CN115130218A - Self-learning prediction method for road noise of automotive suspension structure - Google Patents
Self-learning prediction method for road noise of automotive suspension structure Download PDFInfo
- Publication number
- CN115130218A CN115130218A CN202210755132.XA CN202210755132A CN115130218A CN 115130218 A CN115130218 A CN 115130218A CN 202210755132 A CN202210755132 A CN 202210755132A CN 115130218 A CN115130218 A CN 115130218A
- Authority
- CN
- China
- Prior art keywords
- driving model
- data
- model
- road noise
- suspension
- 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]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Aviation & Aerospace Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Vehicle Body Suspensions (AREA)
Abstract
The invention belongs to the technical field of vehicles, and particularly relates to a self-learning prediction method for road noise of an automotive suspension structure. The invention provides a self-learning prediction method for road noise of an automobile suspension structure, and aims to enhance the robustness and interpretability of the prediction method for the road noise of an automobile. The method is based on a knowledge driving model facing the suspension road noise and a data driving model built by using the LSTM, integrates and iterates knowledge and data, aims to enhance the robustness and the interpretability of the automobile road noise prediction method, and can realize self-learning of the knowledge driving model and the data driving model along with the expansion of data samples, thereby providing effective guidance for automobile NVH engineers and improving the efficiency of automobile road noise prediction analysis.
Description
Technical Field
The invention relates to a self-learning prediction method for road noise of an automobile suspension structure, in particular to a self-learning prediction method for road noise of an automobile suspension structure.
Background
In recent years, the rapid development of the domestic automobile industry not only improves the life rhythm and quality of people, but also drives the development of social productivity. Nowadays, each automobile company is increasingly competitive, and the requirements of automobiles are increased. In order to improve competitiveness, the most important is the whole vehicle performance of the automobile besides the appearance shape attracting eyes. The performance of the whole vehicle relates to each professional department, one department needs to lead the head and the full authority organization and the responsibility, and other professional departments are matched and completed together.
The performance of the whole automobile is also called NVH performance, namely Noise, Vibration and Harshness (Noise, Vibration, Harshness).
The NVH performance of the automobile is always an important evaluation index of a passenger car in the development process, and is also one of the aspects of the most concern of passengers on the driving experience. With the popularization of new energy automobiles, the contribution of road noise to the overall automobile noise is increasingly prominent because the engine noise is not shielded any more. Road noise is the noise response generated by the coaction of the vehicle body panel and the interior acoustic cavity, as road excitation is attenuated by rubber components such as tires and suspension bushings, and vibration is transmitted to the vehicle body. For a normally running automobile, low-frequency noise below 300Hz mainly occurs when the automobile runs at a medium speed, and the running condition is the most common and typical condition in urban roads. When the vehicle runs under such conditions, road noise is the main noise source in the vehicle, and therefore, the analysis and prediction of the whole vehicle road noise is a major concern in the NVH design of the vehicle.
The traditional method for analyzing the noise of the whole automobile comprises a transmission path analysis method, a finite element modeling simulation analysis method, a dynamic modeling method and the like. In the related research data of the automobile road noise disclosed at present, the invention patent of Lu-hong-Wei and Zhu-ya-Wei, etc. "a structural road noise transfer function test method" reduces the influence of strong coherent excitation input and background noise on the structural noise transfer function (CN110243609B) by applying periodic pulse excitation to each wheel in sequence and windowing the measured signal. Wangteng, Lijunming and the like establish a mathematical model of the noise in the vehicle generated by random excitation of the road surface unevenness, and analyze the influence of suspension and tire parameters on the noise in the vehicle based on the model (automobile engineering, 2000(02):93-96+ 142). The research mainly performs mathematical analysis on the vibration mechanism, and the deviation of the calculation result and the measured value is larger because the automobile is simplified into a linear system and many equivalents and simplifications are involved. The method comprises the steps of respectively carrying out subjective evaluation and objective parameter calculation on in-vehicle noise signals under the constant-speed working condition by using yellow sea waves, yellow crow and the like, carrying out correlation analysis on main and objective evaluation results, and establishing an in-vehicle noise sound quality prediction model on the basis of an Adaboost algorithm and combining a BP neural network, an ELM and an SVM, wherein the prediction accuracy and precision of the model are higher than those of GA-BP, GA-ELM and GA-SVM sound quality prediction models (automobile engineering, 2016,38(09): 1120-. The invention is different from the papers and patents in that a whole vehicle dynamics analysis model facing suspension road noise analysis is established, a data driving model based on data driving is established, knowledge and data are fused and iterated, and road noise prediction analysis efficiency is improved while interpretability and robustness of road noise prediction are enhanced.
Disclosure of Invention
Aiming at the problems in the background art, the invention provides a self-learning prediction method of the road noise of an automobile suspension structure, and aims to enhance the robustness and the interpretability of the prediction method of the road noise of an automobile. The method is based on a whole vehicle dynamics analysis model facing the suspension road noise and a data driving model built by using the LSTM, integrates and iterates knowledge and data, aims to enhance the robustness and the interpretability of the automobile road noise prediction method, and can realize self-learning of the knowledge driving model and the data driving model along with the expansion of data samples, thereby providing effective guidance for automobile NVH engineers and improving the efficiency of automobile road noise prediction analysis.
The invention solves the technical problems and provides the following technical scheme:
a self-learning prediction method for road noise of an automobile suspension structure comprises the following steps:
step 1: analyzing a structural road noise transmission path from road surface excitation to the right ear noise of a driver in the automobile corresponding to the suspension according to different structural characteristics of different suspensions of the automobile, and establishing a multi-level decomposition tree of the suspension structural road noise according to a hierarchical standard of 'whole automobile level-system level-subsystem level-part level';
step 2: embedding an intelligent algorithm in the multilevel decomposition tree, and learning the data relationship between target nodes in the multilevel decomposition tree by using the intelligent algorithm to obtain a data driving model;
and step 3: establishing a whole vehicle dynamics analysis model from road excitation to a suspension vehicle body attachment point according to the existing knowledge, and then obtaining a transfer relation from suspension vehicle body attachment point vibration to vehicle interior road noise based on test analysis, thereby constructing a knowledge driving model from the road excitation to the vehicle interior driver right ear noise;
and 4, step 4: based on the established data driving model and knowledge driving model, optimizing the built-in hyper-parameters of the data driving model and the set parameters of the knowledge driving model by using a multi-objective optimization algorithm and taking the evaluation indexes of the two models and the mean square error of the prediction result between the two models as targets, and finally multiplying the prediction results of the two models by the respective set weights and summing the results, wherein the summed result is taken as the final prediction result.
After the technical scheme is adopted, a whole vehicle dynamics analysis model facing to suspension road noise analysis is established, a data driving model based on data driving is established, knowledge and data are fused and iterated, and road noise prediction interpretability and robustness are enhanced while road noise prediction analysis efficiency is improved.
Preferably, the method further comprises a step 5 for following the step 4: and establishing a MySQL database according to test and simulation data by taking a multi-level decomposition tree of the suspension structure road noise and a complete vehicle dynamics analysis model as a data relation basis, and connecting the database with a data driving model and a knowledge driving model in real time.
After the optimal scheme is adopted, after the database is updated, the knowledge driving model and the data driving model can automatically learn new sample data, and the generalization capability of the knowledge driving model and the data driving model is improved.
Preferably, the step 2 comprises the following steps:
the method comprises the steps of sorting and normalizing test and simulation data according to a multi-level decomposition tree of the suspension structure road noise, and then dividing the test and simulation data into a training set, a test set and a verification set according to a certain proportion;
introducing a multi-layer LSTM algorithm and a Dropout method between target nodes between an upper layer and a lower layer in a multi-layer decomposition tree of the road noise of the suspension structure to construct a data driving model, and training the data driving model based on a training set, wherein the LSTM structural formula is as follows:
wherein f is a forgetting gate, g is a memory cell, i is an input gate, o is an output gate, σ is a sigmoid function, and x t For the input of this moment, h t-1 Is a hidden state of the output at the previous moment,
is x t The corresponding weight of the weight is set to be,
is h t-1 Corresponding weight, b (f) F is the corresponding offset;
performing model evaluation on data of the trained data-driven model input test set, and determining the prediction effect of the data-driven model according to the mean square error of the evaluation index of the data-driven model;
and adjusting the LSTM layer number of the data driving model and the neuron loss percentage of Dropout, and repeatedly training, testing and evaluating the data driving model until the mean square error of the data driving model is converged.
After the optimal scheme is adopted, the Dropout method randomly deletes neurons in each layer in the learning process, the deleted neurons do not transmit signals any more, and overfitting of the model is prevented by reducing the complexity of the model.
Based on the multi-layer LSTM algorithm and the Dropout method, the learning capability of the model can be further improved by overlapping the number of layers of the LSTM.
Preferably, the step 3 comprises the following steps:
the automobile system is simplified and equivalent: the vehicle sprung mass and the unsprung mass are both regarded as rigid bodies, the vertical freedom degrees of four wheels, the vertical freedom degrees, the lateral freedom degrees and the pitching freedom degrees of a vehicle body and the local freedom degrees of attachment points of four suspension vehicle bodies are considered, and after the equivalent stiffness and the damping parameters of each part are obtained, a complete vehicle dynamics analytic model with eleven degrees of freedom is established;
the method comprises the following steps of performing frequency response function test in a semi-anechoic chamber, placing a spatial nondirectional sound source at the right ear of a driver by utilizing a reciprocity principle as excitation, measuring the vibration acceleration response of a suspension vehicle body attachment point, and finally obtaining the transfer relationship from the suspension vehicle body attachment point to the right ear noise of the driver in the vehicle by a matrix inversion method, wherein the matrix inversion formula is as follows:
{F N }＝[H MN ] -1 ×{X M }；
in the formula: { F N The input excitation vector of the system is used as the input excitation vector of the system,{X M is the response vector of the response point, H MN A transfer function that is an input to a response;
a complete vehicle dynamics analysis model is built in Simulink, and a transfer function from a suspension vehicle body attachment point obtained through test to the right ear noise of a driver in the vehicle is introduced so as to construct a knowledge driving model.
Preferably, the step 4 comprises the following steps:
based on the data of the verification set in the step 2, the NSGA2 genetic algorithm is used as a multi-objective optimization algorithm, the evaluation indexes of the data driving model and the knowledge driving model and the deviation of the prediction results between the evaluation indexes are used as optimization objectives, and the weight parameters of the memory unit g in the data driving model and the setting of the dynamic stiffness of the vehicle body in the knowledge driving model are iteratively updated, wherein the formula is as follows:
in the formula y d Output results for the data-driven model, t k For true value of data, y k For the output result of the knowledge-driven model, k is the number of samples, W is the weight parameter of the memory unit g, μ 1 、μ 2 、μ 3 、μ 4 Equivalent dynamic stiffness for the attachment point of the suspension body, i 1 、i 2 The optimization interval is set according to the actual engineering;
and according to the prediction effects of the data driving model and the knowledge driving model, the weights of the data driving model and the knowledge driving model are distributed according to a certain proportion, and the prediction results of the data driving model and the knowledge driving model are multiplied by the respective weights and then summed to obtain a final prediction result.
After the technical scheme is adopted, the final prediction result is obtained after the prediction results of the data driving model and the knowledge driving model are weighted, and compared with the case that the data driving model or the knowledge driving model is used independently, the accuracy of prediction is further improved.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
1. the method improves the robustness and accuracy of prediction on the basis of carrying out prediction analysis on the road noise of the automobile suspension structure by combining an intelligent algorithm on the basis of test and simulation data.
2. The method introduces a knowledge model of automobile dynamics to represent the mechanism of the road noise of the suspension structure, fuses knowledge and data, and realizes dual iterative learning of a data driving model and a knowledge driving model.
3. According to the method, the suspension structure road noise database is established, and the data driving model and the knowledge driving model are used for realizing the autonomous learning of the updated data by the data driving model and the knowledge driving model, so that the generalization capability is improved, and the efficiency of the prediction analysis of the suspension structure road noise is improved.
Drawings
FIG. 1 is a technical flow diagram.
Fig. 2 is a flow chart of the NSGA2 algorithm.
Fig. 3 is a schematic drawing of Dropout.
Fig. 4 is a schematic diagram of a macpherson suspension multilevel decomposition tree.
Fig. 5 is a multi-level exploded tree diagram of an inverted E-shaped multi-link suspension.
Fig. 6 is a schematic diagram of the vibration acceleration in the X direction of the front knuckle.
FIG. 7 is a schematic view of a complete vehicle dynamics model with eleven degrees of freedom.
FIG. 8 is a comparison of predicted results.
Fig. 9 is a schematic diagram of a road noise database architecture of an automotive suspension structure.
Detailed Description
In order to make the technical means, features and effects achieved by the present invention easier to understand, the technical solutions in the embodiments of the present invention are clearly and completely described below with reference to the specific embodiments and the drawings in the embodiments of the present invention.
As shown in fig. 1 to 9, the preferred embodiments of the present invention are described below:
referring to fig. 1, the self-learning prediction method for the road noise of the automobile suspension structure comprises the following steps:
step 1: analyzing a structural road noise transmission path from road surface excitation to the right ear noise of a driver in the automobile corresponding to the suspension according to different structural characteristics of different suspensions of the automobile, and establishing a multi-level decomposition tree of the suspension structural road noise according to a hierarchical standard of 'whole automobile level-system level-subsystem level-part level';
step 2: embedding an intelligent algorithm in the multilevel decomposition tree, and learning the data relationship between target nodes in the multilevel decomposition tree by using the intelligent algorithm to obtain a data driving model;
and step 3: establishing a whole vehicle dynamics analysis model from road excitation to a suspension vehicle body attachment point according to the existing knowledge, and then obtaining a transfer relation from the suspension vehicle body attachment point vibration to the in-vehicle road noise based on test analysis, thereby constructing a knowledge driving model from the road excitation to the in-vehicle driver right ear noise;
and 4, step 4: based on the established data driving model and knowledge driving model, optimizing the built-in hyper-parameters of the data driving model and the set parameters of the knowledge driving model by using a multi-objective optimization algorithm and taking the evaluation indexes of the two models and the mean square error of the prediction result between the two models as targets, and finally multiplying the prediction results of the two models by the respective set weights and summing the results, wherein the summed result is taken as the final prediction result.
In the embodiment, a whole vehicle dynamics analysis model for suspension road noise analysis is established, a data driving model based on data driving is established, knowledge and data are fused and iterated, and road noise prediction interpretability and robustness are enhanced while road noise prediction analysis efficiency is improved.
Referring to fig. 1, the present embodiment further includes a step 5 for following the step 4: and establishing a MySQL database according to test and simulation data by taking a multi-level decomposition tree of the suspension structure road noise and a complete vehicle dynamics analysis model as a data relation basis, and connecting the database with a data driving model and a knowledge driving model in real time.
In the embodiment, after the database is updated, the knowledge driving model and the data driving model can independently learn new sample data, and the generalization capability of the knowledge driving model and the data driving model is improved.
Referring to fig. 3, the step 2 includes the following steps:
the method comprises the steps of sorting and normalizing test and simulation data according to a multilayer decomposition tree of the suspension structure road noise, and then dividing the test and simulation data into a training set, a test set and a verification set according to a certain proportion;
introducing a multi-layer LSTM algorithm and a Dropout method between target nodes between an upper layer and a lower layer in a multi-layer decomposition tree of the road noise of the suspension structure to construct a data driving model, and training the data driving model based on a training set, wherein the LSTM structural formula is as follows:
where f is a forgetting gate, g is a memory cell, i is an input gate, o is an output gate, σ is a sigmoid function, xt is an input at this time, h t-1 Is a hidden state of the output at the last moment,
is x t The corresponding weight of the weight is set to be,
is h t-1 Corresponding weight, b (f) F is the corresponding offset;
inputting data of a test set to a trained data driving model for model evaluation, and determining the prediction effect of the data driving model according to the mean square error of the evaluation index of the data driving model;
and adjusting the LSTM layer number of the data driving model and the neuron loss percentage of Dropout, and repeatedly training, testing and evaluating the data driving model until the mean square error of the data driving model is converged.
In this embodiment, the Dropout method randomly deletes neurons in each layer in the learning process, and the deleted neurons do not transmit signals any more, so that overfitting of the model is prevented by reducing the complexity of the model itself.
Based on the multi-layer LSTM algorithm and the Dropout method, the learning capability of the model can be further improved by overlapping the number of layers of the LSTM.
Referring to fig. 1, the step 3 includes the following steps:
the automobile system is simplified and equivalent: the vehicle sprung mass and the unsprung mass are both regarded as rigid bodies, the vertical freedom degrees of four wheels, the freedom degrees of the vertical direction, the side-tipping direction and the pitching direction of a vehicle body and the local freedom degrees of four suspension vehicle body attachment points are considered, and after the equivalent rigidity and the damping parameters of each part are obtained, a complete vehicle dynamics analytic model with eleven degrees of freedom is established;
the method comprises the following steps of performing frequency response function test in a semi-anechoic chamber, placing a spatial nondirectional sound source at the right ear of a driver by utilizing a reciprocity principle as excitation, measuring the vibration acceleration response of a suspension vehicle body attachment point, and finally obtaining the transfer relationship from the suspension vehicle body attachment point to the right ear noise of the driver in the vehicle by a matrix inversion method, wherein the matrix inversion formula is as follows:
{F N }＝[H MN ] -1 ×{X M }；
in the formula: { F N Is the system input excitation vector, { X } M Is the response vector of the response point, H MN A transfer function that is an input to a response;
and (3) building a complete vehicle dynamics analysis model in Simulink, and introducing a transfer function from a suspension vehicle body attachment point obtained by experimental test to the right ear noise of a driver in the vehicle so as to construct a knowledge driving model.
Referring to fig. 2, the step 4 includes the following steps:
based on the data of the verification set in the step 2, the NSGA2 genetic algorithm is used as a multi-objective optimization algorithm, the evaluation indexes of the data driving model and the knowledge driving model and the deviation of the prediction result between the evaluation indexes are used as optimization objectives, and the weight parameters of the memory unit g in the data driving model and the setting of the dynamic stiffness of the vehicle body in the knowledge driving model are iteratively updated, wherein the formula is as follows:
in the formula y d Output results for the data-driven model, t k As true value of the data, y k For the output result of the knowledge-driven model, k is the number of samples, W is the weight parameter of the memory unit g, μ 1 、μ 2 、μ 3 、μ 4 Equivalent dynamic stiffness for the attachment point of the suspension body, i 1 、i 2 The optimization interval is set according to the actual engineering;
and according to the prediction effects of the data driving model and the knowledge driving model, distributing the weights of the data driving model and the knowledge driving model according to a certain proportion, multiplying the prediction results of the data driving model and the knowledge driving model by the respective weights, and then summing the multiplied prediction results to obtain the final prediction result.
In the embodiment, the final prediction result is obtained after the prediction results of the data-driven model and the knowledge-driven model are weighted, and compared with the case that the data-driven model or the knowledge-driven model is used independently, the accuracy of prediction is further improved.
In the embodiment, the robustness and the accuracy of prediction are improved on the basis of carrying out prediction analysis on the road noise of the automobile suspension structure by combining an intelligent algorithm on the basis of test and simulation data.
In the embodiment, a knowledge model of automobile dynamics is introduced to represent a mechanism of road noise of a suspension structure, knowledge is fused with data, and double iterative learning of a data driving model and a knowledge driving model is realized.
In the embodiment, the suspension structure road noise database is established, and the data driving model and the knowledge driving model are used for independently learning the updated data, so that the generalization capability is improved, and the efficiency of the prediction analysis of the suspension structure road noise is improved.
The following further description of the principles of the present invention, in order that those skilled in the art may fully understand the invention, is provided in conjunction with the accompanying drawings, in which:
the invention discloses a self-learning prediction method for road noise of an automobile suspension structure, which is used for predicting the road noise of the automobile suspension structure by taking the model of a rear inverted E-shaped multi-link suspension of a Macpherson suspension as an example.
Referring to fig. 1, the method comprises the following steps:
step 1: referring to fig. 4 and 5, different vibration transmission paths of the suspension are analyzed according to different structural compositions of the macpherson suspension and the inverted-E type multi-link suspension, nodes in the transmission paths are uniformly divided according to a hierarchy of 'vehicle level-system level-subsystem level-part level', a multi-level decomposition tree of the suspension structure road noise of the driver right ear noise in the vehicle from the dynamics parameters of bottom layer elements to the top level is constructed, and a multi-level decomposition tree of the macpherson suspension and a multi-level decomposition tree of the inverted-E type multi-link suspension are constructed.
Step 2: embedding an intelligent algorithm in a multi-level decomposition tree of the suspension structure road noise, and learning the data relationship among all target nodes in the decomposition tree by using the intelligent algorithm so as to construct a data driving model:
2.1 screening and dividing test and simulation data based on the upper and lower hierarchical relations between target points in a multi-level decomposition tree of the suspension structure road noise, sorting all vibration and noise data according to the range of 0Hz to 300Hz and the interval of 1Hz as the research on the suspension structure noise is mainly concerned about the low frequency band, taking the sorted data result as an example of the vibration acceleration of a steering knuckle in the X direction, and dividing the data sample into a training set, a testing set and a verification set according to the proportion of 8:1:1 after disordering the sequence of all data samples, wherein the normalization formula is as follows:
wherein X is the original data, X min And X max Minimum and maximum values, X, respectively, of the original data set nom Is normalized data.
2.2 aiming at each target node in the multi-level decomposition tree of the suspension structure road noise, an LSTM algorithm is introduced to learn the internal relation among the nodes, an input gate i, an output gate o, a forgetting gate f and a memory unit g are introduced to establish an LSTM layer on the basis of the traditional RNN idea, the LSTM layer is superposed on the LSTM layer to improve the learning capacity of the model on time sequence data, then a Dropout method is used for establishing a Dropout layer to set the neuron deletion proportion to be 20% to prevent overfitting of the model, and finally the model is trained on the basis of a training set after all parameters in the model are initialized randomly so as to construct a data driving model of the suspension structure road noise.
2.3 after the model training is finished, inputting data of the test set for model evaluation, and determining the prediction effect of the data-driven model according to the Mean Square Error (MSE) of the evaluation index of the data-driven model, wherein the MSE formula is as follows:
wherein E is the mean square error value, y k Output results for the data-driven model, t k K is the sample size of the data for the true value of the data.
2.4, the mean square error is used as a model evaluation index to find that the model prediction effect is still to be improved, so that the LSTM stacking layer number and the neuron deletion percentage of the Dropout layer in the data driving model are properly adjusted, the comparison results of the model evaluation index and the calculation duration under different parameter adjustments are shown in table 1, after comprehensively considering the model prediction accuracy and the calculation duration, the stacking 3 LSTM layers are finally selected, and the neuron deletion percentage of the Dropout layer is set to be 30% to serve as a final model framework.
TABLE 1 data-driven model pairs under different parameters
And step 3: firstly, establishing a whole vehicle dynamics analysis model from road excitation to a suspension vehicle body attachment point, and then obtaining a transfer relation from suspension vehicle body attachment point vibration to vehicle interior road noise based on test analysis, thereby constructing a knowledge driving model from road excitation to vehicle interior driver right ear noise:
3.1, referring to fig. 7, simplifying and equating an automobile system to a certain extent, regarding sprung mass and unsprung mass of the automobile as rigid bodies, considering the vertical degree of freedom of four wheels, the vertical, lateral and pitching degrees of freedom of the automobile body and the local degrees of freedom of attachment points of four suspension automobile bodies, establishing a complete automobile dynamics analytic model with eleven degrees of freedom after acquiring parameters such as equivalent stiffness, damping and the like of each part required by the model, and transforming the kinetic energy, potential energy and energy consumption formula of the model system based on Lagrange transformation to obtain a differential equation set as follows;
wherein M is the sprung mass; z 1 ,Z 2 ,Z 3 ,Z 4 Exciting the road surface; z5 is the vehicle body displacement; k a1 ,K a2 ,K a3 ,K a4 Equivalent dynamic stiffness for suspension vehicle body attachment points; c a1 ,C a2 ,C a3 ,C a4 Damping suspension body attachment points; z a ,Z b ,Z c ,Z d Is the attachment point displacement; z w1 ,Z w2 ,Z w3 ,Z w4 For displacement between the attachment point to the suspension system; i is X ,I y Roll and pitch moments of inertia; theta and phi are a roll angle and a pitch angle; m is a unit of 1 ,m 2 ,m 3 ,m 4 Is the unsprung mass; z s1 ,Z s2 ,Z s3 ,Z s4 Is the tire displacement; c b1 ,C b2 ,C b3 ,C b4 Damping the suspension system;K b1 ,K b2 ,K b3 ,K b4 is the suspension system stiffness; k t1 ,K t2 ,K t3 ,K t4 Is the tire stiffness; a and b are distances from the front shaft and the rear shaft to the mass center of the whole vehicle respectively; and d is a wheel track.
3.2, performing frequency response function test (FRF) in a semi-anechoic chamber, selecting two excitation points on the left and right of a front suspension and a rear suspension, focusing each excitation point on the X direction, placing a spatial nondirectional sound source at the right ear of a driver by utilizing the reciprocity principle as excitation, measuring the vibration acceleration response of the attachment point of the body of the suspension, and finally obtaining the transmission relation from the attachment point of the body of the suspension to the right ear noise of the driver in the vehicle by a matrix inversion method;
3.3, a complete vehicle dynamics analysis model is built in Simulink, and a transfer function from a suspension vehicle body attachment point obtained through test to the noise of the right ear of a driver in the vehicle is introduced so as to construct a knowledge driving model.
And 4, step 4: based on the established data driving model and the knowledge driving model, a multi-objective optimization algorithm is utilized, the evaluation indexes of the two models and the deviation of the prediction results between the two models are taken as targets, the built-in hyper-parameters of the data driving model and the set parameters of the knowledge driving model are optimized, and finally the prediction results of the two models are multiplied by the respective set weights respectively and then summed to be taken as the final prediction result.
4.1 based on the data of the verification set, optimizing the data driving model and the knowledge model, wherein the input of the data driving model is the bottom layer parameters in the multilevel decomposition tree of the road noise of the suspension structure, such as the dynamic stiffness and the vibration acceleration signal of a bush, the input of the knowledge driving model is the road surface spectrum corresponding to the same test data, the NSGA2 genetic algorithm is used as a multi-objective optimization algorithm, and the prediction evaluation indexes MSE of the data driving model and the knowledge driving model are used as MSE d 、MSE k The mean square error of the prediction result between the two is used as an optimization target, the weight parameters in the memory unit g in the data driving model are respectively expanded by 50 percent up and down to be used as an optimization interval, and the floating of the suspension body dynamic stiffness value in the knowledge driving model is set according to the actual engineering experienceThe range is used as an optimization interval, the iteration frequency is set to be 500, the two models are simultaneously optimized, and the result after optimization shows that the mean square error of the data driving model is reduced by 17 percent and the mean square error of the knowledge driving model is reduced by 13 percent;
4.2 based on the test data of the current vehicle model, the mean square error of the data driving model is 0.078, the mean square error of the knowledge driving model is 0.187, and the comparison shows that the prediction effect of the data driving model is better than that of the knowledge driving model, so the weight of the data driving model and the knowledge driving model is set according to the relation of the evaluation indexes of the prediction results of the data driving model and the knowledge driving model, and finally the prediction results of the data driving model and the knowledge driving model are multiplied by the respective weight and then summed to be the final prediction result, taking a group of latest measured data of the vehicle model as an example, the prediction effect of the single model and the data knowledge fusion dual driving model is shown in fig. 8, and the figure shows that the knowledge data fusion dual driving model shows better performance.
And 5: the method is characterized in that a multi-level decomposition tree of the suspension structure road noise and a complete vehicle dynamics analysis model architecture are used as a basis of data relation, a MySQL relational database system is used for establishing a road noise database of the suspension structure of the vehicle, the overall architecture is shown as figure 9, functions of storing, deleting, modifying, inquiring and the like of vehicle test data and vehicle dynamics model simulation data are achieved through the database, the database is connected with a knowledge driving model and a data driving model in a butt joint mode, the knowledge driving model and the data driving model can automatically train new data samples as training sets after the database is updated, and therefore the knowledge driving model and the data driving model can independently learn new knowledge, and the model has the capability of infinite iteration upgrading.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (5)
1. A self-learning prediction method for road noise of an automotive suspension structure is characterized by comprising the following steps: the method comprises the following steps:
step 1: according to different structural characteristics of different suspensions of an automobile, analyzing a structural road noise transmission path from road surface excitation to right ear noise of a driver in the automobile corresponding to the suspension, and establishing a multi-level decomposition tree of the suspension structural road noise according to a hierarchical standard of 'whole automobile level-system level-subsystem level-part level';
step 2: embedding an intelligent algorithm in the multilevel decomposition tree, and learning the data relationship between target nodes in the multilevel decomposition tree by using the intelligent algorithm to obtain a data driving model;
and step 3: establishing a whole vehicle dynamics analysis model from road excitation to a suspension vehicle body attachment point, and then obtaining a transfer relation from suspension vehicle body attachment point vibration to vehicle interior road noise based on test analysis, thereby constructing a knowledge driving model from road excitation to vehicle interior driver right ear noise;
and 4, step 4: based on the established data driving model and knowledge driving model, optimizing the built-in hyper-parameters of the data driving model and the set parameters of the knowledge driving model by using a multi-objective optimization algorithm and taking the evaluation indexes of the two models and the mean square error of the prediction result between the two models as targets, finally, respectively multiplying the prediction results of the two models by the respective set weights and summing, and taking the summed result as the final prediction result.
2. The self-learning prediction method of the road noise of the automobile suspension structure is characterized by comprising the following steps of: further comprising a step 5 for following said step 4: the method comprises the steps of taking a multi-level decomposition tree of the suspension structure road noise and a complete vehicle dynamics analysis model as a data relation basis, obtaining basic road noise data in a complete vehicle road test or CAE complete vehicle modeling simulation mode, then establishing a MySQL database according to the basic road noise data, and connecting the database with a data driving model and a knowledge driving model in real time.
3. The self-learning prediction method of the road noise of the automobile suspension structure is characterized by comprising the following steps of: the step 2 comprises the following steps:
the method comprises the steps of sorting and normalizing test and simulation data according to a multi-level decomposition tree of the suspension structure road noise, and then dividing the test and simulation data into a training set, a test set and a verification set according to a certain proportion;
introducing a multi-layer LSTM algorithm and a Dropout method between target nodes between an upper layer and a lower layer in a multi-layer decomposition tree of the road noise of the suspension structure to construct a data driving model, and training the data driving model based on a training set, wherein the LSTM structural formula is as follows:
wherein f is a forgetting gate, g is a memory cell, i is an input gate, o is an output gate, σ is a sigmoid function, and x t For the input of this moment, h t-1 Is a hidden state of the output at the previous moment,
is x t The corresponding weight of the weight is set to be,
is h t-1 Corresponding weight, b (f) Is the offset corresponding to f;
inputting data of a test set to a trained data driving model for model evaluation, and determining the prediction effect of the data driving model according to the mean square error of the evaluation index of the data driving model;
and adjusting the LSTM layer number of the data driving model and the neuron loss percentage of Dropout, and repeatedly training, testing and evaluating the data driving model until the mean square error of the data driving model is converged.
4. The self-learning prediction method of the road noise of the automobile suspension structure is characterized by comprising the following steps of: the step 3 comprises the following steps:
the automobile system is simplified and equivalent: the vehicle sprung mass and the unsprung mass are both regarded as rigid bodies, the vertical freedom degrees of four wheels, the freedom degrees of the vertical direction, the side-tipping direction and the pitching direction of a vehicle body and the local freedom degrees of four suspension vehicle body attachment points are considered, and after the equivalent rigidity and the damping parameters of each part are obtained, a complete vehicle dynamics analytic model with eleven degrees of freedom is established;
the method comprises the following steps of performing frequency response function test in a semi-anechoic chamber, placing a spatial nondirectional sound source at the right ear of a driver by utilizing a reciprocity principle as excitation, measuring the vibration acceleration response of a suspension vehicle body attachment point, and finally obtaining the transfer relationship from the suspension vehicle body attachment point to the right ear noise of the driver in the vehicle by a matrix inversion method, wherein the matrix inversion formula is as follows:
{F N }＝[H MN ] -1 ×{X M }；
in the formula: { F N Is the system input excitation vector, { X } M Is the response vector of the response point, H MN A transfer function that is an input to a response;
and (3) building a complete vehicle dynamics analysis model in Simulink, and introducing a transfer function from a suspension vehicle body attachment point obtained by experimental test to the right ear noise of a driver in the vehicle so as to construct a knowledge driving model.
5. The self-learning prediction method of the road noise of the automobile suspension structure is characterized by comprising the following steps of: the step 4 comprises the following steps:
based on the data of the verification set in the step 2, the NSGA2 genetic algorithm is used as a multi-objective optimization algorithm, the evaluation indexes of the data driving model and the knowledge driving model and the deviation of the prediction results between the evaluation indexes are used as optimization objectives, and the weight parameters of the memory unit g in the data driving model and the setting of the dynamic stiffness of the vehicle body in the knowledge driving model are iteratively updated, wherein the formula is as follows:
in the formula y d Output results for the data-driven model, t k As true value of the data, y k For the output result of the knowledge-driven model, k is the number of samples, W is the weight parameter of the memory unit g, μ 1 、μ 2 、μ 3 、μ 4 Equivalent dynamic stiffness for suspension body attachment points, i 1 、i 2 The optimization interval is set according to the actual engineering;
and according to the prediction effects of the data driving model and the knowledge driving model, distributing the weights of the data driving model and the knowledge driving model according to a certain proportion, multiplying the prediction results of the data driving model and the knowledge driving model by the respective weights, and then summing the multiplied prediction results to obtain the final prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210755132.XA CN115130218A (en) | 2022-06-29 | 2022-06-29 | Self-learning prediction method for road noise of automotive suspension structure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210755132.XA CN115130218A (en) | 2022-06-29 | 2022-06-29 | Self-learning prediction method for road noise of automotive suspension structure |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115130218A true CN115130218A (en) | 2022-09-30 |
Family
ID=83382624
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210755132.XA Pending CN115130218A (en) | 2022-06-29 | 2022-06-29 | Self-learning prediction method for road noise of automotive suspension structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115130218A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116779095A (en) * | 2023-04-27 | 2023-09-19 | 西南交通大学 | Knowledge data dual-drive reasonable medication prediction method |
-
2022
- 2022-06-29 CN CN202210755132.XA patent/CN115130218A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116779095A (en) * | 2023-04-27 | 2023-09-19 | 西南交通大学 | Knowledge data dual-drive reasonable medication prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gobbi et al. | Optimal & robust design of a road vehicle suspension system | |
| Soleymani et al. | Adaptive fuzzy controller for vehicle active suspension system based on traffic conditions | |
| Gobbi et al. | Optimising a car chassis | |
| CN109398363B (en) | Pavement grade determination method and device, storage medium and automobile | |
| CN111444623B (en) | Collaborative optimization method and system for damping nonlinear commercial vehicle suspension dynamics | |
| Huang et al. | Multitarget prediction and optimization of pure electric vehicle tire/road airborne noise sound quality based on a knowledge-and data-driven method | |
| CN115130218A (en) | Self-learning prediction method for road noise of automotive suspension structure | |
| CN113886974B (en) | Method for predicting sound path noise of in-vehicle structure | |
| Llopis-Albert et al. | Multiobjective optimization framework for designing a vehicle suspension system. A comparison of optimization algorithms | |
| Zhang et al. | Multiobjective optimization design for lightweight and crash safety of body-in-white based on entropy weighted grey relational analysis and MNSGA-II | |
| CN116205024A (en) | Self-adaptive automatic driving dynamic scene general generation method for high-low dimension evaluation scene | |
| Wang et al. | Active control of hydropneumatic suspension parameters of wheel loaders based on road condition identification | |
| CN113591360B (en) | Magneto-rheological damper structural parameter optimization method based on whole vehicle dynamics model | |
| Jamei et al. | Elicitation and fine-tuning of fuzzy control rules using symbiotic evolution | |
| CN117313248A (en) | Vehicle body structure optimization method and system considering vehicle body performance and collision injury | |
| CN116415494A (en) | Road noise optimization method, system and storage medium | |
| Peckelsen | Objective tyre development: Definition and analysis of tyre characteristics and quantification of their conflicts | |
| Yarmohammadisatri et al. | A family base optimization of a developed nonlinear vehicle suspension model using gray family design algorithm | |
| Nazemi et al. | GA tuned H∞ roll acceleration controller based on series active variable geometry suspension on rough roads | |
| Chen et al. | Multi-objective optimization of interior noise of an automotive body based on different surrogate models and NSGA-II | |
| Shintani et al. | Surrogate modeling of waveform response using singular value decomposition and Bayesian optimization | |
| Xu et al. | Identification of spring-force factors of suspension systems using progressive neural network on a validated computer model | |
| Ghorbany et al. | Global-guidance chaotic multi-objective particle swarm optimization method for pneumatic suspension handling and ride quality enhancement on the basis of a thermodynamic model of a full vehicle | |
| Seeger | Neural Network-Based Wheel Suspension Modeling | |
| Adl et al. | Multi-objective optimal design of a passenger car’s body |
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 |