CN113946908A - Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system - Google Patents
Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system Download PDFInfo
- Publication number
- CN113946908A CN113946908A CN202111137812.7A CN202111137812A CN113946908A CN 113946908 A CN113946908 A CN 113946908A CN 202111137812 A CN202111137812 A CN 202111137812A CN 113946908 A CN113946908 A CN 113946908A
- Authority
- CN
- China
- Prior art keywords
- machine learning
- model
- performance
- learning model
- auxiliary frame
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/23—Design optimisation, verification or simulation using finite element methods [FEM] or finite difference methods [FDM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
Abstract
The invention discloses a machine learning-based auxiliary frame multidisciplinary lightweight optimization method, which comprises the following steps: establishing a sub-frame parameterized model; adopting an adaptive DOE method to create test sample points for analyzing the strength, mode and dynamic stiffness performance of the auxiliary frame; establishing a machine learning model of each performance through an artificial intelligence algorithm, and verifying whether the precision of the machine learning model meets the requirement; performing subframe lightweight optimization on the machine learning model meeting the requirements; and verifying whether all performances meet the requirements. The method solves the problems of complexity and low optimization efficiency of the original automobile auxiliary frame optimization method.
Description
Technical Field
The invention belongs to the field of automobile and mechanical engineering, and particularly relates to a machine learning-based auxiliary frame multidisciplinary lightweight optimization method.
Background
Multidisciplinary optimization is an important means for optimally designing the structure of the auxiliary frame. The multidisciplinary optimization can simultaneously investigate the properties of the auxiliary frame such as strength, mode, dynamic stiffness and the like, and light weight optimization design is carried out on the premise of meeting performance requirements. When the multidisciplinary optimization calculation is carried out, an engineer is required to carry out a plurality of software operations, including a series of complicated software operations such as back auxiliary frame grid division, boundary condition loading, solving condition setting, submitting calculation, result post-processing, optimization scheme checking calculation and the like. The whole process needs a large amount of time for software setting, so that the performance development period of the rear auxiliary frame is prolonged, and the whole process needs a large amount of manual operation, so that different engineers analyze and calculate to obtain different results. There is no standardized and standardized rear subframe lightweight design process. The development of the project of the auxiliary frame is not facilitated. Meanwhile, multidisciplinary optimization is often performed by using commercial software, such as the multidisciplinary optimization software of LSOPT, and the selection of the experimental design, the proxy model creation and the optimization algorithm is limited by the built-in function of the software. In addition, the traditional proxy model algorithm needs more test sample data in order to obtain the precision meeting the engineering requirements, and particularly, when the design variables are increased, the required computing resource overhead is increased.
Disclosure of Invention
The invention aims to provide a machine learning-based auxiliary frame multidisciplinary lightweight optimization method to solve the problems of complexity and low optimization efficiency of an original automobile auxiliary frame optimization method.
The technical scheme adopted by the invention is as follows:
a machine learning-based auxiliary frame multidisciplinary lightweight optimization method comprises the following steps:
establishing a sub-frame parameterized model;
adopting an Adaptive DOE method to create test sample points for analyzing the strength, mode and dynamic stiffness performance of the auxiliary frame;
establishing a machine learning model of each performance through an artificial intelligence algorithm, and verifying whether the precision of the machine learning model meets the requirement;
performing subframe lightweight optimization on the machine learning model meeting the requirements;
and verifying whether all performances meet the requirements.
Further, establishing the subframe parameterized model comprises:
establishing an auxiliary frame strength performance analysis finite element model, inspecting a plurality of working conditions related to strength, and evaluating the stress strength performance under the condition of meeting all the working conditions;
establishing a subframe modal performance analysis finite element model, extracting a frequency characteristic value, and evaluating the first-order modal frequency performance of the subframe;
establishing an auxiliary frame dynamic stiffness analysis finite element model, arranging a plurality of mounting points on an auxiliary frame, and evaluating the dynamic stiffness performance of each mounting point;
and establishing all parameterized models of the auxiliary frame, and setting height and width parameters of all parts of the models.
Further, establishing a full parameterized model of the subframe comprises:
importing the parametric auxiliary frame strength performance analysis finite element model into ANSA software, finding a Morph parameter close to the current auxiliary frame structure in a Morph library, carrying out matching modification according to the current auxiliary frame structure, and creating a strength performance analysis parametric model;
exporting the Morph control area into a Morph parameter template according to the created intensity analysis parameterized model;
importing a parametrically-free modal performance analysis finite element model into ANSA software, importing a Morph parameter template exported in the previous step, and creating a modal performance analysis parameterized model consistent with the strength performance analysis parameterized model;
and importing a dynamic stiffness performance analysis finite element model without parameters into ANSA software, and then importing a Morph parameter template to create a dynamic stiffness performance analysis parameterized model consistent with the modal performance analysis parameterized model.
Further, the process of creating experimental sample points by the Adaptive DOE method includes:
selecting data of initial test design sample points according to the number of design variables, wherein the number of the initial sample points is calculated according to a formula N-2 x d +1, N is the number of the initial sample points, and d is the number of the design variables;
transmitting the data of the test design sample points to the created parameterized model, and calling a Morph tool to generate a finite element solving file of the calculated strength, mode and dynamic stiffness according to the parameterized model;
and the background calls a finite element solver to perform solving calculation on the generated solving file, extracts performance results of strength, mode and dynamic stiffness, and collects all the results to create a data set for the subsequent training of the machine learning model.
Further, creating a machine learning model for each discipline through an artificial intelligence algorithm includes:
and splitting the data sets of all the performance results for training of a machine learning model, selecting one part as training set data and the other part as test set data according to a proportion, and training through the machine learning model to obtain a quality response machine learning model, a stress response machine learning model, a modal response machine learning model and a dynamic stiffness machine learning model.
Further, for the condition that the precision of the machine learning model does not meet the requirement, the number of the sample points of the data set is increased again, and the training of the machine learning model is carried out again, wherein d sample points are added each time.
And further, predicting all the performances through the established machine learning model, and when the quality of the auxiliary frame is minimum, the stress intensity value of the auxiliary frame is smaller than or equal to the intensity performance target value, the modal frequency value of the auxiliary frame is larger than or equal to the modal performance target value, the dynamic stiffness value is larger than or equal to the dynamic stiffness performance target value, and the shape parameter design variable of the auxiliary frame is in an adjustable range, completing the optimization of the multidisciplinary performance of the auxiliary frame.
Further, the plurality of conditions associated with the intensity is considered to include 11 conditions, which are: the method comprises the following steps of falling working condition, steering working condition, braking working condition, rear braking working condition, wheel side direction road tooth collision working condition, wheel forward impact working condition, wheel reverse impact working condition, tunnel braking working condition, vertical working condition, right turning braking working condition and acceleration working condition.
Further, the sub vehicle frame dynamic stiffness analysis model's test mounting point includes: 3 motor mounting points, 4 left control arm mounting points and 4 right control arm mounting points; the parameters covered by the full parameterized model of the subframe include: the height or width of the front cross beam, the rear cross beam and the longitudinal beam.
In another aspect of the present invention, there is also provided a machine learning-based auxiliary frame multidisciplinary lightweight optimization system, including:
the parametric model building module is used for analyzing the strength performance, the modal performance and the dynamic stiffness performance of the auxiliary frame and building an auxiliary frame parametric model;
the system comprises a sample point establishing module, a dynamic stiffness analysis module and a dynamic stiffness analysis module, wherein the sample point establishing module adopts an Adaptive DOE method to establish test sample points for analyzing the strength, the mode and the dynamic stiffness performance of an auxiliary frame;
the machine learning model establishing module is used for establishing a machine learning model of each subject through an artificial intelligence algorithm and verifying whether the precision of the machine learning model meets the requirement or not;
and the model precision verification module is used for carrying out auxiliary frame lightweight optimization on the machine learning model meeting the requirements and verifying whether all performances meet the requirements.
Compared with the prior art, the auxiliary frame multidisciplinary lightweight optimization method and system based on machine learning provided by the invention achieve the following technical effects;
1. firstly, establishing a test sample point for analyzing the strength, the mode and the dynamic stiffness performance of the auxiliary frame by an Adaptive DOE method; and then carrying out model training on the obtained test sample points by a machine learning method, and then verifying the model precision of machine learning. And if the precision DOEs not meet the requirement, increasing the test sample points by an Adaptive DOE method, and then performing machine learning model training by using the increased data set until a model meeting the engineering precision requirement is obtained. And after a machine learning model of each performance is obtained, performing multidisciplinary lightweight process optimization, and finally obtaining the optimized auxiliary frame lightweight process design meeting all performance requirements. By the method, the problem that the design optimization result is not uniform due to the fact that the auxiliary frame is complex to operate and has no standardization, normalization and process in the multidisciplinary optimization design process is solved; meanwhile, the problem that a large number of test design sample points are needed when the number of design variables of a traditional agent model is large is solved, optimization efficiency is greatly improved, calculation time is shortened, the requirement for calculation resources is reduced, and the problems that the development cycle of multidisciplinary optimization design of the auxiliary vehicle is long, the optimization effect is inconsistent and the like are solved.
2. According to the method, the whole auxiliary frame multidisciplinary optimization process is integrated through a self-programming method, the whole optimization process does not need to be operated by using commercial software, all analysis pre-processing, post-processing, calculation solving, optimization and the like are automatically completed through self-programming, and the problems that various commercial software is used, the operation is complex, and the standard and the specification are absent are solved.
Drawings
Fig. 1 is a schematic flow diagram of a sub-frame multidisciplinary lightweight optimization method based on machine learning according to an embodiment of the present invention.
FIG. 2 is a schematic view of a rear subframe mounting point according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of a fully parameterized model of the rear subframe according to an embodiment of the invention.
FIG. 4 is a schematic illustration of a DOE sample point distribution according to an embodiment of the present invention.
FIG. 5 is a graph comparing predicted values and true values of a quality machine learning model according to an embodiment of the present invention.
FIG. 6 is a graph comparing predicted values and actual values of a stress machine learning model according to an embodiment of the present invention.
FIG. 7 is a graph comparing a predicted value and a true value of a modal machine learning model, in accordance with an embodiment of the present invention.
FIG. 8 is a graph comparing predicted values and actual values of a dynamic stiffness machine learning model according to an embodiment of the invention.
FIG. 9 is a distribution diagram of the Adaptive DOE method of embodiments of the present invention after adding sample points.
FIG. 10 is a graph comparing predicted values and true values of an updated dynamic stiffness machine learning model according to an embodiment of the present invention.
FIG. 11 is a graph of the design objective optimization iteration results of an embodiment of the present invention.
Fig. 12 is an architecture diagram of a machine learning based sub-frame multidisciplinary lightweight optimization system according to an embodiment of the present invention.
Detailed Description
The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby. As certain terms are used throughout the description and claims to refer to particular components. As one skilled in the art will appreciate, manufacturers may refer to a component by different names. The present specification and claims do not intend to distinguish between components that differ in name but not function. The following description is of the preferred embodiment for carrying out the invention, and is made for the purpose of illustrating the general principles of the invention and not for the purpose of limiting the scope of the invention. The scope of the present invention is defined by the appended claims.
The invention is described in further detail below with reference to the figures and specific embodiments.
Referring to fig. 1, an embodiment of the invention discloses a machine learning-based auxiliary frame multidisciplinary lightweight optimization method, which comprises the following steps:
step one, establishing an auxiliary frame parameterized model;
secondly, establishing a test sample point for analyzing the strength, the mode and the dynamic stiffness performance of the auxiliary frame by adopting an Adaptive DOE method;
step three, establishing a machine learning model of each performance through an artificial intelligence algorithm, and verifying whether the precision of the machine learning model meets the requirement;
fourthly, performing auxiliary frame lightweight optimization on the machine learning model meeting the requirements;
and step five, verifying whether all performances meet the requirements.
The invention establishes a machine learning model with multidisciplinary performance by an Adaptive DOE method and an artificial intelligence method, and carries out multidisciplinary lightweight process design optimization based on the machine learning model. The method solves the problem of a large number of test sample points required for establishing the traditional proxy model meeting the precision requirement. The precision of the machine learning model is gradually improved through an iterative method, and finally the high-precision machine learning model is trained through a data set constructed by test sample points which are less than those required by a traditional agent model method, so that the problems of long calculation time, huge calculation resource consumption and the like of the traditional optimization method are solved.
The process flow of the present invention will be described in detail below.
Firstly, establishing a sub-frame parameterized model.
The subframe performance analysis mainly comprises strength performance, modal performance and dynamic stiffness performance, and a parameterized model needs to be established for all the performance analysis. Of course, it should be understood by those skilled in the art that the analysis of subframe properties is not limited to the three types of performance parameters listed above.
The method for establishing the auxiliary frame parameterized model comprises the following steps:
1.1, establishing an auxiliary frame strength performance analysis finite element model, inspecting a plurality of working conditions related to strength, and evaluating the stress strength performance under the condition of meeting all the working conditions.
Subframe strength is the ability to meet the requirement that the vehicle not fail during its life. The subframe strength analysis is preferably performed using ABAQUS software or ANSYS software using inertial release, and the strength analysis load is extracted from the subframe multibody dynamics model using Adams software (a mechanical dynamics simulation analysis software).
The intensity analysis mainly examines 11 conditions, including: the method comprises the following steps of falling working conditions, steering working conditions, braking working conditions, rear braking working conditions, lateral road tooth collision working conditions of wheels, forward impact working conditions of the wheels, reverse impact working conditions of the wheels, tunnel braking working conditions, vertical working conditions, right-turn braking working conditions and acceleration working conditions, and of course, more working conditions can be included. The subframe needs to satisfy the strength performance under the above 11 working conditions, that is, the stress strength value of the subframe under all working conditions does not exceed the yield strength of the rear subframe material, and the stress strength analysis result of the subframe needs to satisfy all performance requirements.
And 1.2, establishing a subframe modal performance analysis finite element model, extracting frequency, and evaluating the first-order modal frequency performance of the subframe.
The modal performance of sub vehicle frame has influenced the performance and the quality of whole car, and good sub vehicle frame modal performance can avoid resonating in whole car structure, reduces the noise, improves the NVH performance of whole car. The secondary vehicle modal analysis is carried out by using Nastran software, the frequency calculation is carried out by using a Lanczos method, the Lanczos algorithm is an algorithm for converting a symmetric matrix into a symmetric tri-diagonal matrix through orthogonal similarity transformation, specifically, the natural frequency of the structure can be obtained after squaring through a solution obtained after characteristic value decomposition of a structural vibration equation, and the extraction range is 0-300 Hz. And analyzing by Nastran software to obtain a first-order modal shape of the rear subframe, wherein the first-order modal frequency of the rear subframe meets the performance requirement.
And 1.3, establishing an auxiliary frame dynamic stiffness analysis finite element model, arranging a plurality of mounting points on the auxiliary frame, and evaluating the dynamic stiffness performance of each mounting point.
The dynamic stiffness of sub vehicle frame has influenced whole car and has resisted the ability of warping under dynamic disturbance, and in this embodiment, at 11 mounting points of sub vehicle frame design, motor mounting point A is 3 (mounting point 1 before the motor, 2 back mounting points), and left control arm mounting point B is 4, and right control arm mounting point C is 4. The auxiliary frame mounting point positions are shown in fig. 2, and in practice, the mounting point positions can be set to be different according to different test requirements, and the number of the mounting points can be set to be more. The dynamic stiffness of the control arm mounting point and the dynamic stiffness of the motor suspension mounting point are required to meet performance requirements in the main direction and the secondary direction respectively. For example, a local coordinate system is set for each mounting point, taking a bolt mounting point as an example, taking the axial direction of the bolt as a main direction (assumed as an X axis), and taking other directions (a Y axis and a Z axis) as secondary directions, stress requirements in different directions are different, but performance requirements in all directions should be met.
And 1.4, establishing all parameterized models of the auxiliary frame, and setting height and width parameters of each part of the models.
The full-parametric model of the rear auxiliary frame comprises the height, width and other parameters of the front cross beam, the rear cross beam, the longitudinal beam and other positions. The parameterized rear subframe model is shown in fig. 3, wherein parameter 1 is the height of the upper side of the front beam, parameter 2 is the height of the lower side of the front beam, parameter 3 is the height of the upper side of the rear beam, parameter 4 is the height of the lower side of the rear beam, parameter 5 is the width of the outer side of the longitudinal beam, parameter 6 is the width of the front side of the front beam, and parameter 7 is the width of the rear side of the front beam. The deformation parameters of each position should meet the safety requirements of the test.
The method comprises the following steps of:
1.4.1, importing the auxiliary frame strength performance analysis finite element model without parameters in the step 1.1 into ANSA software (a general CAE pretreatment software). And finally, matching according to the current auxiliary frame structure to form a strength performance parameterized analysis model. Because the parameters in the Morph database are the previously stored Morph parameters and are not necessarily consistent with the current subframe structure (shape), when the parameters are inconsistent with the current subframe structure, only the similar Morph parameters need to be found from the Morph database according to the structure (shape) of the current subframe, and the parameters can be matched with the current subframe structure (shape) to be consistent after simple modification.
1.4.2, according to the parameterized performance intensity analysis model created in the step 1.4.1, the Morph control area is exported to be a Morph parameter template.
1.4.3, importing the parametrically-free modal performance analysis model into ANSA software, and importing the Morph parameter template exported in the previous step into the ANSA software to create a modal performance parametric analysis model which is consistent with the strength performance parametric analysis model created in the previous step. Similarly, a dynamic stiffness performance analysis model without parameters is imported into ANSA software, and then a Morph parameter template is imported to create a dynamic stiffness performance parameterized analysis model.
The strength performance parameterized model, the modal performance parameterized model and the dynamic stiffness performance parameterized model created by 1.4.1, 1.4.2 and 1.4.3 have the same Morph control body, and can be subjected to parameterized control through the same parameter control file and used for generating the strength, modal and dynamic stiffness analysis models during multidisciplinary lightweight optimization. All design constraints are conveniently acquired through modeling of the parameterized model, so that the optimal subframe structure design is obtained.
Secondly, a test sample point is created by an Adaptive DOE method.
Firstly, selecting the data size of initial test design sample points according to the number of design variables, wherein the number of the initial sample points is calculated according to a formula N-2 x d +1, wherein N is the number of the initial sample points, and d is the number of the design variables. The sample points are shown in FIG. 4. DOE (design of experiment) is a mathematical statistical method for arranging experiments and analyzing experimental data; the test design is mainly used for reasonably arranging the tests, and obtaining ideal test results and scientific conclusions with smaller test scale (test times), shorter test period and lower test cost.
And then, transmitting the data of the test design sample points to a control file of the parameterized model created in the step one by a python writing program, and calling a Morph tool by a background to generate a finite element solving file of strength, mode and dynamic stiffness for calculation according to the control file of the parameterized model.
And finally, calling finite element solvers such as abaqus, nanostran and the like through the python background to solve and calculate the generated solution file. And after the calculation is finished, the results of the strength, the mode and the dynamic stiffness are automatically extracted through python, and all the results are collected to create a data set for the subsequent training of the machine learning model.
And thirdly, establishing a machine learning model of each performance through an artificial intelligence algorithm.
And D, splitting the data sets of all the performance results obtained in the step two, and using the data sets for training a machine learning model. All data sets are selected according to a certain proportion, part of the data sets are test sets, part of the data sets are training sets, for example, the data sets are divided into the test sets and the training sets according to the proportion of 1:4, the training set data are used for training a machine learning model, the test set data are used for verifying the precision of the machine learning model, the precision of the model is ensured to reach a preset threshold, for example, the threshold of the precision is set to be 98%, and the requirement is met when the threshold of the precision is greater than or equal to 98%. And training through a machine learning model to obtain a quality response machine learning model, a stress response machine learning model, a modal response machine learning model and a dynamic stiffness machine learning model.
The quality response machine learning model is a model obtained by machine learning training from the quality response of any one of the three models obtained above. After the shape of the model is changed, the stress, the mode and the dynamic stiffness are changed, and meanwhile, the mass is changed. The boundary conditions and solution settings of the three models are different for calculating the strength, the mode and the dynamic stiffness respectively, but the quality of the models is consistent.
The results of the quality machine learning model accuracy verification are shown in fig. 5. Where the solid line represents the true quality value and the dashed line represents the predicted value of the machine learning model. As can be seen from the following Table 1, the maximum error of the predicted value of the quality machine learning model is 0.09%, and the accuracy requirement is met.
TABLE 1 quality machine learning model verification statistics
| Quality verification item | True value | Machine learning prediction | Error of the measurement |
| 1 | 0.02118 | 0.02118 | 0.00％ |
| 2 | 0.02132 | 0.02132 | 0.00％ |
| 3 | 0.02105 | 0.02105 | 0.00％ |
| 4 | 0.0214 | 0.02142 | -0.09％ |
| 5 | 0.0216 | 0.0216 | 0.00％ |
| 6 | 0.02092 | 0.02093 | -0.05％ |
| 7 | 0.02102 | 0.02103 | -0.05％ |
| 8 | 0.0215 | 0.0215 | 0.00％ |
| 9 | 0.02087 | 0.02087 | 0.00％ |
| 10 | 0.02044 | 0.02043 | 0.05％ |
| 11 | 0.02147 | 0.02148 | -0.05％ |
The results of the stress machine learning model accuracy verification are shown in fig. 6. The solid line represents the true stress value, and the data plotted in the dotted line represents the predicted stress value by machine learning. From table 2, it can be found that the maximum error of the predicted value of the stress machine learning model is 1.14%, and the accuracy requirement is met.
TABLE 2 statistics of stress machine learning model validation results
| Stress verification item | True value | Machine learning prediction | Error of the measurement |
| 1 | 150.21 | 150.21 | 0.00％ |
| 2 | 154.26 | 154.13 | 0.08％ |
| 3 | 155.38 | 155.52 | -0.09％ |
| 4 | 150.28 | 152 | -1.14％ |
| 5 | 146.84 | 147.49 | -0.44％ |
| 6 | 159.48 | 158.93 | 0.34％ |
| 7 | 162.47 | 162.38 | 0.06％ |
| 8 | 141.55 | 141.26 | 0.20％ |
| 9 | 167.12 | 167.17 | -0.03％ |
| 10 | 171.46 | 170.73 | 0.43％ |
| 11 | 144.36 | 144.29 | 0.05％ |
The results of the modal machine learning model accuracy verification are shown in fig. 7. The solid line is the true modal frequency value, and the data plotted by the dotted line is the modal frequency value predicted by machine learning. From table 3, it can be found that the maximum error of the prediction value of the modal machine learning model is 0.21%, and the accuracy requirement is met.
TABLE 3 statistics of modal machine learning model validation results
| Modality validation item | True value | Machine learning prediction | Error of the measurement |
| 1 | 146.2 | 146 | 0.14％ |
| 2 | 143.6 | 143.5 | 0.07％ |
| 3 | 143.5 | 143.3 | 0.14％ |
| 4 | 147 | 147.2 | -0.14％ |
| 5 | 146.4 | 146.3 | 0.07％ |
| 6 | 143.8 | 144 | -0.14％ |
| 7 | 141 | 140.7 | 0.21％ |
| 8 | 149.1 | 149.1 | 0.00％ |
| 9 | 140.8 | 140.9 | -0.07％ |
| 10 | 137.8 | 138 | -0.15％ |
| 11 | 147.5 | 147.6 | -0.07％ |
The result of the accuracy verification of the dynamic stiffness machine learning model is shown in fig. 8. Wherein, the solid line represents the real dynamic stiffness value, and the data plotted by the dotted line is the machine learning prediction dynamic stiffness value. From table 4, it can be found that the maximum error of the predicted value of the dynamic stiffness machine learning model is 2.07%, which exceeds the accuracy error range of 2%. And the higher the precision of the machine learning model is, the more accurate the predicted result is, and the accuracy of the subsequent real performance verification result is ensured. The precision of the machine learning model is required to be more than or equal to 98 percent. Therefore, the accuracy of the dynamic stiffness machine learning model does not meet the requirement.
TABLE 4 dynamic stiffness machine learning model verification statistics
When the accuracy of a certain machine learning model does not meet the requirement, the number of sample points of the data set needs to be increased to train the machine learning model again so as to improve the accuracy of the machine learning model. A group of dynamic stiffness calculation sample points are added by using an Adaptive DOE method, the principle of increasing the number of the sample points is to increase d (design variable number) sample points each time, and the training of the machine learning model is performed again after the sample points are increased. The increased sample points are shown in FIG. 9. By the aid of the Adaptive DOE method, only a small number of sample points need to be added each time to gradually improve the accuracy of the machine learning model.
After the sample point data is added, the training of the dynamic stiffness machine learning model is performed again, and the result of the updated accuracy verification of the dynamic stiffness machine learning model is shown in fig. 10. The solid line is the true dynamic stiffness value, and the data plotted by the dotted line is the updated dynamic stiffness value predicted by machine learning. From table 5, it can be found that the maximum error of the updated dynamic stiffness machine learning model prediction value is reduced from 2.07% to 1.29%, and the accuracy requirement is met.
TABLE 5 updated dynamic stiffness machine learning model validation statistics
| Dynamic stiffness verification term | True value | Machine learning prediction | Error of the measurement |
| 1 | 62450.6 | 62914.8 | -0.74％ |
| 2 | 65994.2 | 66116.3 | -0.19％ |
| 3 | 60091.6 | 60898.9 | -1.34％ |
| 4 | 66370.7 | 65655.4 | 1.08％ |
| 5 | 64637 | 64609.2 | 0.04％ |
| 6 | 69109.8 | 70000 | -1.29％ |
| 7 | 64361 | 64487.1 | -0.20％ |
| 8 | 68193.9 | 67908.3 | 0.42％ |
| 9 | 64772.1 | 64433.9 | 0.52％ |
| 10 | 59755.2 | 59704.4 | 0.09％ |
| 11 | 65921.5 | 65861.4 | 0.09％ |
Multidisciplinary lightweight optimization based on machine learning model
The subframe multidisciplinary lightweight optimization problem is mathematically described in the following equation (1).
Wherein, in the formula: m-rear subframe mass, Stress-rear subframe Stress intensity, Freq-rear subframe modal frequency, Si-rear subframe shape parameter design variable, in mm, corresponding to the position deformation quantities, S, of parameters 1, 2, 3, 4, 6, 7 of figure 65And (3) representing the deformation range of the outer side width of the longitudinal beam of the parameter 5, and taking the point with the minimum sub-frame mass as the optimal point when a plurality of points meet the strength, mode and dynamic stiffness performance indexes.
All the performance responses are predicted through the machine learning model in the third step, the genetic algorithm is selected by the optimization algorithm, the whole optimization process is completed through python self-programming, and the result of the designed target optimization iteration process is shown in figure 11. And designing target optimization iteration, wherein the result obtained by each iteration is used as the initial value of the next iteration, so that the closest target result can be obtained.
The method uses the machine learning method for the multidisciplinary lightweight process optimization of the auxiliary frame, improves the model precision, reduces the calculation consumption and obviously shortens the development cycle of the auxiliary frame.
And fifthly, verifying whether all the performances meet the required performance requirements.
And performing verification analysis on all performances of the obtained optimal solution, including the performance verification analysis on the strength, the mode, the dynamic stiffness and the like of the auxiliary frame. The verification result meets all performance requirements, and finally the auxiliary frame light-weight process scheme meeting all performance requirements is obtained.
According to the method, the auxiliary vehicle multidisciplinary lightweight process optimization is carried out through Adaptive DOE and a machine learning method, and a traditional agent model method is replaced. The number of sample points required to obtain a model of the same accuracy is significantly reduced compared to conventional methods. Thereby reducing the demand for computing resources and shortening the development cycle. Meanwhile, the method is completely finished based on a python self-programming method, and a complex and complicated software operation process required by using commercial software in the traditional multidisciplinary optimization is not needed. Further shortening the development period of the whole auxiliary frame. The whole process is completed through python self-programming, complex software operation is not needed, and the optimization efficiency and the optimization effect are higher than those of commercial software, so that the optimized lightweight process scheme meeting all performance requirements is finally obtained.
Referring to fig. 12, in correspondence with the method in the foregoing embodiment, another embodiment of the present invention further provides a machine learning-based subframe multidisciplinary lightweight optimization system, including:
the parametric model building module is used for analyzing the strength performance, the modal performance and the dynamic stiffness performance of the auxiliary frame and building an auxiliary frame parametric model;
the system comprises a sample point establishing module, a dynamic stiffness analysis module and a dynamic stiffness analysis module, wherein the sample point establishing module adopts an Adaptive DOE method to establish test sample points for analyzing the strength, the mode and the dynamic stiffness performance of an auxiliary frame;
the machine learning model establishing module is used for establishing a machine learning model of each subject through an artificial intelligence algorithm and verifying whether the precision of the machine learning model meets the requirement or not;
and the model precision verification module is used for carrying out auxiliary frame lightweight optimization on the machine learning model meeting the requirements and verifying whether all performances meet the requirements.
Each functional module in this embodiment is used to execute the method in the previous embodiment, which has the technical effect achieved by the previous embodiment, and therefore, the details are not described herein.
It should be noted that the above-mentioned embodiments are only preferred embodiments of the present invention, and the scope of the present invention is not limited thereby, and the present invention may be modified in materials and structures, or replaced with technical equivalents, in the constructions of the above-mentioned various components. Therefore, structural equivalents made by using the description and drawings of the present invention or by directly or indirectly applying to other related arts are also encompassed within the scope of the present invention.
Claims (10)
1. A machine learning-based auxiliary frame multidisciplinary lightweight optimization method is characterized by comprising the following steps:
establishing a sub-frame parameterized model;
adopting an Adaptive DOE method to create test sample points for analyzing the strength, mode and dynamic stiffness performance of the auxiliary frame;
establishing a machine learning model of each performance through an artificial intelligence algorithm, and verifying whether the precision of the machine learning model meets the requirement;
performing subframe lightweight optimization based on the machine learning model meeting the requirements;
and verifying whether all performances meet the requirements.
2. The optimization method of claim 1, wherein building a sub-frame parameterized model comprises:
establishing an auxiliary frame strength performance analysis finite element model, inspecting a plurality of working conditions related to strength, and evaluating the stress strength performance under the condition of meeting all the working conditions;
establishing a subframe modal performance analysis finite element model, extracting a frequency characteristic value, and evaluating the first-order modal frequency performance of the subframe;
establishing an auxiliary frame dynamic stiffness analysis finite element model, wherein the auxiliary frame is provided with a plurality of mounting points, and evaluating the dynamic stiffness performance of each mounting point;
and establishing all parameterized models of the auxiliary frame, and setting height and width parameters of all parts of the models.
3. The optimization method of claim 2, wherein establishing the full parameterized model of the subframe comprises:
importing the parametric auxiliary frame strength performance analysis finite element model into ANSA software, finding a Morph parameter close to the current auxiliary frame structure in a Morph library, carrying out matching modification according to the current auxiliary frame structure, and creating a strength performance analysis parametric model;
exporting the Morph control area into a Morph parameter template according to the created intensity analysis parameterized model;
importing a parametrically-free modal performance analysis finite element model into ANSA software, importing a Morph parameter template exported in the previous step, and creating a modal performance analysis parameterized model consistent with the strength performance analysis parameterized model;
and importing a dynamic stiffness performance analysis finite element model without parameters into ANSA software, and then importing a Morph parameter template to create a dynamic stiffness performance analysis parameterized model consistent with the modal performance analysis parameterized model.
4. The optimization method of claim 2 or 3, wherein the process of creating experimental sample points by the Adaptive DOE method comprises:
selecting data of initial test design sample points according to the number of design variables, wherein the number of the initial sample points is calculated according to a formula N-2 x d +1, N is the number of the initial sample points, and d is the number of the design variables;
transmitting the data of the test design sample points to the created parameterized model, and calling a Morph tool to generate a finite element solving file of the calculated strength, mode and dynamic stiffness according to the parameterized model;
and the background calls a finite element solver to perform solving calculation on the generated solving file, extracts performance results of strength, mode and dynamic stiffness, and collects all the results to create a data set for the subsequent training of the machine learning model.
5. The optimization method of claim 4, wherein creating the machine learning model for each discipline through an artificial intelligence algorithm comprises:
and splitting the data sets of all the performance results for training of a machine learning model, selecting one part as training set data and the other part as test set data according to a proportion, and training through the machine learning model to obtain a quality response machine learning model, a stress response machine learning model, a modal response machine learning model and a dynamic stiffness machine learning model.
6. The optimization method of claim 5, wherein for those for which the accuracy of the machine learning model is not satisfactory, the number of sample points of the data set is increased again and the training of the machine learning model is resumed, wherein d sample points are added each time.
7. The optimization method of claim 6, wherein all the properties are predicted by the established machine learning model, and when the subframe mass is minimum, the subframe stress intensity value is less than or equal to the intensity performance target value, the subframe modal frequency value is greater than or equal to the modal performance target value, the dynamic stiffness value is greater than or equal to the dynamic stiffness performance target value, and the subframe shape parameter design variable is within the adjustable range, the optimization of the subframe multidisciplinary performance is completed.
8. The optimization method according to claim 2, wherein the plurality of conditions associated with the intensity is considered to include 11 conditions, which are: the method comprises the following steps of falling working condition, steering working condition, braking working condition, rear braking working condition, wheel side direction road tooth collision working condition, wheel forward impact working condition, wheel reverse impact working condition, tunnel braking working condition, vertical working condition, right turning braking working condition and acceleration working condition.
9. The optimization method according to claim 2 or 7, wherein the test mounting points of the subframe dynamic stiffness analysis model comprise: 3 motor mounting points, 4 left control arm mounting points and 4 right control arm mounting points; the parameters covered by the full parameterized model of the subframe include: the height or width of the front cross beam, the rear cross beam and the longitudinal beam.
10. A sub frame multidisciplinary lightweight optimization system based on machine learning, the system comprising:
the parametric model building module is used for analyzing the strength performance, the modal performance and the dynamic stiffness performance of the auxiliary frame and building an auxiliary frame parametric model;
the system comprises a sample point establishing module, a dynamic stiffness analysis module and a dynamic stiffness analysis module, wherein the sample point establishing module adopts an Adaptive DOE method to establish test sample points for analyzing the strength, the mode and the dynamic stiffness performance of an auxiliary frame;
the machine learning model establishing module is used for establishing a machine learning model of each subject through an artificial intelligence algorithm and verifying whether the precision of the machine learning model meets the requirement or not;
and the model precision verification module is used for carrying out auxiliary frame lightweight optimization on the machine learning model meeting the requirements and verifying whether all performances meet the requirements.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111137812.7A CN113946908A (en) | 2021-09-27 | 2021-09-27 | Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111137812.7A CN113946908A (en) | 2021-09-27 | 2021-09-27 | Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113946908A true CN113946908A (en) | 2022-01-18 |
Family
ID=79329408
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111137812.7A Pending CN113946908A (en) | 2021-09-27 | 2021-09-27 | Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113946908A (en) |
-
2021
- 2021-09-27 CN CN202111137812.7A patent/CN113946908A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115221602A (en) | Vehicle body design method and device based on multi-working-condition topological optimization and storage medium | |
| CN113408055B (en) | Automobile frame structure optimization method | |
| CN107169211A (en) | Automobile body-in-white early stage concept development Topology Optimization Method | |
| US20200394278A1 (en) | Hybrid Finite Element and Artificial Neural Network Method and System for Safety Optimization of Vehicles | |
| CN112287551B (en) | Driving performance system level index decomposition method based on whole vehicle conceptual model | |
| CN116306156B (en) | Vehicle body optimization method and device, storage medium and electronic equipment | |
| CN109255141B (en) | Optimization method for cross section shape of forward conceptual design of automobile body | |
| CN112417586A (en) | Body-in-white optimization processing method, device and system for vehicle and storage medium | |
| US9098673B2 (en) | Structural optimization for vehicle crashworthiness | |
| CN103177165A (en) | Coach body structure design system, coach side overturning simulation test system and method | |
| CN110532701B (en) | Vehicle body sensitivity analysis method based on platformized white vehicle body | |
| CN113946908A (en) | Machine learning-based auxiliary frame multidisciplinary lightweight optimization method and system | |
| CN113158322A (en) | Rim impact analysis method and device and readable storage medium | |
| CN113065186B (en) | Load loading method, device, equipment and storage medium | |
| CN111291446A (en) | Suspension system multidisciplinary optimization design method based on front suspension flutter and idle speed vibration | |
| CN115795678A (en) | Parameter optimization method and storage medium for conceptual design of vehicle body structure | |
| CN111506960B (en) | CAE-based truck front axle lightweight design method | |
| CN113919080A (en) | Machine learning-based quick evaluation method for mechanical property of automobile engine hood | |
| CN114818414A (en) | Simulation method, device, equipment and medium of transverse stabilizer bar | |
| CN114239149A (en) | Commercial vehicle cab body-in-white light weight method based on collision performance optimization | |
| CN114186335B (en) | Optimal design method for automobile frame | |
| CN111597630B (en) | Joint selection method, device, equipment and storage medium | |
| Sharma | CAE driven multi disciplinary optimization of vehicle systems and sub systems | |
| CN114595511A (en) | Topological optimization method for collision force transmission path of automobile front deformable barrier | |
| CN114626143A (en) | Automobile collision analysis optimization method, electronic device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |