New energy automobile body frame lightweight design method based on digital twinning

Abstract

The invention discloses a new energy automobile body frame lightweight design method based on digital twinning, and relates to the technical field of new energy automobile body structure design and computer-aided engineering simulation, and the method comprises the following steps: S1, building a digital twinning body connected with a physical automobile body sensing system; s2, a dynamic precision topology network module is used for dynamically dividing a simulation precision area of the vehicle body frame according to the mechanical energy transmission path and the real-time sensing data. According to the new energy automobile body frame lightweight design method based on digital twinning, intelligent allocation of computing resources is realized through a dynamic precision topology network, and the computing burden of multi-physics coupling simulation is effectively reduced while precise prediction of a key structure is ensured; the sharp contradiction among the model precision, the real-time response and the optimization efficiency is effectively solved, and effective closed-loop iteration of the lightweight design is achieved under the complex constraint condition.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicle body structure design and computer-aided engineering simulation technology, and specifically to a lightweight design method for a new energy vehicle body frame based on digital twins. Background Art

[0002] In the new energy vehicle sector, lightweight body frame design is crucial for improving range and reducing energy consumption. Digital twin technology, with its ability to construct virtual mappings of physical entities and enable interaction between the virtual and the real, provides a powerful tool for optimizing lightweight body frame design. Ideally, a design approach based on high-fidelity digital twins should continuously integrate real-time sensor data from the physical world, such as stress and temperature, and leverage accurate multi-physics coupled simulation models to efficiently perform multi-objective optimization iterations to find the optimal lightweighting solution while meeting numerous stringent performance constraints. However, existing digital twin-based body lightweighting design methods face a fundamental bottleneck in data processing. Building and maintaining high-precision, multi-physics coupled digital twin models requires enormous computing resources, making the simulation process significantly time-consuming. Furthermore, lightweight design itself is a complex multi-objective, multi-constrained optimization problem. Each design iteration requires evaluation using these high-precision models, further increasing the computational burden.

[0003] More importantly, in order to embody the core value of digital twins: virtual-reality synchronization, the system must also have the ability to process and fuse real-time or near-real-time data streams from the physical vehicle body to dynamically update the twin state and drive the optimization process. These three factors: maintaining high model fidelity for accurate predictions, rapidly responding to and fusing real-time data to maintain twin synchronization, and efficiently executing complex multi-objective optimization iterations, create sharp contradictions within limited computing resources and time windows. Existing technologies often find it difficult to effectively coordinate these three factors, forcing compromises on model accuracy, data update timeliness, or optimization efficiency. This results in delayed digital twin updates and lengthy design optimization cycles, ultimately limiting the speed, accuracy, and engineering practicality of lightweight design. Therefore, how to effectively resolve the contradictions between high-precision multi-physics simulation, real-time data fusion efficiency, and complex multi-objective lightweight optimization calculations in digital twin models has become a technical problem that urgently needs to be overcome in this method. Summary of the Invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: A lightweight design method for a new energy vehicle body frame based on digital twins, comprising the following steps:

[0005] Step S1: Establishing a digital twin connected to the physical vehicle body sensing system;

[0006] Step S2: a dynamic precision topology network module is used to dynamically divide the simulation precision area of ​​the vehicle body frame according to the mechanical energy transfer path and real-time sensor data;

[0007] Step S3: dual-track driving optimization engine module, including real-time update track and offline optimization track;

[0008] Step S4: Double verification module, performing residual screening and cross-model consistency check on the output solution.

[0009] Preferably, the execution of the dynamic precision topology network module includes:

[0010] The mechanical energy distribution of the body frame is calculated using a stress flow density algorithm, and continuous regions where the energy intensity exceeds a preset threshold are identified as topologically sensitive regions.

[0011] Cluster analysis of real-time strain sensing data to mark dynamic high-sensitivity areas where stress fluctuations are greater than the set tolerance;

[0012] The topologically sensitive domain is superimposed with the dynamic high-sensitivity area to generate a key precision domain map covering the entire body frame.

[0013] The atlas divides the vehicle body frame into a full-precision area, a medium-precision area, and a proxy model area.

[0014] Preferably, the full-precision area loads a complete finite element model, the medium-precision area adopts a simplified shell element model, and the proxy model area describes the structural response through parameterized equations.

[0015] Preferably, in the dual-track drive optimization engine module:

[0016] The real-time update track is deployed on the edge computing node, which receives the physical vehicle body sensor data stream, predicts the mechanical state change in the full-precision area through a pre-trained neural network, and directly modifies the finite element mesh node displacement in this area;

[0017] The offline optimized track is deployed on a cloud server, and a multi-objective genetic algorithm is run to generate lightweight candidate solutions. Based on the full-precision zone status of the real-time updated track output, the Bayesian optimizer is called to adjust the geometric parameters of the candidate solutions.

[0018] Preferably, the neural network is a hybrid architecture of a long short-term memory network and a three-dimensional convolutional network, the input is time series strain data and spatial temperature distribution, and the output is the displacement increment of six degrees of freedom in the full precision area.

[0019] Preferably, the dual verification module includes:

[0020] First-level residual screening: At fixed intervals, three sub-areas of the proxy model area are randomly selected and switched to the full-precision model to calculate performance indicators. If the deviation from the proxy model prediction value exceeds the allowable tolerance, the proxy model parameter self-learning is triggered and the optimization process is suspended.

[0021] The second cross-model verification: For the final lightweight solution, full-precision virtual collision simulation and reinforcement learning-based agent model reasoning are simultaneously performed to compare the differences in the results of the two in terms of maximum body deformation and first-order modal frequency.

[0022] Preferably, the reinforcement learning-based agent model is independently trained using historical high-precision simulation data and has no shared parameters with the main optimization system.

[0023] Preferably, the virtual collision simulation includes multi-physics field coupling calculations under frontal collision, side collision and torsional conditions.

[0024] Preferably, the method outputs a lightweight solution when the digital twin state update delay is lower than a set threshold and the time taken for a single optimization iteration is shorter than a benchmark duration.

[0025] Preferably, the key precision domain map is dynamically updated every two hours, and the update triggering conditions include: the rate of change of the statistical characteristics of the real-time sensor data exceeds a threshold value, or the optimization engine triggers the identification of a new topology sensitive domain.

[0026] The present invention provides a lightweight design method for the body frame of a new energy vehicle based on digital twins. It has the following beneficial effects:

[0027] This digital twin-based lightweight design method for new energy vehicle body frames achieves intelligent allocation of computing resources through a dynamic precision topological network, strictly limiting high-fidelity simulation to mechanical energy-dominated paths and real-time high-sensitivity areas. While ensuring accurate prediction of key structures, it effectively reduces the computational burden of multi-physics field coupling simulations. Combined with the decoupling design of the dual-track drive optimization engine, the real-time track uses neural network prediction and local correction mechanisms to achieve twin synchronization, while the offline track relies on cloud computing power to perform deep multi-objective optimization, effectively resolving the sharp contradictions among model accuracy, real-time response, and optimization efficiency, enabling lightweight design to achieve effective closed-loop iteration under complex constraints.

[0028] This lightweight design method for new energy vehicle body frames based on digital twins uses a dual verification system to build a full-process quality defense line: residual screening dynamically maintains the credibility of the proxy model through a periodic self-correction mechanism to avoid the risk of model drift caused by long-term operation; cross-model verification forces the parallel execution of high-precision physical simulation and independent data-driven reasoning, and uses a dual-channel arbitration mechanism to eliminate safety misjudgments caused by single model failure; in conjunction with intelligent circuit breaking rules, when update delays exceed the limit, optimization convergence is abnormal, or verification indicators conflict, dangerous solutions are automatically intercepted to ensure that the output results have both lightweight benefits and engineering feasibility, providing full-link technical guarantees for the safety of new energy vehicle bodies. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of module interaction of a lightweight design method for a new energy vehicle body frame based on digital twins of the present invention;

[0030] Figure 2 This is a flow chart of a lightweight design method for a new energy vehicle body frame based on digital twins according to the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] See also Figure 1 and Figure 2 The present invention provides a technical solution: a lightweight design method for a new energy vehicle body frame based on digital twins, comprising the following steps:

[0033] Step S1: Establishing a digital twin connected to the physical vehicle body sensing system;

[0034] Step S2: a dynamic precision topology network module is used to dynamically divide the simulation precision area of ​​the vehicle body frame according to the mechanical energy transfer path and real-time sensor data;

[0035] Step S3: dual-track driving optimization engine module, including real-time update track and offline optimization track;

[0036] Step S4: Double verification module, performing residual screening and cross-model consistency check on the output solution.

[0037] It should be further explained that in the specific implementation process, after establishing the digital twin connected to the physical vehicle body sensor system, the following process is implemented. The specific steps are as follows:

[0038] S01. Construction and execution of dynamic precision topology network, including the following:

[0039] Sensitive domain identification based on mechanical energy transfer paths: The stress flow density algorithm is used to calculate the energy distribution intensity in each area of ​​the vehicle body frame. Continuous structural regions where the energy transfer intensity exceeds a preset threshold are automatically extracted and marked as topologically sensitive domains.

[0040] Dynamic region marking with real-time data integration: Perform spatiotemporal clustering analysis on the incoming real-time strain sensor data stream. If the stress fluctuation amplitude in a certain area continuously exceeds the set tolerance range, it will be added as a dynamic high-sensitivity area.

[0041] Precision domain map generation: The topology-sensitive domain and the dynamic high-sensitivity area are superimposed and integrated to form a key precision domain map covering the entire body frame. This map divides the structure into three types of areas:

[0042] Full-precision area, i.e., topologically sensitive area + dynamic high-sensitivity area: geometric details and material nonlinear properties must be fully preserved;

[0043] Medium-precision area, i.e., secondary energy transfer path: allows simplification of element types and mesh density;

[0044] Proxy model area, i.e. low energy fluctuation area: the finite element model is replaced by parameterized equations.

[0045] S02. The coordinated operation of the dual-track drive optimization engine includes the following:

[0046] Real-time track updates: When new strain and temperature data is transmitted from the physical vehicle body sensors, the edge computing node immediately calls a pre-trained hybrid neural network. A long short-term memory network processes the time series data, while a three-dimensional convolutional network processes the spatial distribution. This allows for direct prediction of displacement and stress increments in the full-precision area. The global finite element solver is bypassed, and the predicted increments are mapped to the corresponding mesh nodes in the full-precision area. Only this area is updated locally, while the models in other areas remain unchanged.

[0047] Offline track optimization: A multi-objective genetic algorithm runs in the background to generate candidate design schemes with lightweight coefficient, structural stiffness, and modal frequency as optimization targets. When the track is updated in real time to output a new full-precision zone state, the Bayesian optimizer is immediately activated to fine-tune the cross-sectional thickness and material distribution parameters of the candidate schemes using this state as the boundary condition.

[0048] S03. Triggering and executing two-factor authentication includes the following:

[0049] First-level residual screening: At fixed intervals, the system automatically selects three sub-regions from the proxy model area and switches to the full-precision finite element model to recalculate performance indicators. If the calculated results of any sub-region deviate from the proxy model prediction by more than the allowable tolerance, the proxy model parameter self-learning process is triggered and all optimization iterations are frozen until the deviation is corrected.

[0050] The second level of cross-model verification: For the final lightweight solution, full-precision virtual collision simulation and independent agent model reasoning are started simultaneously: the virtual collision simulation must cover three working conditions: frontal impact, lateral crush and torsional load; the independent agent model adopts a reinforcement learning architecture, is trained based on historical high-precision simulation data and is isolated from the main system; the solution can only be verified and output when the difference between the two models in key indicators such as the maximum deformation of the vehicle body and the first-order modal frequency is less than the set threshold.

[0051] The implementation of the dynamic precision topology network module includes:

[0052] The mechanical energy distribution of the body frame is calculated using a stress flow density algorithm, and continuous regions where the energy intensity exceeds a preset threshold are identified as topologically sensitive regions.

[0053] Cluster analysis of real-time strain sensing data to mark dynamic high-sensitivity areas where stress fluctuations are greater than the set tolerance;

[0054] The topologically sensitive domain is superimposed with the dynamic high-sensitivity area to generate a key precision domain map covering the entire body frame.

[0055] The atlas divides the vehicle body frame into full-precision area, medium-precision area and proxy model area.

[0056] It should be further explained that, in the specific implementation process, the construction of the dynamic precision topology network is implemented according to the following logic, including:

[0057] S04. Identification of topologically sensitive domains based on physical mechanisms: Standard operating loads, such as bending and torsion, are applied to the initial body frame finite element model. The mechanical energy transfer intensity is calculated element by element using a stress flow density algorithm. This intensity is represented by the product of the element stress tensor and the strain energy density. The largest connected region with an energy intensity exceeding a preset threshold is automatically extracted. If this region contains critical structural nodes, such as welds and mounting holes, its boundary is expanded outward by one unit width to ensure integrity. During the identification process, areas with local stress concentration but low energy transfer contribution, such as small fillets and process holes, are excluded, retaining only the continuous structures that dominate the load transfer path.

[0058] S05. Dynamic high-sensitivity area marking based on real-time data: Continuously receive strain time series data uploaded by the physical vehicle body sensor network, group it by spatial location, and perform sliding window cluster analysis. If the strain fluctuation amplitude of a sensor group continuously exceeds the set tolerance range, and the fluctuation frequency matches the vehicle's current driving state, which includes acceleration and steering, the sensor coverage area is marked as a dynamic high-sensitivity area. When multiple sensor groups in the same area trigger the marking condition simultaneously, they are automatically merged into a single high-sensitivity area and boundary smoothing is performed.

[0059] S06. Generate and resolve conflicts in key precision domain maps: Spatially overlay the topologically sensitive domains and the dynamic high-sensitivity regions, and handle overlaps or separations according to the following rules, including:

[0060] Overlapping area: upgraded to the highest priority full-precision area;

[0061] Separate regions: retain their own independent region types;

[0062] When the boundary gap is smaller than the element size: forced connection to form a continuous area;

[0063] The final graph outputs three types of partitioning instructions, including:

[0064] Execute command in full precision area: load complete nonlinear material model and fine mesh;

[0065] Execution instructions in the medium-precision area: use equivalent homogeneous shell elements and allow mesh coarsening;

[0066] Execution instructions in the proxy model area: activate the parameterized response surface equation and close the finite element calculation.

[0067] The full-precision area loads the complete finite element model, the medium-precision area uses a simplified shell element model, and the proxy model area describes the structural response through parameterized equations.

[0068] It should be further explained that in the specific implementation process, the model construction is implemented according to the following mandatory rules for the three types of areas divided by the key precision domain map:

[0069] S07. Model loading mechanism for full-precision areas: When an area is marked as a full-precision area by the atlas, a nonlinear finite element model containing detailed geometric features is forced to be loaded. The material constitutive relationship must use an experimental calibration curve. Geometric details include ribs and welds.

[0070] If there is a multi-physics coupling effect in the area, such as thermal fatigue, the coupling solver is activated synchronously and the complete boundary condition transfer path is retained; the mesh division density follows the preset curvature adaptation rule: the mesh is automatically refined when the surface curvature radius is less than the threshold, and moderate relaxation is allowed in flat areas.

[0071] S08. Simplified execution logic for medium-precision areas: All secondary energy transfer paths marked as medium-precision areas are uniformly replaced with equivalent shell element models. Specific rules include:

[0072] If the thickness change rate of the original solid structure is lower than the set tolerance, it will be converted into a shell element of constant thickness; if there are local features, such as small holes and shallow grooves, they will be directly geometrically filled when their size is smaller than the average side length of the element; the mesh density is controlled in a hierarchical manner: the boundary layer adjacent to the full-precision area maintains a fine mesh; the interior of the area is allowed to be coarsened to an integer multiple of the original mesh size, but the element aspect ratio is strictly controlled within a reasonable range.

[0073] S09. Parameterization of proxy model areas: For low-sensitivity proxy model areas, the finite element calculation engine is completely disabled and replaced with a predefined parameterized response surface equation. The equation input variables only include the boundary displacement and temperature field of the adjacent full-precision / medium-precision areas. When real-time sensor data detects abnormal fluctuations in this area, such as a single-point strain mutation, the equation call is immediately frozen and the map update process is triggered.

[0074] In the dual-track drive optimization engine module:

[0075] The real-time update track is deployed on the edge computing node, which receives the physical vehicle body sensor data stream, predicts the mechanical state change in the full-precision area through a pre-trained neural network, and directly modifies the finite element mesh node displacement in this area;

[0076] The offline optimized track is deployed on a cloud server, and a multi-objective genetic algorithm is run to generate lightweight candidate solutions. Based on the full-precision zone status of the real-time updated track output, the Bayesian optimizer is called to adjust the geometric parameters of the candidate solutions.

[0077] It should be further explained that, in the specific implementation process, the dual-track drive optimization engine operates in coordination according to the following logic, including the following steps:

[0078] S010. Real-time track updates for fast response at the edge: When physical body sensors transmit new strain and temperature data to the edge computing node, a pre-trained hybrid neural network is immediately invoked. This network synchronously processes the spatiotemporal data stream and outputs full-precision displacement increment predictions for all six degrees of freedom.

[0079] Skip the finite element global equilibrium iteration process and directly map the predicted increments to the grid node coordinates corresponding to the full-precision area. Only the node displacement field in this area is modified while the state of the remaining areas remains frozen. If the neural network prediction confidence is lower than the set threshold, such as the input data exceeds the training range, switch to the simplified finite element solver to perform local recalculation and send a data anomaly alert to the cloud.

[0080] S011. Offline Track Optimization: Deep iteration in the cloud: A multi-objective genetic algorithm is continuously run in the cloud to generate a library of candidate solutions based on vehicle body mass, first-order torsional stiffness, and fatigue life at key points. When the track is updated in real time and a new full-precision zone displacement field is pushed, a three-stage response is immediately triggered, including the following stages:

[0081] Phase 1: Using the real-time updated displacement field as the mandatory boundary condition, verify whether the stress in the full-precision area of ​​the candidate solution exceeds the limit;

[0082] Phase 2: If the number of over-limit solutions exceeds the total ratio threshold, the Bayesian optimizer is activated to adjust the plate thickness distribution of the candidate solutions;

[0083] Phase 3: Re-inject the adjusted solution into the genetic algorithm population to replace the individual with the lowest fitness.

[0084] S012. Status synchronization rules between the two tracks: After each update of the real-time track, the full-precision zone-compressed displacement field including the timestamp is sent to the cloud. The cloud-based optimization track activates the Bayesian fine-tuning process only when it detects that the displacement field change amplitude is greater than the noise tolerance. If the edge sends continuous fault alerts, the cloud-based optimization will suspend optimization and roll back to the previous stable version of the twin state.

[0085] The neural network is a hybrid architecture of a long short-term memory network and a three-dimensional convolutional network. The input is time-series strain data and spatial temperature distribution, and the output is the displacement increment of six degrees of freedom in the full-precision area.

[0086] It should be further explained that, in the specific implementation process, the hybrid neural network implements spatiotemporal data processing and prediction according to the following rules, including the following steps:

[0087] S013. Dynamic screening and structuring of input data: Receive raw data streams uploaded by edge sensors in real time, automatically shield sensor nodes with signal strength below the noise threshold, and retain only valid measurement points. Reorganize valid strain data into a grid based on the vehicle body's spatial position, and fill in missing points through linear interpolation of adjacent nodes. Independently construct a 3D thermal distribution map based on temperature field data, spatially aligned with the strain grid but with separate channels.

[0088] S014. Dual-channel collaborative processing mechanism:

[0089] Time series feature extraction channel: For the strain time series data of each sensor node, a long short-term memory network is used to extract trend features according to a fixed time window. If the feature fluctuation direction is consistent and the amplitude increases within three consecutive time windows, it is marked as a significant change pattern and the weight is increased.

[0090] Spatial feature extraction channel: The reorganized strain grid and temperature distribution map are input into the three-dimensional convolutional network, and local stress concentration areas and thermal-mechanical coupling hotspots are identified through multi-scale convolution kernels; when the spatial gradient exceeds the material yield threshold, high-resolution feature map calculation is forced to start.

[0091] S015. Cross-channel feature fusion and displacement prediction: The trend feature vector output by the temporal channel and the thermal coupling feature map output by the spatial channel are concatenated in the channel dimension. Cross-domain feature fusion is performed through a fully connected layer. If conflicting signals are detected in the fused features, such as temporal compression but spatial stretching, the spatial channel result is prioritized and data review is triggered.

[0092] The final output layer generates displacement increments for the six degrees of freedom in the full-precision zone and adds a confidence score: the confidence is dynamically calculated based on the Mahalanobis distance between the input data and the training set; when the confidence is lower than the set threshold, the prediction result is frozen and the simplified solver is activated.

[0093] The two-factor authentication module includes:

[0094] First-level residual screening: At fixed intervals, three sub-areas of the proxy model area are randomly selected and switched to the full-precision model to calculate performance indicators. If the deviation from the proxy model prediction value exceeds the allowable tolerance, the proxy model parameter self-learning is triggered and the optimization process is suspended.

[0095] The second cross-model verification: For the final lightweight solution, full-precision virtual collision simulation and reinforcement learning-based agent model reasoning are simultaneously performed to compare the differences in the results of the two in terms of maximum body deformation and first-order modal frequency.

[0096] It should be further explained that, during the specific implementation process, the dual verification module performs reliability assurance according to the following rules, including the following steps:

[0097] S016. Periodic triggering of the first residual screening: At fixed working time intervals, the system automatically randomly selects three spatially discrete sub-regions from the proxy model area and immediately switches to the full-precision finite element model to recalculate the equivalent stress and displacement values ​​under the current working condition;

[0098] If the absolute deviation between the full-precision calculation result and the proxy model prediction value in any sub-region exceeds the allowable tolerance, the proxy model is deemed invalid and a three-stage self-correction is performed, including the following three stages:

[0099] Phase 1: Freeze all optimization iterations and retain a snapshot of the current design state;

[0100] Phase 2: Using the full-precision results as a benchmark, reverse fit the proxy model parameters until the residuals converge to within the tolerance range;

[0101] Phase 3: Perform extended verification of the revised model in adjacent areas. Once confirmed, unfreeze the optimization process.

[0102] If the results of three consecutive screenings are within the tolerance range, the interval between the next screenings will be automatically extended.

[0103] S017. Second-level cross-model verification solution output interception: When optimizing the trajectory to generate the final lightweight solution, two independent evaluations are forced to be launched in parallel:

[0104] Full-precision virtual collision simulation: Runs on a high-performance cloud cluster, fully loading geometric nonlinear and material plasticity models, covering three standard working conditions: frontal 100% overlap rigid wall collision, side column collision, and torsional load;

[0105] Reinforcement learning agent model reasoning: Calling the independently trained deep Q network model, inputting the solution design parameters and directly outputting the predicted values ​​of key performance indicators;

[0106] Dual-channel result comparisons implement strict arbitration: For core indicators such as maximum body deformation and first-order torsional modal frequency, if the difference between the two model results is less than a set threshold, the solution is immediately released to the manufacturing system.

[0107] If the difference between key indicators exceeds the limit, the conflicting items are automatically marked and root cause analysis is performed: if the full-precision simulation results do not meet the safety standards, the solution is discarded and the optimization is restarted; if the proxy model prediction deviation is too large, its retraining process is triggered and the solution is retained for review.

[0108] The reinforcement learning-based agent model is independently trained using historical high-precision simulation data and has no shared parameters with the main optimization system.

[0109] It should be further explained that, during the specific implementation process, the construction and operation of the reinforcement learning agent model strictly adhere to the following independence principles:

[0110] S018. Completely isolated collection of training data: Only historical high-precision simulation data is used as training samples. Any intermediate results or real-time sensor data generated during the real-time optimization process are prohibited. Sample screening performs double filtering, including the following two layers of filtering:

[0111] First-level filtering: exclude historical simulation cases with confidence scores lower than the set threshold;

[0112] Second-level filtering: automatically discarded when the sample distribution deviates from the current vehicle model parameters by more than the allowable deviation;

[0113] The final training set covers a balanced distribution of positive and negative samples for frontal collision, side collision and torsional conditions, and the total number of samples remains fixed to avoid data drift.

[0114] S019. Special design of model architecture and training: A deep Q-network architecture is used, but the number of hidden layer nodes and activation function type are strictly different from the hybrid neural network of the main system;

[0115] The training process is performed in three stages:

[0116] Basic training: Minimize the prediction error on historical datasets;

[0117] Adversarial enhancement: injecting noise samples and forcing the model to maintain stable output;

[0118] Safety margin enhancement: Increase the weight of critical samples close to the material failure threshold by ten times;

[0119] All parameters are frozen immediately after training is completed, disabling online updates or fine-tuning.

[0120] S020. Runtime hard isolation measures: Deployed on an independent physical server, design parameters and return results are transmitted only via encrypted APIs between the system and the main digital twin system; input and output variables are subject to formatting constraints, including:

[0121] Input: Only the geometric parameters and material grades of the body frame are received;

[0122] Output: Strictly limited to five core indicators such as maximum deformation of the vehicle body and first-order modal frequency;

[0123] If the input parameters exceed the historical training range, an error code will be returned immediately and the calculation will be rejected.

[0124] Virtual collision simulation includes multi-physics coupling calculations under frontal collision, side collision and torsion conditions.

[0125] It should be further explained that, in the specific implementation process, the construction and execution of the virtual collision simulation conditions follow the following intelligent selection and simplification rules, including the following steps:

[0126] S021. Mandatory coverage of basic conditions: Unconditionally perform full-precision simulation of three standard collision scenarios: 100% overlap frontal collision with a rigid wall: The vehicle adopts the initial velocity and boundary conditions specified in the regulations, but the vehicle body posture dynamically adjusts the pitch angle based on the real-time twin mass distribution;

[0127] Side column collision: The collision point is intelligently selected based on the most vulnerable area based on historical damage data from the vehicle body sensors, and the column diameter is set according to actual road condition statistics;

[0128] Pure torsion condition: The applied load value is a fixed multiple of the maximum design torque of the current vehicle model, but the direction is automatically aligned according to the weak axis of the real-time twin stiffness;

[0129] All working conditions are forced to retain local failure effects such as weld fracture and material tearing, and no simplification or neglect is allowed.

[0130] S022. Dynamic Screening of Derivative Conditions: Frequent non-standard operating conditions for this vehicle model, such as offset collisions and slope collisions, are extracted from the historical accident database and selectively activated according to the following rules: If the safety margin of the current lightweight solution under standard conditions falls below a threshold, the corresponding high-risk derivative condition is automatically added. If the optimization iterations reach a critical point without convergence, a derivative condition strongly related to the structural characteristics of the candidate solution (such as a specialized collision targeting the weight reduction area) is activated.

[0131] The derived working condition allows for moderate simplification: only the complete solution of the full-precision area is retained, and the medium-precision model is used in the non-critical area; the time step of the collision process can be enlarged to an integer multiple of the standard working condition.

[0132] S023. On-demand multiphysics coupling, including:

[0133] Basic coupled field: All load cases must include a joint structural-dynamic solution;

[0134] Conditional triggering of extended physical fields: When the material temperature sensor data continuously exceeds the threshold: activate the thermal-mechanical coupling field, but only implement it in the full-precision area; when the collision speed exceeds the critical value: activate the fluid-structure coupling, and use the parameterized wind pressure model instead of the full CFD; if there are composite material areas on the car body: force-activate the interlayer delamination effect simulation.

[0135] The method outputs a lightweight solution when the digital twin state update delay is lower than the set threshold and the single optimization iteration time is shorter than the benchmark time.

[0136] It should be further explained that during the specific implementation process, the output of the lightweight solution follows a strict three-level decision-making process, including the following steps:

[0137] S024. Dynamic calibration of real-time thresholds: Continuously monitor the digital twin state update latency. When five consecutive update cycles are shorter than the real-time threshold, the threshold is automatically adjusted downward to improve response standards.

[0138] If the resource utilization of edge computing nodes continues to be lower than the warning line, the upward floating threshold allows for a moderate delay, but the floating range is constrained by the safety factor; when any single update timeout reaches the critical multiple, the solution output is immediately suspended and system diagnosis is started.

[0139] S025. Optimize the composite determination of efficiency gain: Calculate the compression ratio of the average time spent in the current iteration relative to the baseline iteration time. When the compression ratio continuously exceeds the set efficiency gain threshold, activate the acceleration output channel;

[0140] Synchronously verify the optimization target convergence curve: If key targets, such as mass and stiffness, have entered a stable convergence stage, the efficiency gain is recognized as valid; if the target value still fluctuates violently, it is determined to be a pseudo-gain and the current compression ratio is ignored; when both the efficiency gain and convergence stability meet the standards, the candidate solution is marked and enters the final verification queue.

[0141] S026. Fuse arbitration of verification results: Double verification is enforced on all solutions entering the queue. Regardless of how excellent the real-time and efficiency indicators are, the solution will be immediately discarded if any of the following situations occurs: any standard operating condition in the full-precision virtual collision simulation exceeds the safety margin; the difference in core indicators across model verification continues to widen; residual screening is incomplete or is in the process of correction; only when the real-time performance meets the standards, the efficiency gain is effective, and all verifications are passed, will the solution be released to the manufacturing system and the current optimization status frozen.

[0142] The critical precision domain map is dynamically updated every two hours. Update trigger conditions include: the rate of change of statistical features of real-time sensor data exceeds the threshold value, or the optimization engine triggers the identification of new topology sensitive domains. It should be further explained that in the specific implementation process, the dynamic update of the critical precision domain map implements intelligent triggering and arbitration rules, including the following steps:

[0143] S027. Real-time data-driven emergency updates: Continuously monitor the data streams from the entire vehicle's sensor network. When the rate of change of statistical characteristics of any sensor group exceeds a threshold, such as a sudden change in the strain mean or spectral energy shift, a local map redrawing is immediately initiated.

[0144] The redrawing range is limited to the three-layer unit radiation area centered on the mutation point, and a spatial correlation test is performed: if three adjacent sensors are synchronously abnormal, the entire subsystem is expanded; if an isolated point mutates, only the point is marked for observation and the response is delayed; the update process uses incremental loading, only replacing the precision instructions of the affected area, keeping the remaining areas frozen to reduce disturbances.

[0145] S028. Preventive updates triggered by optimized paths: When the structural modification plan generated by the offline optimized track involves adjustments to topology-sensitive areas, such as adding weight-reducing holes or removing reinforcement ribs, a pre-update of the map will be automatically triggered before the next iteration cycle.

[0146] The pre-update strategy prioritizes difference comparisons: if the overlap between the new sensitive domain and the original map exceeds a threshold, the boundary is refreshed locally. If a new load path appears, the full model stress flow recalculation is initiated. The updated results must pass lightweight virtual working condition verification, such as static torsion, and can only be injected into the optimization process after passing the verification.

[0147] S029. Safety guarantee for periodic updates: Global graph review is mandatory at fixed working time intervals, but elastic simplification is performed according to system load: When edge computing resources are sufficient: the stress flow algorithm and data clustering are fully run; when resources are tight: only historical high-sensitivity areas are verified, and non-critical areas continue to use the proxy model confidence score; when low-risk deviations are found during the review, such as a slight increase in stress in the proxy model area, an asynchronous correction task queue is generated for delayed processing to ensure uninterrupted real-time optimization.

[0148] S030. Conflict Arbitration for Multi-Source Updates: When emergency updates, preventive updates, and periodic updates are triggered simultaneously, a three-level priority response is implemented, including the following:

[0149] Priority 1: Urgent updates involving safety-critical areas;

[0150] Secondary priority: Optimize preventive updates triggered by path changes;

[0151] Level 3 postponement: global periodic review;

[0152] After arbitration is completed, a synchronization lock instruction is sent to the dual-track engine to ensure optimal state consistency during the update.

[0153] It is important to further explain that, during implementation, after establishing a digital twin of the new energy vehicle body frame and connecting it to the physical sensor network, the system automatically divides the body structure into regions using a unique dynamic precision topological network. Specifically, based on the mechanical energy transfer path, the system first identifies continuous regions that dominate load transfer as topologically sensitive domains. Simultaneously, it analyzes the fluctuation characteristics of real-time strain data and marks areas where stress changes consistently exceed the normal range as dynamically sensitive regions. These two regions are then combined to generate a critical precision domain map, which divides the vehicle body into full precision, medium precision, and proxy model regions. The full precision region forcibly preserves geometric details and material nonlinearities, while the medium precision region uses an equivalent simplified model. The proxy model region completely disables finite element calculations and replaces them with parameterized equations. This map is dynamically updated at regular intervals or when sensor data suddenly changes, prioritizing safety-critical regions.

[0154] The dual-track drive optimization engines operate synchronously to achieve efficient design. The real-time update track is deployed on the edge computing node. When new data is transmitted by the sensor, the displacement increment of the full-precision area is directly predicted through the pre-trained hybrid neural network, and the global solver is skipped to modify only the grid node coordinates in this area. If the prediction confidence is insufficient, it switches to the simplified solver and issues an alarm. The offline optimization track runs a multi-objective genetic algorithm in the cloud to generate candidate solutions. When the real-time track pushes the new full-precision area status, it immediately activates the Bayesian optimizer with this status as the boundary condition to fine-tune the design parameters. On-demand synchronization is achieved between the two tracks through compressed packages of displacement fields with abnormal variation amplitudes.

[0155] All output solutions must be double-verified. The first level of residual screening is started periodically: three sub-areas of the proxy model area are randomly selected and switched to the full-precision model for recalculation. If the deviation of any result from the proxy model prediction value exceeds the limit, the optimization process is frozen and the parameters are corrected by reverse fitting. The second level of verification is mandatory before the solution is output: the full-precision virtual collision simulation and the independently trained reinforcement learning proxy model reasoning are run simultaneously. The collision simulation must include frontal rigid wall collision, side column collision and torsion conditions, and retain local failure effects such as weld fracture. The solution will only be passed if the difference between the two types of models in core indicators such as the maximum deformation of the body and the first-order modal frequency is less than the threshold.

[0156] The final release of the solution must meet three conditions: the digital twin update latency remains below the real-time threshold of dynamic calibration, the optimization iteration efficiency gain is verified as effective through convergence stability, and both double verifications are fully passed. If any link triggers a fuse condition (such as exceeding the collision safety limit or incomplete verification), the solution will be immediately discarded.

[0157] Through a dynamic precision allocation mechanism, full-precision calculations are focused on critical areas that account for less than 20% of the vehicle body volume, resolving the contradiction between high-fidelity simulation and computing resources. Through a dual-track decoupling architecture, twin updates and efficient deep optimization are achieved in parallel, resolving the conflict between real-time performance and design quality. Through a closed-loop verification system, periodic self-correction and dual-model arbitration before output ensure the reliability of the solution.

[0158] It should be further explained that, in the specific implementation process, a lightweight design method for the new energy vehicle body frame based on digital twins includes the following steps:

[0159] Step S1: Construct a dynamic precision topological network: Identify topological sensitive domains based on the mechanical energy transfer path of the vehicle body frame, and simultaneously analyze the fluctuation characteristics of real-time strain data to mark dynamic high-sensitivity areas; superimpose the two to generate a key precision domain map, and divide the vehicle body into full-precision area, medium-precision area, and proxy model area; when the real-time sensor data suddenly changes or the optimized structure is adjusted, the map incremental update is triggered.

[0160] Step S2: Dual-track drive collaborative optimization, including:

[0161] Real-time track edge response: Sensor data is fed into a pre-trained neural network, which directly outputs full-precision displacement increments and modifies corresponding node coordinates. If the prediction confidence is insufficient, a simplified solver is used.

[0162] Offline track cloud iteration: Run a multi-objective genetic algorithm to generate candidate solutions. When receiving the new state of the full-precision zone pushed by the real-time track, immediately start the Bayesian optimizer to fine-tune the design parameters using this state as the boundary condition.

[0163] Step S3: Perform two-factor authentication, including:

[0164] Residual screening: Periodically switch random sub-regions of the proxy model area to the full-precision model for recalculation. If the deviation exceeds the limit, freeze the optimization and reverse fit the proxy model;

[0165] Cross-model verification: Before outputting the solution, a full-precision virtual collision simulation and independent reinforcement learning agent model reasoning are run simultaneously. Verification is passed only when the difference in core indicators is less than the threshold.

[0166] Step S4: Intelligent decision output, including releasing the lightweight solution if and only if: the digital twin update delay is continuously lower than the dynamic calibration threshold, the optimization efficiency gain is verified to be effective through convergence stability, and all double verifications are passed; if any link triggers the fuse condition, such as collision safety exceeds the limit or verification is not completed, the solution will be immediately abandoned.

[0167] Intelligent allocation of computing resources is achieved through a dynamic precision topological network, and high-fidelity simulation is strictly limited to the mechanical energy-dominated path and real-time high-sensitivity areas. While ensuring accurate prediction of key structures, the computational burden of multi-physics field coupling simulation is effectively reduced. Combined with the decoupling design of the dual-track drive optimization engine, the real-time track uses neural network prediction and local correction mechanisms to achieve twin synchronization, and the offline track relies on cloud computing power to perform deep multi-objective optimization, effectively resolving the sharp contradiction between model accuracy, real-time response and optimization efficiency, and enabling lightweight design to achieve effective closed-loop iteration under complex constraints.

[0168] A dual verification system builds a full-process quality defense line: residual screening dynamically maintains the credibility of the proxy model through a periodic self-correction mechanism, avoiding the risk of model drift caused by long-term operation; cross-model verification forces the parallel execution of high-precision physical simulation and independent data-driven reasoning, and uses a dual-channel arbitration mechanism to eliminate safety misjudgments caused by single model failure; in conjunction with intelligent circuit breaking rules, when update delays exceed the limit, optimization convergence is abnormal, or verification indicators conflict, dangerous solutions are automatically intercepted to ensure that the output results have both lightweight benefits and engineering feasibility, providing full-link technical guarantees for the safety of new energy vehicle bodies.

[0169] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0170] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A lightweight design method for a new energy vehicle body frame based on digital twins, characterized in that: The steps include: Step S1: Establishing a digital twin connected to the physical vehicle body sensing system; Step S2: a dynamic precision topology network module is used to dynamically divide the simulation precision area of ​​the vehicle body frame according to the mechanical energy transfer path and real-time sensor data; Step S3: dual-track driving optimization engine module, including real-time update track and offline optimization track; Step S4: Double verification module, performing residual screening and cross-model consistency check on the output solution.

2. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 1, characterized in that: The execution of the dynamic precision topology network module includes: The mechanical energy distribution of the body frame is calculated using a stress flow density algorithm, and continuous regions where the energy intensity exceeds a preset threshold are identified as topologically sensitive regions. Cluster analysis of real-time strain sensing data to mark dynamic high-sensitivity areas where stress fluctuations are greater than the set tolerance; The topologically sensitive domain is superimposed with the dynamic high-sensitivity area to generate a key precision domain map covering the entire body frame. The atlas divides the vehicle body frame into a full-precision area, a medium-precision area, and a proxy model area.

3. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 2 is characterized in that: The full-precision area loads a complete finite element model, the medium-precision area adopts a simplified shell element model, and the proxy model area describes the structural response through parameterized equations.

4. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 1 is characterized in that: In the dual-track drive optimization engine module: The real-time update track is deployed on the edge computing node, which receives the physical vehicle body sensor data stream, predicts the mechanical state change in the full-precision area through a pre-trained neural network, and directly modifies the finite element mesh node displacement in this area; The offline optimized track is deployed on a cloud server, and a multi-objective genetic algorithm is run to generate lightweight candidate solutions. Based on the full-precision zone status of the real-time updated track output, the Bayesian optimizer is called to adjust the geometric parameters of the candidate solutions.

5. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 4 is characterized in that: The neural network is a hybrid architecture of a long short-term memory network and a three-dimensional convolutional network. Its input is time-series strain data and spatial temperature distribution, and its output is the displacement increment of six degrees of freedom in the full-precision area.

6. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 1 is characterized in that: The dual verification module includes: First-level residual screening: At fixed intervals, three sub-areas of the proxy model area are randomly selected and switched to the full-precision model to calculate performance indicators. If the deviation from the proxy model prediction value exceeds the allowable tolerance, the proxy model parameter self-learning is triggered and the optimization process is suspended. The second cross-model verification: For the final lightweight solution, full-precision virtual collision simulation and reinforcement learning-based agent model reasoning are simultaneously performed to compare the differences in the results of the two in terms of maximum body deformation and first-order modal frequency.

7. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 6 is characterized in that: The reinforcement learning-based agent model is independently trained using historical high-precision simulation data and has no shared parameters with the main optimization system.

8. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 6 is characterized in that: The virtual collision simulation includes multi-physics field coupling calculations under frontal collision, side collision and torsional conditions.

9. The lightweight design method for a new energy vehicle body frame based on digital twin according to claim 1, characterized in that: The method outputs a lightweight solution when the digital twin state update delay is lower than a set threshold and the single optimization iteration time is shorter than a benchmark time.

10. A lightweight design method for a new energy vehicle body frame based on digital twins according to any one of claims 2 to 9, characterized in that: The key precision domain map is dynamically updated every two hours, and the update triggering conditions include: the rate of change of the statistical characteristics of the real-time sensor data exceeds the threshold value, or the optimization engine triggers the identification of a new topology sensitive domain.

Images