Machine number fusion-based automobile road noise prediction method and system, and electronic equipment

Abstract

The invention discloses a machine number fusion-based automobile road noise prediction method and system, electronic equipment and a storage medium, and relates to the field of automobile road noise testing, and the method comprises the steps: obtaining road noise characteristics, and constructing a road noise hierarchical decomposition architecture according to the road noise characteristics; an intelligent algorithm is embedded in the hierarchical decomposition architecture, the incidence relation between each hierarchical target node is obtained, and a key physical quantity is introduced to construct a data driving model fused with physical prior; establishing an acoustic-solid coupling finite element model, collecting a noise transfer function, and constructing a TPA model; and fusing the data driving model and the TPA model to generate a prediction result of the in-vehicle road noise. Through a transmission force result based on data driving prediction, an in-vehicle noise prediction model is established through a TPA method by further combining NTF, mechanism and data fusion is realized, and an efficient, accurate and highly interpretable prediction solution is provided for automobile road noise.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automobile road noise testing, and in particular to a machine learning-based automobile road noise prediction method and system and electronic device. BACKGROUND

[0002] Road noise is a key factor affecting the driving experience of a vehicle, not only significantly reducing the subjective comfort of the driver and passengers, but also possibly affecting driving safety due to noise interference, thereby directly affecting the market competitiveness and brand value of the vehicle. Therefore, it is of great significance to deeply study the automobile road noise prediction and optimization method.

[0003] There are several methods for analyzing the road noise of a vehicle, including the transfer path analysis method, finite element modeling simulation analysis, and data-driven method. Among the currently disclosed road noise-related materials, such as

[0004] Chinese patent, invention name: A method for identifying and eliminating in-vehicle booming noise based on TPA analysis model, application number: CN202111166060.7. Combining subjective evaluation and spectral analysis, the TPA model is used to identify the excitation source and transmission path, and to determine the source and contribution of in-vehicle booming noise. This method improves the efficiency and accuracy of troubleshooting, but relies on manual subjective evaluation and TPA model construction, which requires high personnel experience and model accuracy in practical operation.

[0005] Chinese patent, invention name: CAE simulation prediction method for vehicle road vibration noise, application number: CN201610415861.5. By testing the steering knuckle acceleration and transfer function, the wheel hub excitation is derived based on the wheel hub load theory, and is loaded into the vehicle finite element model to realize the simulation prediction of in-vehicle vibration and noise.

[0006] Chinese patent, invention name: A method for predicting in-vehicle structural sound road noise, application number: CN202111266159.4. Based on the suspension NVH performance, the vehicle road noise is hierarchically decomposed, combined with experimental design and road test data, and a support vector regression (SVR) data-driven model is constructed to realize modeling and prediction of in-vehicle structural sound road noise. This method has a clear structure and certain engineering applicability, but the prediction model relies on sample quality and is difficult to fully integrate physical mechanisms, which has the problem of insufficient explainability.

[0007] A patent in China, the invention name: a method for self-learning prediction of road noise of automobile suspension structure, application number: CN202210755132.X, adopts the fusion of knowledge-driven model and data-driven model based on LSTM, improves the robustness and explainability of road noise prediction, and has the ability of continuous self-learning with data expansion, which provides an effective tool for automobile NVH analysis. However, this method relies heavily on large-scale high-quality data samples in the initial stage, and the improvement of explainability and prediction accuracy needs long-term data accumulation and model iteration.

[0008] Therefore, there is an urgent need for a solution to solve the technical problems that the traditional road noise analysis method relies on complex vehicle CAE modeling and simulation, has the limitation of large modeling difficulty and long calculation time. Data-driven methods can effectively learn and predict noise data, but have poor explainability, and the prediction results are difficult to associate with physical mechanisms. SUMMARY

[0009] Therefore, there is an urgent need for a solution to solve the technical problems that the traditional road noise analysis method relies on complex vehicle CAE modeling and simulation, has the limitation of large modeling difficulty and long calculation time. Data-driven methods can effectively learn and predict noise data, but have poor explainability, and the prediction results are difficult to associate with physical mechanisms.

[0010] The present application provides the following solutions:

[0011] According to one aspect of the present application, a machine-number fusion automobile road noise prediction method is provided, comprising the following steps:

[0012] Step S1, according to the road noise characteristics, a road noise hierarchical decomposition architecture is constructed;

[0013] Step S2, embedding intelligent algorithms in the hierarchical decomposition architecture, obtaining the correlation between target nodes of each level, and introducing key physical quantities to construct a data-driven model with fusion physical priori;

[0014] Step S3, establishing a sound-solid coupling finite element model, collecting noise transfer functions, and constructing a TPA model;

[0015] Step S4, fusion of data-driven model and TPA model, generating the prediction result of in-vehicle road noise.

[0016] Further, step S1 includes:

[0017] Step S11, according to the road noise characteristics, the road noise is simplified as 20-300Hz structural sound for processing;

[0018] Step S12, on the basis of identifying different suspension types and structures, combining the main influencing factors of road noise and its transmission path, starting from the target response of the whole vehicle, the road noise is decomposed layer by layer in a top-down manner to construct a hierarchical decomposition architecture.

[0019] Step S12 further comprises:

[0020] The first level of the hierarchical architecture is the noise response at the driver's right ear as the target evaluation index;

[0021] The second level is the vehicle body system, and the parameters include the transmission force on the passive side of each attachment point of the chassis and the vehicle body;

[0022] The sound transmission path from the passive side of the vehicle body attachment point to the driver's right ear is represented by NTF between the first and second levels;

[0023] The third level is the chassis system, and the parameters include the vibration response on the active side of the vehicle body attachment point and IPI;

[0024] The fourth level is the excitation source layer, and the parameters include the vibration excitation signal at the steering knuckle and the dynamic characteristic parameters of the bushing.

[0025] Further, step S2 comprises:

[0026] Step S21, collecting or testing data according to the hierarchical decomposition architecture;

[0027] Step S22, constructing an LSTM model;

[0028] Step S23, constructing PINNs by integrating IPI into the model loss function design in the LSTM output layer;

[0029] Step S24, using a genetic algorithm to optimize the model loss weight and IPI loss weight in the loss function.

[0030] Step S2 further comprises:

[0031] The collected data covers all key parameters in the hierarchical architecture except NTF. Specifically, it includes the passive end transmission force-frequency curve, the active end acceleration response-frequency curve, the noise response at the driver's right ear-frequency curve, and the IPI of the active end-frequency curve.

[0032] Step S22 comprises the following steps:

[0033] Training and constructing a chassis system performance LSTM prediction model of the third level;

[0034] Training and constructing a vehicle body system performance LSTM prediction model of the second level;

[0035] The unknown parameters at each level are predicted by the output of the next level, thus realizing bottom-up hierarchical recursive modeling;

[0036] Step S23 includes the following steps:

[0037] Based on the second-level LSTM prediction model of the vehicle body system performance, IPI is incorporated into the model loss function design to construct the physical information neural network PINNs.

[0038] Furthermore, step S3 includes:

[0039] Step S31: Define the passive side of the vehicle body attachment point as the excitation point of the transfer function, and the noise in the driver's right ear as the response point;

[0040] Step S32: Finite element modeling of interior and vehicle body;

[0041] Step S33: Finite element modeling of the acoustic cavity;

[0042] Step S34: Calculate the NTF (Network Transfer Function) for excitation and response points.

[0043] Furthermore, step S3 also includes:

[0044] Step S32 includes the following steps:

[0045] The attachment models are meshed, and corresponding element properties and material parameters are assigned to them to complete the establishment of the interior and body finite element models. The attachment models include: the body-in-white model and the interior trim;

[0046] Step S33 includes the following steps:

[0047] Using the seat geometry model, body-in-white structure and closure components as input, a finite element model of the acoustic cavity is established. The interior body model and the acoustic cavity model are then assembled to construct an acoustic-structure coupling finite element model.

[0048] Step S34 includes the following sub-steps:

[0049] A unit load of 1 N is applied in the X, Y, and Z directions at each loading point, and frequency response function analysis is performed to calculate the NTF of each response point relative to the loading point.

[0050] Furthermore, step S4 includes:

[0051] Step S41: In the road noise hierarchy architecture, the force transmitted from the steering knuckle excitation to the passive end of the vehicle body is predicted using data-driven methods.

[0052] Step S42: In the road noise hierarchy architecture, the noise transmitted from the passive end of the vehicle body to the driver's right ear is predicted using TPA;

[0053] Step S43: Use an ensemble algorithm to develop a fast prediction interface.

[0054] Furthermore, including:

[0055] Step S41 includes the following steps:

[0056] The third-level deep neural network uses the vibration excitation signal of the steering knuckle as input to predict the vibration acceleration response and IPI of the active end connection point of the chassis system;

[0057] The second-level deep neural network uses the output of the third level and the bushing dynamic stiffness parameters as input to predict the transmitted force at the passive end connection point of the vehicle body system.

[0058] Furthermore, step S42 includes the following sub-steps:

[0059] The first level builds upon the output of the second level, combining the NTF and TPA principles to construct the input and establish a model for predicting the noise response of the driver's right ear.

[0060] According to a second aspect of the present invention, a vehicle road noise prediction system based on machine-data fusion is provided, the system comprising:

[0061] Architecture building module, intelligent model building module, physical model building module, and model fusion prediction module;

[0062] The architecture building module is used to construct a road noise hierarchical decomposition architecture based on road noise characteristics.

[0063] The intelligent model building module is used to embed intelligent algorithms in the hierarchical decomposition architecture, obtain the correlation between target nodes at each level, and introduce key physical quantities to build a data-driven model that integrates physical priors.

[0064] The physical model building module is used to establish an acoustic-structure coupling finite element model, collect noise transfer functions, and build a TPA model.

[0065] The model fusion prediction module is used to fuse the data-driven model and the TPA model to achieve in-vehicle road noise prediction.

[0066] According to three aspects of the present invention, an electronic device is provided, comprising: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0067] The memory stores a computer program, which, when executed by a processor, causes the processor to perform steps of a machine-data fusion method for predicting road noise in vehicles.

[0068] According to four aspects of the present invention, a computer-readable storage medium is provided that stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a machine-data fusion method for predicting vehicle road noise.

[0069] The above solution achieves the following beneficial technical effects:

[0070] This application introduces IPI constraints to construct a PINNs data-driven model, which effectively solves the modeling problem of complex structural paths and significantly improves the interpretability and accuracy of force transmission prediction.

[0071] This application constructs a Physical Information Neural Network (PINNs) by introducing acceleration admittance (IPI) constraints, which significantly improves the interpretability and accuracy of force transmission prediction. Furthermore, a genetic algorithm (GA) is used to optimize the weights of data-driven loss and physical-driven loss in the loss function of PINNs, thereby achieving a balance between prediction accuracy and mechanism consistency.

[0072] This application can obtain the NTF from the passive side of the vehicle body attachment point to the driver's right ear by constructing a finite element model of the interior body and acoustic cavity, thus avoiding the complexity and high computational cost of whole vehicle finite element modeling.

[0073] This application utilizes data-driven prediction of transmission force results and further combines NTF with the TPA method to establish an in-vehicle noise prediction model, thereby achieving the fusion of mechanism and data, and providing an efficient, accurate and highly interpretable prediction solution for automotive road noise. Attached Figure Description

[0074] Figure 1 This is a flowchart of a vehicle road noise prediction method based on machine-data fusion provided in one or more embodiments of the present invention.

[0075] Figure 2 This is a structural diagram of a vehicle road noise prediction system based on machine-data fusion provided in one or more embodiments of the present invention.

[0076] Figure 3 This is a flowchart of a vehicle road noise prediction method based on machine-data fusion, according to a specific embodiment of the present invention.

[0077] Figure 4 This is a diagram of a path noise prediction method based on mechanism data fusion, according to a specific embodiment of the present invention.

[0078] Figure 5 This is a data-driven road noise hierarchical decomposition architecture constructed according to a specific embodiment of the present invention.

[0079] Figure 6 This is a schematic diagram of a PINNs data-driven model constructed by introducing IPI constraints according to a specific embodiment of the present invention.

[0080] Figure 7 This is a schematic diagram of an acoustic-structure coupling finite element model established according to a specific embodiment of the present invention.

[0081] Figure 8 This is a fast data-driven prediction interface established according to a specific embodiment of the present invention.

[0082] Figure 9 This is a block diagram of an electronic device structure for a vehicle road noise prediction method based on machine-data fusion provided in one or more embodiments of the present invention. Detailed Implementation

[0083] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] Figure 1 This is a flowchart of a vehicle road noise prediction method based on machine-data fusion provided in one or more embodiments of the present invention.

[0085] like Figure 1 As shown, the method includes the following steps:

[0086] Step S1: Construct a road noise hierarchical decomposition architecture based on road noise characteristics;

[0087] Step S2: Embed intelligent algorithms in the hierarchical decomposition architecture to obtain the correlation between target nodes at each level, and introduce key physical quantities to build a data-driven model that integrates physical priors;

[0088] Step S3: Establish an acoustic-structure interaction finite element model, collect noise transfer functions, and construct a TPA model;

[0089] Step S4: Integrate the data-driven model and the TPA model to generate prediction results for in-vehicle road noise.

[0090] Specifically, a road noise hierarchical decomposition architecture is constructed. For complex road structures, a data-driven method is used to predict the transmitted force from wheel excitation to the passive side of the vehicle body attachment point. For in-vehicle noise, the TPA method is used for accurate prediction. A road noise prediction method based on mechanism-data fusion is employed, such as... Figure 3 As shown. Due to the complex chassis structure and diverse transmission paths, the data-driven model only needs to select structural parameters related to the predicted performance to construct a hierarchical decomposition architecture, such as... Figure 4 As shown.

[0091] Furthermore, step S1 includes:

[0092] Step S11: Based on the road noise characteristics, simplify the road noise into 20-300Hz structure sound for processing;

[0093] Specifically, under conditions of moderate vehicle speed on rough road surfaces, vehicles easily generate structural road noise below 300Hz. This frequency band falls within the human hearing threshold range (20–20000Hz), significantly impacting in-vehicle acoustic comfort. Therefore, this invention simplifies road noise into structural sound within the 20–300Hz range for modeling and processing.

[0094] Step S12: Based on the understanding of different suspension types and structures, and combined with the main influencing factors of road noise and their transmission paths, starting from the target response of the whole vehicle, the road noise is decomposed layer by layer in a top-down manner to construct a hierarchical decomposition architecture.

[0095] Specifically, based on the clear understanding of different suspension types and structures, starting from the target response of the whole vehicle, road noise is decomposed layer by layer in a top-down manner to construct a hierarchical decomposition architecture of "vehicle level - system level - component level".

[0096] Step S12 also includes:

[0097] The first level of the hierarchical architecture is the noise response at the driver's right ear, which serves as the target evaluation indicator.

[0098] The second level is the body system, whose parameters include the passive force transmitted at each attachment point between the chassis and the body.

[0099] The sound transmission path from the passive side of the vehicle body attachment point to the driver's right ear is represented by the NTF between the first and second levels.

[0100] The third level is the chassis system, whose parameters include the vibration response and IPI on the active side of the body attachment points;

[0101] The fourth level is the excitation source layer, whose parameters include the vibration excitation signal at the steering knuckle and the dynamic characteristic parameters of the bushing.

[0102] Furthermore, step S2 includes:

[0103] Step S21: Collect or test data based on the shelf decomposition architecture;

[0104] Step S22: Construct the LSTM model;

[0105] Step S23: Incorporate IPI into the model loss function design in the LSTM output layer to construct PINNs;

[0106] Step S24: Use a genetic algorithm to optimize the model loss weights and IPI loss weights in the loss function.

[0107] Specifically, the data collected in step S21 covers all key parameters in the hierarchical architecture except for NTF. These include: the passive end force-frequency curve, the active end acceleration response-frequency curve, the noise response-frequency curve at the driver's right ear, and the active end IPI-frequency curve.

[0108] In step S22,

[0109] In the hierarchical decomposition architecture of road noise, a multi-layer LSTM algorithm is introduced between the target nodes in the upper and lower levels to construct a data-driven model, and the data-driven model is trained based on the training set. The LSTM structure formula is as follows:

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] In the formula For the Gate of Oblivion For memory units, For input gate, For output gate, For the sigmoid function, For the input at the next time step, This is the hidden state of the output from the previous moment. for The corresponding weights for The corresponding weights for Corresponding bias;

[0115] Step S2 also includes:

[0116] The collected data covers all key parameters in the hierarchical architecture except for NTF. Specifically, this includes: the passive force-frequency curve, the active acceleration response-frequency curve, the noise response-frequency curve at the driver's right ear, and the active IPI-frequency curve.

[0117] Step S22 includes the following steps:

[0118] Train and build a third-level LSTM prediction model for chassis system performance;

[0119] Train and build a second-level LSTM prediction model for vehicle body system performance;

[0120] The unknown parameters at each level are predicted by the output of the next level, thus realizing bottom-up hierarchical recursive modeling;

[0121] Specifically, the model is evaluated by inputting the trained data-driven model into the test set, and the prediction performance of the data-driven model is determined based on the mean squared error, an evaluation metric for the data-driven model.

[0122] Step S23 includes the following steps:

[0123] Based on the second-level LSTM prediction model of the vehicle body system performance, IPI is incorporated into the model loss function design to construct the physical information neural network PINNs.

[0124] Specifically, the calculation process for step S23 is as follows:

[0125] Acceleration can be represented by impedance because the stronger the impedance of acceleration, the smaller the displacement response of the system. In calculations, the dynamic stiffness at the attachment point is usually represented by the origin acceleration admittance. The formula for its calculation is:

[0126] ;

[0127] In the formula, The stimulus received by the attachment point To excite the angular frequency, For the excitation frequency, To provide dynamic stiffness at the attachment point, For acceleration response.

[0128] PINNs are structurally similar to traditional LSTMs, but the output layer introduces a local loss term based on IPI calculation. By enhancing the training feedback on dynamic stiffness through backpropagation, the model's ability to focus on key local features is improved. The network structure is as follows: Figure 6 As shown. The total loss function of the network is defined as the formula:

[0129] ;

[0130] in, Represented as the total loss function;

[0131] The weighting coefficients, represented as data loss terms, are used to control... The proportion of total losses;

[0132] Represented as the weighting coefficients of the IPI loss term, controlling The proportion of total losses;

[0133] This is represented as a data loss term, used to measure the error between the predicted result and the actual data, reflecting the degree of fit of the model to the original data;

[0134] It is represented as a local loss term calculated based on IPI, used to improve the model's ability to focus on key local features.

[0135] Specifically, in another embodiment, an excitation F is applied to the vehicle chassis attachment point, which includes the connection point between the suspension and the vehicle body, and the excitation frequency of the attachment point is measured. and the corresponding acceleration response At the same time, obtain the dynamic stiffness of the attachment point. .

[0136] use The formula, substituted into the measured excitation frequency and the dynamic stiffness of the attachment point Calculate the attachment point value.

[0137] When the excitation frequency =50Hz, dynamic stiffness =10 6 When N / m, then we can get By comparing IPI at different frequencies and with different dynamic stiffness, the influence of the dynamic stiffness of the attachment point on the acceleration response is analyzed, and its contribution to vehicle road noise is evaluated.

[0138] A PINNs model is constructed, with input data including chassis vibration and attachment point excitation. During model training, road noise prediction is obtained through the total loss function calculation formula, providing a basis for automotive acoustic design optimization.

[0139] in, Ensure that the model fits the input and output data. (Based on IPI calculations) this guides the model to focus on key local features such as the dynamic stiffness of the attachment point. Setting =0.7, =0.3 (the weight can be adjusted according to actual needs), so that the model can learn the patterns in the data while following the physical laws reflected by IPI.

[0140] Specifically, step S24 involves using a genetic algorithm to optimize the model loss weights and IPI loss weights in the loss function. This includes: using the model loss weights and IPI loss weights as optimization variables, and the prediction accuracy of PINNs as the fitness function, performing GA optimization. The specific process includes: first, initializing a weight population containing multiple individuals, each individual consisting of a set of weight parameters for model loss and IPI loss; then, training each individual in PINNs and calculating the fitness function based on the prediction results; next, performing a selection operation on the population based on the fitness values, retaining individuals with excellent performance; subsequently, generating a new generation of weight combinations through crossover and mutation operations to expand the search space and improve global optimization capabilities; the above process continues iterating until the convergence condition is met or a preset number of generations is reached, ultimately obtaining the optimal loss weight combination, thereby improving the performance of the PINNs model in local feature extraction and overall prediction.

[0141] Specifically, by focusing on the noise propagation chain, a PINNs model with IPI physical constraints is used for road noise prediction, and a genetic algorithm (GA) is used to optimize the model loss weights (ω) in the loss function. data ) and IPI loss weight (ω) IPI This improves prediction accuracy.

[0142] Model loss weights (ω) data ) and IPI loss weight (ω) IPI The value range is [0,1], and ω satisfies data +ω IPI =1, reasonably allocate the weight of data fitting and physical laws.

[0143] The fitness function is determined by using the reciprocal of the mean absolute error (MAE) of PINNs in predicting noise in the driver's right ear as the fitness. The smaller the MAE, the higher the fitness, which means the better the model's prediction accuracy.

[0144] Generate an initial population containing a number of individuals, each individual forming a group (ω). data ω IPI For example: Individual 1: ω data =0.4, ω IPI =0.6; Individual 2: ω data =0.8, ω IPI =0.2; The weights of all individuals are randomly generated in the range [0,1], and satisfy ω data +ω IPI =1.

[0145] For each individual, the weights are substituted into the total loss function formula of PINNs. PINNs learns the mapping from "wheel vibration, k, and f" to "driver's right ear noise," and simultaneously... The constraint model is used to obtain the physical properties of the dynamic stiffness of the attachment point.

[0146] The model was validated by creating 20 new test samples, and the MAE of predicted noise and measured noise was calculated and then converted into fitness.

[0147] Sort the individuals by fitness from high to low, retain the 20 best individuals, randomly pair the 20 selected individuals, perform weighted cross-matching on each pair, and supplement the population to 50 individuals. For the newly generated 30 individuals, perform weighted mutation with a 5% probability.

[0148] The optimal weights are obtained by iterating through 100 generations of repeated training and fitness calculation.

[0149] The PINNs model is retrained based on the optimal weights.

[0150] Furthermore, step S3 includes:

[0151] Step S31: Define the passive side of the vehicle body attachment point as the excitation point of the transfer function, and the driver's right ear noise (20-300Hz) as the response point;

[0152] Step S32: Finite element modeling of interior and vehicle body;

[0153] Step S33: Finite element modeling of the acoustic cavity;

[0154] Step S34: Calculate the NTF (Network Transfer Function) for excitation and response points.

[0155] Furthermore, step S3 also includes:

[0156] Step S32 includes the following steps:

[0157] The OptiStruct module of Hypermesh software was used to mesh the accessory models, which included the body-in-white model and interior trim. Corresponding element properties and material parameters were then assigned to complete the finite element model of the interior and body. The accessory models included the body-in-white model and interior trim.

[0158] Step S33 includes the following steps:

[0159] Using the seat geometry model, body-in-white structure and closures, including side doors, sunroof, and tail door, as input, a finite element model of the acoustic cavity is established. The interior body model and the acoustic cavity model are then assembled to construct an acoustic-structure coupling finite element model.

[0160] Step S34 includes the following sub-steps:

[0161] A unit load of 1 N is applied in the X, Y, and Z directions at each loading point, and frequency response function analysis is performed to calculate the NTF of each response point relative to the loading point.

[0162] Specifically, the transmission relationship of noise from the passive end of the vehicle body attachment point to the driver's right ear inside the vehicle is obtained by matrix inversion. The matrix inversion formula is as follows.

[0163] ;

[0164] In the formula: Input the excitation vector to the system. The response vector at the response point, The transfer function from input to response;

[0165] Furthermore, step S4 includes:

[0166] Step S41: In the road noise hierarchy architecture, the force transmitted from the steering knuckle excitation to the passive end of the vehicle body is predicted using data-driven methods.

[0167] Step S42: In the road noise hierarchy architecture, the noise transmitted from the passive end of the vehicle body to the driver's right ear is predicted using TPA;

[0168] Step S43: Use an ensemble algorithm to develop a fast prediction interface.

[0169] Specifically, in the road noise hierarchy architecture (S41), the force transmitted from the steering knuckle excitation to the passive end of the vehicle body is predicted using a data-driven method. Key parameters required in each level, if not directly obtainable, are predicted and compensated by the traditional LSTM model constructed in the lower layer. Specifically, the force transmitted to the passive end of the vehicle body in the second level is modeled and predicted by the third level based on the PINNs model, achieving layer-by-layer mapping and feedback of the structural response.

[0170] Furthermore, including:

[0171] Step S41 includes the following steps:

[0172] The third-level deep neural network uses the vibration excitation signal of the steering knuckle as input to predict the vibration acceleration response and IPI of the active end connection point of the chassis system;

[0173] The second-level deep neural network uses the output of the third level and the bushing dynamic stiffness parameters as input to predict the transmitted force at the passive end connection point of the vehicle body system.

[0174] Specifically, step S42 includes the following sub-steps: further developing a visual, fast prediction interface, such as... Figure 6As shown in the figure, this interface integrates functional modules such as model loading, input parameter setting, prediction result output, and result visualization. Users can conveniently input relevant operating parameters (such as steering knuckle excitation, bushing dynamic stiffness, NTF, etc.) through the graphical interface, and quickly obtain the prediction results of the target response by combining the pre-trained LSTM and PINNs models with the input NTF.

[0175] The first level, based on the output of the second level, combines the NTF and TPA principles to construct the input, establishing a model for predicting the noise response of the driver's right ear. Combining steps S41 and S42, a prediction model integrating the mechanistic model and the data-driven method is constructed, and its prediction results are as follows... Figure 5 As shown in the figure. In comparison, this fusion method is closer to the true value in terms of prediction accuracy, demonstrating superior performance compared to single methods.

[0176] Further step S43, using the PyQt ensemble algorithm, involves developing a fast prediction interface, including: further developing a visual fast prediction interface, such as... Figure 8 As shown in the figure, this interface integrates functional modules such as model loading, input parameter setting, prediction result output, and result visualization. Users can conveniently input relevant operating parameters (such as steering knuckle excitation, bushing dynamic stiffness, NTF, etc.) through the graphical interface, and quickly obtain the prediction results of the target response by combining the pre-trained LSTM and PINNs models with the input NTF.

[0177] A schematic diagram of the established acoustic-structure coupling finite element model is shown below. Figure 7 As shown, the predicted curve is closest to the true value curve, indicating that the method has the highest prediction accuracy for in-vehicle noise and can better restore the frequency characteristics of the noise (such as the position and amplitude of peaks and valleys).

[0178] Figure 2 This is a structural diagram of a multi-vehicle testing application system provided by one or more embodiments of the present invention.

[0179] like Figure 2 The system shown includes:

[0180] Architecture building module, intelligent model building module, physical model building module, and model fusion prediction module;

[0181] The architecture building module is used to construct a road noise hierarchical decomposition architecture based on road noise characteristics.

[0182] The intelligent model building module is used to embed intelligent algorithms in the hierarchical decomposition architecture, obtain the correlation between target nodes at each level, and introduce key physical quantities to build a data-driven model that integrates physical priors.

[0183] The physical model building module is used to establish an acoustic-structure coupling finite element model, collect noise transfer functions, and build a TPA model.

[0184] The model fusion prediction module is used to fuse the data-driven model and the TPA model to achieve in-vehicle road noise prediction.

[0185] It is worth noting that although this device only discloses the architecture construction module, the intelligent model construction module, the physical model construction module, and the model fusion prediction module, it does not mean that this device is limited to the above-mentioned basic functional modules. On the contrary, what this invention intends to express is that, based on the above-mentioned basic functional modules, those skilled in the art can add one or more functional modules in combination with existing technology to form an infinite number of embodiments or technical solutions. That is to say, this system is open rather than closed. Just because this embodiment only discloses a few basic functional modules, it should not be considered that the scope of protection of the claims of this invention is limited to the above-disclosed basic functional modules.

[0186] Figure 9 This is a block diagram of an electronic device structure for a vehicle road noise prediction method based on machine-data fusion provided in one or more embodiments of the present invention.

[0187] like Figure 9 As shown, this application provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0188] The memory stores a computer program that, when executed by a processor, causes the processor to perform steps of a machine-data fusion method for predicting road noise in vehicles.

[0189] This application also provides a computer-readable storage medium storing a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of a machine-data fusion method for predicting vehicle road noise.

[0190] For the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0191] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle road noise prediction method using machine-data fusion, characterized in that, Includes the following steps: Step S1: Obtain road noise characteristics and construct a road noise hierarchical decomposition architecture based on the road noise characteristics; Step S2: Embed intelligent algorithms in the hierarchical decomposition architecture to obtain the correlation between target nodes at each level, and introduce key physical quantities to build a data-driven model that integrates physical priors. Step S3: Establish an acoustic-structure interaction finite element model, collect noise transfer functions, and construct a TPA model; Step S4: Integrate the data-driven model and the TPA model to generate prediction results for in-vehicle road noise.

2. The vehicle road noise prediction method based on machine-data fusion according to claim 1, characterized in that, Step S1 includes: Step S11: Based on the road noise characteristics, simplify the road noise into 20-300Hz structure sound for processing; Step S12: Based on the understanding of different suspension types and structures, and combined with the main influencing factors of road noise and their transmission paths, starting from the target response of the whole vehicle, the road noise is decomposed layer by layer in a top-down manner to construct a hierarchical decomposition architecture. Step S12 further includes: The first level of the hierarchical architecture is the noise response at the driver's right ear, which serves as the target evaluation indicator. The second level is the body system, whose parameters include the passive force transmitted at each attachment point between the chassis and the body. The sound transmission path from the passive side of the vehicle body attachment point to the driver's right ear is represented by the NTF between the first and second levels. The third level is the chassis system, whose parameters include the vibration response and IPI on the active side of the body attachment points; The fourth level is the excitation source layer, whose parameters include the vibration excitation signal at the steering knuckle and the dynamic characteristic parameters of the bushing.

3. The vehicle road noise prediction method based on machine-data fusion according to claim 1, characterized in that, Step S2 includes: Step S21: Collect or test data based on the shelf decomposition architecture; Step S22: Construct the LSTM model; Step S23: Incorporate IPI into the model loss function design in the LSTM output layer to construct PINNs; Step S24: Use a genetic algorithm to optimize the model loss weights and IPI loss weights in the loss function; Step S2 further includes: The collected data covers all key parameters in the hierarchical architecture except for NTF, specifically including: passive end force-frequency curve, active end acceleration response-frequency curve, noise response-frequency curve at the driver's right ear, and active end IPI-frequency curve; Step S22 includes the following steps: Train and build a third-level LSTM prediction model for chassis system performance; Train and build a second-level LSTM prediction model for vehicle body system performance; The unknown parameters at each level are predicted by the output of the next level, thus realizing bottom-up hierarchical recursive modeling; Step S23 includes the following steps: Based on the second-level LSTM prediction model of the vehicle body system performance, IPI is incorporated into the model loss function design to construct the physical information neural network PINNs.

4. The vehicle road noise prediction method based on machine-data fusion according to claim 1, characterized in that, Step S3 includes: Step S31: Define the passive side of the vehicle body attachment point as the excitation point of the transfer function, and the noise in the driver's right ear as the response point; Step S32: Finite element modeling of interior and vehicle body; Step S33: Finite element modeling of the acoustic cavity; Step S34: Calculate the NTF (Network Transfer Function) for excitation and response points.

5. The vehicle road noise prediction method based on machine-data fusion according to claim 4, characterized in that, Step S3 further includes: Step S32 includes the following steps: The attachment model is meshed, and corresponding element properties and material parameters are assigned to it to complete the establishment of the interior body finite element model. The attachment model includes: the body-in-white model and the interior trim; Step S33 includes the following steps: Using the seat geometry model, body-in-white structure and closure components as input, a finite element model of the acoustic cavity is established. The interior body model and the acoustic cavity model are then assembled to construct an acoustic-structure coupling finite element model. Step S34 includes the following sub-steps: A unit load of 1 N is applied in the X, Y, and Z directions at each loading point, and frequency response function analysis is performed to calculate the NTF of each response point relative to the loading point.

6. The vehicle road noise prediction method based on machine-data fusion according to claim 1, characterized in that, Step S4 includes: Step S41: In the road noise hierarchy architecture, the force transmitted from the steering knuckle excitation to the passive end of the vehicle body is predicted using data-driven methods. Step S42: In the road noise hierarchy architecture, the noise transmitted from the passive end of the vehicle body to the driver's right ear is predicted using TPA; Step S43: Use an ensemble algorithm to develop a fast prediction interface.

7. The vehicle road noise prediction method based on machine-data fusion according to claim 6, characterized in that, Step S4 further includes: Step S41 includes the following steps: The third-level deep neural network uses the vibration excitation signal of the steering knuckle as input to predict the vibration acceleration response and IPI of the active end connection point of the chassis system; The second-level deep neural network uses the output of the third level and the bushing dynamic stiffness parameters as input to predict the transmitted force at the passive end connection point of the vehicle body system. Furthermore, step S42 includes the following sub-steps: The first level builds upon the output of the second level, combining the NTF and TPA principles to construct the input and establish a model for predicting the noise response of the driver's right ear.

8. A vehicle road noise prediction system that integrates machine and data, characterized in that, include: Architecture building module, intelligent model building module, physical model building module, and model fusion prediction module; The architecture building module is used to construct a road noise hierarchical decomposition architecture based on road noise characteristics. The intelligent model building module is used to embed intelligent algorithms in the hierarchical decomposition architecture, obtain the correlation between target nodes at each level, and introduce key physical quantities to build a data-driven model that integrates physical priors. The physical model building module is used to establish an acoustic-structure coupling finite element model, collect noise transfer functions, and build a TPA model. The model fusion prediction module is used to fuse the data-driven model and the TPA model to achieve in-vehicle road noise prediction.

9. An electronic device, characterized in that, include: The processor, communication interface, memory, and communication bus are connected, with the processor, communication interface, and memory communicating with each other via the communication bus. The memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of the vehicle road noise prediction method based on machine-data fusion as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program executable by an electronic device, which, when run on the electronic device, causes the electronic device to perform the steps of the vehicle road noise prediction method according to any one of claims 1-7.

Images