Chassis abnormal sound processing method and device, electronic equipment and storage medium

Abstract

The invention provides a chassis abnormal sound processing method and device, electronic equipment and a storage medium, and relates to the technical field of vehicles. The method comprises the following steps: acquiring running data of a vehicle, wherein the running data comprises road condition data, chassis sound data and chassis vibration data in the running process of the vehicle; and then based on the operation data, through a pre-trained abnormal sound prediction model, obtaining and outputting chassis fault information including at least one of the position of a chassis part with abnormal sound, the abnormal sound type, the abnormal sound probability, the risk level and the maintenance suggestion of the vehicle. Wherein the abnormal sound prediction model is obtained by training a deep learning network model on the basis of historical real vehicle operation data and historical real vehicle chassis fault information obtained through a dual-circulation mechanism of a virtual chassis model based on digital twin. Through the above method, the robustness of the abnormal sound prediction model is improved, and the accuracy and efficiency of vehicle chassis abnormal sound detection are improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a method, apparatus, electronic device, and storage medium for dealing with chassis noise. Background Technology

[0002] In the design, development, and lifecycle management of passenger vehicles, chassis noise has always been a core pain point affecting vehicle quality and user satisfaction.

[0003] Existing methods for handling chassis noise mainly include: having engineers listen to the chassis noise phenomenon under different working conditions; or conducting noise road tests based on enterprise standards (such as Belgian roads, washboard roads, etc.) and four-post vibration table tests; or manually determining whether there is abnormal noise in the chassis and the source of the abnormal noise components by using auxiliary diagnostic tools such as stethoscopes or long-handled screwdrivers.

[0004] However, existing methods for dealing with chassis noise rely on manual judgment, which cannot efficiently and accurately control chassis noise. This restricts the early identification and source optimization of noise problems during the vehicle design and development stage, making it difficult to avoid potential noise risks before mass production. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and storage medium for processing chassis abnormal noises, so as to achieve accurate and efficient intelligent diagnosis and early warning of chassis abnormal noises, thereby improving the safety and economy of vehicle operation and maintenance.

[0006] In a first aspect, embodiments of this application provide a method for handling chassis noise, including:

[0007] Acquire vehicle operating data, including road condition data, chassis sound data, and chassis vibration data during vehicle operation;

[0008] Based on the operational data, the chassis fault information of the vehicle is obtained through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions. The abnormal noise prediction model is obtained by training a deep learning network model based on historical real vehicle operation data and historical real vehicle chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model.

[0009] Output the chassis fault information.

[0010] In one possible implementation, the virtual chassis model is constructed based on a multibody dynamics (MBD) model, a finite element analysis (FEA) model, and an acoustic simulation model.

[0011] In one possible implementation, the method further includes:

[0012] The virtual chassis model is driven by the historical prototype test data obtained from the test track of the corresponding vehicle model, and the prototype chassis state response data corresponding to the historical prototype test data is output.

[0013] The chassis status response data of the prototype vehicle is compared with the preset chassis response standard data to obtain the historical prototype vehicle chassis fault information corresponding to the historical prototype vehicle test data.

[0014] Based on the historical prototype chassis fault information, the parameters of the virtual chassis model are adjusted to obtain the processed virtual chassis model;

[0015] Based on the historical real-vehicle operation data corresponding to the vehicle model, the processed virtual chassis model is driven to output the real-vehicle chassis status response data corresponding to the historical real-vehicle operation data;

[0016] The actual vehicle chassis status response data is compared with the chassis response standard data to obtain the historical actual vehicle chassis fault information.

[0017] In one possible implementation, both the historical prototype test data and the historical real-vehicle operation data include road condition data, chassis sound data, and chassis vibration data. The historical prototype test data or the historical real-vehicle operation data drives the virtual chassis model to obtain corresponding chassis state response data, including:

[0018] Based on road condition data, the motion information of each component of the chassis is calculated using the MBD model.

[0019] Based on the motion information, the contact information and friction force change information between the various components of the chassis are obtained through the FEA model simulation.

[0020] Based on chassis sound data and chassis vibration data, the vibration transmission path and sound radiation characteristics are simulated using the acoustic simulation model.

[0021] Based on the motion information, the contact information, the friction force change information, the vibration transmission path, and the sound radiation characteristics, chassis state response data is obtained.

[0022] In one possible implementation, obtaining the vehicle's chassis fault information based on the operational data using a pre-trained abnormal noise prediction model includes:

[0023] The running data is preprocessed to obtain preprocessed running data. The preprocessing includes at least one of missing value filling, normalization, standardization and noise removal.

[0024] Feature extraction is performed on the preprocessed running data to obtain the first feature of the chassis sound data and the second feature of the chassis vibration data;

[0025] The first feature and the second feature are input into the abnormal noise prediction model for prediction to obtain the chassis fault information.

[0026] In one possible implementation, when the chassis fault information includes the probability of abnormal noise, the method further includes:

[0027] If the probability of abnormal noise is greater than a preset threshold, a warning message will be pushed to the user through at least one of the following: steering wheel vibration, in-vehicle display screen, in-vehicle voice prompt, and user terminal.

[0028] Secondly, embodiments of this application provide a chassis noise treatment device, comprising:

[0029] The first processing module is used to acquire vehicle operating data, which includes road condition data, chassis sound data, and chassis vibration data during vehicle operation.

[0030] The second processing module is used to obtain chassis fault information of the vehicle based on the operating data and through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions. The abnormal noise prediction model is obtained by training a deep learning network model based on historical real vehicle operating data and historical real vehicle chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model.

[0031] The third processing module is used to output the chassis fault information.

[0032] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0033] The memory stores computer-executed instructions;

[0034] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0035] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0036] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0037] This application provides a method, apparatus, electronic device, and storage medium for processing chassis abnormal noise. It acquires multi-source data generated during vehicle operation through onboard sensors and other devices, including road condition data reflecting road conditions, chassis sound data reflecting acoustic characteristics, and chassis vibration data reflecting mechanical characteristics. Next, this operational data is input into a pre-trained abnormal noise prediction model to obtain chassis fault information, including at least one of the following: location of the faulty component, type of abnormal noise, probability of occurrence, risk level, and repair suggestions. This abnormal noise prediction model is based on historical real-vehicle operating data and innovatively generates a large amount of accurately labeled historical chassis fault information through a virtual chassis model constructed using a digital twin and its dual-loop mechanism, which is then fully trained on a deep learning network. Finally, the chassis fault information is output to serve subsequent repair decisions or user prompts. The above methods effectively overcome the traditional limitations of obtaining real fault samples, which are difficult and costly. They provide sufficient and high-quality training data for deep learning models, thereby constructing a highly reliable abnormal noise prediction model. This enables accurate localization, type identification, and risk quantification of chassis abnormal noises, realizing the transformation from relying on human experience to data-driven intelligent diagnosis. It improves the efficiency, accuracy, and forward-looking early warning capabilities of vehicle chassis fault detection, effectively ensuring vehicle operation safety and reducing maintenance costs. Attached Figure Description

[0038] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0039] Figure 1 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 1 ;

[0040] Figure 2 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 2 ;

[0041] Figure 3 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 3 ;

[0042] Figure 4 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 4 ;

[0043] Figure 5A schematic diagram of the structure of a chassis noise treatment device provided in this application;

[0044] Figure 6 This is a schematic diagram of the structure of an electronic device provided in this application.

[0045] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0046] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0047] The application background of this application is explained as follows:

[0048] Chassis noise is a common but troublesome problem in the design and development of passenger vehicles. These issues often only surface during later testing phases or after receiving user feedback, leading to design rework, increasing development costs and extending project timelines. As a core component for vehicle power transmission, suspension support, and steering control, the chassis system is susceptible to noise problems due to factors such as friction between its parts, relative motion of rigid connections, or high-frequency movements of electronic actuators.

[0049] The main methods for dealing with chassis noises include the following:

[0050] 1) Subjective testing: Engineers listen to chassis noises under different operating conditions (such as cold start, low speed driving, braking operation, etc.) and combine this with their experience to determine the source of the noise. However, this method relies on human experience, has limited effectiveness, a high barrier to entry, lacks systematicity, and is difficult to develop into a reusable noise detection model.

[0051] 2) Objective Test 1: Simulating real-world road conditions through a noise and vibration test based on enterprise standards. Specifically, drivers drive the vehicle on a dedicated noise and vibration test track equipped with various special road surfaces, such as Belgian roads, manhole cover roads, speed bumps, twisting roads, and cobblestone roads, to simulate different real-world road conditions and stimulate potential noise and vibration issues at various chassis connection points, maximizing the reproduction of chassis noise and vibration problems that drivers might encounter in the real world. However, the drawback of this method is the high time and manufacturing cost of producing a prototype vehicle.

[0052] 3) Objective Test Two: Reproducing chassis vibration under specific road conditions through a four-post vibration table test based on enterprise standards. Specifically, the entire vehicle is placed on a four-post vibration table, and pre-collected real road load data is input to simulate the chassis vibration of a vehicle on a real road. This method can accurately reproduce the same working condition countless times, eliminating interference from inconsistencies in vehicle speed and road surface during actual vehicle testing; it also allows for individual control of the excitation of a specific wheel to help determine whether chassis noise originates from the front or rear axle; furthermore, engineers can perform professional diagnostics in real time inside or under the vehicle. However, this method also suffers from drawbacks such as high time and manufacturing costs for prototype vehicles and chassis systems.

[0053] 4) Using diagnostic tools: Manually determine the presence of abnormal noises in the chassis and the source of the noise components using auxiliary diagnostic tools such as a stethoscope or a long-handled screwdriver. For example, place a stethoscope against key chassis components such as shock absorbers, half-shaft CV joints, and steering gear, and manually determine the source of the noise under different operating conditions to locate the noisy components. Alternatively, manually place a long-handled screwdriver against suspected chassis noise components such as pulleys and suspension linkages, and determine whether the component is loose or worn by listening to the sound. In addition, an on-board diagnostic (OBD) system can be used to read codes such as transmission gear ratio error P0730 or codes related to bearing wear to help determine whether there are chassis noise problems and the source of the noisy components.

[0054] 5) Simulation Testing: During the vehicle design phase, a full-vehicle finite element analysis (FAE) model covering key components of the chassis system is established. After model correction through experimental benchmarking, road noise or abnormal noise road surface excitation loads collected from the prototype vehicle are used to analyze the sensitivity, transfer function, modal characteristics, and other indicators of chassis components. This identifies the design parameters with the greatest impact on abnormal noise, and then proposes targeted optimization schemes such as modifying the structure, adjusting stiffness, damping, and elastic parameters. Finally, the simulation results are verified by comparing the changes in in-vehicle noise before and after optimization or by subjectively evaluating the improvement of chassis abnormal noise. This method aims to determine the existence of chassis abnormal noise problems through a combination of subjective and objective methods, reducing manufacturing and system iteration costs through virtual verification. However, this method requires accumulating a sufficient database of abnormal noise road surfaces and loads, and has certain requirements for simulation test benchmarking experience. At the same time, the simulation computation is large, especially for the accurate simulation of complex nonlinear contact and friction, which requires a lot of computational resources. If the model is too simplified, it may ignore key clearance and friction factors in the early design. In addition, this method lacks probabilistic assessment of random factors such as manufacturing tolerances and changes in material properties.

[0055] 6) CAE-based offline simulation analysis methods: For example, a car manufacturer uses a method combining multibody dynamics simulation and finite element analysis to predict chassis noise risks during passenger vehicle design and development. However, this method cannot achieve data linkage with real vehicle data, the model accuracy relies on manual verification, and it cannot cover all actual user scenarios.

[0056] The above-mentioned methods for testing chassis noise have the following drawbacks:

[0057] 1) Model silos: The simulation model is disconnected from physical vehicle test data. The model has limited accuracy and is fixed once established, making it unable to self-update and correct using massive amounts of test data, resulting in low prediction confidence. 2) Scenario limitations: It is difficult to cover all road conditions and real vehicle driving scenarios. Moreover, the time and manufacturing costs of collecting abnormal noise data by manufacturing physical prototypes for road testing are high. 3) Lack of real-time data drive: It is impossible to continuously optimize the model using data from actual vehicle operation, resulting in lagging prevention and control strategies and only being able to passively respond to abnormal noise problems. 4) Poor adaptability: It is difficult to effectively locate the abnormal noise sources of the chassis system of new generation new energy vehicles. For example, the new structures and functions of intelligent vehicles and drive-by-wire chassis (such as frequent operation of active suspension, such as drive-by-wire braking, active stabilizer bars) have more electronic and electrical components and frequent actuator operation, making the abnormal noise sources more complex. 5) Failure of cross-link coordination: The fragmented mode of front-end design and back-end remediation leads to long rectification cycles and huge costs.

[0058] Against this backdrop, providing a closed-loop intelligent simulation analysis and abnormal noise prediction system covering the entire lifecycle of the vehicle chassis to achieve more efficient and accurate integrated chassis abnormal noise risk prediction is an urgent technical problem to be solved.

[0059] Based on the aforementioned technical problems, the inventors, in the process of researching a chassis noise treatment method covering the entire lifecycle of a vehicle chassis, discovered that the existing Siemens Teamcenter platform supports building a digital twin model of the chassis system for monitoring the chassis's response status. Furthermore, by combining a multibody dynamics model, finite element method, and acoustic simulation model to obtain a virtual chassis model, a dual-loop iterative mechanism is constructed, consisting of an inner loop composed of a simulation-verification closed loop based on prototype vehicle test data and an outer loop composed of a scenario expansion closed loop based on real vehicle test data. This enables dynamic calibration of model parameters, thereby improving the model's generalization ability. Further, based on the virtual chassis model, highly efficient and accurate proactive prediction of chassis noise is achieved, systematically solving the industry problems of model silos, data barriers, and insufficient scenario coverage. Based on this, this application provides a chassis noise treatment method, device, electronic device, and storage medium.

[0060] This application provides a method for handling chassis noise, applicable to the entire lifecycle management of passenger vehicle chassis systems from design, verification, market launch to operation, especially for new systems such as drive-by-wire chassis, active suspension, and drive-by-wire braking in intelligent connected vehicles.

[0061] The technical solution of this application and how it solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0062] Figure 1 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 1 ,like Figure 1 As shown, the method includes:

[0063] S101: Acquire vehicle operating data, including road condition data, chassis sound data, and chassis vibration data during vehicle operation.

[0064] In this step, road condition data refers to the physical characteristics and environmental information describing the road on which the vehicle travels, including key parameters such as road smoothness, slope, road surface material (e.g., asphalt, cement, gravel), and obstacles. Chassis sound data refers to the acoustic signals generated by the vehicle's chassis system during vehicle operation. The sound frequency and amplitude differ between normal and faulty states for different chassis components, which is the core basis for identifying abnormal noises. Chassis vibration data refers to the vibration signals generated by various chassis components due to operation, road surface excitation, or their own malfunctions. The frequency, amplitude, and phase characteristics of the vibrations can reflect the structural state and operational abnormalities of the chassis components.

[0065] For example, triaxial vibration acceleration sensors deployed on structural components such as suspension control arms, steering knuckles, and subframes can measure chassis vibration data in three spatial dimensions, including vibration frequency and amplitude characteristics. Waterproof and shockproof microphone arrays installed in key areas prone to abnormal noise, such as the front and rear wheel wells and the center of the chassis, can be used to collect chassis sound data at a specific sampling frequency (e.g., ≥20kHz) using beamforming technology. Real-time vehicle location information can be obtained through an onboard Global Positioning System (GPS) or BeiDou positioning module, combined with high-precision electronic maps to obtain information including road attributes and road surface material. Furthermore, the location, size, and distribution of road obstacles can be identified in real time using a vehicle-mounted forward-facing camera or LiDAR. In addition, operational data can also include basic information such as vehicle speed, acceleration, and angular motion state measured by an onboard gyroscope. Ultimately, the vehicle operation data obtained through the aforementioned onboard sensors is aggregated to the onboard terminal via the onboard Controller Area Network (CAN) bus or Ethernet, and then transmitted to the cloud platform via 5G / Vehicle to Everything (V2X) or higher-speed network transmission systems, providing a data foundation for subsequent fault analysis and model calculations.

[0066] S102: Based on operational data, obtain chassis fault information of the vehicle through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance recommendations.

[0067] Among them, the abnormal noise prediction model is obtained by training a deep learning network model based on historical operating data and historical chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model.

[0068] In this step, the abnormal noise prediction model is an intelligent diagnostic model built based on a deep learning network model. It is used to take the vehicle's operating data as input and output chassis fault information including at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions, so as to achieve accurate handling of chassis abnormal noise.

[0069] Historical operating data refers to the collection of past data reflecting the vehicle's operating status accumulated through various test scenarios before the abnormal noise prediction model is trained. It is the core input foundation for the model to learn the mapping rules between fault characteristics and data. Historical operating data includes real vehicle test data for the corresponding vehicle model and prototype vehicle test data based on the test track, covering road condition data, chassis sound data, chassis vibration data, etc. under different operating conditions.

[0070] A virtual chassis model based on a digital twin refers to a virtual model that is highly homologous to the physical vehicle chassis in terms of structure, performance, and operating status, and can simulate the normal operating characteristics and fault manifestations of the chassis under various working conditions. The dual-loop mechanism is a closed-loop system that supports the optimization of the virtual chassis model and the accumulation of operating data for scenario expansion, including an inner loop and an outer loop. The inner loop covers the design verification stage, where the virtual chassis model outputs chassis state response data driven by a test library (such as data from a four-post test bench), and the model parameters are adjusted in conjunction with enterprise standards. The outer loop covers the market operation stage, where the virtual chassis model outputs chassis state response data driven by real-vehicle operating data from different vehicles of the same model driven by drivers, and the database is expanded using real-vehicle test data. This dual-loop mechanism collaboratively ensures the accuracy and generalization ability of the virtual chassis model, thereby improving data quality. This allows for the training of a deep learning network model based on historical operating data and historical chassis fault information corresponding to the historical operating data obtained through the virtual chassis model, resulting in a more reliable and robust abnormal noise prediction model.

[0071] For example, the deep learning network model may include Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Generative Adversarial Network (GAN), etc. Among them, the CNN model can be used to extract the spectral features of chassis sound data and chassis vibration data; the LSTM model can be used to learn long-term dependencies in time-series data; the GAN model is used to generate simulated data that approximates real chassis fault information data through the generator, and the discriminator is used for real chassis fault information data and simulated data. The abnormal noise prediction model is optimized through the game between the generator and the discriminator.

[0072] In the process of training the deep learning network model based on historical operating data and historical chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model, the historical operating data accumulated by the dual-loop mechanism (including road condition data, chassis sound data, and chassis vibration data from experimental library data and real vehicle test data) is used as input data, and the corresponding historical chassis fault information (chassis response data output by the virtual model combined with the location of abnormal noise components, abnormal noise type, abnormal noise probability, risk level, and matching maintenance suggestions determined by enterprise standards) is used as label data. Together, they are divided into training set, validation set, and test set to train the deep learning network model. This deep learning network model receives historical operating data through the input layer. After convolution, pooling, and fully connected operations in the hidden layers, it automatically extracts deep features strongly correlated with abnormal noise faults from the historical operating data. These features include abnormal frequency components in chassis sound data, characteristic amplitude and phase changes in vibration data, and correlation patterns between road condition data and specific faults. The output layer then outputs the prediction results, which are compared with historical chassis fault information in the label data. The prediction error is calculated based on the loss function, and the network weights and bias parameters are continuously adjusted through the backpropagation algorithm. The model is iteratively trained until the prediction accuracy, risk level determination accuracy, and maintenance suggestion matching degree of the deep learning network model on the validation set all reach the preset standards. At this point, the training and solidification of the abnormal noise prediction model is complete.

[0073] Throughout the process, the dual-loop mechanism ensures the comprehensiveness, authenticity, and timeliness of the training data, so that the abnormal noise prediction model trained on the deep learning network based on the training data has higher robustness. The synergistic effect of the two ensures the accuracy, completeness, and practicality of chassis fault prediction.

[0074] S103: Output chassis fault information.

[0075] The chassis noise processing method provided in this application involves collecting road condition data, chassis sound data, and chassis vibration data generated during actual vehicle operation using multimodal vehicle-mounted sensors deployed at key parts of the vehicle chassis. This data is then transmitted in real-time to the cloud via vehicle network and wireless communication technology. Based on the operational data, a pre-trained noise prediction model is used to predict and output chassis fault information, including at least one of the following: the location of the chassis components exhibiting the noise, the type of noise, the probability of the noise, the risk level, and repair suggestions. The noise prediction model is trained on a deep learning network model using historical operational data and historical chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model. This method changes the traditional, inefficient fault diagnosis model that relies on manual experience, achieving accurate, efficient, and forward-looking prediction of chassis noise. By applying a dual-loop mechanism based on a virtual chassis model using a digital twin, the abnormal noise prediction model trained on the deep learning network using training data obtained from the virtual chassis model becomes more robust, improving the reliability of chassis abnormal noise prediction. This guides maintenance personnel to quickly locate and resolve problems, shortens maintenance cycles, reduces maintenance costs, and effectively enhances the operational safety and reliability of vehicles.

[0076] Figure 2 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the method for handling the abnormal noise of the chassis is described in detail, and the method further includes:

[0077] S201: Based on the historical prototype test data obtained from the test track of the corresponding vehicle model, the virtual chassis model is driven, and the prototype chassis state response data corresponding to the historical prototype test data is output.

[0078] In this step, historical prototype test data refers to data collected during performance testing and fault simulation of the prototype vehicle at a professional automotive proving ground equipped with standardized test surfaces and operating condition simulation facilities based on enterprise standards. This data, or four-pillar test data, includes road condition data from the proving ground, chassis sound data, and chassis vibration data. The prototype test data covers various road conditions (such as bumpy roads, gravel roads, and sloping roads), extreme operating conditions (such as high-speed emergency braking and extreme cornering), and specific fault conditions (such as artificially induced chassis component wear and loosening). It is characterized by controllable operating conditions, distinct features, and high repeatability.

[0079] Specifically, a virtual chassis model is driven by historical prototype vehicle test data. Based on its own simulation logic and parameter settings from the prototype vehicle test data, the virtual chassis model simulates the chassis operation of the prototype vehicle under corresponding test conditions. This involves reproducing the road surface excitations and component interactions experienced by the prototype vehicle to calculate the dynamic response of the chassis system. The final output of the prototype vehicle chassis state response data represents the simulation feedback results of the virtual chassis model to the prototype vehicle test conditions. This includes motion state data of various chassis components (such as the compression and rebound speed of the suspension system, the speed and torque of the driveshaft, and the deflection angle of the steering knuckle), mechanical response data of the chassis (such as the force on the suspension springs, the damping force variation of the shock absorbers, and the stress distribution and strain values ​​of the chassis structural components), and acoustic response characteristic data of the chassis (such as the acoustic state responses of various chassis components).

[0080] The aforementioned prototype chassis status response data closely matches the actual working conditions of historical prototype tests. It can reflect the normal operating status of the prototype chassis under normal test conditions, and also output the corresponding abnormal state response characteristics in test conditions where specific faults are set in the prototype. This provides an accurate simulation reference for subsequent comparison with real vehicle test results and verification of virtual chassis model parameters.

[0081] S202: Compare the prototype chassis status response data with the preset chassis response standard data to obtain the historical prototype chassis fault information corresponding to the historical prototype test data.

[0082] In this step, chassis response standard data refers to a pre-defined set of benchmark data based on the target vehicle's design specifications, enterprise standards, and a large amount of measured verification data. This includes threshold values ​​for various response parameters of the chassis under normal operating conditions (such as normal vibration frequency range, acoustic signal amplitude standards, and mechanical stress limits), as well as characteristic response data templates under typical fault conditions. It serves as the core reference for determining whether the chassis is in normal condition and whether a fault exists. An example of chassis response standard data is shown in Table 1.

[0083] Table 1. Exemplary chassis response standard data

[0084]

[0085] The chassis status response data of the prototype vehicle is compared with the preset chassis response standard data. Specifically, if a certain type of response data exceeds the normal threshold range or highly matches a certain type of fault feature template, the corresponding fault information is further determined by combining the judgment rules associated in the standard data. For example, by mapping the spatial positioning information of the acoustic response feature data with the position of the faulty component in the standard data, the location of the chassis component with abnormal noise is located (e.g., the "humming" sound feature in the front wheel hub area is matched, and it is determined to be a fault related to the front wheel hub bearing). By analyzing the specific performance of data such as spectral features and time-domain waveforms, the corresponding abnormal noise type is matched (e.g., high-frequency pulse sound pressure peaks correspond to "clicking" sounds, and mid-to-low frequency continuous peaks correspond to "humming" sounds). Based on the degree of data deviation and feature matching degree, the probability of abnormal noise for this fault is calculated. Then, according to the functional importance of the faulty component (e.g., the braking system and steering system are key components) and the degree of impact of the fault on driving safety, the corresponding risk level is determined by referring to the risk level classification rules based on enterprise standards (e.g., braking system faults are judged as high risk, and slight loosening of suspension components is judged as medium risk).

[0086] If no abnormal response data is found after comparison (i.e., all parameters are within the normal threshold range and no fault feature template matches), it is determined to be fault-free, and the corresponding first historical chassis fault information is marked as "no abnormal noise fault". Finally, the above comparison results are structured and organized to form historical prototype chassis fault information containing core contents such as whether there is abnormal noise, the location of chassis components with abnormal noise, the type of abnormal noise, the probability of abnormal noise, and the risk level. This information is associated with the corresponding prototype test data one by one, and together they constitute the input-label data pairs required for deep learning network training.

[0087] S203: Based on historical prototype chassis fault information, adjust the parameters of the virtual chassis model to obtain the processed virtual chassis model.

[0088] Based on feedback from historical prototype chassis fault information, the parameter deviations of the virtual chassis model are automatically corrected, achieving iterative optimization of the model's simulation accuracy. In other words, the deviation between the prototype chassis state response data obtained from the virtual chassis model, driven by prototype test data, and the standard chassis response data is fed back, automatically and iteratively adjusting model parameters (such as bushing stiffness and damping coefficient). After parameter adjustment, closed-loop verification is performed: the original prototype test data is re-input into the adjusted virtual chassis model to generate new chassis state response data, which is then compared with preset standard chassis response data to obtain new fault information. If the deviation between the new fault information and the standard data narrows to a preset range, the parameter tuning is completed, resulting in a processed virtual chassis model. For example, optimization algorithms based on neural networks or genetic algorithms can be used to automatically adjust the parameters of the virtual chassis model, ensuring that the output of the virtual chassis model continuously approaches the standard chassis response data. This enables the model to have self-learning and self-calibration capabilities, and the model accuracy increases with data accumulation, breaking the island effect of traditional models and gradually solving the problem of accuracy stability.

[0089] In one possible implementation, if the probability of abnormal noise is greater than a preset abnormal noise probability threshold, or if the abnormal noise risk level has reached a high risk, the system can autonomously carry out multi-parameter and multi-objective optimization iterations through machine learning algorithms such as reinforcement learning. This will quickly select the optimal improvement solutions, such as adjusting bushing stiffness, changing clearance values, and optimizing material pairing. At the same time, the effectiveness of the solutions will be evaluated, providing engineers with accurate intelligent decision support and helping to efficiently solve abnormal noise problems.

[0090] Taking the need to adjust bushing stiffness as an example, suppose that during the prototype vehicle test, the measured vibration amplitude of the front suspension under bumpy road conditions (20Hz road excitation) was 0.8mm, while the current virtual chassis model, after inputting the same road excitation data, outputs a vibration amplitude of 1.2mm (the simulation value is 40% larger than the actual value), and the first historical chassis fault information shows "vibration response deviation exceeds the standard". It is understandable that, as a buffer component, the stiffness parameter of the front suspension bushing directly affects the vibration transmission efficiency. The greater the bushing stiffness, the more obvious the vibration transmission and the larger the amplitude; the smaller the stiffness, the better the vibration buffering effect and the smaller the amplitude. At this point, the automatic calibration system calculates the parameter adjustment direction using the gradient descent algorithm: reducing bushing stiffness to enhance the buffering effect. Initially, the radial stiffness of the front suspension bushing is adjusted from the initial value of 200 N / mm to 160 N / mm according to the deviation ratio (40%), and then the adjusted parameters are substituted into the model for resimulation. If the newly output vibration amplitude is 0.9 mm (the simulated value is 12.5% ​​larger than the actual value), the stiffness is further reduced to 145 N / mm according to the remaining deviation ratio. After another simulation, the vibration amplitude is 0.82 mm (deviation 2.5%). Finally, after 3 to 5 iterations, the bushing stiffness is adjusted to 142 N / mm. At this point, the deviation between the simulated vibration amplitude and the measured value of 0.8 mm is ≤3% (reaching the preset range), completing the automatic calibration of the bushing stiffness and obtaining the processed virtual chassis model. This ensures that the processed virtual chassis model can more accurately reproduce the operating state and fault characteristics of the real chassis, providing more reliable support for subsequent data processing of the outer loop and deep learning network training.

[0091] S204: Based on the historical real-vehicle operation data of the corresponding vehicle model, the virtual chassis model is processed and outputs the real-vehicle chassis status response data corresponding to the historical real-vehicle operation data.

[0092] In this step, historical real-vehicle operating data refers to data collected by drivers in actual road environments, using a vehicle identical to the prototype. This data covers road conditions in real-world driving scenarios (such as the smoothness, gradient, Belgian roads, speed bumps, washboard roads, twisting roads, mountain roads, and gravel roads), chassis sound data, and chassis vibration data, accurately reflecting the vehicle's operating status in actual use. Furthermore, historical real-vehicle data is used to expand the boundaries of simulation verification, enabling the discovery and prevention of abnormal noise problems that are unthinkable in a laboratory setting, achieving full-scenario coverage.

[0093] It is understandable that the virtual chassis model driven by historical real vehicle operation data outputs real vehicle chassis status response data, which is similar in principle and effect to the virtual chassis model driven by historical prototype test data in S201, and outputs prototype chassis status response data. It will not be elaborated here.

[0094] S205: Compare the actual vehicle chassis status response data with the chassis response standard data to obtain historical actual vehicle chassis fault information.

[0095] Historical real-vehicle operation data breaks through the limitations of historical prototype test data based on the test track, enabling the subsequent chassis noise prediction scenarios to be extended from the test track preset scenarios to actual use scenarios. This allows the noise prediction model, which is trained on a deep learning network model based on historical real-vehicle operation data and historical chassis fault information, to adapt to more complex actual driving conditions, thereby improving the reliability of the noise prediction model.

[0096] It is understandable that comparing the actual vehicle chassis status response data with the chassis response standard data to obtain historical actual vehicle chassis fault information is similar to the technical principle and effect of comparing the prototype vehicle chassis status response data with the preset chassis response standard data in S202 to obtain historical prototype vehicle chassis fault information, and will not be elaborated here.

[0097] The chassis noise processing method provided in this application is a refinement of the dual-loop mechanism based on a digital twin virtual chassis model. Specifically, it includes: First, using historical prototype vehicle test data collected in a professional test track, covering various standards and extreme operating conditions, to drive an initial virtual chassis model and simulate and output corresponding prototype vehicle chassis state response data; then, comparing the prototype vehicle chassis state response data with preset chassis response standard data to obtain historical prototype vehicle chassis fault information, and adjusting the parameters of the virtual chassis model based on this information to obtain a processed virtual chassis model. Next, this processed virtual chassis model is used in the outer loop, driven by massive amounts of historical real-vehicle operating data collected in a real road environment, outputting real-vehicle chassis state response data that better reflects the actual situation. By comparing this data with the chassis response standard data, high-quality historical real-vehicle chassis fault information tags are generated for model training. Using the methods described above, a complete closed loop of "simulation comparison, model parameter tuning, and reapplication" for the virtual chassis model driven by digital twins was constructed and verified. This significantly improved the simulation fidelity and reliability of the virtual chassis model, ensuring that the chassis fault information labels generated by the model have extremely high accuracy. This lays a solid data foundation for the subsequent training of a high-performance abnormal noise prediction model. Furthermore, the dual-loop mechanism effectively overcomes the industry bottlenecks of scarce real-vehicle fault samples and high acquisition costs. Through continuous self-optimization of the model, it significantly enhances the generalization ability and robustness of the abnormal noise prediction model in dealing with complex and ever-changing real-world scenarios. Ultimately, it achieves accurate, efficient, and forward-looking processing of chassis abnormal noise across the entire chain, from source perception and virtual verification to intelligent diagnosis.

[0098] Figure 3 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 3 ,like Figure 2 As shown, based on the above embodiments, historical prototype test data or historical real vehicle operation data drive the virtual chassis model to obtain corresponding chassis state response data, specifically including:

[0099] In one possible implementation, the virtual chassis model is constructed based on the multibody dynamics (MBD) model, the finite element analysis (FEA) model, and the acoustic simulation model.

[0100] Multi-body dynamics (MBD) models are virtual models built based on the principles of multi-body system dynamics. They are used to describe the kinematic relationships, force transmission, and constraints between multiple rigid bodies (such as suspension arms, steering knuckles, wheels, and frames) in a vehicle chassis. MBD models simulate the overall dynamic response of the chassis system (such as displacement, velocity, acceleration, and forces), accurately reproducing the chassis's motion under different operating conditions. Finite element analysis (FEA) models are refined simulation models built based on the finite element analysis method. By discretizing key chassis structural components (such as bushings, shock absorbers, chassis frames, and brake discs) into a finite number of elements, FEA models analyze the local mechanical properties of these components under stress, strain, and deformation. This allows for the capture of changes in the mechanical properties of components caused by structural damage and wear. Acoustics simulation models are simulation models built on acoustic theories (such as wave theory and sound radiation theory) to simulate the acoustic response of a chassis system, including noise generated by component vibration, sound transmission path in the chassis structure, sound radiation characteristics, etc. It can quantify the frequency, amplitude, sound pressure level and other characteristics of the output sound signal, providing support for abnormal noise identification.

[0101] Understandably, historical prototype test data and historical real vehicle operation data both include road condition data, chassis sound data, and chassis vibration data.

[0102] S301: Based on road condition data, the motion information of each component of the chassis is calculated through the MBD model.

[0103] For example, road condition data is first standardized to convert road condition information (such as road surface smoothness, slope, undulation amplitude, impact frequency, etc.) under different scenarios such as urban roads, highways, and bumpy roads into a road excitation input format that can be recognized by the MBD model. This input format can be defined at the contact point between the tire and the ground in the form of displacement excitation or force excitation. The MBD model is based on the principle of multibody dynamics. By solving the system dynamic equations, it simulates the overall motion of the chassis and the interaction between components under the corresponding road excitation. For example, taking the excitation caused by road undulations as an example, the excitation is transmitted to the suspension system through the tire. The MBD model calculates the swing angle of the suspension arm, the spatial displacement of the steering knuckle, and the vertical bounce of the wheel according to the preset hard point coordinates, bushing stiffness, and constraint relationships. For road condition changes corresponding to rapid acceleration, braking, or turning, the dynamic parameters such as linear velocity, angular velocity, linear acceleration, and angular acceleration of each component are output simultaneously.

[0104] Throughout the calculation process, the MBD model outputs motion information of each chassis component in real time at a set time step (e.g., 1ms), covering core data such as displacement, velocity, and acceleration of key chassis components. This reflects the motion state of chassis components under different road conditions and provides an accurate kinematic basis for subsequent analysis of component mechanical response using the finite element model and simulation of abnormal noise characteristics using the acoustic model.

[0105] S302: Based on motion information, the contact information and friction force change information between various chassis components are obtained through FEA model simulation.

[0106] In this step, the component motion information output by the MBD model is imported into the FEA model through the simulation interface, serving as the model's dynamic boundary conditions and excitation sources.

[0107] For example, the displacement change data of the suspension arm is mapped to its connection hard point with the subframe, driving the suspension arm to swing according to the actual motion trajectory; the acceleration data of the wheel is transmitted to the lower end of the shock absorber to simulate the extension and contraction motion of the shock absorber under road excitation. In the solution settings, the FEA model supports detailed simulation of key motion areas, accurately simulating the dynamic contact behavior and frictional changes between components, focusing on gap friction, preloading, assembly state influence, and nonlinear effects.

[0108] The FEA model iteratively calculates based on dynamic equations and contact mechanics theory at a set time step (synchronized with the MBD model, e.g., 1 ms). Within each time step, the FEA model first updates the spatial position of each component based on the input motion excitation, then calculates the contact state between components (whether they are in contact, contact area, contact pressure) and the magnitude of frictional force, thereby solving for the stress distribution, strain field, and local deformation of each chassis component, among other mechanical response data. Simultaneously, it captures real-time dynamic changes in key areas, such as the stiffness decay of bushings during repeated compression-rebound processes, the temperature rise and wear of the brake disc friction surface, and the local stress concentration at the suspension arm hinge due to clearance friction. After the simulation, it outputs the dynamic contact force curves of the refined region and the frictional force variation data over time. This not only accurately reflects the actual working state of the chassis components but also captures the nonlinear mechanical characteristics caused by factors such as clearance friction, insufficient preloading, and assembly deviations, providing mechanical data support for subsequent acoustic model simulation of abnormal noises and identification of fault root causes.

[0109] S303: Based on chassis sound data and chassis vibration data, the vibration transmission path and sound radiation characteristics are obtained through acoustic simulation model.

[0110] Specifically, the chassis sound data and chassis vibration data are first preprocessed to obtain the spectral characteristics of the chassis sound data in the frequency domain, as well as the frequency, amplitude, and phase characteristics of the chassis vibration data, so that the chassis sound data and chassis vibration data meet the excitation requirements of the acoustic simulation model.

[0111] For example, preprocessed chassis vibration data is used as the vibration excitation source for the acoustic simulation model. The model's pre-defined vibration transmission path analysis module, combined with the chassis's structural topology (such as the connection points between the suspension system and the frame, and the fixed positions of transmission components and the vehicle body), simulates the entire process of vibration transmission from a specific excitation source (such as bushing friction, brake disc vibration, or suspension component collision) to the vehicle body via chassis structural components like suspension arms and frame beams, as well as connecting components like bolts and bushings. By calculating the transfer function for each transmission path, which reflects the attenuation or amplification of vibration within the path, the vibration transmission paths are identified, determining which chassis components' vibrations will be transmitted to the vehicle body and ultimately radiate into the passenger compartment. Simultaneously, the vibration transmission efficiency of each path is quantified (such as the transmission loss values ​​of vibrations at different frequencies in each path), clarifying the distribution and transmission patterns of vibration energy within the chassis system.

[0112] Based on vibration velocity distribution data transmitted to the vehicle body or chassis surface, combined with chassis sound data, an acoustic simulation model is used to calculate the sound pressure distribution, sound radiation power, and other sound radiation characteristics of the chassis surface. For example, high-frequency vibrations generated by bushing clearance friction radiate a specific frequency "squeaking" sound through the surface of the metal parts connected to it. The model can output the sound pressure level as a function of frequency and the location coordinates of the sound source on the chassis surface. Meanwhile, vibrations caused by brake disc deformation are transmitted to the steering knuckle through the brake caliper, radiating a low-frequency "thumping" sound. The model can quantify the radiation power and propagation direction of this sound to predict the noise contribution at specific locations inside or outside the vehicle.

[0113] The vibration transmission path and sound radiation characteristics data obtained by the acoustic simulation model clearly reveal the source, transmission process and radiation law of chassis abnormal noise, providing a precise acoustic data basis for subsequent identification of abnormal noise type and location of faulty parts.

[0114] S304: Based on motion information, contact information, friction force change information, vibration transmission path and sound radiation characteristics, chassis state response data is obtained.

[0115] In this step, motion information forms the basis of the overall chassis operating status, reflecting the macroscopic motion characteristics of each chassis component. Contact information and friction force change information reflect the interaction behavior between chassis components, while vibration paths and sound radiation characteristics are the acoustic quantitative representations of chassis noise phenomena. Therefore, the chassis state response data output by the virtual chassis model provides a deep integration and structured presentation of comprehensive information on chassis kinematics, mechanics, and acoustics, offering multi-dimensional and comparable data support for subsequent fault identification.

[0116] The chassis noise processing method provided in this application embodiment involves refining the corresponding chassis state response data by driving a virtual chassis model with historical prototype test data or historical real vehicle operation data. Specifically, it includes: using road condition data to drive a multibody dynamics model to calculate the macroscopic motion information of each chassis component, such as displacement, velocity, and acceleration, thereby accurately replicating the overall dynamic behavior of the real chassis under specific road conditions. Subsequently, this motion information is input as dynamic boundary conditions into a finite element model. This model, through refined structural mechanics simulation, reveals the dynamic contact behavior between components, frictional changes, and the resulting stress-strain distribution, thereby capturing local mechanical characteristics that may lead to abnormal noise, such as bushing stiffness attenuation or abnormal brake disc wear. Simultaneously, chassis sound data and chassis vibration data are input into an acoustic simulation model. This model, based on acoustic theory, simulates the transmission path of vibration in the complex chassis structure and calculates the final sound radiation characteristics, thereby reproducing the generation and propagation process of abnormal noise in a virtual environment and quantifying key acoustic indicators such as frequency and sound pressure level. Ultimately, chassis condition response data, including kinematic information, contact and tribological information, vibration transmission path, and acoustic radiation characteristics, were obtained. This method ensured the high accuracy and interpretability of the chassis condition response data, providing a data foundation for subsequent comparison with standard data to generate reliable fault information labels, and enhancing the quality and reliability of the training data for the abnormal noise prediction model.

[0117] Figure 4 A flowchart illustrating a method for handling chassis noise provided in this application. Figure 4 ,like Figure 4 As shown, in this embodiment... Figure 1 Based on the previous embodiment, in step S102, based on operational data, the vehicle's chassis fault information is obtained through a pre-trained abnormal noise prediction model, specifically including:

[0118] S401: Preprocess the running data to obtain preprocessed running data. Preprocessing includes at least one of missing value filling, normalization, standardization, and noise removal.

[0119] As mentioned in S101, the vehicle operation data acquired by onboard sensors includes road condition data, chassis sound data, and chassis vibration data. This data may contain interference factors such as environmental noise, electromagnetic interference, and random sensor errors, and may also have issues such as missing data and inconsistent numerical ranges. Therefore, it is necessary to preprocess the operation data to obtain high-quality, standardized data, providing a foundation for accurate predictions by the subsequent abnormal noise prediction model.

[0120] For example, noise removal includes using filtering algorithms such as Kalman filtering and wavelet transform to separate and remove irrelevant signals such as environmental wind noise, road clutter, and electromagnetic interference, while retaining valid data related to chassis operation (such as chassis noise data in the 20~20000Hz frequency band). Missing value filling is used to fill in missing values ​​caused by sensor failure or transmission interruption during data acquisition using interpolation, mean filling, etc., to ensure data integrity. Normalization is used to uniformly map the cleaned valid operating data, such as values ​​from different sensors and different dimensions, to a preset range (such as [0,1] or [-1,1]), eliminating the impact of differences in data units. Standardization is used to classify and organize operating data according to operating conditions such as vehicle type, driving speed, ambient temperature, and vehicle load, further completing data standardization and providing a unified reference benchmark for data under the same operating conditions, improving data comparability and consistency. Based on this, the operating data is matched with historical real-vehicle operating data from the virtual chassis model using timestamps and spatial coordinates to ensure the temporal synchronization of data from different sources. Finally, using threshold judgment, isolated forest algorithm, statistical analysis and other anomaly detection methods, the preprocessed operating data is screened to automatically identify and remove or mark abnormal data that exceeds the reasonable range. Then, the preprocessed operating data is output after processing such as filling missing values, normalization, standardization, noise removal and anomaly detection. This preprocessed operating data effectively retains the core characteristics of the chassis operating status and has consistency, integrity and reliability, so as to improve the accuracy of the abnormal noise prediction model prediction results.

[0121] S402: Perform feature extraction on the preprocessed running data to obtain the first feature of the chassis sound data and the second feature of the chassis vibration data.

[0122] Feature extraction is performed on the preprocessed operating data to obtain key features that reflect the correlation between chassis status and faults, providing core input for the abnormal noise prediction model.

[0123] For example, for the preprocessed chassis sound data, Fourier transform, short-time Fourier transform, or Mel frequency cepstral transduction can be used to convert the time-domain chassis audio signal into frequency-domain data to obtain the spectral characteristics of the chassis sound data, including characteristic frequency points (such as specific frequency peaks corresponding to abnormal noises, the 2~8kHz high-frequency band corresponding to "squeaking" sounds, and the 100~500Hz mid-low-frequency band corresponding to "humming" sounds), frequency amplitude distribution, power spectral density, harmonic components, and octave characteristics, etc.; and extract the timbre features such as tone and amplitude changes of the chassis sound data to obtain the first feature of the chassis sound data.

[0124] For the preprocessed chassis vibration data, the root mean square (RMS) value is calculated to characterize the average energy level of the vibration signal (the RMS value is in a stable range under normal operating conditions, and rises significantly during faults due to increased vibration energy). Simultaneously, peak values ​​are extracted to capture the impact signals of instantaneous, severe vibrations caused by component collisions or loosening during vibration. Furthermore, in addition to the RMS and peak values, the time-domain characteristics of the chassis vibration data include parameters such as mean, peak factor, kurtosis, and waveform factor, comprehensively depicting the amplitude distribution, fluctuation patterns, and impact characteristics of the vibration signal over time.

[0125] Based on this, a Fast Fourier Transform (FFT) is performed on the time-domain vibration signal to convert it from the time domain to the frequency domain, obtaining a frequency-amplitude distribution spectrum. The dominant frequency with the largest amplitude is then identified and extracted. It is understandable that different chassis components (such as suspension springs, drive shafts, and wheel bearings) have their inherent dominant frequencies. Faults can lead to dominant frequency shifts, new abnormal frequency components, or a significant increase in the amplitude of the dominant frequency. Furthermore, the frequency domain characteristics of chassis vibration data also include the energy proportions of each frequency band, such as the energy distribution ratios of the low-frequency band (10~100Hz), mid-frequency band (100~1000Hz), and high-frequency band (above 1000Hz). In addition, the vibration signal is demodulated using autocorrelation function analysis and envelope spectrum analysis to extract modulation frequencies and sideband features. This is particularly suitable for identifying modulation-type vibration anomalies caused by gear meshing and bearing rolling. For example, bearing ball wear will produce sidebands at specific intervals in the envelope spectrum, thus enabling accurate detection of early, subtle vibration faults.

[0126] Furthermore, structural features corresponding to the source of abnormal noises are extracted using a virtual chassis model. Then, deep learning techniques such as CNN and Recurrent Neural Network (RNN) are applied to the data to automatically capture complex abnormal noise patterns. Finally, key information from multiple dimensions, including acoustics, vibration, and structure, is integrated to form a structured abnormal noise feature vector.

[0127] S403: Input the first feature and the second feature into the abnormal noise prediction model for prediction to obtain chassis fault information.

[0128] In this step, the abnormal noise prediction model is built on a deep learning network and has been fully trained with historical fault data accumulated through the dual-loop mechanism of the virtual chassis model, thus learning the accurate mapping pattern between feature data and chassis faults.

[0129] Specifically, the abnormal noise prediction model uses the characteristics of operational data to classify noises using end-to-end deep learning network models such as CNN and LSTM. It then combines this with a virtual chassis model to simulate the source of the abnormal noise, locate faulty components, and output a health report containing chassis fault information. This health report includes at least one of the following: the location of the chassis component with the abnormal noise (coordinates of a specific chassis component), the type of abnormal noise, the probability of the abnormal noise, the risk level, and a repair recommendation. The abnormal noise type includes gear meshing noise, bearing wear noise, etc.; the risk level includes low risk, medium risk, and high risk; and the repair recommendation could be to replace the left rear suspension bearing, etc. Furthermore, the chassis fault information also includes the location, risk level, and probability of abnormal noise of potentially faulty components (such as worn stabilizer bar bushings).

[0130] Furthermore, when conducting probabilistic risk assessments of chassis components exhibiting abnormal noise, uncertainties such as assembly tolerances and material property variations during design, manufacturing, and use can be considered. Extensive iterative calculations can be performed using Monte Carlo simulations, with each iteration employing randomly generated parameters within the tolerance range. This allows for the quantification and statistical analysis of the probability and intensity of abnormal noise occurrences under different operating conditions, such as the probability of the abnormal noise sound pressure level exceeding a threshold under specific conditions, thus achieving a precise quantitative assessment of abnormal noise risk.

[0131] The chassis noise processing method provided in this application is based on operational data and uses a pre-trained noise prediction model to obtain detailed chassis fault information of the vehicle. Specifically, it includes: firstly, preprocessing the raw operational data by performing a series of operations such as filling missing values, normalization, standardization, and noise removal to effectively purify the data and eliminate dimensional differences and interference, thereby obtaining high-quality, standardized input data. Subsequently, deep feature extraction is performed on the preprocessed operational data. The spectral characteristics of the chassis sound data are analyzed to form the first feature, while the time-domain and frequency-domain features of the chassis vibration data are obtained to form the second feature, thus comprehensively capturing the acoustic and dynamic characteristics of the fault. Finally, the multimodal features based on the preprocessed operational data are input into the pre-trained noise prediction model. This model uses its built-in deep learning network to intelligently analyze and recognize the features, ultimately outputting accurate structured chassis fault information, covering the location of the component with the noise, the type of noise, the probability of occurrence, the potential risk level, and specific repair suggestions. The above methods enable rapid, accurate, and automated diagnosis and localization of chassis noise faults, which not only greatly improves maintenance efficiency and reduces reliance on human experience, but also provides strong intelligent support for safe vehicle operation through forward-looking risk level assessment.

[0132] Based on the above embodiments, when the chassis fault information includes the probability of abnormal noise, the method further includes: if the probability of abnormal noise is greater than a preset threshold for abnormal noise, then a warning message is pushed to the user through at least one of steering wheel vibration, in-vehicle display screen, vehicle voice prompt, and user terminal.

[0133] Understandably, if the probability of abnormal noise exceeds a preset threshold (e.g., 0.8), it indicates that the possibility of a chassis malfunction has reached a level requiring user attention, or the risk level of the abnormal noise has reached high risk, in which case a warning push mechanism will be immediately activated. Warning messages will be delivered in at least one easily perceptible way for the user, such as through steering wheel vibration feedback (e.g., slight vibrations at a specific frequency) allowing the user to intuitively perceive the noise while driving; or through visual icons or pop-ups on the in-vehicle display screen; or through clear voice announcements on the vehicle's infotainment system informing the user of the abnormal noise risk; or through push notifications sent to the linked user terminal, ensuring that the user receives the message promptly regardless of whether they are driving.

[0134] Figure 5 A schematic diagram of a chassis noise treatment device provided in this application is shown below. Figure 5 As shown, the chassis noise treatment device 50 provided in this embodiment includes:

[0135] The first processing module 501 is used to acquire vehicle operating data, including road condition data, chassis sound data and chassis vibration data during vehicle operation.

[0136] The second processing module 502 is used to obtain chassis fault information of the vehicle based on the running data and through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions. The abnormal noise prediction model is obtained by training a deep learning network model based on historical real vehicle running data and historical real vehicle chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model.

[0137] The third processing module 503 is used to output chassis fault information.

[0138] In one possible implementation, the virtual chassis model is constructed based on the multibody dynamics (MBD) model, the finite element analysis (FEA) model, and the acoustic simulation model.

[0139] In one possible implementation, the chassis noise handling device 50 further includes a fourth processing module 504, for:

[0140] The virtual chassis model is driven by historical prototype test data obtained from the test track for the corresponding vehicle model, and the prototype chassis state response data corresponding to the historical prototype test data is output.

[0141] The chassis status response data of the prototype vehicle is compared with the preset chassis response standard data to obtain the historical chassis fault information of the prototype vehicle corresponding to the historical prototype vehicle test data.

[0142] Based on historical prototype chassis fault information, the parameters of the virtual chassis model are adjusted to obtain the processed virtual chassis model.

[0143] The virtual chassis model is driven by historical real-vehicle operation data of the corresponding vehicle model, and outputs real-vehicle chassis status response data corresponding to the historical real-vehicle operation data.

[0144] By comparing the actual vehicle chassis status response data with the chassis response standard data, historical actual vehicle chassis fault information is obtained.

[0145] In one possible implementation, both historical prototype test data and historical real-vehicle operation data include road condition data, chassis sound data, and chassis vibration data. The fourth processing module 504 is specifically used for:

[0146] Based on road condition data, motion information of various chassis components is calculated using the MBD model;

[0147] Based on motion information, the contact information and friction force change information between various chassis components are obtained through FEA model simulation.

[0148] Based on chassis sound data and chassis vibration data, the vibration transmission path and sound radiation characteristics are obtained through acoustic simulation model.

[0149] Based on motion information, contact information, friction force change information, vibration transmission path, and sound radiation characteristics, chassis state response data are obtained.

[0150] In one possible implementation, the second processing module 502 is specifically used for:

[0151] The running data is preprocessed to obtain preprocessed running data. The preprocessing includes at least one of missing value filling, normalization, standardization and noise removal.

[0152] Feature extraction is performed on the preprocessed running data to obtain the first feature of the chassis sound data and the second feature of the chassis vibration data.

[0153] The first and second features are input into the abnormal noise prediction model for prediction to obtain chassis fault information.

[0154] In one possible implementation, when the chassis fault information includes the probability of abnormal noise, the chassis abnormal noise processing device 50 further includes a fifth processing module 505, used for:

[0155] If the probability of abnormal noise exceeds the preset threshold, a warning message will be pushed to the user through at least one of the following methods: steering wheel vibration, in-vehicle display screen, in-vehicle voice prompts, and user terminal.

[0156] The chassis noise treatment device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described again in this embodiment.

[0157] Figure 6 A schematic diagram of the structure of an electronic device provided in this application, such as... Figure 6 As shown, the electronic device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the electronic device 60 further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0158] In a specific implementation, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to perform the above-described method.

[0159] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0160] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0161] The memory may include random access memory (RAM) in high-speed memory, and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0162] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0163] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0164] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0165] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0166] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside within an ASIC. Alternatively, the processor and the readable storage medium can exist as discrete components in a device.

[0167] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0168] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0169] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0170] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0171] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0172] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for treating chassis noise, characterized in that, include: Acquire vehicle operating data, including road condition data, chassis sound data, and chassis vibration data during vehicle operation; Based on the operational data, the chassis fault information of the vehicle is obtained through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions. The abnormal noise prediction model is obtained by training a deep learning network model based on historical real vehicle operation data and historical real vehicle chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model. Output the chassis fault information.

2. The method according to claim 1, characterized in that, The virtual chassis model is constructed based on the multibody dynamics (MBD) model, the finite element analysis (FEA) model, and the acoustic simulation model.

3. The method according to claim 2, characterized in that, The method further includes: The virtual chassis model is driven by the historical prototype test data obtained from the test track of the corresponding vehicle model, and the prototype chassis state response data corresponding to the historical prototype test data is output. The chassis status response data of the prototype vehicle is compared with the preset chassis response standard data to obtain the historical prototype vehicle chassis fault information corresponding to the historical prototype vehicle test data. Based on the historical prototype chassis fault information, the parameters of the virtual chassis model are adjusted to obtain the processed virtual chassis model; Based on the historical real-vehicle operation data corresponding to the vehicle model, the processed virtual chassis model is driven to output the real-vehicle chassis status response data corresponding to the historical real-vehicle operation data; The actual vehicle chassis status response data is compared with the chassis response standard data to obtain the historical actual vehicle chassis fault information.

4. The method according to claim 3, characterized in that, Both the historical prototype test data and the historical real-vehicle operation data include road condition data, chassis sound data, and chassis vibration data. The historical prototype test data or the historical real-vehicle operation data drives the virtual chassis model to obtain corresponding chassis state response data, including: Based on road condition data, the motion information of each component of the chassis is calculated using the MBD model. Based on the motion information, the contact information and friction force change information between the various components of the chassis are obtained through the FEA model simulation. Based on chassis sound data and chassis vibration data, the vibration transmission path and sound radiation characteristics are simulated using the acoustic simulation model. Based on the motion information, the contact information, the friction force change information, the vibration transmission path, and the sound radiation characteristics, chassis state response data is obtained.

5. The method according to any one of claims 1 to 3, characterized in that, Based on the operational data, the chassis fault information of the vehicle is obtained through a pre-trained abnormal noise prediction model, including: The running data is preprocessed to obtain preprocessed running data. The preprocessing includes at least one of missing value filling, normalization, standardization and noise removal. Feature extraction is performed on the preprocessed running data to obtain the first feature of the chassis sound data and the second feature of the chassis vibration data; The first feature and the second feature are input into the abnormal noise prediction model for prediction to obtain the chassis fault information.

6. The method according to any one of claims 1 to 3, characterized in that, When the chassis fault information includes the probability of abnormal noise, the method further includes: If the probability of abnormal noise is greater than a preset threshold, a warning message will be pushed to the user through at least one of the following: steering wheel vibration, in-vehicle display screen, in-vehicle voice prompt, and user terminal.

7. A device for treating chassis noise, characterized in that, include: The first processing module is used to acquire vehicle operating data, which includes road condition data, chassis sound data, and chassis vibration data during vehicle operation. The second processing module is used to obtain chassis fault information of the vehicle based on the operating data and through a pre-trained abnormal noise prediction model. The chassis fault information includes at least one of the following: the location of the chassis component with abnormal noise, the type of abnormal noise, the probability of abnormal noise, the risk level, and maintenance suggestions. The abnormal noise prediction model is obtained by training a deep learning network model based on historical real vehicle operating data and historical real vehicle chassis fault information obtained through a dual-loop mechanism based on a digital twin virtual chassis model. The third processing module is used to output the chassis fault information.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 6.

Images