A vehicle body structure optimization method and system considering vehicle body performance and crash damage

Abstract

This invention belongs to the field of vehicle body structure design technology, and provides a method and system for optimizing vehicle body structure considering vehicle performance and collision damage. The method includes: generating several sets of solutions in the solution space, where each set of solutions consists of the thickness of each plate component; for each set of solutions, performing a genetic operation, and then using a neural network to predict the driver injury value, vehicle mass, and first-order mode of the white body; if, compared to before the genetic operation, the predicted vehicle mass of a certain set of solutions does not increase and the first-order mode of the white body does not decrease, then that set of solutions is retained; based on the driver injury value, the retained solution is used to update the optimal solution; determining whether the iteration terminates, if not, returning to perform the genetic operation; if so, decoding the optimal solution into the thickness of each plate component. This not only reduces the driver injury value but also improves the vehicle body's NVH performance, avoiding the shortcomings of single-objective optimization that sometimes sacrifices other aspects.

Description

Technical Field

[0001] This invention belongs to the field of vehicle body structure design technology, and in particular relates to a method and system for optimizing vehicle body structure that takes into account vehicle performance and collision damage. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Early vehicle development largely relied on real-vehicle crash tests. Due to this heavy dependence on experience and testing, accuracy was low, development cycles were lengthy, and costs were extremely high. To shorten development cycles and reduce costs, analytical crash testing, multi-rigid-body dynamics, and the finite element method (FEM) were subsequently proposed. Among these, the FEM, due to its high accuracy, ease of model modification, and robust results, has been widely used in research related to automotive crash safety. With continuous research and exploration by scholars both domestically and internationally, significant research achievements have been made in automotive crash safety.

[0004] A 25% offset frontal collision, due to its different energy absorption mechanism, is more likely to cause injury than a frontal collision, making it one of the most challenging scenarios for vehicle collision safety. With ongoing research, many scholars have achieved some results in addressing this issue. By analyzing the energy transfer path during a collision and optimizing the vehicle's structure and material thickness, significant improvements have been made, resulting in a certain degree of enhanced vehicle safety during collisions.

[0005] In related studies on vehicle body structure optimization, only the vehicle's collision safety was considered during the optimization process, while other performance aspects were ignored. Improving collision safety can lead to a decline in other vehicle performance aspects. Summary of the Invention

[0006] To address the technical problems mentioned above, this invention provides a method and system for optimizing vehicle body structure that considers vehicle body performance and collision damage. It considers the NVH performance of the vehicle body, uses the first-order mode and mass of the vehicle body as constraints, and aims to minimize driver injury. The invention explores a method that combines neural networks and genetic algorithms to optimize vehicle body design, which not only reduces driver injury but also improves the NVH performance of the vehicle body, avoiding the shortcomings of single-objective optimization that often result in unintended consequences.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] The first aspect of the present invention provides a method for optimizing a vehicle body structure that takes into account vehicle body performance and collision damage.

[0009] A method for optimizing a vehicle body structure that considers vehicle body performance and collision damage includes:

[0010] Several sets of solutions are generated in the solution space. For each set of solutions, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body. Based on the driver injury value, a set of solutions is selected as the optimal solution. Each set of solutions consists of the thickness of each plate.

[0011] For each set of solutions, after performing the genetic operation, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body. If, compared to before the genetic operation, the predicted vehicle mass of a certain set of solutions does not increase and the first-order mode of the white body does not decrease, then that set of solutions is retained. Based on the driver injury value, the retained solution is used to update the optimal solution. It is then determined whether the iteration terminates. If not, the genetic operation is performed again. If so, the optimal solution is decoded into the thickness of each plate.

[0012] Furthermore, the driver injury value is an indicator of head injury to the driver in an offset collision.

[0013] Furthermore, the first-order modes of the body-in-white include: the first-order torsional mode and the first-order bending mode of the body-in-white.

[0014] Furthermore, the panels include: front longitudinal beam outer panel, front longitudinal beam inner panel, upper longitudinal beam, longitudinal beam connecting beam outer panel, longitudinal beam connecting beam inner panel, front bulkhead, side panel, front pillar upper inner panel, front pillar lower inner panel, roof, middle pillar inner panel and / or rear tail panel.

[0015] Furthermore, the neural network employs a recurrent neural network with local memory units and local feedback connections.

[0016] Furthermore, based on the vehicle body's noise, vibration, and acoustic roughness, through whole-vehicle collision simulation analysis and panel acoustic sensitivity analysis, it was determined that each solution is composed of the thickness of each panel.

[0017] Furthermore, the genetic operations include crossover and mutation.

[0018] A second aspect of the present invention provides a vehicle body structure optimization system that takes into account vehicle body performance and collision damage.

[0019] A vehicle body structure optimization system that considers vehicle body performance and collision damage includes:

[0020] The initialization module is configured to generate several sets of solutions in the solution space, and for each set of solutions, use a neural network to predict the driver injury value, vehicle mass and first-order mode of the white body, and select a set of solutions as the optimal solution based on the driver injury value; wherein each set of solutions consists of the thickness of each plate.

[0021] The optimization module is configured as follows: for each set of solutions, after performing a genetic operation, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body; if the predicted vehicle mass does not increase and the first-order mode of the white body does not decrease compared to before the genetic operation, then the set of solutions is retained; based on the driver injury value, the retained solution is used to update the optimal solution; it is determined whether the iteration terminates. If not, the genetic operation is performed again; if so, the optimal solution is decoded into the thickness of each plate.

[0022] A third aspect of the present invention provides a computer-readable storage medium.

[0023] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the deep neural network training computational performance prediction method described in the first aspect above.

[0024] A fourth aspect of the present invention provides a computer device.

[0025] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the deep neural network training computational performance prediction method described in the first aspect above.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] This invention considers the NVH performance of the vehicle body, takes the first-order mode and mass of the vehicle body as constraints, and takes minimizing the driver injury value as the optimization objective. It explores a method that combines neural networks and genetic algorithms to optimize the vehicle body design. This method not only reduces the driver injury value, but also improves the NVH performance of the vehicle body. It avoids the shortcomings of single-objective optimization that may lead to unintended consequences, and provides a reference method for carrying out multidisciplinary optimization design research on vehicle body structure. Attached Figure Description

[0028] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0029] Figure 1 This is a schematic diagram of the 25% offset collision model shown in Embodiment 1 of the present invention;

[0030] Figure 2 This is a schematic diagram of the BIP vehicle body model shown in Embodiment 1 of the present invention;

[0031] Figure 3 This is a schematic diagram of a vehicle-dummy collision model shown in Embodiment 1 of the present invention;

[0032] Figure 4 This is a schematic diagram of the front deformation and overall vehicle attitude of a vehicle in a 25% offset collision, as shown in Embodiment 1 of the present invention.

[0033] Figure 5 This is a graph showing the change in energy and increase in mass during a 25% offset collision, as illustrated in Embodiment 1 of the present invention.

[0034] Figure 6 This is a graph showing the three-dimensional acceleration measurement curves of the occupant's head as illustrated in Embodiment 1 of the present invention;

[0035] Figure 7 This is a composite acceleration curve of the occupant's head shown in Embodiment 1 of the present invention;

[0036] Figure 8 This is the first torsional mode shape diagram of the vehicle body structure shown in Embodiment 1 of the present invention;

[0037] Figure 9 This is the first-order bending mode shape diagram of the vehicle body structure shown in Embodiment 1 of the present invention;

[0038] Figure 10(a) is a schematic diagram of the first-order torsional mode sensitivity shown in Embodiment 1 of the present invention;

[0039] Figure 10(b) is a schematic diagram of the first-order bending mode sensitivity shown in Embodiment 1 of the present invention;

[0040] Figure 10(c) is a schematic diagram of the vehicle body mass sensitivity shown in Embodiment 1 of the present invention;

[0041] Figure 11 This is a diagram of the Elman neural network topology shown in Embodiment 1 of the present invention;

[0042] Figure 12 This is a flowchart of the genetic algorithm optimization process shown in Embodiment 1 of the present invention. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0044] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0045] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0046] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of the present invention. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0047] Example 1

[0048] This embodiment provides a method for optimizing vehicle body structure that takes into account vehicle performance and collision damage.

[0049] This embodiment provides a vehicle body structure optimization method that considers vehicle body performance and collision damage. Based on a summary of vehicle collision-related research, taking a microcar as an example, when optimizing collision safety, the NVH (noise, vibration, and harshness) performance of the vehicle body is considered. The first-order mode and vehicle mass of the vehicle body are used as constraints, and the optimization objective is to minimize the head injury value of the driver. This invention explores a multidisciplinary optimization design method that uses neural network algorithms to comprehensively consider vehicle collision safety and NVH performance.

[0050] 1. Establish a finite element model.

[0051] (1.1) Finite element model of vehicle collision.

[0052] Road traffic accidents are mostly head-on collisions, with head-on collisions with an overlap rate of less than 25% having a high incidence rate. Because their energy absorption and force transmission methods differ from other head-on collisions, they are more likely to cause death or injury to occupants. In 2016, the C-IASI evaluation procedure added a 25% small offset crash test. However, the evaluation results of the past three years show that less than one-third of vehicle models can achieve an excellent rating. There is an urgent need to optimize some models on the market and test their crash safety using a 25% offset crash test. This study selects a 25% offset crash test at a speed of 64 km / h from the C-IASI test to analyze the injuries suffered by occupants.

[0053] like Figure 1 As shown, the model has a total mass of 962.914 kg, 1,950,979 nodes, and 1,931,233 elements, of which 61,366 are triangular elements, accounting for approximately 4.12%. Based on the accuracy requirements of the collision simulation, an 8mm mesh size was selected for the target elements, and the overall vehicle mesh range was 4-10mm. The vehicle model was preprocessed using Hypermesh software and solved using the LS-DYNA solver.

[0054] (1.2) Finite element model of the vehicle body.

[0055] There are generally two models available for modal analysis of the body-in-white (BIW). BIW refers to the welded body sections, bolted collision energy-absorbing structures, excluding glass, doors, engine hood, sunroof, trunk lid, and fenders. For example... Figure 2 As shown, BIP is based on BIW with the addition of front and rear windshields and triangular windows. For BIP, the overall torsion and overall bending modes can usually be identified.

[0056] The vehicle body NVH analysis model is simpler than the crash model, retaining only the body panels, subframe, body welds, weld seams, and adhesives. During NVH performance analysis, the deformation of body components is considered linear elastic deformation; therefore, the material card type for body panels, closures, and welds is MAT1, with material parameters including density, elastic modulus, and Poisson's ratio. To improve accuracy, the CNRB rigid elements in the LS-DYNA crash model were modified to RBE2 elements from OptiStruct. Furthermore, spot welds in the OptiStruct model are modeled using ACM (Shell Gap) elements with the same elastic properties as the welded components. The model contains 550,602 shell elements, 2,976 ACM weld elements, and 1,496 RBE2 elements.

[0057] 2. Whole vehicle collision simulation analysis.

[0058] (2.1) Simulation model of 25% offset collision of the whole vehicle.

[0059] Since the invention of the first crash test dummy, vehicle crash test dummies have evolved from a single dummy model to a diverse range of models for different types and groups. Different dummy models allow for the acquisition of test data on different groups during collisions. Among these, the most widely used are the Hybrid III dummy and the THOR dummy. The THOR dummy is a successor to the Hybrid III dummy, addressing its shortcomings while incorporating numerous improvements, including the addition of sensors for the face, neck, and chest. Therefore, compared to the Hybrid III dummy, the THOR dummy can more realistically and accurately reflect the injuries suffered by occupants during an actual collision.

[0060] During the collision, a THOR dummy model at the 50th percentile was used to simulate the injuries suffered by the driver in a real collision. For example... Figure 3 The image shows a vehicle dummy model. The THOR-50M dummy model has the same material properties, physical structure, and mechanical response characteristics as its real-life counterpart. The restraint system includes a two-dimensional seatbelt model, a steering wheel-mounted airbag, and side curtain airbags.

[0061] For the frontal 25% overlap rigid barrier collision simulation, the vehicle model collides with the rigid barrier at a speed of 64 km / h. The LS-DYNA solver is used, and the solution time is set to 150 ms.

[0062] (2.2) Collision result analysis.

[0063] In traffic accidents, frontal collisions are often the most frequent and deadly. Because frontal collisions are typically accompanied by high speeds and significant kinetic energy, components such as the engine are subjected to a strong impact, squeezing the front bulkhead and compressing the survival space for front-seat passengers. Furthermore, the immense impact inertial force often results in severe head and chest injuries to occupants, frequently leading to death.

[0064] In a 25% offset crash test, the front of the vehicle overlaps with the obstacle by 25%. Under these conditions, the impact at the collision point is significant due to the small contact area between the front of the vehicle and the obstacle, posing a severe challenge to the vehicle's crash safety in extreme situations. C-NCAP tests use a 40% offset deformable obstacle for offset small overlap crashes. In contrast, this embodiment, based on C-IASI's 25% offset crash test, uses a rigid obstacle for offset crash analysis, which is more suited to simulating collisions between the vehicle and walls, utility poles, etc.

[0065] According to C-IASI, the collision scenario uses a non-deformable barrier, and the initial vehicle collision velocity is 64 km / h. The calculation time is set to 150 ms using a control card. After configuring other control cards such as energy control, time step, and file generation, the solution file is exported and submitted to the LsDyna solver for solving.

[0066] Figure 4 The image shows the deformation and attitude of the vehicle's front end during the collision process within 150ms. Throughout the collision, the left front component experienced significant deformation, and the front of the left A-pillar underwent localized deformation. Approximately 100ms after the impact, the force concentrated on the left side, resulting in substantial lateral movement of the rear of the vehicle. The effectiveness of the model under a 25% offset collision scenario is also analyzed using energy and mass changes during the collision process.

[0067] like Figure 5 As shown in the energy curve, the total energy remains essentially constant throughout the collision. At the start of the collision, the total energy is primarily kinetic energy. Around 70 ms, the kinetic energy reaches its minimum, while the internal energy reaches its maximum. The mass increase during the collision is approximately 12.5 kg, representing about 0.82% of the total mass. The energy and mass changes in the 25% offset collision model are both within the normal range. The energy change in this calculation model is reasonable, and the hourglass energy ratio and mass increase are both within the normal range, thus it can be used for subsequent research and analysis.

[0068] Head injuries are one of the main forms of injury in traffic accidents. Different degrees of head injuries can cause serious injury or even death. This embodiment mainly studies the head injuries suffered by occupants in traffic accidents. Injuries to occupants in traffic accidents can be mainly divided into two situations: one is direct impact of the head with the vehicle interior, and the other is brain tissue injury caused by inertial forces acting on the head.

[0069] HIC 36 It is currently the most commonly used indicator for evaluating human head injuries in experiments, and the calculation formula is shown below:

[0070]

[0071] In formula (1), t1 and t2 are the start and end times of the selected time interval in the entire simulation, and a(t) is the composite acceleration at the center of gravity of the human head. Currently, the time interval between t1 and t2 in experiments is generally 36 ms, mainly because the HIC value is highest within a 36 ms time interval, which better reflects the injuries suffered by occupants in real accidents. Table 1 shows the human head injury index HIC. 36 The corresponding AIS level relationship.

[0072] Table 1. Human Head Injury Index (HIC) 36 Corresponding AIS level

[0073] 1 130-519 Headache and dizziness minor injury 2 520-899 Temporary loss of consciousness; linear fracture Moderate injury 3 900-1254 Loss of consciousness within 1-6 hours; depressed fracture Severe injury 4 1255-1574 Loss of consciousness within 6-24 hours; open fracture severe injury 5 1574-1859 Prolonged loss of consciousness; cerebral hematoma Severe injury, life-threatening 6 >1860 Death or beyond rescue lethal

[0074] The measurement results of the accelerometers in the X, Y, and Z directions at the measurement points on the dummy's head are exported, and the changes in acceleration of the head in the three directions during the collision are shown below. Figure 6 As shown. X-axis acceleration represents the forward and backward motion of the head; the X-axis acceleration curve peaks within 70-80 ms and then gradually decays. Y-axis acceleration represents the left and right motion of the head, and its curve is approximately ±30 m / s². 2 Internal fluctuations occurred, but due to significant lateral movement of the vehicle body during the collision, the peak Y-axis acceleration of 69.3 m / s² was reached approximately 148 ms at the end of the collision. 2 The Z-axis acceleration manifests as the pitch motion of the head, with its acceleration curve reaching a reverse peak of -62.42 m / s² at 61.5 ms. 2 Due to the reaction force, it reaches a positive peak speed of 64.56 m / s at 85.1 ms. 2 It then gradually diminishes.

[0075] By synthesizing the three-dimensional accelerations and taking the root of the sum of the squares of the three-dimensional acceleration curves, the composite acceleration curve can be obtained, as shown in the figure. Figure 7 As shown.

[0076] The composite acceleration curve is calculated according to formula (1) to obtain the driver's HIC during the collision process. 36 The value, within the range of 59.30ms-95.30ms, is HIC. 36 The value reached a peak of 845.96, according to the Human Head Injury Index (HII). 36 The corresponding AIS grading table shows that the corresponding head injury level is moderate, with the specific injury description being temporary loss of consciousness and linear fracture.

[0077] 3. Simulation analysis of vehicle body NVH performance.

[0078] (3.1) Modal analysis.

[0079] When performing NVH performance analysis on the vehicle body, the first-order overall mode of the body-in-white should be considered first. If the frequency value of the first-order mode is too low, it is easily excited by low-frequency excitations such as those from the engine, causing body vibration and noise, thus affecting the ride comfort of the vehicle. Therefore, the reduction of the first-order mode frequency of the body-in-white should be avoided during the modal planning stage. The modal solution file should be submitted to Optistruct for calculation. Some modal values ​​and mode shapes are shown in Table 2, and the first-order mode shapes are shown in... Figure 8 and Figure 9The body-in-white weighs 306.86 kg.

[0080] Table 2. First eight non-zero modal frequencies and mode shapes of the vehicle body structure

[0081] First stage 28.99 Water tank support lateral mode Second stage 29.63 Longitudinal mode of the rear tail plate of the water tank bracket Third stage 30.87 Longitudinal mode of the rear tail plate of the water tank bracket Fourth stage 37.24 Water tank support lateral mode Fifth stage 39.75 First-order torsional mode Sixth level 43.34 First-order bending mode Seventh level 45.84 Local modes of water tank support Eighth level 48.76 Torsional mode

[0082] (3.2) Acoustic sensitivity analysis of the board.

[0083] One of the major causes of in-vehicle noise is the vibration modes of the body panels. In modal analysis, the points with zero or very small displacements are called modal nodes. Optimizing the structure to distribute the vehicle mass as close as possible to these modal nodes can significantly improve body vibration. Without altering the body shape design, changing the thickness of the body panels not only affects the vehicle's NVH performance but also has a significant impact on collision safety. Body panel thickness, as a crucial structural parameter, is considered as a design variable to obtain an optimal combination of panel thicknesses, thus optimizing collision safety while considering NVH performance.

[0084] Modal analysis was used to determine the first-order torsional and bending modes of the vehicle body as constraints for in-vehicle noise optimization. Lightweight design in automobiles can significantly improve fuel economy while reducing costs; therefore, NVH performance cannot be improved by reducing lightweighting performance during the optimization process. Thus, body mass was added as an optimization constraint. Body panels were selected and numbered, and sensitivity analysis was performed on panel thickness based on the bending and torsional modes and body mass.

[0085] Table 3. Sensitivity Analysis Variables

[0086]

[0087]

[0088] Based on the input variables defined in Table 3, the first-order torsional mode, first-order bending mode, and body mass of the white body are defined as the output responses. With all factors having two levels, the Plackett-Burman sampling method can calculate the main effects of the factors with the fewest number of trials. In the sensitivity analysis, the interaction effects between design variables are ignored, and the influence of variations in design variables between their upper and lower limits on each response is examined. The Plackett-Burman sampling method is used.

[0089] In the HyperGraph post-processing software, the sensitivity of each design variable to the response and its sign can be viewed by examining the Pareto plot, as shown in Figures 10(a), 10(b), and 10(c).

[0090] The vertical scales in Figures 10(a), 10(b), and 10(c) all represent sensitivity. Increasing the thickness of the body panels improves the modal sensitivity of the body, therefore the modal sensitivity is generally positive. Panel thickness is directly proportional to panel mass, so the entire body mass has a positive sensitivity. When selecting design variables, the first-order torsional mode, the first-order bending mode, and the two panels with higher body mass sensitivity should be considered comprehensively. The side panels have a significant impact on both modal and mass sensitivity.

[0091] 4. Occupant injury value prediction.

[0092] (4.1) Experimental design.

[0093] Through simulation analysis of a 25% offset crash and analysis of the vehicle's NVH performance, the objectives, constraints, and design variables for occupant injury prediction were ultimately determined. The objective function is the driver's head injury risk (HIC) in an offset crash. 36 Damage values. The constraints are: first-order torsional mode of the body-in-white; first-order bending mode; body mass (including only the body-in-white and closing components). Twelve plates, including the outer front longitudinal beam, the inner front longitudinal beam, and the upper longitudinal beam, are used as variables in the subsequent experimental design, and their variable values ​​are shown in Table 4.

[0094] Table 4. Experimental design variables.

[0095]

[0096]

[0097] In experimental design, there are various sampling methods to choose from, commonly including full factorial design, Latin hypercube, Taguchi method, and Hammersley sampling. The advantage of Hammersley sampling is that it can provide a reliable estimate of the output statistics with a small number of samples. It also achieves a good uniform distribution on a k-dimensional hypercube, and the identification process of Hammersley sampling is relatively efficient. It only requires sampling on the input and output signals, without needing to identify the internal state of the system. Hammersley sampling is chosen for Design of Experiments (DOE) to obtain the subsequent training and test sets.

[0098] (4.2) Elman Neural Network.

[0099] A neural network is a network composed of numerous interconnected neurons. Based on the information flow during network operation, neural networks can be divided into two basic types: feedforward neural networks and feedback neural networks. Feedforward networks rely on hidden layers and nonlinear transfer functions to achieve complex nonlinear mapping functions. The output of a feedforward network depends only on the current output and the weight matrix, and is independent of previous network outputs. In contrast, feedback neural networks receive delayed input or output data feedback, making them a feedback dynamic system; therefore, feedback neural networks are also called recurrent neural networks or regression networks.

[0100] The Elman neural network (a recurrent neural network with local memory units and local feedback connections) topology is as follows: Figure 11 As shown, the Elman neural network adds a support layer to the hidden layers of a feedforward network to act as a one-step delay operator for memory purposes. An Elman neural network generally consists of four layers: an input layer, hidden layers, a support layer, and an output layer. As illustrated, the connection between the input, hidden, and output layers is similar to that of a feedforward network. The units in the input layer act as signal transmitters, and the units in the output layer act as weighting agents. The units in the hidden layers can be linear or nonlinear functions, and the support layer is used to remember the output of the hidden layer units from the previous moment and return it as input to the network. Because of the self-connected nature of the Elman neural network, it is highly sensitive to past data. Furthermore, the addition of the internal feedback network enhances the network's ability to process dynamic information. Therefore, the Elman neural network can approximate any nonlinear mapping with arbitrary precision.

[0101] The structure of an Elman neural network is as follows: Figure 11 As shown, its nonlinear spatial expression can be expressed as:

[0102] y(k)=g(w 3 x(k)) (2)

[0103] x(k)=f(w 1 x c (k)+w 2 (u(k-1))) (3)

[0104] x c (k)=x(k-1) (4)

[0105] In equation (2), y is the m-dimensional output vector; x is the n-dimensional intermediate layer node unit vector; u is the r-dimensional input vector; x c w is an n-dimensional state vector of the receiving layer. 3 The weights from the intermediate layer to the output layer; w 2The weights from the input layer to the intermediate layer; w 1 denoted as , where is the weight from the receiving layer to the intermediate layer; g(*) is the transfer function of the output layer neurons, which is a linear combination of the outputs of the intermediate layers; f(*) is the transfer function of the intermediate layer neurons.

[0106] 5. Analysis of Elman prediction results.

[0107] Neural network for driver head HIC in a collision 36 The values, first-order modes of the body-in-white, and body mass are predicted. Table 5 shows the driver head HIC. 36 To compare the predicted values ​​with the simulation results, and to compare the prediction accuracy with other neural network algorithms, radial basis function neural networks, generalized regression neural networks (GRNN), and backpropagation neural networks were used to predict the head injury values ​​of drivers. The mean square error, root mean square error, mean absolute error, and mean absolute percentage error were used to evaluate the prediction error, as shown in Table 5.

[0108] Table 5. Comparison of Errors in Algorithm Prediction of Head Injury Values

[0109]

[0110]

[0111] As can be seen from the data in Table 5, the Eleman neural network's predicted values ​​are highly consistent with the actual values, and its prediction error is smaller compared to other neural networks. The degree of consistency between predicted and actual values, from highest to lowest, is Eleman, GRNN, REF, and BP.

[0112] Mean Square Error (MSE) represents the sum of squares of the distances between each predicted value and the actual value. The smaller the value, the smaller the error and the higher the prediction accuracy.

[0113]

[0114] Root Mean Squared Error (RMSE) is the square root of MSE, used to measure the deviation between the observed value and the true value.

[0115]

[0116] Compared to mean absolute error, mean absolute error (MAE) is better reflected because the deviation is absolute and there is no cancellation between positive and negative values.

[0117]

[0118] MAPE (Mean Absolute Percentage Error) provides a more intuitive representation of forecast accuracy, somewhat similar to the concept of percentage increase. It compares the difference between the predicted and actual values ​​to the actual value, showing the percentage difference.

[0119]

[0120] The prediction errors of the first-order mode of the body-in-white and the body mass predicted by the Elman neural network algorithm are shown in Table 6.

[0121] Table 6. Prediction Error of Elman Neural Network Algorithm

[0122] First-order torsional mode of the white body / Hz <![CDATA[1.883×10 -2 ]]> <![CDATA[1.372×10 -1 ]]> <![CDATA[1.183×10 -1 ]]> <![CDATA[2.985×10 -1 ]]> First-order bending mode of the white body / Hz <![CDATA[2.845×10 -1 ]]> <![CDATA[5.334×10 -1 ]]> <![CDATA[3.941×10 -1 ]]> <![CDATA[9.306×10 -1 ]]> Vehicle body weight / kg <![CDATA[2.424×10 -5 ]]> <![CDATA[4.923×10 -3 ]]> <![CDATA[3.603×10 -3 ]]> 1.167

[0123] Tables 5 and 6 show that the Elman neural network has the smallest prediction error for driver head injury values, significantly outperforming the other three neural network algorithms in terms of mean square error, root mean square error, mean absolute error, and mean absolute percentage error. The Elman neural network algorithm also exhibits small prediction errors when predicting vehicle mass and the first-order mode of the body-in-white, meeting the prediction requirements. Considering manufacturing precision, the step size for each input variable in the solution space is set to 0.1, resulting in a total of 244,140,625 feasible solutions. The Elman neural network is used to predict the outputs corresponding to the feasible solutions: driver injury value, vehicle mass, and first-order mode. With the objective of minimizing driver head injury values ​​and constraints of not increasing vehicle mass and not decreasing the first-order mode of the body-in-white, the optimal solution is sought within the feasible solution set. The mathematical expression is:

[0124]

[0125] Given the enormous solution set, iterative computation is the primary method for solving this problem numerically. However, general iterative methods are prone to getting trapped in local minima, resulting in "infinite loops" that prevent further iteration. Genetic algorithms, a global optimization algorithm, effectively overcome this drawback. Compared to traditional optimization methods, genetic algorithms, based on biological evolution, exhibit excellent convergence, low computation time, and high robustness. Therefore, genetic algorithms are chosen to search for optimal solutions within the solution space. Figure 12The flowchart shown is for the algorithm. First, several sets of solutions (the thickness of each plate) are generated within the solution space (i.e., the upper and lower limits shown in Table 4). These solutions are then processed by a trained Elman neural network to obtain predicted values ​​for each set. Solutions that satisfy the constraints are retained until the population size meets the requirements. Fitness is calculated for each individual in the population. The fitness is the difference between the driver's head injury value corresponding to the initial solution and the predicted injury values ​​of each solution. The larger the difference, the better the fitness and the superior the individual. After performing genetic, crossover, and mutation operations on the individuals, the Elman neural network makes predictions, and this optimization process is repeated. The initial solution is the first globally optimal solution. If a solution better than the current optimal solution is found, the optimal solution is replaced until the iteration terminates.

[0126] The optimal solution was obtained by optimizing the solution space and finding the optimal solution that satisfies the constraints: [1.7,2,1,1.3,1.5,0.7,0.8,1.3,0.6,0.8,1.2,0.8]. To verify the quality of the Elman neural network algorithm's final predicted solution, the plate thickness of the optimal solution was assigned to the finite element model as an attribute, and the model was submitted to the Optistruct and LS-DYNA solvers for simulation calculation. The optimization results are shown in Table 7. (HIC of the driver's head) 36 The value decreased by 173.43, the vehicle weight decreased, but remained basically at the original level. The first-order mode has been optimized compared to the original data.

[0127] Table 7. Comparison of Responses Before and After Optimization for Each Constraint

[0128] <![CDATA[Driver's head HIC 36 value]]>< 845.96 627.1895 672.53 6.74 Vehicle body weight / kg 306.86 304.921 306.62 0.55 First-order torsional mode of the white body / Hz 39.749 40.209 39.972 -0.59 First-order bending mode of the white body / Hz 43.338 44.504 44.800 0.66

[0129] A vehicle body simulation model and a full-vehicle collision model were established using Hypermesh software. NVH (Noise, Vibration, and Harshness) and frontal offset collision simulation analyses were conducted. An optimization design method that comprehensively considers vehicle NVH performance and frontal offset collision performance using an Elman neural network is proposed. Results demonstrate that while ensuring the vehicle body's NVH performance, the overall frontal offset collision performance is significantly improved, avoiding the shortcomings of single-objective optimization that compromises other aspects. This confirms that improving both ride comfort and safety is achievable in vehicle body design. The Elman neural network algorithm's prediction errors for each response are all within 7%, meeting the prediction accuracy requirements. This provides a reference for the application of advanced intelligent algorithms in predicting vehicle body NVH performance and collision performance, and can be used by engineers in practical engineering projects.

[0130] Excellent vehicle body structure design has always been a major topic in passive safety research. A reliable vehicle body structure design can improve the lower limit of vehicle collision safety, laying a solid foundation for the subsequent deployment of safety systems. Frontal collisions are the most common type of traffic accident. Among them, frontal collisions with an overlap rate of less than 25% have a high incidence rate. Because their energy absorption and force transmission methods differ from other types of collisions, they are more likely to cause death or injury to occupants. Therefore, exploring a low-cost, lightweight optimization scheme that can reduce occupant injuries is of significant practical importance. To comprehensively consider various vehicle body performance aspects and improve vehicle collision safety during the body design stage, this embodiment takes a microcar as an example. Considering the NVH performance of the body, when optimizing collision safety, the thickness of the body panels determined by collision analysis and sensitivity analysis is used as the input variable. An Elman neural network is used to predict the first-order mode of the body, body mass, and driver head injury value. Using the trained Elman neural network, with the first-order mode of the body and body mass as constraints, and minimizing the driver head injury value as the optimization objective, a genetic algorithm is used to optimize the solution space of the selected input variables. The optimal solution obtained, while satisfying the performance constraints, reduces the head injury value of the driver by approximately 173.43 compared to the original solution, and improves the NVH performance of the vehicle body. This provides a valuable method for conducting multidisciplinary optimization design research on vehicle body structures.

[0131] Example 2

[0132] This embodiment provides a vehicle body structure optimization system that takes into account vehicle body performance and collision damage.

[0133] A vehicle body structure optimization system that considers vehicle body performance and collision damage, such as Figure 7 As shown, it includes:

[0134] The initialization module is configured to generate several sets of solutions in the solution space, and for each set of solutions, use a neural network to predict the driver injury value, vehicle mass and first-order mode of the white body, and select a set of solutions as the optimal solution based on the driver injury value; wherein each set of solutions consists of the thickness of each plate.

[0135] The optimization module is configured as follows: for each set of solutions, after performing a genetic operation, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body; if the predicted vehicle mass does not increase and the first-order mode of the white body does not decrease compared to before the genetic operation, then the set of solutions is retained; based on the driver injury value, the retained solution is used to update the optimal solution; it is determined whether the iteration terminates. If not, the genetic operation is performed again; if so, the optimal solution is decoded into the thickness of each plate.

[0136] It should be noted that the above modules implement the same examples and application scenarios as the steps in Embodiment 1, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of a system, can be executed in a computer system such as a set of computer-executable instructions.

[0137] Example 3

[0138] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the deep neural network training computational performance prediction method described in Embodiment 1 above.

[0139] Example 4

[0140] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the deep neural network training computational performance prediction method described in Embodiment 1 above.

[0141] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing a vehicle body structure considering vehicle body performance and collision damage, characterized in that, include: Several sets of solutions are generated in the solution space. For each set of solutions, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body. Based on the driver injury value, a set of solutions is selected as the optimal solution. Each set of solutions consists of the thickness of each plate. The first-order modes of the body-in-white include: the first-order torsional mode and the first-order bending mode of the body-in-white; the neural network adopts a recurrent neural network with local memory units and local feedback connections; For each set of solutions, after performing the genetic operation, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body. If, compared to before the genetic operation, the predicted vehicle mass of a certain set of solutions does not increase and the first-order mode of the white body does not decrease, then that set of solutions is retained. Based on the driver injury value, the retained solution is used to update the optimal solution. It is then determined whether the iteration terminates. If not, the genetic operation is performed again. If so, the optimal solution is decoded into the thickness of each plate. Among these constraints, the objective is to minimize the driver's head injury value, while maintaining the vehicle's mass and preventing a decrease in the first-order mode of the body-in-white. The optimal solution is then sought within the feasible solution set, and the mathematical expression is as follows: In the formula, This represents the driver's head injury value. For the first-order torsional mode of the white body, For the first-order bending mode of the white body, For vehicle body weight; In genetic algorithms, the fitness function is defined as the difference between the driver's head injury value corresponding to the initial solution and the injury value predicted by each solution.

2. The vehicle body structure optimization method considering vehicle body performance and collision damage according to claim 1, characterized in that, The driver injury value is an indicator of head injury to the driver in an offset collision.

3. The vehicle body structure optimization method considering vehicle body performance and collision damage according to claim 1, characterized in that, The panels include: front longitudinal beam outer panel, front longitudinal beam inner panel, upper longitudinal beam, longitudinal beam connecting beam outer panel, longitudinal beam connecting beam inner panel, front bulkhead, side panel, front pillar upper inner panel, front pillar lower inner panel, roof, middle pillar inner panel and / or rear tail panel.

4. The vehicle body structure optimization method considering vehicle body performance and collision damage according to claim 1, characterized in that, Based on the noise, vibration, and acoustic roughness of the vehicle body, through whole-vehicle collision simulation analysis and panel acoustic sensitivity analysis, it is determined that each solution is composed of the thickness of each panel.

5. The vehicle body structure optimization method considering vehicle body performance and collision damage according to claim 1, characterized in that, The genetic operations include crossover and mutation.

6. A vehicle body structure optimization system considering vehicle body performance and collision damage, characterized in that, include: The initialization module is configured to generate several sets of solutions in the solution space, and for each set of solutions, use a neural network to predict the driver injury value, vehicle mass and first-order mode of the white body, and select a set of solutions as the optimal solution based on the driver injury value; wherein each set of solutions consists of the thickness of each plate. The first-order modes of the body-in-white include: the first-order torsional mode and the first-order bending mode of the body-in-white; the neural network adopts a recurrent neural network with local memory units and local feedback connections; The optimization module is configured as follows: for each set of solutions, after performing a genetic operation, a neural network is used to predict the driver injury value, vehicle mass, and first-order mode of the white body; if, compared with before the genetic operation, the predicted vehicle mass of a certain set of solutions does not increase and the first-order mode of the white body does not decrease, then the set of solutions is retained; based on the driver injury value, the retained solution is used to update the optimal solution; it is determined whether the iteration terminates; if not, the genetic operation is performed again; if so, the optimal solution is decoded into the thickness of each plate. Among these constraints, the objective is to minimize the driver's head injury value, while maintaining the vehicle's mass and preventing a decrease in the first-order mode of the body-in-white. The optimal solution is then sought within the feasible solution set, and the mathematical expression is as follows: In the formula, This represents the driver's head injury value. For the first-order torsional mode of the white body, For the first-order bending mode of the white body, For vehicle body weight; In genetic algorithms, the fitness function is defined as the difference between the driver's head injury value corresponding to the initial solution and the injury value predicted by each solution.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the vehicle structure optimization method that takes into account vehicle performance and collision damage as described in any one of claims 1-5.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the vehicle structure optimization method considering vehicle performance and collision damage as described in any one of claims 1-5.

Images