameters of the reduced-order model based on the starting condition, and constructing a sample database of the model parameters, wherein the model parameters include input parameters and output parameters;
[0039] A setting module, used to divide the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio;
[0040] A construction module is used to establish a reduced-order model, perform model training on the reduced-order model through modeling parameters, and perform verification analysis on the trained reduced-order model through verification parameters to obtain estimated output parameters;
[0041] The evaluation module compares the estimated output parameters with the output parameters in the modeling parameters to evaluate the quality of the reduced-order model.
[0042] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention performs big data order reduction modeling based on the previous starting parameters of the developed powertrain, establishes a big data model for the parameters related to the starting vibration and suspension, sets the starting vibration target and designs the suspension layout position through big data order reduction prediction, and provides a reliable basis for the selection of suspension stiffness adjustment. There are many parameters involved, and the number of powertrain suspensions is 3-4, each suspension has three directional positions, and each suspension has three directional stiffness. The big data order reduction prediction method can be used to apply machine learning to multi-dimensional parameters, and the suspension stiffness is predicted with the starting vibration as the boundary. The suspension NVH simulation design can obtain a rapid prediction effect through the big data order reduction model, which can improve efficiency and provide a cost reduction solution. The prediction of the starting vibration through different algorithms of the present invention and the big data order reduction model established through historical data can reversely deduce the design solution that meets the seat guide vibration acceleration target perceived by the customer, shorten the research and development time, and reduce the subsequent adjustment cost through design optimization, thereby reducing the prototype vehicle development cost. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 It is a flow chart of a starting vibration analysis method based on a reduced-order model in an embodiment of the present invention;
[0044] Figure 2 A schematic diagram of the degree of fit of training data for a Kriging algorithm training model according to an embodiment of the present invention;
[0045] Figure 3 This is a schematic diagram of the Kriging algorithm training model verification data consistency in an embodiment of the present invention;
[0046] Figure 4 Schematic diagram of the degree of consistency of training data for the MLP-neural network algorithm training model in an embodiment of the present invention;
[0047] Figure 5 This is a schematic diagram of the MLP-neural network algorithm training model verification data consistency in an embodiment of the present invention;
[0048] Figure 6 It is a connection block diagram of a starting vibration analysis device based on a reduced-order model in an embodiment of the present invention;
[0049] In the figure: 10, acquisition module; 20, setting module; 30, construction module; 40, evaluation module. DETAILED DESCRIPTION
[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0051] In the description of the present invention, unless otherwise specified, "plurality" means two or more than two; the orientations or positional relationships indicated by the terms "upper", "lower", "left", "right", "inner", "outer", "front end", "rear end", "head", "tail", etc. are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0052] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0053] Please refer to the attached drawings of the specification, the present invention provides a technical solution: a starting vibration analysis method based on a reduced-order model, comprising the following steps:
[0054] S102, based on the starting condition, obtaining model parameters of the reduced-order model, and constructing a sample database of the model parameters, wherein the model parameters include input parameters and output parameters;
[0055] Among them, this step firstly separates the model parameters related to the starting vibration under the starting condition, and defines the model parameters, wherein the model parameters include the starting related parameters, namely the input parameters, and the starting response parameters, namely the output parameters.
[0056] Specifically, based on the starting condition, the model parameters of the reduced-order model are obtained, and a sample database of the model parameters is constructed, which also includes:
[0057] Design tests based on starting conditions and process test data;
[0058] The model parameters related to the starting vibration in the processed test data are obtained, and a sample database of the model parameters is constructed.
[0059] Among them, in this embodiment, it is preferred that the DOE test method and the LHS test method can be used to build the database respectively.
[0060] Specifically, the test includes simulation analysis and calculation of the whole vehicle under starting conditions and actual measurement of the whole vehicle;
[0061] The model parameters obtained include simulation analysis data of the vehicle simulation analysis calculation model under starting conditions and vehicle actual measurement data, a ratio of the simulation analysis data to the vehicle actual measurement data is preset, and a sample database of model parameters is constructed according to the ratio;
[0062] Among them, the input parameters and output parameters used by the reduced-order model come from the vehicle simulation analysis calculation model of the starting condition and the vehicle measured data. The modeling method of the simulation analysis calculation model can refer to patent CN202310071621.8 A model-based low-frequency vibration optimization simulation control method.
[0063] For example, in this embodiment, the reduced-order model parameter collection requires 1,000 parameter groups, each of which includes input parameters and corresponding output parameters. At least 100 pieces of measured data are required, and at least 900 pieces of simulation parameters of the simulation analysis calculation model are required to ensure prediction accuracy.
[0064] For example, according to the input parameters and the corresponding parameter gradients in the simulation parameters, they are input into the simulation analysis model, and then the output parameters in the simulation parameters are calculated, that is, the output seat rail vibration acceleration value RSS value, as shown in Table 1 below, which is the input parameter in the sample database;
[0065] Table 1 Input parameters
[0066]
[0067] Specifically, the model parameters include input parameters and output parameters, wherein the input parameters include starter torque peak, starter torque slope, engine cylinder pressure, powertrain suspension stiffness, powertrain suspension position and shock absorber stiffness damping, and the output parameters include seat rail vibration acceleration.
[0068] Specifically, under the vehicle starting condition, the seat rail vibration acceleration value is obtained;
[0069] Filter the seat rail vibration acceleration value;
[0070] The seat rail vibration acceleration value obtained is the average of the square of the peak and trough differences in the X, Y, and Z axis directions in the vehicle coordinate system.
[0071] Among them, the vibration acceleration value of the seat rail in the vehicle starting condition, the industry evaluates that this vibration size needs to be filtered. Therefore, in this embodiment, 5-32Hz filtering is preferred. The RSS value of the seat rail vibration acceleration value is the average of the squares of the peak and trough differences (PP values) in the X, Y, and Z directions in the vehicle coordinate system.
[0072] Data cleaning requires cleaning up unreasonable parameters, as such data will seriously affect data regression.
[0073] S104, dividing the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio;
[0074] In this embodiment, preferably, 80% of the parameter groups in the sample database are randomly used for training, and the remaining 20% of the parameter groups in the sample database are used as verification parameters for verifying the reduced-order model.
[0075] S106, establishing a reduced-order model, training the reduced-order model through modeling parameters, and performing verification analysis on the trained reduced-order model through verification parameters to obtain estimated output parameters;
[0076] Specifically, the modeling parameters are used to train the reduced-order model by calling the Kriging algorithm and / or the neural network regression model; the verification parameters are analyzed in the trained reduced-order model to obtain the estimated seat rail vibration acceleration value, and the estimated vibration value is compared with the actual starting vibration value in the sample database to evaluate the quality of the reduced-order model.
[0077] Among them, the model training calls the following algorithms respectively: Kriging algorithm (Kriging), neural network regression model (MLP-neural network algorithm);
[0078] like Figure 2 As shown in the figure, the Kriging algorithm training model and training data are consistent, the green is the database result, and the yellow square data is the reduced-order model, with a consistency of more than 95%; Figure 3 As shown, the Kriging algorithm training model verifies the data consistency. The green is the database result, and the yellow square data is the reduced-order model. The consistency is over 98%;
[0079] like Figure 4 As shown in the figure, the MLP-neural network algorithm training model training data fits well, the green is the database result, the yellow square data is the reduced-order model, and the fit is over 92%; Figure 5 As shown, the MLP-neural network algorithm training model verifies the data consistency. The green is the database result, and the yellow square data is the reduced-order model. The consistency exceeds 90%;
[0080] Kriging uses a regression algorithm. The prediction results obtained by the difference are highly consistent with the actual results. The neural network is calculated through the hidden layer nodes. The overall prediction results are consistent with the actual results, with only a very small number of points exceeding the deviation. Therefore, the reduced-order model can be trained using the Kriging algorithm or the neural network regression model, or the Kriging algorithm and the neural network regression model can be used to train the reduced-order model respectively, and then the two results can be compared and verified to determine and select a more suitable reduced-order model.
[0081] S108. Compare the estimated output parameters with the output parameters in the modeling parameters to evaluate the quality of the reduced-order model.
[0082] Specifically, the estimated vibration value is compared with the actual starting vibration value to evaluate the quality of the reduced-order model, which also includes:
[0083] The average normalized error calculation and Pearson coefficient are used to evaluate whether the quality of the reduced-order model meets the qualification standard;
[0084] The calculation formula for the average normalized error is:
[0085]
[0086] Where: MNE is the mean normalized error, f(x i ) is the estimated value of the reduced-order model, y i is the original actual value, y max and min are the maximum and minimum values of the model output parameters respectively;
[0087] The calculation formula of Pearson coefficient is:
[0088]
[0089] Where: R 2 is the Pearson coefficient, RSS is the sum of squares of deviations between the true value and the estimated value caused by the error, and TSS is the total variance.
[0090] Among them, mean normalized error (MNE): This is the ratio of the absolute error in the prediction over the range of all response values in the validation dataset, R 2 is a measure of how well the model explains the variance in the data; this is also known as the Pearson coefficient.
[0091] Specifically, the quality qualification standard of the reduced-order model is set, the first threshold of MNE is set according to the qualification standard, and R 2 The second threshold value of
[0092] The MNE value is not greater than the first threshold, and R 2 When it is not less than the second threshold, the quality of the reduced-order model meets the qualification standard;
[0093] Select the lower MNE among the reduced-order models that meet the quality criteria while maintaining a high R 2 The reduced-order model of the value is used as a large data reduced-order model.
[0094] In this embodiment, the first threshold is preferably 10%, the second threshold is 80%, the MNE value is not greater than 10%, and R 2 When it is not less than 80%, the quality of the reduced-order model meets the qualification standard.
[0095] It can be understood that a good reduced-order model has a lower MNE while maintaining a higher R2 value. The model validation results of the two algorithms are compared, and the one with a smaller mean normalized error (MNE) and a larger R2 is selected as the big data reduced-order model.
[0096] Further, a third threshold value of the seat rail vibration acceleration value is set and recorded as A;
[0097] Export the big data reduced-order model into an m file recognizable by MATLAB, and set the vibration acceleration value of the seat rail to ≤A;
[0098] The lower deviation of the suspension and shock absorber stiffness is taken as the lower boundary of stiffness, and the upper deviation of the suspension and shock absorber stiffness is taken as the upper boundary of stiffness;
[0099] According to the actual project powertrain suspension position parameters and suspension stiffness boundaries, the suspension position coordinates, suspension stiffness, shock absorber stiffness and damping are obtained through the big data reduction model.
[0100] It can be understood that the big data reduced-order model is exported into an m file recognizable by MATLAB, the starting vibration is set to ≤A m / s^2, and the powertrain suspension is used as the position parameter and the suspension stiffness boundary according to the actual project. The suspension position coordinates, suspension stiffness, shock absorber stiffness and damping are obtained through the prediction model; because the powertrain suspension and shock absorber have stiffness fatigue requirements, the lower deviation of the suspension and shock absorber stiffness is used as the lower boundary of stiffness, and because the large stiffness of the suspension directly affects the acceleration vibration isolation rate of the suspension, the upper deviation of the suspension is used as the upper boundary of stiffness; at the same time, the powertrain suspension position has a certain boundary for the system to choose.
[0101] The present invention uses historical test and simulation data to perform big data analysis, and compiles complex NVH technical requirements into a reduced-order model through model parameters and phenomena. This method can use the reduced-order model to evaluate NVH technical parameters and identify and predict NVH risks in the conceptual design stage, thereby reducing costs and improving development efficiency from the early stages of design. This technology demonstrates the use of a reduced-order metamodel NVH development framework to predict the vibration performance of seat guide rails under starting conditions. The reduced-order model established by combining test data with simulation data based on a physical model can be applied to NVH design in the early stages of conceptual design, and the reduced-order model can be used to Design parameters corresponding to NVH performance are designed, and historical data can be mined to develop a predictive regression model; the starting vibration is studied using the big data reduction modeling method, which is convenient for finding a reasonable suspension layout position in the early conceptual design stage. The correlation between the suspension position and the starting vibration is found through the big data reduction model, and the structural performance selection of motor-related parameters is studied. The stability of the starting vibration is studied and a reasonable design for reducing the starting vibration is summarized; for example, the starting vibration of a vehicle model may be completed through adjustment. The traditional adjustment process requires testing 9 sets of solutions through orthogonal experimental design, and the suspension stiffness interval is expected to be 30N / mm. However, through the machine learning process, the suspension solution is not limited by the test manufacturing cycle and cost, the suspension stiffness solution can be unrestricted, and the minimum stiffness interval can be 1N / mm. The suspension NVH solution can be found more accurately through big data, and the NVH performance of the whole vehicle can be predicted using the big data reduction modeling method. In the absence of detailed data parameters in the conceptual design, the starting vibration level of a new vehicle model can be predicted through the content of the historical database.
[0102] On the other hand, a starting vibration analysis device based on a reduced-order model is provided, comprising:
[0103] An acquisition module 10 is used to acquire model parameters of the reduced-order model based on the starting condition and to construct a sample database of the model parameters, wherein the model parameters include input parameters and output parameters;
[0104] Among them, based on the starting conditions, the test is designed and the test data is processed;
[0105] Obtain model parameters related to starting vibration in the processed test data and construct a sample database of model parameters;
[0106] The test includes simulation analysis and calculation of the whole vehicle under starting conditions and actual measurement of the whole vehicle;
[0107] The model parameters obtained include simulation analysis data of the vehicle simulation analysis calculation model under starting conditions and vehicle actual measurement data, a ratio of the simulation analysis data to the vehicle actual measurement data is preset, and a sample database of model parameters is constructed according to the ratio;
[0108] The model parameters include input parameters and output parameters, wherein the input parameters include starter torque peak value, starter torque slope, engine cylinder pressure, powertrain mounting stiffness, powertrain mounting position and shock absorber stiffness damping, and the output parameters include seat rail vibration acceleration;
[0109] Under the vehicle starting condition, obtain the seat rail vibration acceleration value;
[0110] Filter the seat rail vibration acceleration value;
[0111] The seat rail vibration acceleration value obtained is the average of the square of the peak and trough differences in the X, Y, and Z axis directions in the vehicle coordinate system.
[0112] A setting module 20, for dividing the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio;
[0113] In this embodiment, the module 20 is preferably configured to retrieve 80% of the parameter groups in the sample database for training at random, and the remaining 20% of the parameter groups in the sample database are used as verification parameters for verifying the reduced-order model.
[0114] A construction module 30 is used to establish a reduced-order model, perform model training on the reduced-order model through modeling parameters, and perform verification analysis on the trained reduced-order model through verification parameters to obtain estimated output parameters;
[0115] The modeling parameters are used to train the reduced-order model by calling the Kriging algorithm and / or the neural network regression model; the verification parameters are analyzed in the trained reduced-order model to obtain the estimated seat rail vibration acceleration value, and the estimated vibration value is compared with the actual starting vibration value in the sample database to evaluate the quality of the reduced-order model.
[0116] The evaluation module 40 compares the estimated output parameters with the output parameters in the modeling parameters to evaluate the quality of the reduced-order model.
[0117] The evaluation module 40 evaluates whether the quality of the reduced-order model meets the qualified standard by calculating the average normalized error and the Pearson coefficient;
[0118] Set the quality qualification standard of the reduced-order model, set the first threshold of MNE according to the qualification standard, and R 2 The second threshold value of
[0119] The MNE value is not greater than the first threshold, and R 2 When it is not less than the second threshold, the quality of the reduced-order model meets the qualification standard;
[0120] Select the lower MNE among the reduced-order models that meet the quality criteria while maintaining a high R2 The reduced-order model of the value is used as a large data reduced-order model;
[0121] Setting a third threshold of the seat rail vibration acceleration value and recording it as A;
[0122] Export the big data reduced-order model into an m file recognizable by MATLAB, and set the vibration acceleration value of the seat rail to ≤A;
[0123] The lower deviation of the suspension and shock absorber stiffness is taken as the lower boundary of stiffness, and the upper deviation of the suspension and shock absorber stiffness is taken as the upper boundary of stiffness;
[0124] According to the actual project powertrain suspension position parameters and suspension stiffness boundaries, the suspension position coordinates, suspension stiffness, shock absorber stiffness and damping are obtained through the big data reduction model.
[0125] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A starting vibration analysis method based on a reduced-order model, characterized in that: include: Based on the starting condition, model parameters of the reduced-order model are obtained, and a sample database of the model parameters is constructed, wherein the model parameters include input parameters and output parameters; Dividing the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio; Establish a reduced-order model, train the reduced-order model through modeling parameters, and verify and analyze the trained reduced-order model through verification parameters to obtain estimated output parameters; The quality of the reduced-order model was assessed by comparing the estimated output parameters with the output parameters in the modeling parameters.
2. A starting vibration analysis method based on a reduced-order model according to claim 1, characterized in that: Based on the starting condition, the model parameters of the reduced-order model are obtained, and a sample database of the model parameters is constructed, which also includes: Design tests based on starting conditions and process test data; The model parameters related to the starting vibration in the processed test data are obtained, and a sample database of the model parameters is constructed.
3. A starting vibration analysis method based on a reduced-order model according to claim 2, characterized in that: Design the test based on the starting conditions and process the test data, including: The test includes simulation analysis and calculation of the whole vehicle under starting conditions and actual measurement of the whole vehicle; The model parameters obtained include simulation analysis data of the vehicle simulation analysis calculation model under starting conditions and vehicle measured data, a ratio of the simulation analysis data to the vehicle measured data is preset, and a sample database of model parameters is constructed according to the ratio.
4. A starting vibration analysis method based on a reduced-order model according to claim 3, characterized in that: The model parameters include input parameters and output parameters, wherein the input parameters include starter torque peak, starter torque slope, engine cylinder pressure, powertrain suspension stiffness, powertrain suspension position and shock absorber stiffness damping, and the output parameters include seat rail vibration acceleration.
5. The starting vibration analysis method based on reduced-order model according to claim 4 is characterized in that: Also includes: Under the vehicle starting condition, obtain the seat rail vibration acceleration value; Filter the seat rail vibration acceleration value; The seat rail vibration acceleration value obtained is the average of the square of the peak and trough differences in the X, Y, and Z axis directions in the vehicle coordinate system.
6. A starting vibration analysis method based on a reduced-order model according to claim 5, characterized in that: Establish a reduced-order model, train the reduced-order model through modeling parameters, and verify and analyze the trained reduced-order model through verification parameters to obtain estimated output parameters, including: Using the modeling parameters, the reduced-order model is trained by calling a Kriging algorithm and / or a neural network regression model; The verification parameters are analyzed in the trained reduced-order model to obtain the estimated seat rail vibration acceleration values. The estimated vibration values are compared with the actual starting vibration values in the sample database to evaluate the quality of the reduced-order model.
7. A starting vibration analysis method based on a reduced-order model according to claim 6, characterized in that: Compare the estimated vibration values with the actual starting vibration values to evaluate the quality of the reduced order model, including: The average normalized error calculation and Pearson coefficient are used to evaluate whether the quality of the reduced-order model meets the qualification standard; The calculation formula for the average normalized error is: Where: MNE is the mean normalized error, f(x i ) is the estimated value of the reduced-order model, y i is the original actual value, y max and min are the maximum and minimum values of the model output parameters respectively; The calculation formula of Pearson coefficient is: Where: R 2 is the Pearson coefficient, RSS is the sum of squares of deviations between the true value and the estimated value caused by the error, and TSS is the total variance.
8. The starting vibration analysis method based on reduced-order model according to claim 7, characterized in that: include: Set the quality qualification standard of the reduced-order model, set the first threshold of MNE according to the qualification standard, and R 2 The second threshold value of The MNE value is not greater than the first threshold, and R 2 When it is not less than the second threshold, the quality of the reduced-order model meets the qualification standard; Select the lower MNE among the reduced-order models that meet the quality criteria while maintaining a high R 2 The reduced-order model of the value is used as a large data reduced-order model.
9. A starting vibration analysis method based on a reduced-order model according to claim 8, characterized in that: include: Setting a third threshold of the seat rail vibration acceleration value and recording it as A; Export the big data reduced-order model into an m file recognizable by MATLAB, and set the vibration acceleration value of the seat rail to ≤A; The lower deviation of the suspension and shock absorber stiffness is taken as the lower boundary of stiffness, and the upper deviation of the suspension and shock absorber stiffness is taken as the upper boundary of stiffness; According to the actual project powertrain suspension position parameters and suspension stiffness boundaries, the suspension position coordinates, suspension stiffness, shock absorber stiffness and damping are obtained through the big data reduction model.
10. A starting vibration analysis device based on a reduced-order model, characterized in that: include: An acquisition module, used for acquiring model parameters of the reduced-order model based on the starting condition, and constructing a sample database of the model parameters, wherein the model parameters include input parameters and output parameters; A setting module, used to divide the model parameters in the sample database into modeling parameters for modeling and verification parameters for verification according to a preset ratio; A construction module is used to establish a reduced-order model, perform model training on the reduced-order model through modeling parameters, and perform verification analysis on the trained reduced-order model through verification parameters to obtain estimated output parameters; The evaluation module compares the estimated output parameters with the output parameters in the modeling parameters to evaluate the quality of the reduced-order model.
Citation Information
Patent Citations
Model-based low-frequency vibration optimization simulation control method
CN116305794A