A collaborative design method, apparatus, equipment, medium, and product for occupant restraint systems.

Abstract

This application discloses a collaborative design method, apparatus, equipment, medium, and product for occupant restraint systems, relating to the field of vehicle safety technology. The method includes: acquiring an initial simulation model and physical test data; identifying deterministic design variables and uncertain parameters of the system; calibrating the model based on the test data to obtain a baseline simulation model; quantifying the parameter value range and correlation through an adaptive multidimensional parallelepiped interval model to obtain joint uncertainty domain information including the parameter midpoint, radius, and correlation coefficient matrix, with the matrix dynamically updated; generating sample points based on the baseline model and joint uncertainty domain information, obtaining system response data, and constructing a surrogate model; optimizing the design variables through the surrogate model to obtain the optimal design scheme; fixing this scheme and calibrating the parameters with the objectives of minimizing the time-series response error and maximizing the Bayesian factor; and finally combining the two to determine the final system design data, achieving a high-confidence design.

Description

Technical Field

[0001] This application relates to the field of vehicle safety technology, and in particular to a collaborative design method, device, equipment, medium and product for occupant restraint systems. Background Technology

[0002] As a key component of automotive passive safety, the occupant restraint system encompasses devices such as airbags, seat belts, and seats. Its design performance directly determines the effectiveness of occupant injury protection during a collision, and it is a crucial element in improving overall vehicle safety levels and meeting safety certification standards such as C-NCAP. With the automotive industry moving towards intelligent and integrated systems, user demands for vehicle collision safety are constantly increasing. Simultaneously, the requirements for design efficiency, model reliability, and engineering practicality in vehicle development are becoming increasingly stringent. Therefore, achieving high-performance, highly robust design of the occupant restraint system and high-precision calibration of simulation models while considering the uncertainties of actual engineering has become a major research and application topic in the field of automotive passive safety.

[0003] Currently, the mainstream occupant restraint system design methodology in the industry primarily utilizes Computer-Aided Engineering (CAE). This involves constructing multibody dynamics and finite element simulation models, combined with deterministic optimization algorithms, to conduct an iterative development process of "design, simulation, and optimization." This method identifies deterministic design variables such as seatbelt force-limiting loads and airbag vent diameters, optimizes performance based on fixed simulation model parameters, and calibrates the simulation model parameters using engineer experience or global sampling methods. This approach has, to some extent, improved the performance of the restraint system and has become an indispensable technical tool in vehicle development. However, existing design and calibration methods still suffer from several fundamental shortcomings in practical engineering applications, such as overly idealized simulations, inaccurate results, and low efficiency, making it difficult to meet the requirements of high-confidence designs.

[0004] Therefore, there is an urgent need for a design method that can achieve high confidence levels. Summary of the Invention

[0005] This application provides a collaborative design method, apparatus, equipment, medium, and product for occupant restraint systems, which can achieve high-confidence design.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] Firstly, this application provides a collaborative design method for occupant restraint systems, including:

[0008] Acquire the initial simulation model and physical test data corresponding to the occupant constraint system, and identify the deterministic design variables and uncertain parameters in the occupant constraint system;

[0009] The initial simulation model was calibrated based on physical test data to obtain a baseline simulation model;

[0010] Based on the adaptive multidimensional parallelepiped interval model, the value range of the uncertainty parameters and the correlation between the uncertainty parameters are quantified to obtain joint uncertainty domain information. The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter and the correlation coefficient matrix between the uncertainty parameters. The correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism.

[0011] Based on the baseline simulation model and joint uncertainty domain information, an experimental design method is used to generate sample points in the input space; the input space consists of deterministic design variables and uncertainty parameters; the system response data of the sample points is obtained through simulation, and a surrogate model is constructed based on the system response data;

[0012] Based on the parameter correlations represented by the surrogate model and the joint uncertainty domain information, the deterministic design variables are optimized to obtain the optimal design scheme;

[0013] Using the optimal design scheme as a fixed input, and with the objectives of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters, the uncertainty parameters are calibrated to obtain the optimal calibration parameters.

[0014] Based on the optimal design scheme and optimal calibration parameters, the final design data of the occupant restraint system are determined.

[0015] In some possible implementations, the uncertainty parameter includes at least one of interval uncertainty parameter, random uncertainty parameter, and fuzzy uncertainty parameter; the method also includes:

[0016] Based on the degree of influence of interval uncertainty parameters, random uncertainty parameters, and fuzzy uncertainty parameters on the occupant constraint system, the uncertainty contribution index corresponding to each uncertainty parameter is obtained;

[0017] The allocation of computing resources is adjusted based on the uncertainty contribution index.

[0018] In some possible implementations, the uncertainty parameters also include pre-collision attitude parameters, which characterize the dynamic initial attitude of the occupants relative to the vehicle interior at the moment of collision.

[0019] In some possible implementations, the online adaptive update mechanism dynamically adjusts the correlation coefficient matrix based on historical correlation information, the latest empirical correlation information, and prior correction information based on physical mechanisms.

[0020] In some possible implementations, based on surrogate models and joint uncertainty domain information, the deterministic design variables are optimized considering parameter correlations to obtain the optimal design scheme, including:

[0021] Based on the correlation coefficient matrix in the joint uncertainty domain information and the adaptive multidimensional parallelepiped interval model, the original interval uncertainty parameters with correlation are mapped to independent standard interval variables;

[0022] Based on the surrogate model and standard interval variables, the fast interval probability approximation algorithm is used to calculate the degree of reliability constraint satisfaction corresponding to the deterministic design variables;

[0023] Based on the degree of satisfaction of reliability constraints, the optimal design scheme is obtained by iteratively optimizing the deterministic design variables using a multi-objective evolutionary algorithm.

[0024] In some possible implementations, timing response error is comprehensively evaluated using a shift distance metric, which includes amplitude error, phase error, and shape error components.

[0025] Secondly, this application provides a collaborative design device for an occupant restraint system, comprising:

[0026] The acquisition module is used to acquire the initial simulation model and physical test data corresponding to the occupant constraint system, as well as to identify the deterministic design variables and uncertain parameters in the occupant constraint system;

[0027] The calibration module is used to calibrate the initial simulation model based on physical test data to obtain a baseline simulation model;

[0028] The information determination module is used to quantify the value range of uncertainty parameters and the correlation between uncertainty parameters based on the adaptive multidimensional parallelepiped interval model, so as to obtain joint uncertainty domain information. The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter and the correlation coefficient matrix between each uncertainty parameter. The correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism.

[0029] The model building module is used to generate sample points in the input space based on the baseline simulation model and joint uncertainty domain information, using experimental design methods. The input space consists of deterministic design variables and uncertainty parameters. The system response data of the sample points is obtained through simulation, and a surrogate model is built based on the system response data.

[0030] The optimization module is used to optimize deterministic design variables based on the parameter correlations characterized by the surrogate model and joint uncertainty domain information to obtain the optimal design scheme. The optimal design scheme is used as a fixed input to calibrate the uncertainty parameters with the objectives of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters, so as to obtain the optimal calibration parameters.

[0031] The design module is used to determine the final design data of the occupant constraint system based on the optimal design scheme and optimal calibration parameters.

[0032] Thirdly, this application provides a computing device, including a memory and a processor;

[0033] The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of the first aspects.

[0034] Fourthly, this application provides a computer-readable storage medium for storing a computer program for performing the method as described in any one of the first aspects.

[0035] Fifthly, this application provides a computer program product comprising one or more computer instructions, wherein when the computer instructions are executed by a computer, the computer performs the method as described in any one of the first aspects.

[0036] As can be seen from the above technical solution, this application has at least the following beneficial effects:

[0037] In this application, by simultaneously acquiring the initial simulation model and physical test data, a basic model carrier and real physical reference are provided for subsequent model calibration and simulation analysis, effectively avoiding the problem of simulation and reality being disconnected due to the lack of measured data support. At the same time, deterministic design variables and uncertain parameters are comprehensively identified, clarifying the parameters that can be actively controlled in the design and the key variable parameters that affect the stability of system performance. This achieves the sorting and classification of all system parameters, defining a clear parameter range for subsequent targeted optimization and calibration work, and avoiding design loopholes caused by incomplete parameter identification.

[0038] The initial simulation model was calibrated using physical test data to obtain a baseline simulation model. Measured data was then used to correct parameters such as contact parameters and material properties, ensuring that the simulation model's output response closely matched the actual physical test results. This significantly improved the simulation model's realism, reliability, and accuracy. The calibrated baseline simulation model served as a unified benchmark model for all subsequent uncertainty quantification, proxy model construction, and optimization design, effectively guaranteeing the validity and consistency of simulation analysis results in each subsequent stage. This laid a reliable model foundation for the entire collaborative design process, avoiding deviations in subsequent design results caused by distortion of the base model.

[0039] Based on the adaptive multidimensional parallelepiped interval model, the value range of uncertainty parameters and the correlation between parameters are quantified, and joint uncertainty domain information is obtained. This model characterizes the fluctuation range of uncertainty parameters and the intrinsic correlation between them, overcoming the problem of uncertainty assessment distortion caused by neglecting parameter correlation in traditional methods. Simultaneously, the correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism, integrating historical information, the latest empirical data, and physical mechanism knowledge. This allows the joint uncertainty domain information to reflect the actual distribution and correlation changes of parameters in real time and accurately, providing a parameter characterization basis for subsequent high-dimensional uncertainty analysis. Furthermore, the joint uncertainty domain information integrates information such as the parameter midpoint, radius, and correlation coefficient matrix, achieving a systematic and standardized characterization of uncertainty parameters, facilitating unified access and analysis in subsequent stages.

[0040] Based on the baseline simulation model and joint uncertainty domain information, sample points are generated through experimental design methods, and system response data is obtained through simulation. A surrogate model is then constructed. A scientific sampling method is used to achieve efficient exploration of the high-dimensional input space, covering all parameter features with a small number of sample points, which significantly reduces the sample size for simulation calculations. The system response data obtained through automated simulation provides a complete input and output dataset for the construction of the surrogate model, ensuring the effectiveness of model training. The constructed high-precision surrogate model can quickly and accurately predict the system response, replacing the computationally expensive and time-consuming baseline simulation model. This realizes the transformation from computationally intensive physical simulation to rapid mathematical evaluation, effectively solving the curse of dimensionality problem in high-dimensional uncertainty analysis, and significantly improving the computational efficiency of subsequent optimization and calibration stages, ensuring the feasibility of the entire collaborative design method in engineering practice.

[0041] Based on the surrogate model and joint uncertainty domain information, optimization considering parameter correlation is carried out on deterministic design variables to obtain the optimal design scheme. Incorporating parameter correlation into the optimization considerations makes the optimization process closely resemble the real-world scenario of interrelated uncertain parameters in engineering practice, effectively avoiding the performance degradation or even failure of design schemes in practical applications caused by neglecting parameter correlation in traditional deterministic optimization. Relying on the rapid computational capability of the surrogate model and combined with efficient optimization algorithms, accurate optimization of deterministic design variables is achieved, significantly improving optimization efficiency while ensuring optimization accuracy. The final optimal design scheme can stably meet various safety performance constraints under the influence of mixed uncertain parameters, taking into account both the performance optimization of the occupant constraint system and the robustness of engineering applications, and setting the optimal design benchmark for subsequent parameter calibration.

[0042] After fixing the optimal design scheme, the uncertain parameters are calibrated with the dual objectives of minimizing the time-series response error and maximizing the Bayesian factor, and the optimal calibration parameters are obtained. Through dual-objective optimization, the simulation accuracy and statistical reliability are synergistically improved, overcoming the limitations of traditional calibration methods that only pursue simulation curve fitting and lack statistical confidence assessment. The time-series response error is comprehensively evaluated using the moving distance index, which comprehensively considers the deviations in multiple dimensions such as amplitude, phase, and shape, and significantly improves the accuracy of the assessment of the consistency between simulation and experimental data. The statistical reliability of the calibration parameters is quantified by the Bayesian factor, providing rigorous probabilistic statistical support for the calibration results, and ensuring that the calibration parameters are supported by sufficient physical experimental data. The final optimal calibration parameters enable the simulation model to reproduce the real physical response with high quality under the optimal design scheme, greatly improving the reliability and credibility of the simulation model, providing accurate parameter support for system performance prediction and safety decision-making, and ultimately achieving a high-confidence design.

[0043] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0044] Figure 1A flowchart illustrating a collaborative design method for an occupant restraint system provided in an embodiment of this application;

[0045] Figure 2 A structural diagram of a collaborative design device for an occupant restraint system provided in an embodiment of this application;

[0046] Figure 3 This is a schematic diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0047] The terms "first," "second," and "third," etc., used in this application specification and accompanying drawings are used to distinguish different objects, not to limit a specific order.

[0048] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0049] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the related technologies is given first:

[0050] Currently, the mainstream design methodology for occupant restraint systems in the industry primarily utilizes Computer-Aided Engineering (CAE). This approach involves constructing multibody dynamics and finite element simulation models, combined with deterministic optimization algorithms, to conduct an iterative development process of design, simulation, and optimization. By identifying deterministic design variables such as seatbelt force-limiting loads and airbag vent diameters, this method optimizes performance based on fixed simulation model parameters. Simultaneously, it relies on engineer experience or global sampling methods to calibrate the simulation model's parameters, thereby improving the performance of the restraint system to a certain extent and becoming an indispensable technical means in vehicle development.

[0051] However, existing design and calibration methods still have many fundamental shortcomings in practical engineering applications, making it difficult to meet the design requirements of high confidence and high robustness. First, traditional methods are based on idealized deterministic simulation assumptions, ignoring the widespread engineering uncertainties such as material property dispersion, manufacturing and assembly tolerances, and fluctuations in contact friction coefficients. Furthermore, they fail to consider the complex correlations between various uncertainty parameters, leading to performance degradation or even failure of optimized solutions under actual working conditions. Second, the model calibration process relies heavily on manual trial and error based on experience, lacking quantitative evaluation of the statistical confidence of the calibration results, resulting in significant hidden dangers in guiding high-risk safety decisions using simulation models. Third, existing methods cannot systematically handle multi-source heterogeneous uncertainties such as interval, random, and fuzzy uncertainties, nor do they consider the impact of occupant dynamic attitude on the system response before a collision, leading to insufficient scenario adaptability of the design solutions. Fourth, high-dimensional uncertainty analysis suffers from low computational efficiency; traditional precise calculation methods struggle to achieve a balance between accuracy and efficiency, limiting the large-scale application of these methods in practical engineering.

[0052] In view of this, embodiments of this application provide a collaborative design method for an occupant restraint system. To make the technical solution of this application clearer and easier to understand, the collaborative design method for an occupant restraint system provided in embodiments of this application is described below. Figure 1 As shown in the figure, this is a flowchart illustrating a collaborative design method for an occupant restraint system provided in an embodiment of this application. The collaborative design method for the occupant restraint system includes:

[0053] S101. Obtain the initial simulation model and physical test data corresponding to the occupant constraint system, and identify the deterministic design variables and uncertain parameters in the occupant constraint system.

[0054] Uncertainty parameters include at least one of interval uncertainty parameters, random uncertainty parameters, fuzzy uncertainty parameters, and pre-collision attitude parameters. Pre-collision attitude parameters characterize the dynamic initial attitude of the occupants relative to the vehicle interior at the moment of collision.

[0055] Optionally, based on the degree of influence of interval uncertainty parameters, random uncertainty parameters, and fuzzy uncertainty parameters on the occupant constraint system, an uncertainty contribution index corresponding to each uncertainty parameter can be obtained; and the allocation of computing resources can be adjusted based on the uncertainty contribution index.

[0056] For example, the Uncertainty Contribution Index (UCI) can be calculated as follows:

[0057]

[0058] in, This represents the contribution of the i-th type of uncertainty parameter to the total uncertainty. The value ranges from 0 to 1. The larger the value, the greater the impact of this type of uncertainty on the system fluctuation. This represents the variance (or the square of the interval radius, etc.) of the occupant constraint system response R when only considering the i-th type of uncertainty parameter (such as one of interval uncertainty, random uncertainty, or fuzzy uncertainty). This value is obtained through a surrogate model combined with sampling analysis. This represents the total fluctuation measure of the system response R caused by all categories of uncertainty parameters; j represents the traversal index of all uncertainty categories, used to sum the fluctuation measures of all uncertainty sources to obtain the total fluctuation; R represents the system response of the occupant restraint system, which usually refers to key occupant injury indicators such as head injury criteria, chest compression, tibial force, or other performance outputs.

[0059] Subsequently, based on the calculated UCI values, the impact of different types of uncertainty parameters on the occupant constraint system is quantified. Based on this, the allocation of computational resources in the subsequent design and calibration process is dynamically adjusted. For example, for uncertainty parameters with high UCI values ​​and large impact, more computational resources are allocated for refined analysis; for uncertainty parameters with low UCI values ​​and small impact, the calculation process is appropriately simplified or the sampling precision is reduced, thereby improving the overall computational efficiency while ensuring the accuracy of the analysis.

[0060] The initial simulation model refers to the computer simulation model of the occupant restraint system built based on multibody dynamics and the finite element method, which has not been calibrated with physical test data. The model includes major components such as occupants (e.g., Hybrid III 50th dummy), seat belt system (including pretensioners and force limiters), airbags, and vehicle interior, and serves as the basic prototype for subsequent model calibration and simulation analysis.

[0061] Physical test data is time-series data of occupant dynamic response collected through real vehicle crash tests or skid tests. It includes key indicators such as head acceleration, chest compression, tibia force, and head injury criterion (HIC). It serves as a true reference for calibrating simulation models and verifying model accuracy.

[0062] Deterministic design variables are parameters that can be actively controlled during the design and optimization phase of the occupant restraint system and can be precisely realized during the manufacturing process. They are the main design parameters that designers can adjust, and typically include the rated load of the seat belt force limiter, the diameter of the airbag vent, the airbag ignition sequence, and the pretensioner trigger time.

[0063] Uncertainty parameters are parameters whose values ​​cannot be precisely determined or have inherent variability due to cognitive limitations, manufacturing tolerances, material property dispersion, changes in usage conditions, etc. They are key factors affecting the actual performance stability of occupant restraint systems and mainly include the following four categories:

[0064] Interval uncertainty parameters are parameters whose upper and lower bounds can be estimated, but whose exact probability distribution is unknown. Typical examples include the range of friction coefficients between a dummy and a restraint system, the range of stiffness fluctuations in seat belt webbing, and the range of permeability of airbag fabric.

[0065] Random uncertainty parameters are parameters whose fluctuations follow known or hypothetical statistical laws and can be described by joint probability density functions or probability distribution functions. Typical examples include the normal distribution fluctuation of gas production rate in a gas generator and the random deviation of sensor trigger time.

[0066] Fuzzy uncertainty parameters are parameters that are inherently fuzzy and have unclear boundaries, requiring membership functions to characterize them. Typical examples include fuzzy evaluations of occupant muscle tension (relaxed, normal, tense) and fuzzy descriptions of the aging state of interior materials.

[0067] Pre-collision attitude parameters can be extracted from pre-collision simulation systems or on-board monitoring data. They are used to characterize the dynamic initial attitude of the occupant relative to the vehicle interior at the moment of collision, such as the three-dimensional displacement and rotation of the head relative to the hip point. They directly affect the occupant's dynamic response during the collision.

[0068] The uncertainty contribution index is used to quantify the degree of influence of various uncertain parameters, such as interval, random, and fuzzy, on the response fluctuation of the occupant constraint system. It is calculated as the ratio of the system response variance (or the square of the interval radius, etc.) caused by a certain type of uncertainty parameter to the total system response fluctuation caused by all uncertain parameters. It can intuitively reflect the influence weight of various uncertain parameters on system performance.

[0069] For example, to achieve comprehensive acquisition of simulation and experimental data, identification of design and uncertainty parameters, and to rationally allocate computational resources through uncertainty contribution indicators, the specific implementation process is as follows: First, an initial simulation model of the occupant restraint system, including major components such as occupants, seat belts, airbags, and vehicle interior, is built. Physical experimental data on key occupant dynamic responses, such as head acceleration and chest compression, are collected through real-vehicle collision or slide tests, providing a basic model and real reference for subsequent model calibration. Second, two main types of parameters in the system are comprehensively identified: one type is deterministic design variables that designers can actively control and that can be precisely achieved in manufacturing; the other type is uncertain parameters whose values ​​vary or cannot be precisely determined. Among these, uncertain parameters must at least include interval uncertain parameters and random uncertain parameters. One of the following parameters is required: deterministic parameters, fuzzy uncertainty parameters, and pre-collision attitude parameters. The pre-collision attitude parameters must accurately characterize the dynamic initial attitude of the occupants relative to the vehicle interior at the moment of collision, thus comprehensively covering various parameters that affect system performance in actual engineering. Finally, the influence degree analysis of uncertainty parameters can be carried out according to engineering requirements. By calculating the uncertainty contribution index corresponding to various types of uncertainty parameters such as interval, random, and fuzzy, the influence weight of different types of uncertainty parameters on the occupant constraint system response can be quantified. Based on this index, the allocation of computing resources in the subsequent design and calibration process can be adjusted in a targeted manner. More computing resources are allocated to uncertainty parameters with high contribution and large impact for refined analysis, while the calculation of parameters with low contribution and small impact is appropriately simplified, thereby improving the computational efficiency of the overall design process while ensuring the accuracy of the analysis.

[0070] Optionally, the process for identifying deterministic design variables and uncertainty parameters is as follows:

[0071] 1) Deterministic design variables (d) are parameters that can be actively controlled during the design and optimization phases, and whose values ​​can be precisely achieved during manufacturing. These parameters form a vector d=[ , ,…, ] and determine its design space. ≤d≤ Typical deterministic design variables include, but are not limited to, the rated load of the seatbelt force limiter, the diameter of the airbag vent, the airbag ignition sequence, and the pretensioner trigger time. In this vector d, each element represents a specific deterministic design variable, such as the rated load of the seatbelt force limiter or the diameter of the airbag vent; D represents the total number of deterministic design variables, i.e., the dimension of the vector. This is a lower bound vector for deterministic design variables, where each element corresponds to the minimum allowable value of a design variable (such as the minimum rated load of a seat belt force limiter). This is an upper limit vector for deterministic design variables, where each element corresponds to the maximum allowable value of a design variable (such as the maximum diameter of the airbag vent).

[0072] 2) Uncertainty parameters are parameters whose values ​​cannot be precisely determined or have inherent variability due to limitations in knowledge, manufacturing tolerances, material dispersion, or changes in usage conditions. Uncertainty parameters can be further classified as:

[0073] A. Uncertainty parameter X=[ , ,..., ] T Let be a column vector consisting of q uncertain parameters. The upper and lower bounds of the parameters can be estimated, but their exact probability distributions are unknown. The lower bound of each uncertain parameter is . The upper bound of the value is Typical examples include: the range of friction coefficients between the dummy and the restraint system, the fluctuation range of seat belt webbing stiffness, and the range of airbag fabric permeability. These uncertainties often stem from insufficient understanding or simplification of parametric models; q represents the total number of uncertain parameters.

[0074] B. A random uncertainty parameter Y, whose joint probability distribution function or probability density function is known or can be assumed. Its fluctuations follow known or assumed statistical laws and can be described by the joint probability density function or probability distribution function. Typical examples include: the normal distribution fluctuation of gas generator production rate, and the random deviation of sensor trigger time. This type of uncertainty usually originates from inherent manufacturing or environmental random factors.

[0075] C. The fuzzy uncertainty parameter Z is inherently fuzzy in its description, with unclear boundaries, and requires characterization using membership functions. Typical examples include: fuzzy evaluation of occupant muscle tension (relaxed, normal, tense), and fuzzy description of the aging state of interior materials.

[0076] D. Pre-collision attitude parameters ( This parameter, extracted from pre-collision simulation systems or onboard monitoring data, characterizes the dynamic initial posture of the occupant relative to the vehicle interior at the moment of collision, such as the three-dimensional displacement and rotation of the head relative to the hip point. This parameter incorporates the pre-collision dynamic scene into the optimization and evaluation framework, enhancing the method's adaptability to active safety and integrated safety systems.

[0077] S102. Based on the physical test data, the initial simulation model is calibrated to obtain the baseline simulation model.

[0078] Calibration refers to the process of adjusting the internal parameters of the initial simulation model, such as contact parameters and material properties, so that the degree of agreement between the occupant dynamic response data output by the simulation model and the physical test data reaches the preset accuracy standard. It is an important step in improving the realism and credibility of the simulation model.

[0079] The baseline simulation model refers to the occupant constraint system simulation model that closely matches the actual physical test response after calibration with physical test data. It serves as the foundation and benchmark simulation model for all subsequent analysis stages, such as uncertainty quantification, surrogate model construction, and optimization design.

[0080] For example, in order to complete the accuracy calibration and benchmark establishment of the simulation model, physical test data collected from real vehicle collision or sliding test can be used as the real reference. By iteratively adjusting key internal parameters such as contact stiffness, friction coefficient, and material properties in the initial simulation model, the accuracy of the occupant dynamic response time series data output by the model and the physical test data can reach the preset standard. Finally, a baseline simulation model that highly matches the response of the actual physical system is obtained, providing a reliable model foundation for subsequent full-process simulation analysis and optimization design.

[0081] Optionally, the initial simulation model is calibrated based on physical test data to ensure that the cross-correlation coefficient (CCC) between the simulation response and the physical test data is greater than 0.9, and the amplitude error of the head injury criterion (HIC) and chest compression is less than 5%, thereby obtaining a validated baseline simulation model.

[0082] The cross-correlation coefficient (CCC) is used to evaluate the degree of agreement between simulated values ​​and physical experimental values. Its definition is as follows:

[0083]

[0084] in, Cross-correlation coefficient (CCC), with a value range of [ [1,1], the closer to 1, the higher the simulation value. Compared with test values The higher the consistency, such as >0.9, meaning the simulation and experimental data achieve an extremely high degree of agreement; These are experimental values. These are simulation values. yes (Experimental data) and Covariance of (simulation data); For experimental values The variance; Simulation values The variance. For experimental values The mean; Simulation values The mean.

[0085] This formula provides objective and quantitative model validation criteria. Its calculation results directly determine the usability of the baseline model and serve as the basis for all subsequent analyses and optimizations.

[0086] S103. Based on the adaptive multidimensional parallelepiped interval model, the value range of the uncertainty parameters and the correlation between the uncertainty parameters are quantified to obtain the joint uncertainty domain information.

[0087] The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter, as well as the correlation coefficient matrix between each uncertainty parameter; the correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism.

[0088] Optionally, the online adaptive update mechanism dynamically adjusts the correlation coefficient matrix based on historical correlation information, the latest empirical correlation information, and prior correction information based on physical mechanisms.

[0089] The adaptive multi-dimensional parallelel-hexahedron interval model, or MP-I model, is a mathematical model that is constructed for interval uncertainty parameters and can accurately quantify the correlation and dynamic evolution between parameters. It defines the joint uncertainty domain of related interval variables as a multi-dimensional parallelepiped that can be moved, rotated, and stretched in the parameter space. It is the main model for quantifying parameter correlation in this method.

[0090] Joint uncertainty domain information is a comprehensive set of information used to fully characterize the range of values ​​of uncertainty parameters and the correlation characteristics between parameters. It includes at least the midpoint, radius and correlation coefficient matrix of each uncertainty parameter, and is an important input for subsequent uncertainty analysis and optimization design.

[0091] The midpoint of an uncertainty parameter is the center value of the range of values ​​for a single uncertainty parameter. It is calculated from the mean of the upper and lower bounds of the parameter and is a quantitative indicator that represents the center position of the parameter's value.

[0092] The radius of uncertainty parameter is half the range of values ​​for a single uncertainty parameter. It is calculated from half the difference between the upper and lower bounds of the parameter and is a quantitative indicator that characterizes the range of fluctuation of the parameter value.

[0093] The correlation coefficient matrix is ​​a symmetric matrix with elements ranging from -1 to 1. It is used to quantify the degree of linear correlation between any two uncertain parameters and is a quantitative indicator characterizing the correlation between parameters.

[0094] The online adaptive update mechanism is an algorithmic mechanism used to dynamically adjust the correlation coefficient matrix. It can integrate historical data, the latest sampled data, and physical domain knowledge, so that the correlation coefficient matrix can reflect the actual changes in the correlation between parameters in real time and avoid the analysis distortion caused by a fixed matrix.

[0095] Historical correlation information refers to the correlation information between uncertain parameters contained in the correlation coefficient matrix of the previous iteration stage, and serves as the basic reference for dynamic updates of correlation.

[0096] The latest empirical correlation information is based on the sampling data generated during the current optimization and calibration processes. The empirical correlation coefficient matrix contains correlation information that reflects the latest actual correlation status between parameters.

[0097] The prior correction information based on physical mechanisms is a correlation coefficient correction term derived from engineering physics laws and domain prior knowledge. It is used to constrain pure data-driven correlation analysis and prevent model overfitting or physical inconsistencies.

[0098] For example, in order to achieve accurate and dynamic quantification of the value range of uncertain parameters and the correlation between parameters, a quantitative analysis of the identified uncertain parameters can be carried out based on the constructed adaptive multidimensional parallelepiped interval model. On the one hand, the value range of each uncertain parameter is determined, and the midpoint and radius corresponding to each parameter are calculated. On the other hand, the linear correlation between each uncertain parameter is accurately quantified, and a correlation coefficient matrix characterizing the correlation characteristics between parameters is constructed. The parameter midpoint, radius and correlation coefficient matrix are integrated into joint uncertainty domain information.

[0099] For the correlation coefficient matrix, an online adaptive update mechanism is adopted for dynamic adjustment. This mechanism can integrate historical correlation information, the latest empirical correlation information obtained based on the current sampled data, and prior correction information derived from physical mechanisms and prior knowledge. By dynamically balancing the contribution weights of the three types of information, the correlation coefficient matrix can be iteratively updated, so that the joint uncertainty domain information can reflect the actual distribution and correlation characteristics of uncertainty parameters in real time and accurately, providing a precise basis for uncertainty parameter characterization for subsequent surrogate model construction and optimization design.

[0100] Optionally, the adaptive multidimensional parallelepiped interval model can accurately quantify the correlation and dynamic evolution between uncertainty parameters X. The adaptive multidimensional parallelepiped interval model can be understood as defining the joint uncertainty domain information of the correlated uncertainty parameters X as a movable, rotatable, and scalable multidimensional parallelepiped in the parameter space, mathematically expressed as:

[0101]

[0102] in, For joint uncertainty domain information; The vector representing the midpoint of the interval determines the center position of the parallelepiped; It is directly calculated from the identified interval boundaries and represents the center of the uncertainty domain; It is a diagonal matrix of interval radii, such as Its diagonal elements =( - ) The interval radius of each parameter determines the semi-axis length (scaling) of the parallelepiped in each principal axis direction; S is the correlation transformation matrix, which determines the rotation and skew of the parallelepiped, thus accurately characterizing the correlation between the parameters. , Scaling matrix This is used to transform the correlation coefficient matrix ρ into a suitable scale. These are the main diagonal elements of the scaling matrix T, corresponding to the scaling coefficients of the 1st to qth uncertainty parameters, respectively. The matrix ρ contains information about the linear correlation coefficients between the parameters (ρ is the correlation coefficient matrix, a symmetric matrix whose elements...). ∈[ 1,1] represents the uncertainty parameter. and The degree of linear correlation between them); S is constructed by ρ (such as by Cholesky decomposition or principal component analysis) to map independent standardized interval variables U to correlated X, realizing the rotation and skew of the shape of the uncertain domain, thereby accurately characterizing the correlation; It is in The vector of independent standard interval variables that take values ​​within the space is used to drive the generation of the entire uncertainty domain and is the basic variable for model analysis; q is the number of uncertainty parameters.

[0103] To overcome the limitation of fixed correlation coefficient models in failing to reflect the dynamic evolution of parameter correlation during actual analysis, an online adaptive update mechanism for the correlation coefficient matrix ρ is proposed, which can be expressed as:

[0104]

[0105] in, This is the correlation coefficient matrix for the (k+1)th iteration; The correlation coefficient matrix for the k-th iteration is the correlation information obtained from the previous iteration and serves as the basis for this update. The empirical correlation coefficient matrix is ​​calculated based on the current sampling data (such as newly generated samples during optimization or calibration), representing the latest observed correlation information; It is a correlation coefficient correction term derived from physical mechanisms or prior knowledge, used to introduce domain knowledge constraints to prevent overfitting or physical inconsistencies in models driven purely by data. , The adaptive weighting coefficients are used to balance the contributions of historical information, new observation data, and physical knowledge. Their values ​​can be determined by model selection criteria such as the Akaike information content criterion and the Bayesian information criterion, or by dynamic optimization based on online learning algorithms based on prediction errors, to ensure that the model can both track changes in correlation and maintain stability.

[0106] For handling uncertain parameters related to spatiotemporal evolution (such as airbag pressure varying over time), the model is extended to a spatiotemporal MP-I model:

[0107]

[0108] in, This represents the uncertainty parameter at time t, whose value changes dynamically with time. The midpoint vector of the interval represents the central location of the uncertain parameter's value; It is a diagonal matrix of interval radii, whose diagonal elements are the interval half lengths of each parameter, which determine the degree of scaling of the uncertainty domain in each principal axis direction; This is a spatial correlation matrix based on location relationships, describing the correlation between parameters at different locations in space; This is a matrix of time-dependent kernel functions (such as exponential kernels and Gaussian kernels) that describes the decay of correlation between different time points. 0 This represents element-wise multiplication (Hadamard product). For spatiotemporally coupled interval variables, This is an additional random perturbation term. This extension enables the model to describe more complex dynamic uncertainties.

[0109] S104. Based on the baseline simulation model and joint uncertainty domain information, an experimental design method is used to generate sample points in the input space consisting of deterministic design variables and mixed uncertainty parameters. The system response data is obtained through simulation, and a surrogate model is constructed.

[0110] The experimental design method is a scientific sampling method used to efficiently explore high-dimensional input space and obtain representative sample points. In this step, the optimal Latin hypercube sampling (OLHS) is selected, which can minimize the spatial correlation of sample points and maximize the spatial filling while ensuring uniform projection of each dimension, so as to achieve a comprehensive exploration of high-dimensional space with a small number of samples.

[0111] The input space is a multidimensional parameter space composed of deterministic design variables and mixed uncertainty parameters of the occupant constraint system. It includes all parameter dimensions involved in the design and analysis and is the range boundary for experimental design and sample point generation.

[0112] Sample points are several sets of parameter combinations selected within the input space through experimental design methods. Each set of sample points corresponds to a set of deterministic design variables and mixed uncertainty parameters, and serves as the input basis for calling the simulation model for calculation.

[0113] System response data is the output index data of various occupant restraint systems obtained after substituting sample points into the baseline simulation model for collision simulation calculation. It mainly consists of time-series response data of occupant injury indicators such as head injury criteria, chest compression, and tibia index.

[0114] A surrogate model, also known as a meta-model or response surface model, is a mathematical model that can quickly predict the system response based on the input parameters of sample points and the system response data. In this step, high-precision models such as Kriging, Support Vector Regression (SVR), and Gaussian Process Regression (GPR) can be used to replace the computationally expensive baseline simulation model.

[0115] For example, by designing experimental sampling and constructing surrogate models, a transformation from high-cost physical simulation to rapid mathematical prediction can be achieved. Specifically, the following steps are taken: using a validated baseline simulation model as the computational foundation, and a joint input space including deterministic design variables and mixed uncertainty parameters as the sampling range, an experimental design method employing optimal Latin hypercube sampling is used to generate sample points. The number of sample points is determined based on the total dimension of the input space to meet the requirement of sample sufficiency. An automated script then batch-substitutes all sample points into the baseline simulation model to perform collision simulation calculations, obtaining system response data such as occupant injury indicators for each sample point, forming the input-output dataset for model training. For each independent system response indicator, a dedicated high-precision surrogate model is constructed based on the aforementioned dataset. The model type is selected according to the response characteristics, choosing Kriging, SVR, or GPR models. After training and validation, a surrogate model capable of quickly and accurately predicting the system response under arbitrary parameter combinations is obtained, providing an efficient computational tool for subsequent reliability optimization and parameter calibration, significantly reducing the computational cost of high-dimensional uncertainty analysis.

[0116] Optionally, based on the joint uncertainty domain information, a high-dimensional mixed input space can be constructed, such as a high-dimensional mixed input space composed of the following parameter categories:

[0117] Deterministic design variable d: dimension n, design space for each variable is d ;

[0118] Uncertainty parameter X: dimension q, the range of each parameter is as follows =[ , ];

[0119] The random uncertainty parameter Y has a dimension r and follows a known or assumed joint probability distribution.

[0120] The fuzzy uncertainty parameter Z: dimension s, described by the membership function;

[0121] Pre-collision attitude parameters P pre Dimension t is extracted from pre-collision simulation or monitoring data.

[0122] Furthermore, the total dimension of the high-dimensional hybrid input space is D = n + q + r + s + t, which is typically high-dimensional (e.g., ...). ).

[0123] Furthermore, the Optimal Latin Hypercube Sampling (OLHS) method can be used to generate training sample points. OLHS, while ensuring uniform projection distribution across each input dimension, minimizes the spatial correlation between sample points and maximizes spatial filling (e.g., based on the maximum-minimum distance criterion) through optimization algorithms (such as simulated annealing and genetic algorithms), thereby achieving a more comprehensive and uniform exploration of the high-dimensional space with fewer samples. The number of sampling points N is determined based on empirical criteria. To ensure sufficient sample size, and considering the need for subsequent model validation, N is usually set between 10D and 20D.

[0124] Furthermore, samples can be generated and simulations performed. For example, based on the coordinates of sample points generated by OLHS, validated baseline simulation models (such as MADYMO, LS-DYNA, etc.) can be called in batches via automated scripts to perform collision simulation calculations and obtain the occupant injury index response time-series data corresponding to each sample point, including but not limited to: head injury criteria. Chest compression Chest composite acceleration Thigh strength Safety indicators such as the Tibial Index (TI) are included. All response data constitute the dataset for training and testing the surrogate model.

[0125] Response to each individual injury metric (e.g.) Each specialized surrogate model is constructed based on the training set data. Depending on the response characteristics, one of the following high-precision models can be selected:

[0126] The Kriging model is suitable for highly nonlinear responses with spatial correlation. This model treats the response as a Gaussian stochastic process:

[0127]

[0128] in, The system response (such as occupant injury indicators in a collision) under input variable x (e.g., design variable, uncertainty parameter). This is a global trend item; It is a zero-mean Gaussian process that satisfies zero mean and covariance determined by the kernel function. Define a Gaussian process; the preferred kernel function is Matérn5 / 2, which strikes a good balance between smoothness and flexibility and can effectively capture local variations.

[0129] Support Vector Regression (SVR) models are suitable for high-dimensional, small-sample nonlinear regression problems. SVR uses kernel tricks to map the input to a high-dimensional feature space and constructs the optimal separating hyperplane in this space. A radial basis function kernel is used, and its hyperparameters (penalty coefficient C, kernel width γ) are optimized through grid search combined with cross-validation.

[0130] Gaussian Process Regression (GPR) models, like Kriging, belong to the Bayesian nonparametric model family and provide predicted mean and variance. By selecting appropriate mean functions (such as constant or linear) and covariance functions (such as RBF or Matérn), they can flexibly adapt to different response characteristics.

[0131] Model accuracy verification can also be performed, which can specifically include:

[0132] 1) Training and validation set partitioning: The total sample points are randomly divided into a training set and an independent test set in a ratio (e.g., 7:3); the training set is used for model parameter learning, and the test set is used to evaluate the model's generalization ability.

[0133] 2) Accuracy Validation Metrics and Strict Standards: The following metrics are calculated on independent test sets, and all surrogate models must simultaneously meet them:

[0134] A coefficient of determination R² > 0.9 measures the degree of linear correlation between model predictions and actual values; the closer it is to 1, the stronger the model's explanatory power.

[0135] If the mean absolute error (MAE) is less than 5%, it can be calculated. This is then normalized to a percentage relative to the true range of the indicator; where, This represents the total number of samples in the test set, i.e., the number of sample points in the independent test set. The true system response value of the i-th test sample (such as the true value of performance indicators of the occupant restraint system, such as head injury criterion and chest compression). The model predicts the response value for the i-th test sample, which is the prediction result output by a surrogate model such as Gaussian process regression (GPR).

[0136] For time-series response curves (such as chest acceleration), additional comprehensive error indicators such as movement distance are calculated, and corresponding thresholds are set (such as movement distance < 0.1) to ensure the fidelity of the dynamic response pattern.

[0137] Model Iteration Update Mechanism: In the subsequent optimization and calibration process, if new high-fidelity simulation data is obtained in areas where the surrogate model has high prediction uncertainty (such as large prediction variance), these data can be added to the training set as incremental samples to update and retrain the surrogate model online, gradually improving its prediction accuracy in key areas and forming an active learning loop.

[0138] S105. Based on the surrogate model and joint uncertainty domain information, optimize the deterministic design variables by considering parameter correlation to obtain the optimal design scheme.

[0139] One possible approach is to map correlated original interval uncertainty parameters into independent standard interval variables based on the correlation coefficient matrix and adaptive multidimensional parallelepiped interval model in the joint uncertainty domain information; based on the surrogate model and standard interval variables, a fast interval probability approximation algorithm is used to calculate the degree of reliability constraint satisfaction corresponding to the deterministic design variables; based on the degree of reliability constraint satisfaction, a multi-objective evolutionary algorithm is combined to iteratively optimize the deterministic design variables to obtain the optimal design scheme.

[0140] The degree of reliability constraint satisfaction characterizes the probability level of the design scheme corresponding to deterministic design variables, which satisfies the safety performance constraints of the occupant restraint system under the influence of mixed uncertainty parameters. It is calculated by the fast interval probability approximation algorithm and serves as the constraint basis for optimization design.

[0141] Independent standard interval variables are interval variables that are uniformly distributed and independent within the range of [-1,1], obtained by transforming the original interval uncertainty parameters with correlation through an adaptive multidimensional parallelepiped interval model. This can greatly simplify the calculation process of subsequent reliability analysis.

[0142] The fast interval probability approximation algorithm is an efficient calculation method based on Chebyshev's inequality. It can quickly estimate the probability that the system response meets the interval threshold constraint by using information such as the mean and variance of the response function. Under the premise of controllable accuracy loss (<2%), it can improve the calculation efficiency by 5-10 times.

[0143] Multi-objective evolutionary algorithms are intelligent optimization algorithms used to solve multi-objective optimization problems. In this step, algorithms such as NSGA-II and MOEA / D are preferred. They can obtain a non-dominated Pareto optimal solution set through population iterative evolution, which is suitable for the needs of collaborative optimization of multiple damage indices in occupant constraint systems.

[0144] Iterative optimization is the solution process of a multi-objective evolutionary algorithm. It generates offspring by performing operations such as selection, crossover, and mutation on the population in the design space, and iteratively optimizes the population based on non-dominated sorting and crowding degree calculation, gradually guiding the population to converge toward the Pareto optimal solution set, thereby achieving the search for the optimal design scheme.

[0145] The optimal design scheme is a combination of deterministic design variables selected from the Pareto optimal solution set obtained by the multi-objective evolutionary algorithm through multi-attribute decision-making methods such as TOPSIS and minimum distance method, which can stably satisfy safety constraints and have the optimal occupant injury index under the influence of uncertainty.

[0146] For example, to achieve robust optimization of deterministic design variables considering parameter correlation, the original interval uncertainty parameters can be transformed into independent variables within the range [-1,1] based on the correlation coefficient matrix in the joint uncertainty domain information and combined with the adaptive multidimensional parallelepiped interval model, simplifying reliability analysis calculations. Then, using the constructed high-precision surrogate model as the calculation basis, the standard interval variables and the deterministic design variables to be evaluated are substituted into the model, and a fast interval probability approximation algorithm is used to quickly calculate the degree to which the system response corresponding to the design variable satisfies the reliability constraint of the safety performance constraint. Finally, the degree to which the reliability constraint is satisfied is used as the optimization constraint, and the occupant injury index is used as the optimization objective. Multi-objective evolutionary algorithms such as NSGA-II and MOEA / D are used to iteratively optimize the deterministic design variables. Pareto optimal solution sets are obtained through population initialization, fitness evaluation, and evolutionary iteration. Then, multi-attribute decision-making methods are used to select the optimal design scheme that balances performance and robustness from the solution set, ensuring that the scheme can still stably meet the safety performance requirements under the influence of uncertainties in actual engineering.

[0147] Optionally, in order to find the optimal occupant constraint system design that can stably meet various safety performance constraints under the influence of mixed uncertainties (especially correlated interval uncertainties), it is necessary to deeply integrate uncertainty quantification and efficient reliability assessment methods into the traditional multi-objective optimization framework, forming a closed-loop optimization design process oriented towards engineering robustness.

[0148] Using the adaptive multidimensional parallelepiped interval (MP-I) model, the correlated original interval uncertainty parameter X is mapped to a set of independent and uniformly distributed standard interval variables U in the range [-1, 1]. This transformation process is reversible, and its mathematical expression is as follows:

[0149]

[0150] in, It is the inverse of the correlation transformation matrix S, which carries information about the correlation coefficient matrix ρ and is used to eliminate the correlation between variables; It is a diagonal matrix with the reciprocal of the interval radius as its diagonal elements, used for standardizing and scaling variables; Let be the vector of the midpoint of the interval.

[0151] The geometric meaning of this transformation is to transform the uncertain domain of a tilted and rotated multidimensional parallelepiped. This is mapped to a standard hypercube region. After this transformation, all interval variables become independent within the U space, and the range of each component is unified to [-1, 1], thus laying a solid foundation for the subsequent application of efficient reliability analysis algorithms within the standard interval space.

[0152] In reliability optimization, it is necessary to frequently assess the likelihood that constraints (such as chest compression not exceeding a certain safe range) will be met, i.e., to calculate whether the system response function G(X) does not exceed the safe range threshold. =[ , The interval probability Pr{G(X)≤] };in, This is the lower bound of the safe interval, and the minimum allowable value of the safe interval. The upper bound of the safety interval is defined as the maximum allowable value within the safety interval (e.g., the upper limit of chest compression), representing the safety constraints that the system performance must meet. Traditional precise calculation methods (such as the Monte Carlo method) involve enormous computational costs in high-dimensional cases. Therefore, this invention proposes a fast approximation algorithm based on the Chebyshev inequality, which improves computational efficiency by 5-10 times while ensuring controllable accuracy loss (<2%).

[0153] This method utilizes the mean and variance information of the response function G(X) to directly estimate the likelihood using a modified Chebyshev inequality, avoiding large-scale sampling. The specific formula is as follows:

[0154]

[0155] Where Pr is the interval probability; This is the system response function; Let G(X) be the mean of the response function; Let G(X) be the variance of the system response function. The response is the mean / central reference value; This refers to parameters related to the deviation correction amount / interval width. Chebyshev coefficients are adaptive adjustment parameters, calibrated using a small number of precise calculation samples (such as Monte Carlo) to balance approximate accuracy and conservatism.

[0156] These first and second moment information are rapidly obtained through a combination of a pre-constructed high-precision surrogate model and point estimation (PEM) or Taylor expansion, without the need for additional simulation; that is, in the fast interval probability approximation algorithm, It is also equivalent to For interval-type thresholds, the safe interval thresholds are transformed into an expression that matches the statistical characteristics of the system response, which facilitates the calculation of interval probability by combining Chebyshev's inequality. The response is the mean / central reference value; This refers to parameters related to the deviation correction amount / interval width.

[0157] Based on the aforementioned independence transformation and rapid probability calculation, a multi-objective optimization model for interval reliability is established. This model aims to find a set of deterministic design variables d such that, under the influence of uncertainty, the expected performance (represented by the midpoint of the interval) of multiple occupant injury indicators is optimal, while simultaneously satisfying all safety constraints with high probability (probability). Its mathematical model is as follows:

[0158]

[0159] in, To perform multi-objective optimization to minimize the deterministic design variable d, the objective is to reduce the expected performance of each occupant injury index. The midpoint of the interval under the influence of uncertainty for the j-th objective function (such as a certain damage index) is calculated through a surrogate model and interval analysis, where j=1,2,...,m, and m is the total number of objective functions; Pr is the interval probability, calculated using a fast algorithm. Let be the k-th constraint function (system response function), which is a function of the deterministic design variable d and the uncertainty parameter X, representing the k-th occupant injury index; Let be the allowed interval limit for the k-th constraint; The preset reliability threshold for the k-th constraint (e.g., 0.95); It is the index of the constraint function, and K is the total number of safety constraints.

[0160] The above model was solved using multi-objective evolutionary algorithms such as NSGA-II and MOEA / D to obtain the Pareto optimal solution set.

[0161] Finally, based on the minimum distance method, TOPSIS, or multi-attribute decision method, the final robust optimal design scheme is selected from the Pareto optimal solution set.

[0162] 1) The optimization solution process may include:

[0163] Population initialization, that is, randomly generating an initial design scheme population within the design space;

[0164] Fitness assessment, which involves rapidly calculating all objective function values ​​for each individual (design) in the population using a surrogate model. ) and constraint violation degree (based on probability);

[0165] Evolutionary iteration refers to the algorithm generating offspring through selection, crossover, and mutation operations, and iteratively evolving the population based on mechanisms such as non-dominated sorting and crowding calculation, guiding the search to converge toward the Pareto optimal solution set;

[0166] The convergent output, i.e. the final output of the algorithm, is a set of Pareto optimal solutions. The solutions in this set are non-dominated in the target space and represent different performance-robustness trade-offs.

[0167] 2) The final decision on the solution may include:

[0168] Selecting the final implementation scheme from the Pareto optimal solution set is a multi-attribute decision problem. Methods such as TOPSIS (Topological Solution Ranking) or the minimum distance method can be used. Taking TOPSIS as an example:

[0169] a. First, determine the positive ideal solution (the virtual point formed by the optimal value of each objective) and the negative ideal solution (the virtual point formed by the worst value of each objective) for each optimization objective.

[0170] b. Then calculate the Euclidean distance between each Pareto solution and the positive and negative ideal solutions;

[0171] c. Finally, select the solution that is closest to the ideal solution as the final robust optimal design. Decision-makers can also assign different weights to different objectives based on engineering experience to reflect design preferences.

[0172] This step enables the completion of the entire process from uncertainty quantification to robust optimization decision-making, and allows for the systematic design of reliable solutions for occupant constraint systems under complex real-world uncertainties.

[0173] S106. Using the optimal design scheme as a fixed input, with the goal of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters, the uncertainty parameters are calibrated to obtain the optimal calibration parameters.

[0174] The timing response error is comprehensively evaluated using the shift distance index, which includes amplitude error, phase error, and shape error components.

[0175] The calibration parameters are the uncertainties of the occupant constraint system that need to be identified with high precision. These parameters are difficult to measure directly and accurately, but they have a significant impact on the simulation results and are the objects of this calibration.

[0176] Timing response error is the deviation between the dynamic response timing data output by the simulation model and the same type of timing data collected by physical experiments. It is an indicator for measuring the degree of agreement between simulation and experimental data.

[0177] The moving distance index is a quantitative indicator used to comprehensively evaluate the overall deviation between time series response curves. It integrates three types of error components: amplitude, phase, and shape, and can comprehensively reflect the matching degree between simulation and experimental time series data in terms of numerical magnitude, time synchronization, and curve trend.

[0178] Amplitude error is one of the components of the distance index, referring to the deviation in numerical value between the simulation and experimental time-series response curves. In this method, distance calculation and normalization based on Dynamic Time Warping (DTW) can overcome the evaluation bias caused by misalignment of time-series data.

[0179] Phase error is one of the components of the distance travel index. It refers to the time difference between the occurrence of key feature points of the simulation and experimental time-series response curves, reflecting the synchronicity of the curves in the time dimension.

[0180] Morphological error is one component of the distance travel index. It refers to the difference in the distribution of the first derivative of the simulation and experimental time-series response curves, reflecting the degree of matching between the overall trend and slope characteristics of the curves.

[0181] The Bayes factor is an index used to quantify the statistical confidence of calibration parameters. It is the ratio of the marginal likelihood of the simulation model with calibration parameters as input to the baseline model under given physical experimental data. The larger the value, the higher the support of the experimental data obtained by the calibration parameters.

[0182] The optimal calibration parameters are the best combination of uncertainty parameters obtained under the dual objectives of minimizing the time-series response error and maximizing the Bayes factor, which combines simulation accuracy and statistical reliability.

[0183] For example, to achieve accurate identification of uncertain parameters based on dual objectives, the optimal design scheme obtained in the previous steps can be substituted into the simulation model as a fixed input, isolating the influence of design variables on the calibration process and focusing the calibration on uncertain parameters. The calibration solution for uncertain parameters is carried out with the dual optimization objectives of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayesian factor of the calibration parameters. The time-series response error is comprehensively evaluated using the moving distance index, which integrates three error components: amplitude, phase, and shape, and can comprehensively and accurately reflect the overall deviation between the simulation and experimental time-series data. The dual objectives are synergistically optimized through a multi-objective optimization solution strategy to obtain the optimal combination of calibration parameters that achieves the best trade-off between simulation accuracy and statistical reliability. This combination of parameters ensures that the simulation model can reproduce the real response of the physical test with high quality and has sufficient statistical test data support, which greatly improves the reliability of the simulation model. Optionally, high-precision model calibration based on multi-objective Bayesian inference, after obtaining the optimal design scheme, fixes the design variables and shifts the optimization focus from finding the optimal design to identifying the most reliable model. This mainly targets a specific set of uncertain parameters (primarily uncertain parameters) and, based on physical experimental data, performs high-precision, high-statistical-confidence parameter calibration to ensure that the simulation model can reproduce the real physical response with high quality. First, the calibration problem is defined, and a benchmark is established. Specifically, this may include: the calibration objective, i.e., calibrating the uncertain parameters. These parameters are usually difficult to measure directly and accurately, but have a significant impact on the simulation results (e.g., contact friction coefficient, material property parameters).

[0184] Fixed conditions are used, that is, the optimal design scheme obtained through optimization is used as the fixed input for simulation, the influence of design variables is isolated, and the calibration process focuses on the identification of uncertain parameters;

[0185] Benchmark data refers to key dynamic response time-series data acquired through real-vehicle crash tests or high-precision slide tests. As a calibration standard, select a response that is sensitive to parameters and has clear engineering significance, such as the composite acceleration curve of the chest, the displacement curve of the pelvis, or the tibial force curve of the lower leg; the data needs to be preprocessed, including filtering (such as SAE CFC60 / 180 / 600), alignment (time zero point synchronization) and normalization.

[0186] Furthermore, a multi-objective calibration problem can be constructed, which may specifically include:

[0187] Objective 1 (accuracy objective), namely minimizing the combined error between simulation and experimental time-series data, adopts the movement distance index, which incorporates the magnitude error. Phase error And shape error. Among them, amplitude error Dynamic time warping distance calculation is employed to overcome timing misalignment issues, and normalization processing is performed. For the simulation response... With test response Its moving distance The calculation process is as follows:

[0188]

[0189] in, This is the amplitude error (based on the difference in the curve envelope area). Phase error (based on the time difference of key feature points); The normalized morphological error (based on the difference in the distribution of the first derivative of the curve). , , The corresponding weights can be set according to their importance.

[0190] Objective 2 (Confidence Objective): Maximizing the Bayes Factor (B) as a statistical measure of the confidence level of the uncertainty parameter X. The Bayes Factor is defined as the ratio of the marginal likelihoods of different parameter hypotheses given experimental data V1. ,in, The Bayes factor is a statistical measure of the confidence level of the uncertainty parameter X. It is a simulation model with X as a parameter. The baseline model (such as the prior mean model) is used, and the marginal likelihood is approximated by numerical integration or MCMC (Markov Chain Monte Carlo) sampling. It is the marginal likelihood (or evidence) of model M, representing the average probability of producing the observed data V2 under model M, which is obtained by numerical integration or MCMC sampling approximation over all possible model predictions (considering the inherent randomness of the model); Candidate models Generate the marginal likelihood of the observed data V2; As a benchmark model Generate the marginal likelihood of the observed data V2.

[0191] Marginal likelihood typically lacks an analytical solution and requires approximation using numerical methods, such as nested sampling or harmonic averaging estimation after MCMC sampling. According to the Kass and Raftery criterion, >3 indicates that there is "positive evidence" to support it. Superior , >10 indicates "strong evidence" >100 indicates "decisive evidence".

[0192] Therefore, the calibration problem is transformed into That is, within the feasible region of the uncertainty parameter X. Within this framework, we simultaneously minimize the movement distance error and maximize the Bayesian factor (equivalent to minimizing its negative value); where, The moving distance error represents the fitting error between the model's predictions and the experimental / observational data, and needs to be minimized. To transform maximizing the Bayesian factor B(X) into minimizing its negative value, a multi-objective minimization optimization framework is adapted. The Bayesian factor is a statistical measure of the reliability of the uncertainty parameter X. The larger the value, the better the parametric model can explain the observed data.

[0193] Furthermore, the problem can be solved and implemented based on efficient solution strategies. Solving the aforementioned multi-objective calibration problem faces challenges, such as the high cost of objective function computation (requiring the use of simulation or surrogate models) and the involvement of high-dimensional integrals in Bayesian factor calculation. This invention proposes two efficient solution strategies, as follows:

[0194] Strategy A, Multi-objective Bayesian Optimization (MOBO), treats the calibration problem as a black-box optimization problem. It utilizes a Gaussian Process (GP) to establish surrogate models for both objective functions, constructing an iterative optimization framework assisted by these surrogate models. The specific process is as follows: In each iteration, based on the current GP model and a multi-objective acquisition function (such as Expected HypervolumeImprovement, EHVI), a candidate parameter point with the greatest potential to improve the current Pareto optimal solution set is recommended. A high-fidelity simulation evaluation is performed on this point, and new data is added to the training set to update the GP model. This process is repeated until convergence. This method efficiently explores the parameter space with a minimal number of simulations.

[0195] Strategy B, a hybrid sampling strategy combining evolutionary algorithms and MCMC, combines global search with local fine-grained sampling. First, a multi-objective evolutionary algorithm (such as NSGA-II) is used for global exploration to quickly obtain a broad-coverage approximation of the Pareto optimal solution set. Then, for promising regions on the Pareto optimal solution set, Markov Chain Monte Carlo (MCMC) sampling, such as No-U-Turn Sampler (NUTS), is initiated to accurately depict the posterior distribution of the parameters and reliably calculate the Bayesian factor. The specific process is as follows: NSGA-II provides a diverse initial population; during the evolutionary process, the constructed surrogate model is used to quickly evaluate the objective function; after convergence, representative solutions on the Pareto optimal solution set are selected as multiple starting points for MCMC sampling, and multiple Markov chains are run in parallel to deeply explore their neighborhoods and calculate high-precision marginal likelihoods.

[0196] Finally, the decision regarding the calibration results is made, and the results are applied. The solution yields a Pareto optimal calibration solution set, where each solution represents a combination of parameters with different trade-offs between accuracy and confidence. The final calibration value can be selected according to specific requirements. If the goal is to achieve the highest statistical confidence, then the solution with the largest Bayesian factor should be directly selected as the optimal choice. This ensures that the calibration parameters are supported by the strongest experimental data. If a balance between accuracy and confidence is required, the Pareto optimal solution set can be selected based on additional criteria (such as a threshold for a key error), or a comprehensive optimal solution can be selected using methods such as TOPSIS.

[0197] The final calibration value of the final output This represents a complete simulation model that has undergone robust optimization and high-confidence calibration, and can be directly used for engineering development and safety assessment. Not only does this model achieve optimal performance at the design point, but its uncertainty parameter X has also undergone rigorous statistical testing, significantly improving the model's credibility and reliability in guiding safety decisions, making performance predictions, and extrapolating.

[0198] S107. Based on the optimal design scheme and optimal calibration parameters, determine the final design data of the occupant restraint system.

[0199] The final design data is a complete set of parameter data that can directly guide the engineering research and development, production and manufacturing and performance verification of the occupant restraint system. It is composed of the deterministic design variable values ​​corresponding to the optimal design scheme and the uncertain parameter identification values ​​corresponding to the optimal calibration parameters. It is the final output of this collaborative design method.

[0200] For example, in order to integrate and determine the design data and provide a direct basis for practical engineering applications, the optimal design scheme obtained by multi-objective robust optimization considering parameter correlation in the previous steps, and the optimal calibration parameters obtained by bi-objective Bayesian calibration, can be integrated. The optimal design scheme is the deterministic design variable value of the occupant restraint system that can be actively controlled and precisely manufactured, while the optimal calibration parameters are the accurately identified uncertain parameters that have a significant impact on the simulation results and have been statistically verified. Together, they constitute the final design data of the occupant restraint system.

[0201] This set of design data has undergone robustness verification under the influence of engineering uncertainties and is supported by high-precision, high-statistical-confidence simulation models. It can be directly applied to the actual development, production calibration and collision safety performance verification of occupant restraint systems, ensuring that the system has excellent and stable safety performance in actual working conditions.

[0202] It should be noted that this solution achieves deep integration and collaboration between design and calibration, that is, it integrates the robust optimization design of the occupant constraint system with the high-precision, high-confidence calibration of the simulation model. By first optimizing to obtain the optimal design scheme, and then performing Bayesian inference on the uncertainty parameter X based on the optimal design scheme, it avoids the problem of mismatch between the design results and the calibration model in the traditional step-by-step process, and ensures that the final solution (optimal design scheme, final calibration value) has high performance and high reliability.

[0203] It also improves the accuracy of uncertainty quantification and robust optimization; by introducing an adaptive MP-I model, it achieves accurate description and online updating of the dynamic correlation between interval parameters in occupant constraint system optimization. Combined with a hierarchical hybrid uncertainty framework, it systematically handles heterogeneous uncertainties such as cognitive, stochastic, and fuzzy uncertainties, making the reliability assessment and robust optimization results closer to engineering practice, effectively avoiding assessment distortion caused by ignoring correlations or the simplification of uncertainty types, and improving the robustness of the design scheme.

[0204] It also improves the computational efficiency of high-dimensional uncertainty analysis; for the problem of dimensionality curse in interval probability calculation and mixed uncertainty propagation, the proposed fast interval probability approximation algorithm and UCI-based model simplification strategy greatly improve the efficiency of related calculation links while ensuring controllable accuracy loss (e.g., <2%), making the method computationally feasible for handling complex engineering problems.

[0205] Furthermore, a statistical reliability standard for model calibration was established; the Bayesian factor was used as one of the calibration targets, forming a new paradigm for occupant constraint system model calibration from "empirical fitting" to "statistical reliability." This method not only pursues an intuitive match between simulation curves and experimental curves, but also provides a rigorous confidence quantification for calibration parameters from a probabilistic statistical perspective, greatly enhancing the reliability of simulation models in safety performance prediction and decision support.

[0206] It also enhances the method's adaptability to different scenarios by introducing pre-collision attitude parameters. The spatiotemporal correlation MP-I model enables this scheme to effectively address the dynamic impact of the pre-collision phase in integrated safety systems, as well as to model uncertainties with spatiotemporal variation characteristics, thus expanding the application potential of the method in scenario-based safety assessment and optimization for future intelligent vehicles.

[0207] Based on the above, the collaborative design method for occupant constraint systems provides a basic model carrier and real physical reference for subsequent model calibration and simulation analysis by simultaneously acquiring initial simulation models and physical test data. This effectively avoids the problem of simulation being disconnected from reality due to the lack of measured data support. At the same time, it comprehensively identifies deterministic design variables and uncertain parameters, clarifies the parameters that can be actively controlled in the design and the key variable parameters that affect the stability of system performance, and realizes the sorting and classification of all system parameters. This defines a clear parameter range for subsequent targeted optimization and calibration work, avoiding design loopholes caused by incomplete parameter identification.

[0208] The initial simulation model was calibrated using physical test data to obtain a baseline simulation model. Measured data was then used to correct parameters such as contact parameters and material properties, ensuring that the simulation model's output response closely matched the actual physical test results. This significantly improved the simulation model's realism, reliability, and accuracy. The calibrated baseline simulation model served as a unified benchmark model for all subsequent uncertainty quantification, proxy model construction, and optimization design, effectively guaranteeing the validity and consistency of simulation analysis results in each subsequent stage. This laid a reliable model foundation for the entire collaborative design process, avoiding deviations in subsequent design results caused by distortion of the base model.

[0209] Based on the adaptive multidimensional parallelepiped interval model, the value range of uncertainty parameters and the correlation between parameters are quantified, and joint uncertainty domain information is obtained. This model characterizes the fluctuation range of uncertainty parameters and the intrinsic correlation between them, overcoming the problem of uncertainty assessment distortion caused by neglecting parameter correlation in traditional methods. Simultaneously, the correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism, integrating historical information, the latest empirical data, and physical mechanism knowledge. This allows the joint uncertainty domain information to reflect the actual distribution and correlation changes of parameters in real time and accurately, providing a parameter characterization basis for subsequent high-dimensional uncertainty analysis. Furthermore, the joint uncertainty domain information integrates information such as the parameter midpoint, radius, and correlation coefficient matrix, achieving a systematic and standardized characterization of uncertainty parameters, facilitating unified access and analysis in subsequent stages.

[0210] Based on the baseline simulation model and joint uncertainty domain information, sample points are generated through experimental design methods, and system response data is obtained through simulation. A surrogate model is then constructed. A scientific sampling method is used to achieve efficient exploration of the high-dimensional input space, covering all parameter features with a small number of sample points, which significantly reduces the sample size for simulation calculations. The system response data obtained through automated simulation provides a complete input and output dataset for the construction of the surrogate model, ensuring the effectiveness of model training. The constructed high-precision surrogate model can quickly and accurately predict the system response, replacing the computationally expensive and time-consuming baseline simulation model. This realizes the transformation from computationally intensive physical simulation to rapid mathematical evaluation, effectively solving the curse of dimensionality problem in high-dimensional uncertainty analysis, and significantly improving the computational efficiency of subsequent optimization and calibration stages, ensuring the feasibility of the entire collaborative design method in engineering practice.

[0211] Based on the surrogate model and joint uncertainty domain information, optimization considering parameter correlation is carried out on deterministic design variables to obtain the optimal design scheme. Incorporating parameter correlation into the optimization considerations makes the optimization process closely resemble the real-world scenario of interrelated uncertain parameters in engineering practice, effectively avoiding the performance degradation or even failure of design schemes in practical applications caused by neglecting parameter correlation in traditional deterministic optimization. Relying on the rapid computational capability of the surrogate model and combined with efficient optimization algorithms, accurate optimization of deterministic design variables is achieved, significantly improving optimization efficiency while ensuring optimization accuracy. The final optimal design scheme can stably meet various safety performance constraints under the influence of mixed uncertain parameters, taking into account both the performance optimization of the occupant constraint system and the robustness of engineering applications, and setting the optimal design benchmark for subsequent parameter calibration.

[0212] After fixing the optimal design scheme, the uncertain parameters are calibrated with the dual objectives of minimizing the time-series response error and maximizing the Bayesian factor, and the optimal calibration parameters are obtained. Through dual-objective optimization, the simulation accuracy and statistical reliability are synergistically improved, overcoming the limitations of traditional calibration methods that only pursue simulation curve fitting and lack statistical confidence assessment. The time-series response error is comprehensively evaluated using the moving distance index, which comprehensively considers the deviations in multiple dimensions such as amplitude, phase, and shape, and significantly improves the accuracy of the assessment of the consistency between simulation and experimental data. The statistical reliability of the calibration parameters is quantified by the Bayesian factor, providing rigorous probabilistic statistical support for the calibration results, and ensuring that the calibration parameters are supported by sufficient physical experimental data. The final optimal calibration parameters enable the simulation model to reproduce the real physical response with high quality under the optimal design scheme, greatly improving the reliability and credibility of the simulation model, providing accurate parameter support for system performance prediction and safety decision-making, and ultimately achieving a high-confidence design.

[0213] The above text combined Figure 1 The collaborative design method of the occupant restraint system provided in the embodiments of this application has been described in detail. The apparatus and equipment provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0214] This application also provides a collaborative design device for occupant restraint systems, such as... Figure 2 As shown in the figure, this is a schematic diagram of a collaborative design device for an occupant restraint system provided in an embodiment of this application. The device includes:

[0215] The acquisition module 201 is used to acquire the initial simulation model and physical test data corresponding to the occupant constraint system, as well as to identify the deterministic design variables and uncertain parameters in the occupant constraint system;

[0216] Calibration module 202 is used to calibrate the initial simulation model based on physical test data to obtain a baseline simulation model;

[0217] The information determination module 203 is used to quantify the value range of the uncertainty parameters and the correlation between the uncertainty parameters based on the adaptive multidimensional parallelepiped interval model, so as to obtain the joint uncertainty domain information. The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter and the correlation coefficient matrix between the uncertainty parameters. The correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism.

[0218] The model building module 204 is used to generate sample points in the input space based on the baseline simulation model and joint uncertainty domain information, using experimental design methods; wherein, the input space consists of deterministic design variables and uncertainty parameters; the system response data of the sample points is obtained through simulation, and a surrogate model is built based on the system response data;

[0219] The optimization module 205 is used to optimize the deterministic design variables based on the parameter correlation characterized by the surrogate model and the joint uncertainty domain information to obtain the optimal design scheme; taking the optimal design scheme as a fixed input, the uncertainty parameters are calibrated with the goal of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters to obtain the optimal calibration parameters;

[0220] Design module 206 is used to determine the final design data of the occupant constraint system based on the optimal design scheme and optimal calibration parameters.

[0221] In some possible implementations, the collaborative design apparatus for the occupant restraint system also includes:

[0222] The resource adjustment module is used to obtain the uncertainty contribution index corresponding to each uncertainty parameter based on the degree of influence of interval uncertainty parameters, random uncertainty parameters, and fuzzy uncertainty parameters on the occupant constraint system; and to adjust the computational resource allocation based on the uncertainty contribution index.

[0223] In some possible implementations, the uncertainty parameters also include pre-collision attitude parameters, which characterize the dynamic initial attitude of the occupants relative to the vehicle interior at the moment of collision.

[0224] In some possible implementations, the online adaptive update mechanism dynamically adjusts the correlation coefficient matrix based on historical correlation information, the latest empirical correlation information, and prior correction information based on physical mechanisms.

[0225] In some possible implementations, the design module is specifically used for:

[0226] Based on the correlation coefficient matrix in the joint uncertainty domain information and the adaptive multidimensional parallelepiped interval model, the original interval uncertainty parameters with correlation are mapped to independent standard interval variables. Based on the surrogate model and the standard interval variables, the fast interval probability approximation algorithm is used to calculate the degree of reliability constraint satisfaction corresponding to the deterministic design variables. Based on the degree of reliability constraint satisfaction, the deterministic design variables are iteratively optimized by combining a multi-objective evolutionary algorithm to obtain the optimal design scheme.

[0227] In some possible implementations, timing response error is comprehensively evaluated using a shift distance metric, which includes amplitude error, phase error, and shape error components.

[0228] The collaborative design apparatus for an occupant restraint system according to the embodiments of this application can correspond to the execution of the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the collaborative design apparatus for an occupant restraint system are respectively for implementing Figure 1For the sake of brevity, the corresponding processes of each method in the illustrated embodiments will not be described in detail here.

[0229] This application also provides a computing device. For example... Figure 3 As shown in the figure, this is a schematic diagram of a computing device provided in an embodiment of this application. The computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0230] Bus 401 can be 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 representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0231] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0232] Communication interface 403 is used for communication with external devices.

[0233] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0234] The memory 404 stores executable code, and the processor 402 executes the executable code to perform the aforementioned co-design method for the occupant restraint system.

[0235] Specifically, in achieving Figure 2 In the case of the illustrated embodiment, and Figure 2When the modules or units of the collaborative design device for the occupant restraint system described in the embodiment are implemented in software, the execution... Figure 2 The software or program code required for the functions of each module / unit can be partially or entirely stored in memory 404. Processor 402 executes the program code corresponding to each unit stored in memory 404, and executes the aforementioned collaborative design method for the occupant restraint system.

[0236] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the aforementioned co-design method for the occupant restraint system.

[0237] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0238] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0239] When the computer program product is executed by a computer, the computer executes any of the aforementioned collaborative design methods for occupant restraint systems. The computer program product can be a software installation package; when any of the aforementioned collaborative design methods for occupant restraint systems is required, the computer program product can be downloaded and executed on the computer.

[0240] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0241] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A collaborative design method for an occupant constraint system, characterized in that, The method includes: Acquire the initial simulation model and physical test data corresponding to the occupant constraint system, and identify the deterministic design variables and uncertainty parameters in the occupant constraint system; The initial simulation model is calibrated based on the physical test data to obtain a baseline simulation model; Based on the adaptive multidimensional parallelepiped interval model, the value ranges of the uncertainty parameters and the correlations between the uncertainty parameters are quantified to obtain joint uncertainty domain information. The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter, as well as the correlation coefficient matrix between the uncertainty parameters. The correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism. This online adaptive update mechanism dynamically adjusts the correlation coefficient matrix based on historical correlation information, the latest empirical correlation information, and prior correction information based on physical mechanisms. Based on the baseline simulation model and the joint uncertainty domain information, experimental design methods are used to generate sample points in the input space; wherein, the input space consists of deterministic design variables and uncertainty parameters; The system response data of the sample points is obtained through simulation, and a proxy model is constructed based on the system response data; Based on the parameter correlation represented by the surrogate model and the joint uncertainty domain information, the deterministic design variables are optimized to obtain the optimal design scheme; Using the optimal design scheme as a fixed input, and with the objectives of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters, the uncertainty parameters are calibrated to obtain the optimal calibration parameters. Based on the optimal design scheme and the optimal calibration parameters, the final design data of the occupant restraint system is determined.

2. The method according to claim 1, characterized in that, The uncertainty parameter includes at least one of interval uncertainty parameter, random uncertainty parameter, and fuzzy uncertainty parameter; the method further includes: Based on the degree of influence of interval uncertainty parameters, random uncertainty parameters, and fuzzy uncertainty parameters on the occupant constraint system, the uncertainty contribution index corresponding to each uncertainty parameter is obtained; The allocation of computing resources is adjusted based on the uncertainty contribution index.

3. The method according to claim 2, characterized in that, The uncertainty parameters also include pre-collision attitude parameters, which characterize the dynamic initial attitude of the occupants relative to the vehicle interior at the moment of collision.

4. The method according to claim 1, characterized in that, The optimization of the deterministic design variables, considering parameter correlations, based on the surrogate model and the joint uncertainty domain information, to obtain the optimal design scheme includes: Based on the correlation coefficient matrix in the joint uncertainty domain information and the adaptive multidimensional parallelepiped interval model, the original interval uncertainty parameters with correlation are mapped to independent standard interval variables; Based on the surrogate model and the standard interval variables, the reliability constraint satisfaction degree corresponding to the deterministic design variables is calculated using a fast interval probability approximation algorithm; the fast interval probability approximation algorithm is a fast approximation algorithm based on the Chebyshev inequality. Based on the degree to which the reliability constraints are satisfied, the deterministic design variables are iteratively optimized using a multi-objective evolutionary algorithm to obtain the optimal design scheme.

5. The method according to claim 1, characterized in that, The timing response error is comprehensively evaluated using the moving distance index, which includes amplitude error, phase error and shape error components.

6. A collaborative design device for an occupant restraint system, characterized in that, The device includes: The acquisition module is used to acquire the initial simulation model and physical test data corresponding to the occupant restraint system, and to identify the deterministic design variables and uncertainty parameters in the occupant restraint system; The calibration module is used to calibrate the initial simulation model based on the physical test data to obtain a baseline simulation model; The information determination module is used to quantify the value range of the uncertainty parameters and the correlation between the uncertainty parameters based on an adaptive multidimensional parallelepiped interval model, thereby obtaining joint uncertainty domain information. The joint uncertainty domain information includes at least the midpoint and radius of each uncertainty parameter, as well as the correlation coefficient matrix between the uncertainty parameters. The correlation coefficient matrix is ​​dynamically updated through an online adaptive update mechanism. This online adaptive update mechanism dynamically adjusts the correlation coefficient matrix based on historical correlation information, the latest empirical correlation information, and prior correction information based on physical mechanisms. The model building module is used to generate sample points in the input space based on the baseline simulation model and the joint uncertainty domain information using experimental design methods; wherein the input space consists of deterministic design variables and uncertainty parameters; the system response data of the sample points is obtained through simulation, and a surrogate model is constructed based on the system response data; The optimization module is used to optimize the deterministic design variables based on the parameter correlation characterized by the surrogate model and the joint uncertainty domain information to obtain the optimal design scheme; and to calibrate the uncertainty parameters by taking the optimal design scheme as a fixed input, with the goal of minimizing the time-series response error between the system response data and the physical test data and maximizing the Bayes factor of the calibration parameters, to obtain the optimal calibration parameters. The design module is used to determine the final design data of the occupant restraint system based on the optimal design scheme and the optimal calibration parameters.

7. A computing device, characterized in that, Including memory and processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes one or more computer instructions, which, when executed by a computer, perform the method as described in any one of claims 1 to 5.

Images