Passenger and vehicle safety analysis method and system based on multi-dimensional collision simulation

Abstract

The invention relates to a passenger and vehicle safety analysis method and system based on multi-dimensional collision simulation, and belongs to the technical field of intelligent traffic control, and the analysis method comprises the steps: collecting vehicle structure parameters, passenger physiological data and collision environment parameters; biomechanical feature vectors are generated based on passenger physiological data, and a basic coupling model is constructed in combination with vehicle structure parameters; obtaining an initial collision speed according to the collision environment parameters, and dividing collision stages through a stage decision tree to obtain a collision stage division result; based on a collision stage division result, performing staged simulation on the basic coupling model through a dynamic solver, and outputting a multi-dimensional dynamic data set; generating a three-dimensional safety margin distribution map based on the multi-dimensional dynamic data set; and executing protection strategy decision logic according to the three-dimensional safety margin distribution diagram, and outputting constraint system control parameters. According to the method, the passive safety system can be promoted to develop from empirical design to the intelligent direction of data driving and scene self-adaption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent traffic control, in particular to a passenger and vehicle safety analysis method and system based on multi-dimensional collision simulation. BACKGROUND

[0002] With the rapid development of light-weight structure, high-strength material application and automatic driving technology, the traditional collision safety evaluation system based on standard dummies and fixed working conditions has been difficult to meet the needs of modern vehicles for fine, personalized and forward-looking safety performance. Especially under the promotion of increasingly stringent safety evaluation regulations, vehicle enterprises not only need to ensure that the vehicle meets the standards in various collision tests, but also need to deeply understand the passenger injury formation mechanism, optimize the response characteristics of the restraint system, and identify potential risks in the early development stage to reduce the cost of later real vehicle testing and recall risks. Therefore, building a comprehensive simulation analysis platform that can integrate vehicle structure response, human biomechanical behavior and external collision environment has become one of the core challenges of current automotive safety technology research and development.

[0003] Currently, the evaluation of vehicle collision safety performance generally adopts a serial simulation process of "structure first and then human body": first, the impact response of the vehicle body is analyzed based on the finite element method to obtain the boundary conditions such as passenger compartment deformation and acceleration field; then these data are input to drive the dummy model to calculate the biomechanical response. Although this step-by-step modeling method realizes the correlation analysis of the vehicle and the human body to some extent, it is still a decoupling process and does not fully consider the dynamic feedback mechanism between the human body and the structure. For example, when the passenger's chest force path changes significantly due to body size difference or forward-leaning posture, which affects the load transfer efficiency of the seat belt and airbag, this reverse effect is not effectively included in the solution process of the vehicle body response. Moreover, the current simulation model often has difficulty in balancing the calculation efficiency and physical fidelity in the time scale processing of different stages of the whole collision process. SUMMARY

[0004] In order to solve the above technical problems, the present application provides a passenger and vehicle safety analysis method and system based on multi-dimensional collision simulation.

[0005] In the first aspect, the present application provides a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, which adopts the following technical solution:

[0006] The passenger and vehicle safety analysis method based on multi-dimensional collision simulation comprises:

[0007] Collecting vehicle structure parameters, passenger physiological data and collision environment parameters;

[0008] Based on the passenger physiological data, a biomechanical feature vector is generated, and a basic coupled model containing mechanical and biological dimensions is constructed by combining the vehicle structural parameters.

[0009] The initial collision velocity is obtained based on the collision environment parameters, and the collision stages are divided using a stage decision tree to obtain the collision stage division results.

[0010] Based on the collision phase division results, a phased simulation is performed on the basic coupling model using a dynamic solver to output a multi-dimensional dynamic dataset.

[0011] A three-dimensional safety margin distribution map is generated based on the aforementioned multidimensional dynamic dataset;

[0012] Based on the three-dimensional safety margin distribution map, the protection strategy decision logic is executed, and the constraint system control parameters are output.

[0013] By adopting the above technical solution, passenger physiological differences, pre-collision posture, and vehicle structural response are deeply integrated into a unified dynamic system. Collision stages are dynamically divided based on the dominant physical mechanisms, and multiple solvers, including explicit, implicit, and random sampling solvers, are accurately matched, significantly improving the spatiotemporal fidelity of simulations in large structural deformations, biological tissue responses, and secondary risk prediction. The generated three-dimensional safety margin distribution map innovatively correlates engineering indicators with human injury in the same assessment space, revealing not only the nonlinear relationship between structural integrity and occupant safety but also supporting quantitative margin analysis of the design state, thereby driving the automatic output of differentiated protection strategies. The technical solution of this application can significantly improve the accuracy of occupant injury prediction, especially enhancing the protection capabilities for sensitive groups such as the elderly and children. Simultaneously, it directly transforms simulation results into parameter commands that can be embedded in the vehicle control system, promoting the development of passive safety systems from experience-based design to data-driven, scenario-adaptive intelligent systems.

[0014] Secondly, this application provides a passenger and vehicle safety analysis system based on multi-dimensional collision simulation, employing the following technical solution:

[0015] A passenger and vehicle safety analysis system based on multi-dimensional collision simulation, the analysis system comprising:

[0016] The data acquisition module is used to collect vehicle structural parameters, passenger physiological data, and collision environment parameters.

[0017] The model building module is used to generate biomechanical feature vectors based on the passenger physiological data and to build a basic coupled model containing mechanical and biological dimensions by combining vehicle structural parameters.

[0018] The collision phase division module is used to obtain the initial collision velocity based on the collision environment parameters, and divide the collision phases through a phase decision tree to obtain the collision phase division result.

[0019] The phased simulation module is used to perform phased simulation on the basic coupling model based on the collision phase division results, and output a multi-dimensional dynamic dataset.

[0020] The distribution map generation module is used to generate a three-dimensional safety margin distribution map based on the multidimensional dynamic dataset.

[0021] The protection strategy decision module is used to execute protection strategy decision logic based on the three-dimensional safety margin distribution map and output constraint system control parameters.

[0022] Thirdly, this application provides a computer device, which adopts the following technical solution:

[0023] A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0024] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0025] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect. Attached Figure Description

[0026] Figure 1 This is a first flowchart illustrating a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, according to one embodiment of this application.

[0027] Figure 2 This is a second flowchart illustrating a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, which is one embodiment of this application.

[0028] Figure 3 This is a schematic diagram of the third process of a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, which is one embodiment of this application.

[0029] Figure 4 This is a schematic diagram of the fourth process of a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, according to one embodiment of this application.

[0030] Figure 5 This is a schematic diagram of the fifth step of a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, according to one embodiment of this application.

[0031] Figure 6 This is a schematic diagram of the sixth process of a passenger and vehicle safety analysis method based on multi-dimensional collision simulation, which is one embodiment of this application. Detailed Implementation

[0032] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-6 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0033] This application discloses a passenger and vehicle safety analysis method based on multi-dimensional collision simulation.

[0034] Reference Figure 1 A passenger and vehicle safety analysis method based on multi-dimensional collision simulation, the analysis method includes:

[0035] Step S101: Collect vehicle structural parameters, passenger physiological data, and collision environment parameters;

[0036] Among them, vehicle structural parameters include the properties of body materials (such as the yield strength and elastic modulus of high-strength steel and aluminum alloy), the geometric configuration of energy absorption zones (such as the cross-sectional dimensions of longitudinal beams and the arrangement of front anti-collision beams), and the configuration of restraint systems (seat belt pretension, airbag volume and inflation rate). These parameters directly determine the energy absorption capacity and structural deformation mode of the vehicle under impact load.

[0037] Passenger physiological data includes, but is not limited to, age, gender, height, weight, and pre-collision posture angles (such as seat tilt angle and head-to-headrest distance). This information not only affects the inertial response path of the human body in a collision, but more importantly, it determines the mechanical tolerance limits of bones, muscles, and internal organs. For example, due to decreased bone density and soft tissue degeneration, the cervical spine injury threshold of the elderly is significantly lower than that of young adults; and a forward-leaning seating posture shortens the buffer distance between the occupant and the steering wheel, increasing the risk of head impact in a frontal collision.

[0038] The collision environment parameters refer to external boundary conditions such as initial collision velocity, collision type (frontal, side, offset), collision angle, and road surface friction coefficient, which together constitute the initial excitation field of the simulation. This step constructs a "personalized + contextualized" input system, breaking through the limitations of using standard dummy models and fixed working conditions in traditional simulations, enabling subsequent analysis to have individual adaptability and the ability to reproduce realistic scenarios.

[0039] Step S102: Generate biomechanical feature vectors based on passenger physiological data, and construct a basic coupled model containing mechanical and biological dimensions by combining vehicle structural parameters;

[0040] The biomechanical feature vector is not simply a stack of physiological parameters, but rather a weighted mapping function calibrated through medical research and biomechanical experiments. This transforms static parameters such as age and body posture into dynamic response indicators that can participate in dynamic calculations. For example, the age correction factor can be quantified using empirical formulas for the brittleness index and ligament laxity, while posture angles are determined through inverse kinematics inversion to identify the initial torques and center of gravity distribution of each joint. This vector is then embedded into a finite element human body model (such as GHBMC or THUMS), endowing it with individualized material properties and geometry.

[0041] Simultaneously, vehicle structural parameters are imported in the form of CAD / CAE models, forming a multibody dynamics framework that combines rigid and flexible bodies. These two components are coupled at key interfaces such as the seat-hip area, steering wheel-chest area, and seatbelt-shoulder area using contact force algorithms (e.g., the penalty function method or the Lagrange multiplier method), thus creating a truly integrated human-vehicle simulation system. This "mechanical-biological dual-dimensional coupling" breaks the traditional sequential process of "vehicle first, then human" in safety assessments, enabling real-time interactive feedback between structural deformation and human response. For example, when the A-pillar intrusion causes the steering wheel to shift backward, the system can simultaneously calculate the impact of this displacement on the chest compression of occupants of different body types, thereby dynamically adjusting the damage prediction results.

[0042] Step S103: Obtain the initial collision velocity based on the collision environment parameters, and divide the collision into stages using a stage decision tree to obtain the collision stage division result;

[0043] The initial collision velocity is not always given directly as an input parameter, but often needs to be derived from a deeper level of scenario description. For example, in offset or oblique collisions, the vehicle may strike an obstacle at a certain angle. In this case, the effective collision velocity should be the velocity component along the main impact direction, i.e., v. eff =v0·cosθ, where v0 is the vehicle speed and θ is the collision angle. Similarly, in rear-end collisions or multi-vehicle pile-up scenarios, the initial velocity must also consider the motion state of the preceding vehicle and the relative velocity vector. Furthermore, the road friction coefficient also affects the vehicle's braking behavior before the collision, thus altering the actual velocity value at the moment of impact. Therefore, the system needs to integrate all collision environment parameters and, through kinematic inversion or pre-simulation estimation modules, accurately reconstruct the kinetic energy level at the instant of the actual impact.

[0044] Subsequently, the stage decision tree is a logical judgment structure based on rules and thresholds. In one embodiment of this application, a vehicle collision can be divided into three stages with different dominant mechanisms: the initial impact stage (0–20ms), the intermediate interaction stage (20–100ms), and the late evolution stage (>100ms). It should be noted that these time intervals are not absolutely fixed, but are dynamically adjusted due to various factors such as initial velocity, collision type, and structural stiffness. For example, in a high-speed frontal collision, the structural crushing process is faster, and the initial stage may only last 15ms; while in a low-speed offset collision, due to the energy transfer delay caused by local contact, this stage may be extended to 25ms. Therefore, a simple hard division cannot be used, but an intelligent judgment mechanism must be introduced. The stage decision tree first uses the initial collision velocity as the primary criterion: when the velocity is higher than a certain threshold (e.g., 50km / h), it is judged as a high-energy impact, and a compact time division strategy is initiated; when the velocity is lower than the threshold, the time window is widened to avoid misjudgment. Secondly, based on the collision type, it is further divided into two categories: frontal collisions focus on the longitudinal beam compression rate, and when it exceeds the set acceleration gradient, it is considered the end of the initial stage; side collisions focus on the abrupt change in the B-pillar intrusion speed as a stage transition marker.

[0045] In addition, the system can also use the strain rate change rate (i.e. strain acceleration) of key measurement points on the vehicle body as an auxiliary criterion to identify the critical moment when the structure transitions from elastic response to plastic yielding, thereby accurately locating the end point of the first stage. Once these conditions are met, the decision tree outputs a "end of initial collision stage" signal, triggering the solver switch for the next stage.

[0046] Step S104: Based on the collision stage division results, perform staged simulation on the basic coupling model through a dynamic solver and output a multi-dimensional dynamic dataset.

[0047] Specifically, an explicit dynamics algorithm is used to solve the vehicle structure response in the initial stage of the collision; an implicit integral algorithm is used to solve the passenger biomechanical response in the middle stage of the collision; and a random sampling algorithm is used to predict the risk of secondary collisions in the later stage of the collision.

[0048] Specifically, the collision process is essentially a highly nonlinear, multi-timescale transient dynamics problem, with significantly different dominant physical mechanisms at different stages. Using a single numerical algorithm cannot balance efficiency and accuracy. Therefore, this application proposes a three-stage partitioning mechanism based on time thresholds: 0–20ms is the initial collision stage, dominated by high-speed impact, where the vehicle body structure undergoes severe plastic deformation, and energy is mainly dissipated through longitudinal beam crushing and firewall intrusion. At this stage, the system stiffness changes drastically, and inertial effects dominate; therefore, an explicit dynamics algorithm (such as the central difference method) is used for solving the problem. This algorithm does not require a global stiffness matrix; each step only depends on the state of the previous time step, making it suitable for handling highly nonlinear contact and large deformation problems. Although the time step is small, its parallel efficiency is high, and it can accurately capture instantaneous events such as structural fracture and weld failure.

[0049] Subsequently, during the mid-stage of the collision (20–100 ms), the macroscopic deformation of the vehicle body stabilizes, while the occupants begin to interact complexly with the restraint system—seatbelt locking, airbag deployment, and the occupants' forward thrust and rebound occur successively. This stage involves slow but highly precise biomechanical responses such as soft tissue creep, joint flexion, and relative organ displacement, implying a large number of internal degrees of freedom and local nonlinearities, making implicit integration algorithms (such as the Newmark-β or Hilber-Hughes-Taylor methods) suitable. These algorithms are unconditionally stable, allow for large time steps, and effectively suppress numerical oscillations, making them particularly suitable for solving the vibrations of flexible bodies and the responses of viscoelastic materials.

[0050] Finally, in the late-stage collision phase exceeding 100 ms, the primary collision ends, but there remains a risk of secondary collisions due to occupant swaying, release of residual structural stress, or vehicle rollover, such as head impact with side windows, neck sprains, or fuel leaks causing fires. These events are highly uncertain and have low probability characteristics, making them difficult to predict directly using deterministic equations. Therefore, a random sampling algorithm (such as Monte Carlo or Latin hypercube sampling) is introduced. Based on the known final state of the primary collision, statistical distribution perturbations are applied to factors such as residual kinetic energy, occupant attitude disturbances, and environmental interference, generating thousands of possible evolution paths. The probability density function of high-risk scenarios is then extracted from these paths.

[0051] The resulting "multidimensional dynamic dataset" not only includes traditional acceleration and displacement curves, but also integrates cross-scale information such as structural strain cloud maps, organ stress time histories, and joint angular velocity spectra, providing a comprehensive basis for subsequent safety assessments.

[0052] Step S105: Generate a three-dimensional safety margin distribution map based on the multidimensional dynamic dataset;

[0053] Specifically, the three-dimensional safety margin distribution map is not a simple chart, but a new assessment space that integrates engineering safety theory and human factors principles. The three axes respectively map the vehicle body deformation rate (the degree of geometric distortion of key areas (such as the passenger compartment) per unit time, reflecting structural integrity), biological injury indicators (normalized values ​​of indicators such as the HIC head injury criterion, Nij neck injury index, and rib fracture probability, characterizing the level of human injury), and survival space compression rate (the reduction ratio of the minimum clearance distance between the occupant's head and chest and interior components after a collision relative to the initial value, reflecting the available buffer zone). By projecting the simulation results onto this three-dimensional space, the danger quadrant of "high deformation-high injury-low space" can be intuitively identified, and some seemingly structurally intact but internally damaged "hidden high-risk" conditions can also be discovered due to the special posture of the occupants.

[0054] More importantly, this distribution map supports isosurface plotting and gradient analysis, enabling the identification of a safety envelope that meets survival thresholds, thereby quantifying the "margin" distance between the current design state and the safety boundary. For example, although a certain vehicle model did not exceed the structural failure threshold in an offset collision, its trajectory points were close to the high-damage zone, suggesting the need to optimize airbag deployment timing to reduce the risk of concussion. This three-dimensional mapping mechanism upgrades vehicle passive safety from a single-index evaluation to a system-level situational awareness tool for the first time, greatly enhancing the directionality of design iteration.

[0055] Step S106: Execute the protection strategy decision logic based on the three-dimensional safety margin distribution map and output the constraint system control parameters.

[0056] This decision-making logic is not a simple threshold trigger, but rather a dynamic parameter adjustment based on spatial location and safety margin. For example, when the system detects that the amount of biological damage exceeds a preset first threshold (e.g., 0.29, corresponding to a 50% probability of AIS2 level injury), it indicates that the occupant faces a high risk of soft tissue or traumatic brain injury. At this time, a "flexible protection" strategy is adopted: the airbag inflation pressure is reduced to avoid excessive compression of the chest and abdomen by high-pressure gas in close-range collisions, which is especially suitable for children or elderly passengers; at the same time, the pretensioner is triggered earlier so that the seat belt is tensioned at the beginning of the collision, reducing the occupant's forward momentum and reducing the relative speed with the airbag.

[0057] On the other hand, if the survival space compression rate exceeds the second threshold (e.g., 40%), it indicates severe intrusion into the occupant compartment, posing a threat of rigid impact to the head or neck. In this case, a "space reconstruction" strategy is activated: the headrest active displacement mechanism is activated, using a built-in motor or gas generator to push the headrest forward and upward, filling the gap behind the neck and preventing whiplash injuries. This dual-threshold differentiated response mechanism embodies the intelligent protection concept of "tailoring measures to injuries," avoiding the over-constraint or insufficient protection problems caused by the "one-size-fits-all" response of traditional systems. The output control parameters can be directly connected to the vehicle domain controller to guide the real-time intervention of next-generation active safety systems.

[0058] In the above embodiments, passenger physiological differences, pre-collision posture, and vehicle structural response are deeply integrated into a unified dynamic system. Collision stages are dynamically divided based on the dominant physical mechanisms, and multiple solvers, including explicit, implicit, and random sampling solvers, are accurately matched, significantly improving the spatiotemporal fidelity of simulations in predicting large structural deformations, biological tissue responses, and secondary risks. The generated three-dimensional safety margin distribution map innovatively correlates engineering indicators with human injury in the same assessment space, revealing not only the nonlinear relationship between structural integrity and occupant safety but also supporting quantitative margin analysis of the design state, thereby driving the automatic output of differentiated protection strategies. The technical solution of this application can significantly improve the accuracy of occupant injury prediction, especially enhancing the protection capabilities for sensitive groups such as the elderly and children. Simultaneously, it directly transforms simulation results into parameter commands that can be embedded in the vehicle control system, promoting the development of passive safety systems from experience-based design to data-driven, scenario-adaptive intelligent systems.

[0059] Reference Figure 2 As one implementation of step S102, the steps of generating biomechanical feature vectors based on passenger physiological data and constructing a basic coupled model containing mechanical and biological dimensions by combining vehicle structural parameters include:

[0060] Step S201: Receive passenger physiological data, including age, body mass index, and pre-collision posture angle;

[0061] Traditional passive safety systems often use standardized dummy models (such as 50th percentile adult males) for simulation and testing, ignoring the wide variability of real drivers and passengers in terms of age distribution, body size, and sitting posture. This solution explicitly uses age, body mass index (BMI), and pre-collision posture angle as core input parameters, precisely to break this limitation.

[0062] Specifically, age not only reflects the maturity and degree of degeneration of the skeletal system, but is also closely related to key biomechanical properties such as bone density, ligament elasticity, and muscle reaction speed. For example, decreased bone density in the elderly leads to increased bone fragility, making them more prone to rib fractures or cervical spine injuries under the same impact load; children, due to their larger head proportions and weaker neck support, face a higher risk of whiplash injuries. Body mass index directly affects the distribution ratio of body fat and muscle tissue, thus influencing the force transmission path during seatbelt restraint and chest compression behavior. Pre-collision posture angles (including trunk tilt angle, shoulder height, and head-to-headrest distance) determine the initial position and center of gravity of the human body at the moment of impact, significantly affecting its trajectory under inertial forces. These data can be collected in real time by onboard sensors (such as pressure seats, infrared posture recognition cameras, and identity recognition systems), forming the basic input for subsequent personalized modeling, enabling safety analysis to move from "group average" to "individual adaptation."

[0063] Step S202: Perform feature extraction operations on passenger physiological data and generate standardized biomechanical feature vectors based on pre-configured dimension weight coefficients.

[0064] This extraction and calculation process is not a simple numerical normalization, but a feature mapping mechanism supported by medical research and biomechanical experiments. For example, the bone mineral density correction factor can be calculated using the formula for calculating the age-related bone mineral density correction factor. This calculation process is essentially an empirical model of the decline in human bone strength with age.

[0065] Specifically, the formula for calculating the bone mineral density correction factor is: The baseline bone mineral density constant k1 can be set to the average bone mineral density value of a healthy 25-year-old adult (approximately 1.25 g / cm²), while the decay factor c is derived from epidemiological statistics (e.g., a decrease of 0.8% per year). When 65-year-old passenger information is entered, the system automatically calculates its bone mineral density correction factor to be 0.73, meaning that its bone strength is only 73% of that of a young adult. This value will directly affect the material property settings of the skeletal units in the subsequent human body model.

[0066] Similarly, the body mass index is mapped to the thickness of abdominal soft tissue or the sensitivity to visceral displacement through the nonlinear function g(bmi), and the posture angle is transformed into the initial torque of the joint and the contact boundary conditions through inverse kinematics processing.

[0067] Ultimately, these weighted and normalized components are combined into a unified biomechanical feature vector under the adjustment of dimensional weight coefficients α, β, and γ, forming a high-dimensional state descriptor that can characterize an individual's susceptibility to injury, specifically represented as follows:

[0068] ;

[0069] This vector not only preserves the essential characteristics of the original physiological information, but also achieves cross-scale and cross-modal data fusion through mathematical transformation, providing a unified semantic space for subsequent coupling with the vehicle structure model. f, g, and h are the normalization functions of the corresponding parameters.

[0070] Step S203: Obtain vehicle structural parameters, including body material properties, constraint system configuration, and coordinates of key structural nodes;

[0071] The vehicle body material properties include the elastic modulus, Poisson's ratio, yield strength, and plastic hardening curve of high-strength steel, aluminum alloys, and composite materials. These parameters determine the deformation mode and energy absorption capacity of the structure under impact loads. The restraint system configuration includes the technical parameters of active intervention devices such as seatbelt pretensioner type, airbag deployment time and pressure curve, and steering wheel crumple zone mechanism. Together, they constitute the first line of defense for occupant protection. The coordinates of key structural nodes are used to define the geometric topology of the vehicle body finite element model, ensuring accurate reproduction of the spatial relative positions of each component in the simulation. These parameters are typically derived from the vehicle's CAE database or BOM list, ensuring high engineering traceability.

[0072] Step S204: Input the biomechanical feature vector and vehicle structural parameters into the dual-channel coupling architecture;

[0073] The first channel uses a finite element discretization algorithm to convert vehicle structural parameters into nodal stress-strain matrices and constructs a finite element mesh model of the vehicle body; the second channel uses a viscoelastic constitutive model to convert biomechanical feature vectors into tissue response tensors and constructs a human tissue dynamics network.

[0074] Specifically, the dual-channel coupled architecture is essentially a parallelized and specialized modeling paradigm, distinct from the traditional single-solution approach that unifies the modeling of humans and vehicles. The first channel focuses on high-fidelity modeling of the mechanical system: using the finite element discretization algorithm, the continuous vehicle body structure is divided into millions of tiny units (such as shell elements and solid elements). Each node is assigned corresponding material properties and boundary conditions, forming a large matrix M containing information such as node displacement, stress, and strain. vehicle The model can accurately capture key failure behaviors during a collision, such as longitudinal beam crushing, A-pillar intrusion, and floor heave. The second channel focuses on the dynamic response of internal human tissues: based on viscoelastic constitutive models (such as the generalized Maxwell model or Zener model), biomechanical feature vectors are mapped to the time-dependent mechanical behaviors of different tissues (skin, muscle, bone, internal organs), constructing a dynamic network T that reflects tissue creep, stress relaxation, and energy dissipation characteristics. tissueThis network can not only simulate bone fractures, but also predict deep injury mechanisms such as brain tissue shear strain and liver rupture.

[0075] Understandably, the two channels operate independently, avoiding numerical instability and error accumulation caused by scale differences in traditional serial modeling, while also supporting heterogeneous computing acceleration. For example, the mechanical channel can be solved in parallel on a GPU cluster, while the biological channel is deployed on a dedicated processor with high memory bandwidth, greatly improving the overall simulation efficiency.

[0076] Step S205: The vehicle body finite element mesh model and the human tissue dynamics network are fused through a cross-dimensional correlator to output a basic coupled model that includes mechanical and biological dimensions.

[0077] This fusion process is not a simple data concatenation, but rather a physically consistent mapping guided by a pre-trained weight matrix. The interaction equation C in the cross-dimensional correlation algorithm... couple =W1M vehicle +W2T tissue Essentially, it is a weighted coupling operator, where W1 and W2 are transformation weight matrices trained with a large amount of real vehicle crash test and dummy test data, used to coordinate the energy transfer and force balance between the response fields of mechanical and biological systems with different dimensions and time scales.

[0078] For example, in a scenario of chest impact with a steering wheel, W1 controls how the steering wheel reaction force is applied to the chest wall surface, while W2 determines how this external force is internally distributed among the ribs, lungs, and heart. This data-driven fusion mechanism retains the constraints of physical laws (such as Newton's third law) while introducing empirical corrections to compensate for the shortcomings of theoretical models, significantly improving the correlation between simulation results and real injury data. The final output "mechanical-biological coupling model" is a spatiotemporally consistent joint dynamic system capable of synchronously tracking the vehicle deformation process and the evolution of human tissue damage at millisecond-level time resolution, providing a highly reliable analytical basis for subsequent collision consequence assessment and protection strategy generation.

[0079] The above implementation achieves a fundamental shift in vehicle passive safety analysis from "static generalization" to "dynamic personalization." This technical solution improves simulation accuracy and establishes a new occupant-centered safety assessment paradigm. It fully considers the impact of individual physiological differences on injury risk and accurately reproduces the nonlinear response of the vehicle structure in a collision. The dual-channel parallel architecture effectively solves the computational bottleneck in multi-scale coupling, and the pre-trained weight matrix enables seamless integration of cross-disciplinary data. This application can not only be used for safety performance optimization during new vehicle development but also provide real-time risk warnings and personalized constraint control commands for intelligent cockpits and autonomous driving systems, demonstrating significant engineering application value and industrialization prospects.

[0080] As a further implementation of the passenger and vehicle safety analysis method for multi-dimensional collision simulation, the staged simulation of the basic coupled model using a dynamic solver includes:

[0081] When the collision phase is in the initial stage, the explicit dynamics solver is invoked to solve for the vehicle body structure response;

[0082] When the collision phase is in the middle stage, the implicit integral solver is invoked to solve for the passenger's biomechanical response.

[0083] When the collision phase is in the late stage, the random path predictor is invoked to predict the risk of secondary collisions.

[0084] Specifically, the collision process is essentially a highly nonlinear, multi-timescale transient dynamics problem, with significant differences in the dominant physical mechanisms at different stages. It is difficult to balance efficiency and accuracy using a single numerical algorithm.

[0085] In one embodiment of this application, during the initial collision phase of 0–20 ms, the system experiences rapid energy release, extremely high structural deformation rates, and numerous highly nonlinear events such as contact-separation and material failure. If an implicit algorithm is used at this time, iterative solutions to the inverse stiffness matrix are required, resulting in extremely high computational costs and a high risk of divergence due to sudden stiffness changes. Therefore, this solution utilizes an explicit dynamics solver, the core of which is the use of the central difference method to achieve time stepping. The calculation can proceed solely based on the state of the previous time step, without the need to form a global stiffness matrix, making it particularly suitable for handling high-speed transient problems. Although the algorithm has an extremely small time step (typically on the microsecond scale), it boasts high parallel efficiency and can stably capture key deformation modes such as longitudinal beam wrinkling and firewall intrusion.

[0086] Subsequently, during the mid-stage of the collision (20–100 ms), the macroscopic deformation of the vehicle body stabilizes, while the occupants begin to interact deeply with the restraint system: seatbelt locking, airbag inflation, chest compression, and neck flexion occur successively. These biomechanical responses involve slow but highly accurate processes such as soft tissue creep and joint flexion, making implicit integrators (such as the Newmark-β method) suitable. These algorithms are unconditionally stable, allow for large time steps, and can control numerical damping by adjusting integration parameters (e.g., γ=0.5, β=0.25), effectively suppressing the interference of high-frequency oscillations on the calculation of biological tissue stress, thus more accurately predicting key indicators such as the HIC head injury index and the Nij neck injury criterion.

[0087] Finally, after 100ms, the primary collision process ends, but the vehicle may still be at risk of rolling over, skidding, or secondary collisions with obstacles. Such events are highly uncertain and difficult to describe using deterministic equations. Therefore, the system switches to a stochastic path predictor, employing a Monte Carlo sampling method. Based on the known final state, random perturbations (such as slight attitude changes, road friction fluctuations, and occupant residual kinetic energy distribution) are introduced to generate thousands of possible evolution paths. The proportion of these paths leading to dangerous situations such as head impact with the side window or excessive cervical flexion is statistically analyzed, thus quantifying the probability of a secondary collision. This transition from determinism to probabilistics signifies a shift in simulation objectives from "reproducing the process" to "predicting risks."

[0088] It should be noted that by aligning the output datasets of the solvers at each stage in both time and space, a multidimensional dynamic dataset with a unified timestamp can be generated. Since explicit solvers use microsecond-level step sizes, while implicit solvers may advance at millisecond-level, and random sampling generates discrete results, direct concatenation would lead to time series misalignment and spatial node mismatch.

[0089] To address this, the system establishes a unified timeline and resamples the original data for each stage using cubic spline interpolation. This interpolation method not only maintains curve smoothness but also preserves key features such as peak acceleration, avoiding oscillation distortion caused by linear interpolation. The final output data structure is a three-dimensional tensor with dimensions of "time," "spatial nodes," and "physical quantities." For example, at a given moment, the equivalent stress of a vehicle body node, the shear strain of the corresponding thoracic vertebral segment, and the probability of head impact can be queried simultaneously. This spatiotemporal alignment mechanism ensures that subsequent analysis modules (such as safety margin assessment and control command generation) can make comprehensive judgments based on a consistent time benchmark, avoiding misjudgments caused by asynchronous data.

[0090] The above embodiments achieve refined, differentiated, and forward-looking analysis of the entire vehicle collision process. This application not only improves simulation accuracy but also establishes a unified analysis framework across time scales, physical fields, and determinism and stochasticity: in the initial stage, an efficient explicit algorithm captures the transient response of the structure; in the middle stage, a high-precision implicit method analyzes the human injury mechanism; in the later stage, a probabilistic model assesses residual risk; and a spatiotemporal alignment technique integrates heterogeneous outputs into a unified data stream. This method significantly outperforms traditional single-algorithm simulations and is particularly suitable for scenarios that are difficult to cover with empirical rules, such as complex offset collisions and multi-body chain accidents, providing a reliable data foundation for real-time decision-making in next-generation intelligent safety systems.

[0091] Reference Figure 3 As one implementation of step S105, the step of generating a three-dimensional safety margin distribution map based on a multidimensional dynamic dataset includes:

[0092] Step S301: Receive a multidimensional dynamic dataset, which includes time series data of vehicle body structural deformation, biological damage indicators, and living space compression rate.

[0093] This dataset is not a static parameter, but rather a high spatiotemporal resolution output from a preceding dynamic phased simulation, possessing complete temporal evolution characteristics.

[0094] Specifically, vehicle body structural deformation typically refers to the maximum intrusion displacement or cumulative plastic strain of key areas (such as A-pillars, sill beams, and firewalls) during a collision, reflecting the vehicle structure's ability to resist deformation. Biological injury indicators encompass medically relevant measures certified by international standards (such as FMVSS and EuroNCAP), including the Neck Injury Criterion (NIC), Head Injury Index (HIC), and Chest Compression Factor (CFC), used to assess the level of physiological injury occupants may suffer in a collision. Survival space compression rate specifically refers to the reduction in the minimum clearance distance between the occupant's head / chest and interior components (such as the steering wheel, dashboard, and side windows) during a collision relative to the initial state, directly related to whether a rigid impact occurs. These data are organized in time-series format, recording the dynamic changes throughout the entire collision process, forming the raw material for subsequent analysis. The key to this step is ensuring the integrity and consistency of the data source—that is, all physical quantities are generated based on the same simulation conditions, the same timeline, and the same coupled model, avoiding mapping distortion due to data fragmentation.

[0095] Step S302: Perform normalization processing on the multidimensional dynamic dataset to convert physical quantities of different dimensions into standardized parameters;

[0096] Since structural deformation is measured in millimeters, biological damage indicators are mostly dimensionless but vary greatly in magnitude (e.g., HIC can reach over 1000, while NIC is usually below 1.0), and survival space compression rate is a percentage. Directly using these for spatial mapping can lead to a single dimension dominating the overall distribution pattern, causing visual and computational discrepancies. Therefore, mathematical transformations are necessary to unify them to similar numerical ranges. This scheme uses a combination of range normalization and linear scaling to map each physical quantity to [0,1] or a reasonably extended range. For example, plastic strain rate can be linearly stretched using its minimum and maximum observed values ​​in typical collisions to ensure comparability between different vehicle models; biological damage indicators are corrected by introducing an age decay factor before normalization—as mentioned in claim 2, elderly people are more susceptible to injury under the same mechanical load due to tissue degeneration, so the basic damage value needs to be multiplied by a (1-λ·age) factor to amplify it and reflect individual differences. This physiological characteristic correction mechanism makes safety assessment no longer limited to "structural performance superiority or inferiority," but truly shifts to "protective effect on people," improving the human-centered adaptability of the evaluation system. The normalized parameters are no longer isolated engineering values, but become standardized state variables that can participate in spatial geometric calculations, laying a mathematical foundation for subsequent 3D modeling.

[0097] Step S303: Establish a three-dimensional mapping coordinate system; wherein the X-axis maps the vehicle body deformation rate, the Y-axis maps the biological damage amount corrected by physiological characteristics, and the Z-axis maps the survival space compression rate.

[0098] Specifically, the design of this coordinate system has profound engineering and medical implications: the X-axis represents the structural integrity of the vehicle, that is, the ability of the vehicle body to maintain the geometric stability of the passenger compartment under impact. The higher the plastic strain rate, the more likely the material has entered the irreversible yielding stage, and the higher the risk of structural failure. The Y-axis represents the degree of human injury. After being corrected for physiological parameters such as age and BMI, it can more accurately reflect the actual probability of injury of a specific passenger, rather than the average response of dummy tests. The Z-axis represents the usable survival space, that is, the buffer zone in which the occupant can move freely without hard contact during a collision. The larger its compression ratio, the more likely the restraint system has failed to effectively control the forward motion or the more serious the structural intrusion.

[0099] These three dimensions together constitute the three pillars for measuring vehicle safety performance: whether the vehicle can withstand the impact, whether occupants will be injured, and whether there is room to maneuver. By arranging these three key factors orthogonally, a three-dimensional safety space is formed, and any collision event can be abstracted as a trajectory point or evolution path in this space. For example, although a certain vehicle model may have relatively small structural deformation (low X-value) in an offset collision, its safety status is still in the high-risk zone due to excessive chest compression caused by premature airbag deployment (high Y-value). This three-dimensional mapping mechanism achieves a unified expression of mechanical engineering indicators and biomedical consequences for the first time, greatly enhancing the systematicness and explanatory power of safety assessment.

[0100] Step S304: Based on the three-dimensional mapped coordinate system, the standardized parameters are converted into three-dimensional spatial point cloud data according to the mapping relationship, and a continuous three-dimensional safety margin distribution surface is generated through spatial interpolation algorithm, and a three-dimensional safety margin distribution map is output.

[0101] In this model, the combination of three parameters at each time step (i.e., the vehicle deformation rate, the corrected biological damage amount, and the survival space compression rate at a certain moment) is mapped to a coordinate point (x, y, z) in three-dimensional space. As time progresses, this forms a continuous point cloud trajectory, depicting the dynamic evolution of the safety state during the collision. These point clouds not only reflect the instantaneous state but can also identify high-frequency risk areas through density distribution or distinguish pattern characteristics of different collision types through cluster analysis. For example, in a frontal collision, the point cloud may rise rapidly along the X-axis and then tend to plateau; while in a side collision, it may increase rapidly along the Z-axis, indicating that lateral intrusion leads to a sharp compression of the survival space. The generation of point cloud data makes the originally abstract numbers visible, measurable, and comparable, providing a discrete sampling basis for the subsequent construction of continuous surfaces.

[0102] Furthermore, since the simulation data is sampled finitely in both time and space, directly connecting the point cloud would create a jagged, discontinuous surface, making it difficult to use for accurate judgment and threshold comparison. Therefore, the system employs radial basis function interpolation (RBF), specifically using thin-plate splines as the basis function. This method exhibits excellent smoothness and shape preservation capabilities when handling irregularly distributed point clouds, effectively suppressing oscillations and maintaining local features. Its core idea is to construct a global smoothing function that precisely matches the original value at each known sampling point, while making reasonable inferences at other locations using a weighted distance function. The generated surface not only reflects the safety trend of the current simulation results but can also be used for interpolation to predict potential risks under unsimulated conditions, forming a "safety envelope" covering various collision conditions. The undulations of this surface intuitively reflect the changes in safety margin under different states: the lower the surface, the closer the system is to the danger boundary; locally concave areas indicate the existence of design weaknesses under specific conditions.

[0103] The aforementioned implementation plan transforms vehicle collision safety assessment from fragmented indicators to a systematic spatial model. By establishing a multi-dimensional safety evaluation system that integrates structural mechanics, biomedicine, and spatial geometry, it considers both the vehicle's inherent crashworthiness and the impact of individual occupant differences; it reflects both instantaneous states and captures the evolutionary trend throughout the entire process; and through the combination of continuous surfaces and safety thresholds, it ultimately achieves quantitative risk identification and boundary warning.

[0104] Reference Figure 4As one implementation of step S106, the step of executing the protection strategy decision logic based on the three-dimensional safety margin distribution map and outputting the constraint system control parameters includes:

[0105] Step S401: Receive a three-dimensional safety margin distribution map, which includes a three-dimensional mesh model that maps vehicle body deformation rate, biological damage amount and survival space compression rate.

[0106] In this model, the X-axis represents the plastic strain rate of the vehicle body structure under impact, reflecting the degree of deformation of the vehicle's energy-absorbing structure and its ability to maintain the integrity of the passenger compartment. The Y-axis represents the biological damage amount corrected for physiological characteristics, taking into account the influence of factors such as age and body shape on individual tolerance limits, and truly reflecting the injury risk of a specific passenger. The Z-axis represents the survival space compression rate, reflecting the buffer zone in which occupants can move freely without rigid impact during a collision. These three dimensions together constitute an orthogonal safety state space, and any collision scenario can be mapped to a point or trajectory path in this space. This three-dimensional mesh model stores the safety margin values ​​at each location in the form of discrete nodes, and continuous estimates for any location can be obtained through interpolation, possessing a high degree of spatial resolution.

[0107] Step S402: Divide the mesh model into safe domain, critical domain, and danger domain subspaces according to the preset safety threshold;

[0108] This division is not a simple threshold truncation, but a geometric judgment based on a spherical or ellipsoidal safety envelope, reflecting a holistic consideration of the coupling effect of multiple physical quantities.

[0109] In this embodiment of the application, the Euclidean norm can be defined. As a comprehensive risk index, and set As a dividing radius, the system can divide the entire three-dimensional space into three levels: when the comprehensive risk index is less than or equal to 0.7, it is determined to be the "safe domain", indicating that the occupants are adequately protected under the current working conditions and no additional intervention is required; between 0.7 and 1.0 is the "critical domain", indicating that there is a potential risk and an early warning or slight adjustment needs to be initiated; exceeding 1.0 enters the "danger domain", indicating that structural failure, biological damage or space compression has reached a critical level and strong intervention measures must be taken.

[0110] Understandably, this vector-based partitioning method avoids the one-sidedness of traditional single-indicator over-limit judgments. For example, even if structural deformation is small (low x value), if biological damage is high (high y value) and space is severely compressed (high z value), the overall risk may still exceed the safety boundary. This partitioning mechanism enables the system to measure complex coupled risks with a unified standard, providing clear triggering conditions for subsequent differentiated response strategies.

[0111] Step S403: Identify key risk feature points in the critical domain and danger domain subspaces, and extract the corresponding physical quantity over-limit identifiers;

[0112] Not all points located in high-risk areas are equally important. The system needs to further screen out the key feature points that have the greatest impact on occupant safety and are most representative, such as the peak point of biological damage, the point of most severe structural intrusion, or the point of minimum survival space. By analyzing the coordinates of these points, the system can identify which physical quantity(s) are driving the increase in risk and generate corresponding "over-limit identifiers".

[0113] For example, if the Y-axis value (biological damage) at a certain point is significantly higher than the threshold y max If other dimensions are still within a controllable range, a "y_over" flag is generated; if the Z-axis (living space compression rate) exceeds the limit, a "z_over" flag is generated. These flags are not only Boolean switching signals, but can also carry over-limit amplitude information for subsequent quantitative adjustment of control intensity.

[0114] Step S404: Determine the collision type classification result based on the collision environment parameters, and select the multi-objective optimization weight set;

[0115] Among these, the importance weights of each safety dimension differ significantly under different collision types. For example, in a frontal collision, the front structure of the vehicle undertakes the main energy absorption task, therefore the mechanical dimension (structural integrity) has the highest weight (w). m =0.6), while biological damage and survival space are relatively less important; in side collisions, the B-pillar strength is limited, and lateral intrusion directly threatens the occupant's torso and head, increasing the survival space dimension weight to 0.4, making it a key control target; while in rear-end collisions, the main risk comes from whiplash injuries to the neck, increasing the biological dimension weight to 0.5, making it the primary protection target. The system determines the current operating condition through a front-end collision type identification module (such as radar and inertial navigation fusion judgment) and automatically loads the corresponding weight set. This dynamic weighting mechanism ensures that the optimization target always matches the actual threat, avoiding the failure of "one-size-fits-all" control strategies in complex scenarios, and is a key link in achieving intelligent protection.

[0116] Step S405: Perform protection strategy decision tree operation based on the physical quantity over-limit identifier and the multi-objective optimization weight set to generate constraint system control parameters.

[0117] This decision tree is not a simple if-else logic, but a multi-objective optimization engine that integrates physical priority, control coupling, and engineering feasibility. For example, when the biological damage exceeds the limit and the biological dimension has a high weight, the system will activate a "flexible protection" strategy: on the one hand, it will reduce the airbag inflation pressure. This prevents excessive pressure on the chest and abdomen from high-pressure gas during close-range collisions, making it especially suitable for elderly or child passengers; on the other hand, it allows for earlier activation of the seatbelt pretensioners. This effectively restrains the occupants in the early stages of a collision, reducing forward momentum.

[0118] If the survival space compression rate exceeds the limit and the spatial dimension weight exceeds the threshold (e.g., 0.3), the "spatial reconstruction" strategy is activated: an active headrest displacement command is issued. This pushes the headrest forward and upward, filling the gap at the back of the neck and preventing cervical hyperextension injury. The generation of these control parameters is not an isolated operation, but a coordinated adjustment based on weighted coordination to ensure that various intervention measures are coordinated in terms of timing and intensity, avoiding conflicts or waste of resources.

[0119] In the above implementation, the risk level is determined based on the overall situation of the three-dimensional safety space, the target is dynamically adjusted and optimized in combination with the collision type, and finally, executable multi-system collaborative instructions are generated. This technical solution significantly improves the adaptability and protection effectiveness of the restraint system, and is particularly suitable for active safety integration in smart cockpits, advanced driver assistance systems (ADAS), and autonomous driving platforms, providing a scalable and deployable engineering solution for the future development of automotive safety technology.

[0120] Reference Figure 5 As a further implementation of the passenger and vehicle safety analysis method, after the step of outputting the constraint system control parameters, the method further includes:

[0121] Step S501: Monitor the relative speed and distance between the vehicle and the obstacle in real time, and calculate the predicted collision time.

[0122] This step relies on deep fusion of multi-source sensors to ensure high-precision, low-latency collision situational awareness even in complex traffic environments. Millimeter-wave radar, with its strong penetration and high speed measurement accuracy, continuously acquires the relative velocity vector v between the vehicle and obstacles ahead. rel Even in severe weather conditions such as rain, snow, fog, and haze, it maintains stable output; the stereo vision camera accurately measures the distance d between the two objects using the principle of binocular parallax, possessing excellent spatial resolution and capable of recognizing obstacle outlines and lateral positions; the inertial measurement unit (IMU) records the vehicle's own longitudinal acceleration a in real time. ego This is used to correct dynamic changes during vehicle braking or acceleration. These heterogeneous data are not simply superimposed, but dynamically fused and state estimated using the Extended Kalman Filter (EKF) algorithm.

[0123] Specifically, EKF, as a nonlinear recursive filtering method, can optimally estimate the system's state variables (such as relative position, velocity, and acceleration) and continuously update the collision time (TTC) in the presence of noise and uncertainty. The specific calculation formula is as follows:

[0124] ;

[0125] In the above formula, v rel Here, d represents the relative velocity vector obtained by the millimeter-wave radar, d represents the distance measured by the stereo vision camera, and a represents the distance. ego The vehicle acceleration is monitored by the inertial measurement unit (IMU). The resulting time series {TTCt} has a time resolution of 5 ms, forming a continuously evolving collision countdown curve, providing a precise time reference for the graded response strategy.

[0126] Step S502: When the predicted collision time is less than the preset time threshold, a preloading command is generated based on the constraint system control parameters.

[0127] This process is not a simple threshold triggering, but a multi-level response logic based on a finite state machine (FSM) to ensure that control actions match the risk level.

[0128] For example, the system defines three key states: when the TTC is greater than 500ms, it is in a "sleep state" and all actuators remain in low-power standby mode; when the TTC enters the range of 150ms to 500ms, it switches to a "warning state" and starts the parameter preload mechanism, sending initialization commands to the airbag control module and seat belt pretensioner in advance to activate their internal circuits, charge energy storage elements, and pre-open valves, eliminating cold start delay; when the TTC is further shortened to less than 150ms, it enters an "emergency state" and the system determines that a collision is unavoidable, immediately performs full parameter loading, and completes the final control configuration.

[0129] During this process, the preload command is not generated as a fixed value output, but is dynamically modulated according to real-time operating conditions: for example, the airbag inflation pressure P inflate The seatbelt pretensioner's travel S is increased proportionally based on relative speed, reflecting the adaptability principle of "high speed, high energy; low speed, gentle protection"; pretension An exponential decay function is used, which gradually increases as the distance decreases, to ensure effective constraint at different collision distances.

[0130] More importantly, the system also integrates passenger posture detection data (such as skeletal key points captured by a TOF camera). If it detects that an occupant is leaning forward, letting go of their hands, or in a non-standard sitting posture, it will automatically reduce the airbag inflation pressure by about 15% to prevent secondary injuries to the face or chest caused by the impact of high-pressure gas during close deployment. This context-aware dynamic optimization mechanism makes the pre-loading strategy both safe and humane, avoiding new risks brought about by "overprotection".

[0131] Step S503: Send the preload command to the restraint system actuator via the vehicle bus;

[0132] Restraint systems (such as airbag controllers, pretensioner modules, and active headrest motors) are typically distributed across different ECUs, each using different communication protocols. This solution employs Time-Triggered Ethernet (TTEthernet) as the underlying communication protocol, combined with Time Division Multiple Access (TDMA) to achieve deterministic transmission. Each 1ms communication cycle is divided into multiple fixed time slots: the first 200μs are dedicated to airbag control command transmission, 200–400μs are allocated to the seatbelt pretensioner, and the remaining time slots are used by other non-critical ECUs. The data frame structure is strictly designed according to the AUTOSAR standard, including a 12-bit frame ID (e.g., 0xD2 identifies airbag commands), a 32-bit timestamp (accurate to 0.1ms), a 16-bit normalized parameter value, and an 8-bit CRC checksum, ensuring the uniqueness, timeliness, and integrity of the commands. Because TTEthernet supports nanosecond-level clock synchronization, all nodes can perform operations under a unified time base, avoiding misalignment caused by clock drift.

[0133] In step S504, the restraint system actuator adjusts the pressure level of the airbag inflator and the travel of the seat belt pretensioner according to the preload command.

[0134] The airbag inflation system employs a three-stage pressure control mechanism: when TTC > 100ms, the primary solid gas generator ignites, inflating the airbag to a standby pressure of 20kPa, avoiding response lag in a fully vented state and preventing accidental deployment; when TTC drops to the 50–100ms range, the secondary pressurizing agent is ready, increasing the pressure to 80kPa, shortening the time required for full deployment; when TTC ≤ 50ms, the system triggers the final detonation command, achieving instantaneous inflation to 300kPa. This process relies on a two-stage solid gas generator (SSG) technology, with a primary agent response time of only 8ms and a secondary delay of 2ms. Combined with the precise ignition timing of the electronic control unit, pressure curves tailored to different collision intensities can be generated.

[0135] For the seatbelt pretensioner, a stroke control algorithm based on PID feedback is used. The controller determines the target stroke S according to the command. cmd The actual stroke S fed back by the motor encoderactual The deviation integral is used to dynamically adjust the motor rotation angle θ. motor This ensures a smooth and precise tightening process. Mechanically, a planetary gear reduction mechanism (18:1 ratio) converts the high-speed motor's rotation into linear pull of the rack and pinion, completing a 150mm stroke in 0.3 seconds and generating a peak tensile force of up to 4500N, effectively restraining the occupant's forward momentum. In the preloaded state, the pretensioner has already partially tightened the seatbelt, eliminating webbing slack and allowing it to function immediately in the event of a collision, significantly improving restraint efficiency.

[0136] In the above embodiments, the graded pressure control and dynamic stroke adjustment mechanism enable the restraint system to adaptively adjust according to the collision intensity and occupant status, which not only enhances the protection capability, but also reduces the risk of injury caused by over-deployment in small collisions, improves the timeliness and adaptability of protective actions, and realizes the forward-moving, intelligent and precise upgrade of the vehicle's passive safety system.

[0137] Reference Figure 6 As a further implementation of the vehicle safety analysis method, before constructing the basic coupled model that includes mechanical and biological dimensions, the following steps are also included:

[0138] Step S601: Obtain the vehicle deformation field distribution map and passenger injury medical report from the historical crash test.

[0139] Among them, the vehicle deformation field distribution map is usually generated by high-speed digital image correlation technology (DIC) or high-fidelity finite element simulation. It presents the dynamic response of the vehicle body every millisecond during the collision in a gridded or pixelated form. Each spatial node not only contains kinematic parameters such as displacement and velocity, but also covers material nonlinear behavior indicators such as plastic strain rate, energy absorption density, and stress tensor, forming a four-dimensional spatiotemporal data cube (time × space × direction × physical quantity).

[0140] Meanwhile, the passenger injury medical report integrates multi-source information: on the one hand, it comes from the high-precision sensors built into the collision dummy, which record engineered injury indicators (such as HIC, Nij, CFC) such as head acceleration, neck torque, and chest compression; on the other hand, it combines clinical medical images (such as CT and MRI) to classify organ damage in real injury cases or biomechanical tests, and performs structured coding according to the simplified injury scoring standard (AIS) to form a mapping chain from "mechanical input" to "physiological output".

[0141] Step S602: Enhance the dataset by synthesizing the association between passenger physiological data and vehicle structural parameters using a generative adversarial network;

[0142] This step employs a dual-channel conditional generative adversarial network (DC-GAN), whose generator receives two input branches: one is the vehicle body structural parameters, represented as a three-dimensional deformation gradient tensor, capturing strain distribution characteristics in the X / Y / Z directions; the other is a physiological data vector, including key biological indicators such as the neck injury criterion (NIC) and rib fracture probability. The two branches interact and merge in a cross-domain attention module to learn the implicit nonlinear mapping relationship between mechanical load and human response.

[0143] For example, when a car door anti-collision beam yields at a specific location, will the shear stress on the spleen due to inertial forward thrust exceed the tissue rupture threshold? The discriminator not only judges whether the generated sample is "realistic," but also introduces a biomechanical constraint layer to verify whether the generated human injury conforms to known tissue mechanics criteria (such as the Yamada model's limitation on the maximum tensile rate of muscles and the empirical range of brain tissue shear modulus).

[0144] Crucially, this application also designed a damage correlation loss function, which mandates that the statistical correlation between plastic strain rate and AIS damage level in the generated data must be highly consistent with the real data, with the error controlled within 5%, thereby avoiding the generation of physically implausible false samples such as "severe structural deformation but no damage" or "minor collision but fatal injury".

[0145] Step S603: Inject the association augmentation dataset into the training process of the basic coupling model;

[0146] Specifically, the training process employs a phased knowledge distillation framework: the first phase pre-trains the basic coupled model (i.e., the student model) using real collision data to establish preliminary mechatronic response capabilities; the second phase introduces a teacher-student architecture, inputting the generated augmented data into a fully trained teacher model, which outputs biological damage predictions and their confidence scores. The student model learns the soft-label distribution of the teacher model through the KL divergence loss function, rather than directly fitting hard labels. This approach conveys more probabilistic information and improves generalization ability.

[0147] Specifically, the system introduces a confidence filtering mechanism, only including a sample in the distillation process if the teacher model's prediction for that sample has high confidence, thus preventing low-quality or marginalized synthetic data from contaminating the student model. Furthermore, to prevent augmented data from dominating the training process due to its large quantity, the system applies a dynamic weighting coefficient of 0.3 to 0.5 to this data, achieving a balanced use of real and synthetic data.

[0148] Step S604: Update the weight matrices of the mechanical and biological dimensions in the basic coupled model.

[0149] The weight matrix is ​​not a single set of parameters, but rather a cross-domain interactive core embedded in a dual-path neural network architecture: one path is a 3D convolutional network that processes the spatiotemporal evolution characteristics of the vehicle body deformation field; the other path is a graph neural network (GNN) that models the topological connections and force transmission paths between human bones, organs, and soft tissues. Weight updates are mainly concentrated in the interaction layer between the two, identifying the most influential coupling factors for damage prediction by calculating the Shapley value of each feature in the augmented dataset relative to the overall loss function.

[0150] For example, if acceleration signals in the pelvic region are identified as a key precursor to lower limb fractures, their corresponding connectivity weights will receive a significant boost; while roof deformation, although potentially affecting overall stiffness, has a weaker correlation with concussion, so the weights of related pathways will be moderately attenuated. Simultaneously, the system implements bio-specific parameter tuning: for enhanced samples from the elderly population, the weight coefficient of osteoporosis factors is automatically increased (+Δw_age) to enhance the model's sensitivity to fragility fractures; for side-impact scenarios, the parameter learning of the thoracic viscoelastic constitutive model is strengthened to more accurately simulate the dynamic responses of the ribs and lungs.

[0151] The above implementation achieves high-performance evolution of the mechano-biological coupling model under data-scarce conditions. It effectively expands limited real experimental data into thousands of high-fidelity enhanced samples. Through physical constraints and selective learning mechanisms, it ensures that the model significantly improves its predictive capabilities for special populations and extreme working conditions without sacrificing adherence to fundamental mechanical laws. This technical solution improves the prediction accuracy of key indicators such as neck injury and visceral rupture, providing a highly reliable simulation foundation for intelligent safety systems, personalized protection strategies, and virtual verification platforms.

[0152] This application also discloses a passenger and vehicle safety analysis system based on multi-dimensional collision simulation.

[0153] A passenger and vehicle safety analysis system based on multi-dimensional collision simulation. The analysis system includes:

[0154] The data acquisition module is used to collect vehicle structural parameters, passenger physiological data, and collision environment parameters.

[0155] The model building module is used to generate biomechanical feature vectors based on passenger physiological data and combine them with vehicle structural parameters to build a basic coupled model that includes mechanical and biological dimensions.

[0156] The collision phase division module is used to obtain the initial collision velocity based on the collision environment parameters, and divide the collision phases through a phase decision tree to obtain the collision phase division results.

[0157] The phased simulation module is used to perform phased simulations on the basic coupled model based on the collision phase division results, and output a multi-dimensional dynamic dataset.

[0158] The distribution map generation module is used to generate a three-dimensional safety margin distribution map based on a multi-dimensional dynamic dataset.

[0159] The protection strategy decision module is used to execute protection strategy decision logic based on the three-dimensional safety margin distribution map and output constraint system control parameters.

[0160] The passenger and vehicle safety analysis system based on multi-dimensional collision simulation in this application embodiment can implement any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiments.

[0161] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0162] This application also discloses a computer device.

[0163] A computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a passenger and vehicle safety analysis method based on multi-dimensional collision simulation as described above.

[0164] This application also discloses a computer-readable storage medium.

[0165] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the passenger and vehicle safety analysis methods based on multi-dimensional collision simulation.

[0166] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0167] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0168] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce a good effect.

[0169] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A passenger and vehicle safety analysis method based on multi-dimensional collision simulation, characterized in that, The analytical method includes: Collect vehicle structural parameters, passenger physiological data, and collision environment parameters; Based on the passenger physiological data, a biomechanical feature vector is generated, and a basic coupled model containing mechanical and biological dimensions is constructed by combining the vehicle structural parameters. The initial collision velocity is obtained based on the collision environment parameters, and the collision stages are divided using a stage decision tree to obtain the collision stage division results. Based on the collision phase division results, a phased simulation is performed on the basic coupling model using a dynamic solver to output a multi-dimensional dynamic dataset. A three-dimensional safety margin distribution map is generated based on the aforementioned multidimensional dynamic dataset; Based on the three-dimensional safety margin distribution map, the protection strategy decision logic is executed, and the constraint system control parameters are output.

2. The passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to claim 1, characterized in that, The steps of generating biomechanical feature vectors based on the passenger physiological data and constructing a basic coupled model containing mechanical and biological dimensions by combining vehicle structural parameters include: Receive passenger physiological data, including age, body mass index, and pre-collision posture angle; Feature extraction operations are performed on the passenger physiological data, and a standardized biomechanical feature vector is generated based on pre-configured dimension weight coefficients. Obtain vehicle structural parameters, including body material properties, constraint system configuration, and coordinates of key structural nodes; The biomechanical feature vectors and vehicle structural parameters are input into a dual-channel coupled architecture. The first channel converts the vehicle structural parameters into nodal stress-strain matrices using a finite element discretization algorithm to construct a finite element mesh model of the vehicle body. The second channel converts the biomechanical feature vectors into tissue response tensors using a viscoelastic constitutive model to construct a human tissue dynamics network. By fusing the vehicle body finite element mesh model with the human tissue dynamics network through a cross-dimensional correlator, a basic coupled model containing mechanical and biological dimensions is output.

3. The passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to claim 2, characterized in that, Performing a phased simulation of the basic coupled model using a dynamic solver includes: When the collision phase is in the initial stage, the explicit dynamics solver is invoked to solve for the vehicle body structure response; When the collision phase is in the middle stage, the implicit integral solver is invoked to solve for the passenger's biomechanical response. When the collision phase is in the late stage, the random path predictor is invoked to predict the risk of secondary collisions.

4. The passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to claim 1, characterized in that, The steps for generating a three-dimensional safety margin distribution map based on the multidimensional dynamic dataset include: Receive the multidimensional dynamic dataset, which includes time series data of vehicle body structural deformation, biological damage indicators, and living space compression rate; Normalization is performed on the multidimensional dynamic dataset to convert physical quantities of different dimensions into standardized parameters; A three-dimensional mapping coordinate system is established; where the X-axis maps the vehicle body deformation rate, the Y-axis maps the biological damage amount corrected by physiological characteristics, and the Z-axis maps the survival space compression rate. Based on the three-dimensional mapped coordinate system, the standardized parameters are converted into three-dimensional spatial point cloud data according to the mapping relationship, and a continuous three-dimensional safety margin distribution surface is generated by spatial interpolation algorithm, outputting a three-dimensional safety margin distribution map.

5. The passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to claim 4, characterized in that, The steps for executing the protection strategy decision logic based on the three-dimensional safety margin distribution map and outputting the constraint system control parameters include: Receive a three-dimensional safety margin distribution map, which includes a three-dimensional mesh model that maps vehicle body deformation rate, biological damage amount and survival space compression rate; The mesh model is divided into safe domain, critical domain, and danger domain subspaces according to a preset safety threshold. Key risk feature points are identified within the critical domain and danger domain subspaces, and corresponding physical quantity over-limit identifiers are extracted. The collision type classification result is determined based on the collision environment parameters, and a multi-objective optimization weight set is selected. Based on the physical quantity over-limit identifier and the multi-objective optimization weight set, a protection strategy decision tree operation is performed to generate the constraint system control parameters.

6. The passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to claim 5, characterized in that, Following the step of outputting the control parameters of the constraint system, the following is also included: Real-time monitoring of the relative speed and distance between vehicles and obstacles, and calculation of collision time prediction; When the predicted collision time is less than a preset time threshold, a preloading instruction is generated based on the constraint system control parameters. The preload command is sent to the constraint system actuator via the vehicle bus; The restraint system actuator adjusts the airbag inflation device pressure level and seat belt pretensioner stroke according to the preload command.

7. A passenger and vehicle safety analysis method based on multi-dimensional collision simulation according to any one of claims 1 to 6, characterized in that, Before constructing the basic coupled model that includes mechanical and biological dimensions, the following steps are also included: Obtain vehicle deformation field distribution maps and passenger injury medical reports from historical crash tests; Enhanced datasets are created by synthesizing the correlation between passenger physiological data and vehicle structural parameters using generative adversarial networks; The association-enhanced dataset is then injected into the training process of the basic coupled model; Update the weight matrices for the mechanical and biological dimensions in the basic coupled model.

8. A passenger and vehicle safety analysis system based on multi-dimensional collision simulation, characterized in that, The analysis system includes: The data acquisition module is used to collect vehicle structural parameters, passenger physiological data, and collision environment parameters. The model building module is used to generate biomechanical feature vectors based on the passenger physiological data and to build a basic coupled model containing mechanical and biological dimensions by combining vehicle structural parameters. The collision phase division module is used to obtain the initial collision velocity based on the collision environment parameters, and divide the collision phases through a phase decision tree to obtain the collision phase division result. The phased simulation module is used to perform phased simulation on the basic coupling model based on the collision phase division results, and output a multi-dimensional dynamic dataset. The distribution map generation module is used to generate a three-dimensional safety margin distribution map based on the multidimensional dynamic dataset. The protection strategy decision module is used to execute protection strategy decision logic based on the three-dimensional safety margin distribution map and output constraint system control parameters.

9. A computer device, characterized in that: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images