A method for predicting the strength of a knuckle arm structure of a special vehicle

Abstract

This invention relates to the field of vehicle structure monitoring technology and discloses a method for predicting the structural strength of a steering knuckle arm for special vehicles. The method includes the following steps: acquiring historical driving state parameters and road environment characteristic parameters of special vehicles; processing these parameters using a vehicle dynamics model to obtain the load spectrum of the steering knuckle arm connection point; generating and integrating a multi-fidelity training sample library using finite element analysis and rapid simulation; training the multi-fidelity training sample library using a cooperative radial basis function (CoRBF) surrogate model algorithm to obtain a Co-RBF model; acquiring vehicle load data in real time and inputting it into the Co-RBF model to output the predicted key strength indicators of the steering knuckle arm under the current load. This invention establishes a rapid load-stress mapping relationship through the designed Co-RBF model, reducing the prediction time from several hours to seconds, thereby achieving online real-time evaluation and interactive analysis of the structural state.

Description

Technical Field

[0001] This invention relates to the field of vehicle structure monitoring technology, specifically a method for predicting the structural strength of a steering knuckle arm for special vehicles. Background Technology

[0002] The steering knuckle arm is a key load-bearing component of special vehicles, and its structural strength directly determines the driving safety of the entire vehicle. Therefore, accurate and real-time strength monitoring and prediction of the steering knuckle arm is a core aspect of the research, development and maintenance of special vehicles.

[0003] In existing technologies, the analysis and monitoring of steering knuckle arm strength are mostly achieved by combining finite element simulation with strain sensors. This involves additional strain sensors on the steering knuckle arm to collect real-time strain data, which is then transmitted to a cloud server and combined with a finite element model to perform stress inversion and predict the structural strength of the steering knuckle arm. While this prediction method provides high computational accuracy, it suffers from time-sensitivity issues. Traditional finite element simulation processes often require several hours of computation time. Essentially, the aforementioned simulation process is an offline analysis method, only capable of post-hoc verification of pre-collected or assumed historical operating conditions, making it difficult to match with the real-time state parameters generated during actual vehicle operation. When a vehicle is traveling under complex and ever-changing task profiles, the load on the steering knuckle arm changes rapidly. The severe lag in traditional finite element calculations makes it difficult to keep up with the pace of these changes, completely failing to achieve second-level response and online early warning for dangerous stress states. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting the structural strength of a steering knuckle arm in special vehicles, in order to solve the problems existing in the current method of predicting the structural strength of a steering knuckle arm by combining finite element simulation with strain sensors.

[0005] The technical solution of this invention is: A method for predicting the structural strength of a steering knuckle arm in a special vehicle includes the following steps: Acquire historical driving status parameters of special vehicles and road environment characteristic parameters; By processing the historical special vehicle driving state parameters and road environment characteristic parameters obtained using the vehicle dynamics model, the load spectrum of the steering knuckle arm connection point is obtained. The load spectrum is analyzed using high-fidelity finite element method to obtain the corresponding high-precision stress response and generate a high-fidelity sample set. The load spectrum is then rapidly simulated using a low-fidelity model to obtain the corresponding low-precision response and generate a low-fidelity sample set. An initial RBF model is constructed based on a low-fidelity sample set. The initial RBF model is then trained using the high-fidelity sample set combined with the collaborative radial basis function multifidelity surrogate model algorithm. The global trend of the low-fidelity model is then fused with the local precise information of the high-fidelity sample to obtain the Co-RBF model. Real-time vehicle load data is acquired and input into the Co-RBF model, which outputs the predicted key strength indicators of the steering knuckle arm under the current load.

[0006] Preferably, as a further improvement of the present invention, the mathematical expression of the cooperative radial basis function multifidelity surrogate model algorithm is: y_ H (x) = ρ·y_ L (x)+δ(x), Where δ(x) = Σ_{k=1}^{n_ H}w_ k ·φ(||x-x_ H ^(k)||) In the formula, y_ H (x) represents the high-precision stress response, ρ is the global scale factor, and y_ L (x) represents the low-precision stress response, δ(x) is the local correction function, and n_ H w_ represents the total number of high-fidelity samples. k φ is the undetermined weighting coefficient, x is the load spectrum input variable, φ(·) is the selected radial basis function, and ||·|| represents the Euclidean distance.

[0007] Preferably, as a further improvement of the present invention, the training of the initial RBF model using the cooperative radial basis function multifidelity surrogate model algorithm includes the following steps: Substitute the high-fidelity sample data and the corresponding low-fidelity predicted values ​​into the mathematical expression of the model algorithm to construct a system of linear equations; The least squares method is used for fitting, and the optimal global scale factor and undetermined weight coefficients are obtained to minimize the model prediction error.

[0008] Preferably, as a further improvement of the present invention, the method of obtaining real-time load according to monitoring requirements includes: The vehicle's current driving status parameters are collected in real time by onboard sensors, and the current road environment feature parameters are obtained through a pre-calibrated road feature recognition algorithm. The collected driving state parameters and the acquired road environment feature parameters are input into the intelligent working condition matching module in the vehicle dynamics model. The built-in working condition-parameter mapping library is used to quickly compare the current real-time data with the template library according to the pattern recognition algorithm. It automatically identifies the most similar working condition category of the current vehicle and calls the pre-stored working condition load spectrum as the real-time vehicle load data.

[0009] Preferably, as a further improvement of the present invention, during the process of outputting the predicted key strength index of the steering knuckle arm under the current load by the Co-RBF multi-fidelity proxy model, the predicted key strength index of the steering knuckle arm is sent to a three-dimensional visualization platform. The three-dimensional visualization platform maps the data onto the lightweight three-dimensional model of the steering knuckle arm, renders a dynamically changing stress color cloud map and updates the numerical display and the stress-time history curve of key nodes, and realizes decision-making through a human-computer interaction interface.

[0010] Preferably, as a further improvement of the present invention, the human-machine interface implements decision-making measures including working condition switching, model interaction operation, early warning threshold setting and over-limit alarm, and completes the operation and maintenance decision of the steering knuckle based on visualized information.

[0011] Preferably, as a further improvement of the present invention, the vehicle dynamics model is constructed based on vehicle dynamics simulation software, which is any one of Trucksim, CarSim, ADAMS, and Simulink.

[0012] Compared with the prior art, the beneficial effects of the present invention are: Real-time prediction and interaction to meet online monitoring needs: By establishing a fast load-stress mapping relationship through the Co-RBF model, the prediction time is shortened from several hours (usually 2~6 hours) to seconds (≤2 seconds), so as to realize online real-time assessment and interactive analysis of structural status. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the steps in the method for predicting the structural strength of a steering knuckle arm of a special vehicle according to the present invention.

[0014] Figure 2 This is a schematic diagram of the system framework for a method for predicting the structural strength of a steering knuckle arm of a special vehicle according to the present invention. Detailed Implementation

[0015] The following is combined Figures 1-2The specific embodiments of the present invention will be described in detail below. In the description of the invention, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0016] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature; in the description of the invention, unless otherwise stated, "a plurality of" means two or more.

[0017] Example 1 like Figures 1-2 As shown, this invention provides a method and system for predicting the structural strength of a steering knuckle arm for special vehicles, comprising the following steps: The generation stage of simulation loads: S11. Obtain historical special vehicle driving status parameters and road environment characteristic parameters.

[0018] Specifically, an onboard sensor array (including but not limited to vehicle speed sensor, inertial measurement unit, GPS, and steering wheel angle sensor) is used to collect the vehicle's current driving state parameters (such as steering wheel angle, vehicle speed, longitudinal / lateral acceleration, yaw rate, and suspension travel). Simultaneously, a pre-calibrated road feature recognition algorithm (such as road roughness inversion based on wheel acceleration spectrum) is used to obtain current road environment feature parameters (such as road surface type, estimated road friction coefficient, and road roughness level).

[0019] S12. Use vehicle dynamics simulation software to establish a dynamic model of the target special vehicle. Based on the historical special vehicle driving state parameters and road environment characteristic parameters, design and execute a set of working conditions covering typical and extreme cases, and batch simulate and export the spectral load (load time series) of the steering knuckle arm connection point.

[0020] The vehicle dynamics simulation software can be any one of Trucksim, CarSim, ADAMS, and Simulink; the simulation setup includes at least one of high-speed steering, emergency braking, and off-road impact. Vehicle dynamics simulation software is essentially a highly parameterized physics solver. Taking Trucksim as an example, its working principle can be summarized as follows:

[0021] 1) Built-in "parametric physical world" (pre-encapsulated key mathematical models) Vehicle model library: Includes parameters such as vehicle mass, moment of inertia, suspension characteristics, tire models, steering system transmission ratio, and braking system characteristics.

[0022] Road environment library: Includes road surface geometry (slope, curvature, elevation) and road surface adhesion characteristics (friction coefficient varies with road surface type). The software has built-in various standard road models: straight road, sloping road, washboard road, pothole road, gravel road (achieved by setting random unevenness power spectral density), split road surface (different friction coefficients for the left and right wheels), etc.

[0023] Driver model: The driver's ability to track the path, reaction time, etc. can be set for closed-loop simulation.

[0024] 2) Input parameter = Define "Current Scene" The "new vehicle motion parameters" (such as steering wheel angle, vehicle speed, road surface type, coefficient of friction, etc.) set by the user through the interactive interface are essentially specifying specific boundary and initial conditions for the aforementioned physical model. These parameters are not the simulation software itself, but rather the instructions that drive the simulation software to run.

[0025] 3) Simulation calculation = solving the "physical equations" After receiving the parameters, the vehicle dynamics simulation software solves the complex multibody dynamics differential equations of the vehicle in real time and outputs the physical quantities of interest, including the load-time history (including triaxial forces, torques, etc.) at the steering knuckle arm connection point. The spectral load of the steering knuckle arm connection point is exported through batch simulation.

[0026] These simulated load data serve two core functions: first, as a "sample source" for the offline training phase, providing low-cost, high-coverage input data for the subsequent construction of surrogate models; second, as a "real-time driving signal" for the online prediction phase, simulating the actual vehicle's operating state and inputting it into the prediction model to obtain real-time intensity status. This module is also responsible for the unified storage, management, and retrieval of all simulated load data, subsequently generated sample data, and model parameters.

[0027] Sample library construction phase: S2. Obtain the corresponding high-precision stress response by performing high-fidelity finite element analysis on the load spectrum and generate a high-fidelity sample set. Obtain the corresponding low-precision response by performing rapid simulation on the load spectrum using a low-fidelity model and generate a low-fidelity sample set.

[0028] Specifically, using the output load spectrum as input, batch simulation of the steering knuckle arm parametric model is performed using high-fidelity finite element analysis software (such as ANSYS) to obtain accurate stress / strain field results, forming "high-fidelity sample pairs." Simultaneously, simplified models or coarse meshes can be used for rapid simulation, generating a large number of "low-fidelity samples" to capture global response trends. The load spectrum output by the load generation module is used as the input variable (denoted as x). High-fidelity finite element analysis (FEA) is then used to obtain the corresponding high-precision stress response (denoted as y_). H (x)), forming a high-fidelity sample set {x_ H ^(i)y_ H ^(i)}, where i=1 to n_ H n_ H This represents the total number of high-fidelity samples; simultaneously, a simplified physical model or coarse-grid finite element method is used for rapid calculation to obtain the corresponding low-precision response (denoted as y_). L (x)), forming a low-fidelity sample set {x_} that covers a wider range of input spaces. L ^(j)y_ L ^(j)}, where j=1 to n_ L And n_ L >>n_ H ,

[0029] n_ L The total number of low-fidelity samples (i.e., sample size) is represented by the symbol "n_". L >>n_ H " represents the number of low-fidelity samples n_ L In engineering practice, it should be significantly larger than the number of high-fidelity samples, n_. H Typically, the ratio of n_L / n_H is required to be no less than 5 to 10 times. The specific multiple needs to be determined comprehensively based on the degree of nonlinearity of the problem, the dimensionality of the input variables, and the computational resources.

[0030] Model training and validation phase: S3. Construct an initial RBF model based on the low-fidelity sample set, and use the high-fidelity sample set combined with the collaborative radial basis function multi-fidelity surrogate model algorithm to learn and train the initial RBF model. The global trend of the low-fidelity model and the local accurate information of the high-fidelity sample are fused to obtain the Co-RBF model.

[0031] Specifically, the mathematical expression for the multifidelity surrogate model algorithm with the same radial basis function is: y_ H (x) = ρ·y_ L (x)+δ(x), Where δ(x) = Σ_{k=1}^{n_ H}w_k ·φ(||x-x_ H ^(k)||) In the formula, y_ H (x) represents the high-precision stress response, ρ is the global scale factor, and y_ L (x) represents the low-precision stress response, δ(x) is the local correction function, and n_ H w_ represents the total number of high-fidelity samples. k φ is the undetermined weighting coefficient, x is the load spectrum input variable, φ(·) is the selected radial basis function, and ||·|| represents the Euclidean distance.

[0032] The Collaborative Radial Basis Function (RBF) multi-fidelity surrogate model algorithm deeply fuses the global trend of the low-fidelity model with the local precision information of the high-fidelity samples through a global scaling factor and a local correction function. In practice, a limited number of high-fidelity samples are used to correct the initial RBF model built on a large number of low-fidelity samples, ultimately forming a lightweight mathematical model that maintains high accuracy while achieving millisecond-level prediction speed. The training process determines the model parameters through an optimization algorithm to minimize the prediction error. The training process involves solving for the parameters ρ and {w_ in the mathematical expression. k The process of transferring high-fidelity sample data {y_ H ^(i)} and the corresponding low-fidelity prediction value {y_ L (x_) H Substituting ^(i))} into the above model, a system of linear equations is constructed. The least squares method is typically used for fitting to minimize the prediction error at high-fidelity sample points. This process can be intuitively understood as: first using the low-fidelity model y_ L (x) captures the response trend, and then uses a small amount of high-fidelity data to "stretch" and "finely sculpt" this trend by optimizing ρ and δ(x), thereby obtaining a fusion model Co-RBF model that combines the efficiency of a low-fidelity model with the accuracy of a high-fidelity model.

[0033] Real-time load driving and intensity prediction stage: S4. Deploy the pre-trained Co-RBF model online. During online operation, obtain real-time loads according to monitoring requirements and input them into the pre-trained Co-RBF fusion model to output the key strength indicators of the steering knuckle arm prediction.

[0034] When the system is running online, the acquisition of payload data adopts a dual-mode collaborative mechanism of "intelligent matching as the main method and user-defined methods as the auxiliary method", as follows: 1) Automatic matching mode (default mode, used for real-time monitoring) The system collects real-time vehicle driving status parameters (such as vehicle speed, longitudinal / lateral acceleration, yaw rate, steering angle, suspension travel, etc.) through an onboard sensor array (including but not limited to vehicle speed sensor, inertial measurement unit, GPS, steering wheel angle sensor) or by connecting to the vehicle's CAN bus. Simultaneously, it obtains current road environment characteristic parameters (such as estimated road friction coefficient and road roughness level) through a pre-calibrated road feature recognition algorithm (such as road roughness inversion based on wheel acceleration spectrum).

[0035] This real-time sensing data is input into the "intelligent operating condition matching module." This module maintains an "operating condition-parameter mapping library," which pre-stores feature parameter templates corresponding to typical operating conditions (such as high-speed steering, emergency braking, off-road impact, and washboard road driving). The module uses pattern recognition algorithms (such as K-nearest neighbors, support vector machines, or lightweight neural networks) to quickly compare the current real-time data with the template library, automatically identifying the "most similar operating condition category" currently experienced by the vehicle.

[0036] The recognition result triggers one of the following two sub-modes: Sub-mode A (Direct Invocation): If the current operating condition highly matches the pre-stored template (similarity higher than a preset threshold, such as 90%), the system directly invokes the pre-simulation load spectrum file corresponding to that template as input to the Co-RBF model. This mode has the fastest response speed (microseconds) and is suitable for routine monitoring.

[0037] Sub-mode B (Real-time Correction Generation): If the current operating condition deviates from the template (similarity below the threshold but above the second threshold), the system will use the identified difference parameters (e.g., the current vehicle speed is 10 km / h higher than the template speed) as correction values ​​and send them to the background dynamics simulation service. This service will then drive the system to perform a local rapid resimulation based on the closest template operating condition (adjusting only the difference parameters, with minimal computational cost) to generate the corrected load spectrum. This mode balances accuracy and real-time performance.

[0038] 2) User-defined mode (auxiliary mode, used for offline analysis or special scenarios) Users can manually set special vehicle driving status parameters (such as steering wheel angle, vehicle speed, and load) and road environment parameters (such as road surface type and friction coefficient) through a visual interactive interface, driving the simulation software to generate load spectra for any custom working condition in real time. This mode is mainly used for virtual verification in the design phase, exploratory analysis of unknown working conditions, or offline evaluation of special scenarios, and does not participate in the online real-time monitoring process.

[0039] In vehicle dynamics simulation, motion parameters, as referenced in Table 1, are typically categorized as follows: Table 1. Reference Table for Action Parameters Users set new vehicle motion parameters through an interactive interface, essentially parameterizing complex vehicle dynamics simulations and making them accessible to users through a user-friendly graphical interface. This enables real-time control of virtual vehicle behavior and on-demand generation of loads. This makes the invention not only a passive monitoring system but also an active, exploratory virtual experimental platform.

[0040] "Real-time generation / retrieval of load data tailored to the current complex road conditions through two modes" corresponds to the following two operating scenarios shown in Table 2 in this invention: Table 2. Reference Table for Operating Modes Regardless of the mode, the final input to the proxy model is the simulated load data, rather than the strain signals collected by strain sensors using existing technology. This eliminates the need for traditional strain-type physical sensors glued to the surface of the steering knuckle arm, avoiding the need to directly place strain gauges on load-bearing components and resolving issues such as complex sensor wiring, susceptibility to damage, and high maintenance costs. Therefore, using vehicle dynamics simulation software to generate load data as the simulated load data saves on the purchase, installation, and maintenance costs of strain sensors, while also avoiding the risk of strain sensor failure under harsh conditions, resulting in simpler and more stable system deployment.

[0041] The method for users to set new vehicle motion parameters through an interactive interface (developed based on Unity3D) is as follows: Interactive interface design: Develop a visual parameter input panel in Unity3D, including numerical input controls (slider, input box, drop-down menu, switch), real-time feedback, and parameter templates.

[0042] Parameter Packaging and Transmission: When the user adjusts the parameters in the interface and clicks the "Apply Parameters and Simulate" button, the system performs parameter verification, parameter serialization, and communication transmission.

[0043] Real-time dynamic model driven: The backend server receives parameter data, including parsing parameters, calling simulation interfaces, and performing simulation calculations.

[0044] Load data feedback and real-time prediction: This includes load data output, data reception, triggering prediction, and visualization updates.

[0045] 3) Collaboration between automatic and custom functions User-defined working conditions and their generated load spectra, once confirmed, can be incorporated into the "Working Condition-Parameter Mapping Library" as new templates for subsequent automatic matching modes, enabling the system to learn itself and continuously enrich the template library.

[0046] The system first calls the fast low-fidelity model to calculate y_ L(x_new), and then combine it with the pre-trained scaling factor ρ and correction function δ(x_new), through the formula y_ H (x_new)=ρ·y_ L (x_new) + δ(x_new) calculates the high-fidelity stress prediction value. The model completes the calculation and outputs the full-field stress / strain distribution prediction results of the steering knuckle arm under the current load, the maximum stress value and its location, the safety factor, and other key strength indicators.

[0047] As an optimized implementation, in order to realize the three-dimensional visualization and human-computer interaction process of setting the steering knuckle arm structural strength prediction, the key strength indicators of the steering knuckle arm prediction can be sent to the three-dimensional visualization platform. The data is mapped onto the lightweight three-dimensional model of the steering knuckle arm, rendering a dynamically changing stress color cloud map and updating the numerical display and the stress-time history curve of key nodes. Decisions can be made through the human-computer interaction interface, allowing users to make decisions based on intuitive information, such as adjusting vehicle operation and planning maintenance.

[0048] First, the high-precision steering knuckle CAD model undergoes lightweighting (generating a lightweight version that highly represents the original precision model visually and mechanically under strict geometric error and feature preservation criteria, thus providing a solid and efficient geometric foundation for real-time, high-fidelity 3D visualization monitoring). This significantly reduces the number of triangular faces to ensure smooth real-time rendering while preserving the geometric features of stress concentration areas (such as fillets and grooves), ensuring both visualization accuracy and realistic mechanical representation. An improved Quadric ErrorMetric (QEM) edge-folding algorithm is used as the core of the lightweighting process. The standard QEM algorithm only calculates the sum of squares of the geometric position errors caused by an edge folding operation when deciding whether to fold it. By introducing "feature importance" as a core weight into this error calculation, the algorithm "penalizes" edge-folding operations located in high-importance feature regions, making them less likely to be executed. The lightweight model is then imported into the Unity3D scene.

[0049] The system receives stress / strain data from the prediction module in real time and maps it onto a 3D model, dynamically updating and displaying it as a color cloud map to intuitively show stress concentration areas and evolution processes. Simultaneously, the platform provides a human-computer interaction interface, allowing users to: ① select and switch between different simulation conditions (load input); ② view the stress-time history curves of key nodes; ③ set stress warning thresholds, with the system automatically triggering audible and visual alarms when predicted values ​​exceed limits; and ④ perform operations such as rotating, scaling, and sectioning the 3D model to examine the structural state from multiple angles.

[0050] In this embodiment, unlike traditional monitoring interfaces that primarily rely on data reports or graphs, the present invention develops a high-fidelity 3D visualization platform based on Unity3D. This platform can render stress-strain color cloud maps and dynamic deformations of the steering knuckle arm in real time, and provides interactive functions such as historical data playback, virtual disassembly and assembly, and early warning prompts, greatly improving the intuitiveness of status perception and the efficiency of decision support.

[0051] As an optimized implementation, this invention also supports load source expansion, allowing access to real vehicle CAN bus data and historical test data to achieve a hybrid driving mode combining simulation and actual measurement. Simultaneously, it supports online learning and updating of the Co-RBF model, automatically correcting the model by integrating new simulation or actual measurement data to maintain long-term prediction accuracy. When vehicle performance degrades or unforeseen operating conditions occur, the system can automatically correct the prediction model by integrating new simulation or actual measurement data, thereby maintaining long-term prediction accuracy and exhibiting stronger environmental adaptability and long-term robustness.

[0052] The above-disclosed embodiments are merely preferred embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for predicting the structural strength of a steering knuckle arm in a special vehicle, characterized in that, Includes the following steps: Acquire historical driving status parameters of special vehicles and road environment characteristic parameters; By processing the historical special vehicle driving state parameters and road environment characteristic parameters obtained using the vehicle dynamics model, the load spectrum of the steering knuckle arm connection point is obtained. The load spectrum is analyzed using high-fidelity finite element method to obtain the corresponding high-precision stress response and generate a high-fidelity sample set. The load spectrum is then rapidly simulated using a low-fidelity model to obtain the corresponding low-precision response and generate a low-fidelity sample set. An initial RBF model is constructed based on a low-fidelity sample set. The initial RBF model is then trained using the high-fidelity sample set combined with the collaborative radial basis function multifidelity surrogate model algorithm. The global trend of the low-fidelity model is then fused with the local precise information of the high-fidelity sample to obtain the Co-RBF model. Real-time vehicle load data is acquired and input into the Co-RBF model, which outputs the predicted key strength indicators of the steering knuckle arm under the current load.

2. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 1, characterized in that, The mathematical expression for the cooperative radial basis function multifidelity surrogate model algorithm is: y_ H (x)=ρ·y_ L (x)+δ(x), Where, δ(x)=Σ_{k=1}^{n_ H }w_ k ·φ(||x-x_ H ^(k)||), In the formula, y_ H (x) represents the high-precision stress response, ρ is the global scale factor, and y_ L (x) represents the low-precision stress response, δ(x) is the local correction function, and n_ H w_ represents the total number of high-fidelity samples. k φ is the undetermined weighting coefficient, x is the load spectrum input variable, φ(·) is the selected radial basis function, and ||·|| represents the Euclidean distance.

3. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 2, characterized in that, The training of the initial RBF model using the cooperative radial basis function multifidelity surrogate model algorithm includes the following steps: Substitute the high-fidelity sample data and the corresponding low-fidelity predicted values ​​into the mathematical expression of the model algorithm to construct a system of linear equations; The least squares method is used for fitting, and the optimal global scale factor and undetermined weight coefficients are obtained to minimize the model prediction error.

4. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 1, characterized in that, The methods for acquiring vehicle load data in real time include: The vehicle's current driving status parameters are collected in real time by onboard sensors, and the current road environment feature parameters are obtained through a pre-calibrated road feature recognition algorithm. The collected driving state parameters and the acquired road environment feature parameters are input into the vehicle dynamics model. The vehicle dynamics model includes an intelligent working condition matching module with a built-in working condition-parameter mapping library. The intelligent working condition matching module uses a pattern recognition algorithm to quickly compare the current real-time data with the template library, automatically identify the most similar working condition category of the current vehicle, and call the pre-stored working condition load spectrum as the real-time vehicle load data.

5. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 1, characterized in that, During the process of outputting the predicted key strength indicators of the steering knuckle arm under the current load in the Co-RBF multi-fidelity proxy model, the predicted key strength indicators of the steering knuckle arm are sent to the 3D visualization platform. The 3D visualization platform maps the data onto the lightweight 3D model of the steering knuckle arm, renders the dynamically changing stress color cloud map and updates the numerical display and the stress-time history curve of key nodes, and realizes decision-making through the human-computer interaction interface.

6. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 5, characterized in that, The human-machine interface enables decision-making through measures including operating condition switching, model interaction, early warning threshold setting, and over-limit alarms, and completes steering knuckle boom maintenance decisions based on visualized information.

7. The method for predicting the structural strength of a steering knuckle arm for special vehicles according to claim 1, characterized in that, The vehicle dynamics model is built based on vehicle dynamics simulation software, which can be any one of Trucksim, CarSim, ADAMS, and Simulink.

Images