Transient impact structure optimization method and device and storage medium

Abstract

The invention provides an optimization method and device of a transient impact structure and a storage medium, and belongs to the field of mechanical design, and the method comprises the steps: carrying out the modeling according to a preset overall structure cell external size of the transient impact structure and a preset curve, and obtaining a cell model of the transient impact structure; performing analogue simulation on the cell element model, and training a preset initial curve model according to the obtained feature data set to obtain a target curve model; and inputting the randomly generated cell structure into the target curve model, carrying out iterative evolution on the cell structure according to the obtained simulation response curve and a preset optimization target, and optimizing the transient impact structure according to the strain response curve corresponding to the cell structure. According to the scheme provided by the invention, through local control and smooth characteristics of the B spline curve, wide structural design can be promoted through minimum control parameters, and the universality and generalization capability of designing the transient impact structure are ensured.

Description

Technical Field

[0001] This application belongs to the field of mechanical design technology and relates to an optimization method, device and storage medium for transient impact structures. Background Technology

[0002] In many engineering scenarios such as aerospace and transportation, especially in the case of dynamic impact loads on structures and equipment, material failure, equipment malfunction, and personal injury can occur. Therefore, protective designs for structures under transient dynamic impact loads are necessary.

[0003] In the existing technology, the design for transient load protection is mostly based on simple configurations, and configuration optimization is carried out by combining simulation and experiment. However, the following problems are still faced: (1) In the traditional structural design method, the structural shape parameters should not be large. Therefore, the designed structure itself lacks design freedom and is mostly a simple basic configuration, such as circle, square, hexagon, etc. It is difficult to find a suitable target object, the design domain is limited, and there is a lack of methods to describe complex structures with fewer parameters; (2) Considering transient impact loads, the nonlinear factors of the material itself make the traditional topology optimization method based on sensitivity solution difficult to calculate. Therefore, an effective structural optimization method is urgently needed for impact protection problems; (3) Due to the increase in the design parameter dimensions and the number of finite element solution calculations, the overall optimization time will increase exponentially, and there is a lack of solution methods for complex configurations under transient impact. Summary of the Invention

[0004] This application provides an optimization method, device, and storage medium for transient impact structures, which solves the problems of lack of design freedom, nonlinear factors of the material itself, and excessive optimization time in the prior art when designing transient impact structures.

[0005] To achieve the above objectives, this application adopts the following technical solution: In a first aspect, embodiments of this application provide an optimization method for a transient impact structure, the method comprising: Based on the pre-set overall structural cell outer dimensions of the transient impact structure, and combined with the pre-set curve, a cell model of the transient impact structure is obtained. The cell model was simulated to obtain a feature dataset; Based on the feature dataset, the pre-set initial curve model is trained to obtain the target curve model; The randomly generated cell structure is input into the target curve model to obtain the simulated response curve of the cell structure; Based on the simulated response curve and the pre-set optimization objective, the cell structure is iteratively evolved until the updated cell structure satisfies the iteration conditions or the optimization objective. The transient impact structure is optimized based on the strain response curve corresponding to the cell structure.

[0006] Optionally, the step of modeling the transient impact structure by combining the pre-set overall structural cell outer dimensions of the transient impact structure with a pre-set curve to obtain the cell model of the transient impact structure includes: Obtain the overall structural cell outer dimensions of the transient impact structure; Based on the overall structural cell outer dimensions, and combined with the pre-set spline curves, the boundary curves of the cell structure are generated. Based on the boundary curve and the pre-set cell height, the cell model is obtained by modeling.

[0007] Optionally, the simulation of the cell model to obtain the feature dataset includes: Based on the orthogonality and space-filling property of the cell description parameters of the transient impact structure, the topological description feature parameters of the transient impact structure are obtained; The cell model is simulated according to the pre-set load curve and the contact mode corresponding to the boundary curve to obtain the transient data of the cell model; Based on the transient data, the time-domain strain information of the transient impact structure is obtained; The feature dataset is obtained by combining the topological description feature parameters and the time-domain strain information.

[0008] Optionally, training a pre-set initial curve model based on the feature dataset to obtain a target curve model includes: The initial curve model is constructed based on a pre-defined input layer, output layer, and at least one hidden layer; The training set from the dataset is input into the initial curve model to obtain a temporary response curve; Based on the temporary response curve and the loss function, the initial curve model is iteratively trained to obtain the target curve model.

[0009] Optionally, the step of inputting the randomly generated cell structure into the target curve model to obtain the simulated response curve of the cell structure includes: Based on the cells corresponding to the feature dataset, multiple initial cell structures are randomly generated; An initial cell population is formed based on multiple initial cell structures; Each initial cell structure in the initial cell population is input into the target curve model to obtain the simulated response curve corresponding to each initial cell structure.

[0010] Optionally, the step of iteratively evolving the cell structure based on the simulated response curve and a pre-set optimization objective until the updated cell structure satisfies the iteration conditions or the optimization objective includes: Define the structural optimization objective; Based on the structure optimization objective, and combined with the simulated response curves corresponding to each initial cell structure in the initial cell population, multiple candidate cell structures are determined. Based on the multiple candidate cell structures, multiple random cell structures are randomly generated, and a temporary cell population is obtained through evolution. The random cell structures in the temporary cell population are input into the target curve model again to obtain the simulated response curve corresponding to each random cell structure, until the simulated response curve satisfies the optimization objective, or the number of iterations satisfies the iteration condition.

[0011] Optionally, determining the structural optimization objective includes: Based on the cell structure, calculate the dynamic mechanical parameters of the cell structure; Based on the dynamic mechanical parameters and the transmission peak load of the cell structure, the structural optimization target is obtained.

[0012] Optionally, optimizing the transient impact structure based on the strain response curve corresponding to the cell structure includes: According to the structural optimization objective, the strain response curve is selected from the simulated response curves corresponding to the multiple random cell structures; Based on the strain response curve, determine the corresponding target cell structure; The transient impact structure is optimized based on the parameters corresponding to the target cell structure.

[0013] Secondly, embodiments of this application provide an electronic device, including a processor, a memory, and a communication bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory through the communication bus, and the processor executes the machine-readable instructions to implement the method described in any one of the first aspects.

[0014] Thirdly, embodiments of this application provide a storage medium storing a computer program, which, when executed by a processor, performs the method described in any one of the first aspects.

[0015] This application provides an optimization method for transient impact structures. Considering the generalization of structural parametric design, unlike traditional simple structural design methods, this application uses MMV (Mean Multivariable Model), or B-spline curve, for structural parametric description. B-spline curves are a type of Bézier curve, constructed through polynomial expansion and defined by control points and node vectors. With its unique local control and smoothing properties, B-spline curves allow shape manipulation through node selection, thereby facilitating a wide range of structural designs with minimal control parameters. This inherent characteristic ensures the versatility and generalization ability of the surrogate model. Furthermore, to reduce the impact of the number of parameters on the neural network, near-orthogonal Latin hypercube sampling is used for feature configuration extraction, improving space-filling performance while ensuring parameter orthogonality. By introducing NOLHS (Normally Orthogonal Models), feature parameters are effectively extracted to provide feature data vectors under multiple design parameters, reducing the high computational cost caused by parameter selection and successfully achieving feature configuration sample extraction for complex topological configurations. Attached Figure Description

[0016] Figure 1A A schematic flowchart illustrating an optimization method for a transient impact structure provided in this application embodiment; Figure 1B A schematic flowchart illustrating another method for optimizing a transient impact structure provided in an embodiment of this application; Figure 2A A schematic diagram of a B-spline configuration provided in an embodiment of this application; Figure 2B A schematic diagram of another B-spline configuration provided in an embodiment of this application; Figure 2C A Boolean operation graph for multiple B-spline curves is provided in the embodiments of this application; Figure 3A A schematic diagram of a B-spline configuration provided in an embodiment of this application; Figure 3B A schematic diagram of another B-spline configuration provided in this application embodiment; Figure 4A A simulation and experimental comparison diagram is provided for an embodiment of this application; Figure 4B Another simulation and experimental comparison diagram provided for an embodiment of this application; Figure 5 A schematic diagram of an experimental apparatus provided in an embodiment of this application; Figure 6 A diagram of an MLP network architecture provided in this application embodiment; Figure 7A is a performance prediction diagram of a novel structure provided in an embodiment of this application; Figure 7B is a performance prediction diagram of a novel structure provided in an embodiment of this application; Figure 7C is a performance prediction diagram of a novel structure provided in an embodiment of this application; Figure 7D is a performance prediction diagram of a novel structure provided in an embodiment of this application; Figure 8A is an optimization iteration diagram provided in an embodiment of this application; Figure 8B is another optimization iteration diagram provided by an embodiment of this application; Figure 8C is another optimization iteration diagram provided by an embodiment of this application; Figure 8D is another optimization iteration diagram provided by the embodiments of this application. Detailed Implementation

[0017] In the following description, specific details such as particular system architectures and technologies are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known structural design techniques, simulation algorithms, and electronic devices are omitted so as not to obscure the description of this application with unnecessary detail.

[0018] The terminology used in the following embodiments is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “the,” “the,” and “the” are intended to also include expressions such as “one or more,” unless the context clearly indicates otherwise.

[0019] In numerous engineering scenarios, including aerospace, transportation, and individual soldier protection, especially those involving dynamic impact loads on structures and equipment, material failure, equipment malfunction, and personnel injury can occur. Therefore, structural protection under transient dynamic impact loads is an urgent problem that the engineering community needs to solve.

[0020] Transient impact problems are characterized by large contact forces, instantaneous loading, and wide load bandwidths. They also involve nonlinear factors such as material plasticity and strain rate effects, making calculations difficult to converge and time-consuming. Furthermore, considering the multiple iterative processes involved in transient impact problems, structural optimization becomes even more challenging.

[0021] Currently, for transient load protection design, simple configurations are mostly used, and configuration optimization is carried out by combining simulation and experiment. However, the following problems are still faced: (1) In traditional structural design methods, the structural shape parameters should not be large. Therefore, the designed structure itself lacks design freedom and is mostly a simple basic configuration, such as circle, square, hexagon, etc. It is difficult to find a suitable target object, the design domain is limited, and there is a lack of methods to describe complex structures using fewer parameters; (2) Considering transient impact loads, the nonlinear factors of the material itself make the traditional topology optimization method based on sensitivity solution difficult to calculate. Therefore, an effective structural optimization method is urgently needed for impact protection problems; (3) Due to the increase in the design parameter dimensions and the number of finite element solution calculations, the overall optimization time will increase exponentially, and there is a lack of solution methods for complex configurations under transient impact.

[0022] Regarding problem (1), in recent years, relevant scholars have carried out some work to improve the positive correlation between configuration complexity and the number of design parameters. In order to achieve a simplified representation of complex configurations, people have proposed using levelset functions (LSF), moving morphable components (MMC), and moving morphable voids (MMV). The MMV topology description method, namely B-spline curve description, is used for structural design. By selecting control points and basis functions, the corresponding structural contour is obtained, avoiding the problem that traditional structural design can only design simple configurations and that there are many design parameters for complex configurations. This method can enhance the inherent generalization of the structural model and enhance its wide applicability in structural optimization. However, at present, relevant scholars mostly apply it to static topology optimization problems, only considering the static load of the structure and not the transient response, and lack efficient calculation methods. Applying B-splines to transient complex configuration optimization problems, combined with the framework proposed in this invention, can use fewer parameters to describe complex structures, while also possessing computational efficiency. This design method demonstrates its potential in materials science and engineering applications, providing new ideas and feasibility for structural design and optimization under transient impact.

[0023] For problem (2), previous researchers have mostly dealt with static linear elasticity problems, updating the design parameters by calculating the sensitivity of the design parameters to the objective function until the iteration converges. This method is suitable for simple problems. When the object under consideration contains strong nonlinear factors such as shocks, the sensitivity calculation becomes difficult or even impossible to solve, requiring non-gradient optimization methods to iterate the configuration parameters. In recent years, non-gradient algorithms have flourished, including genetic algorithms (GA), particle swarm optimization (PSO), ant colony algorithms (ACA), and simulated annealing (SA). In order to consider multiple objectives in the optimization process, the non-dominated sorting genetic algorithm II (NSGA-II) is used as the optimization algorithm here. NSGA-II is an improved multi-objective genetic algorithm. By introducing non-dominated sorting and crowding calculation, it calculates each Pareto order while preserving the genetic diversity of individuals in each generation. Through natural selection, the superior individuals are selected for iteration. The optimization algorithm based on NSGA-II not only solves the problem of nonlinear solution difficulties caused by traditional sensitivity calculation, but also provides technical support for multi-objective optimization under impact load.

[0024] For problem (3), although fewer parameters can be used to describe complex configurations in problem (1), the number of parameters is still higher than that of traditional optimization parameters. Considering the number of individuals and iterations in the optimization iteration, the overall instantaneous explicit solution time will increase exponentially, which is impossible to achieve in actual engineering applications. In recent years, with the continuous development of machine learning, it has been gradually discovered and applied in interdisciplinary fields and has achieved many results, making it an effective research method. Multilayer perceptron (MLP) is a common feedforward neural network model, mainly composed of three parts: input layer, hidden layer and output layer. It is composed of multiple neurons, and through the connection between layers, weight and bias settings, a nonlinear mapping relationship between input and output is established. Based on MLP, by building a surrogate model, the computation time can be greatly reduced while ensuring accuracy. It is used to replace finite element calculation and reduce the high time cost brought by computation. However, at present, scholars mostly build surrogate models for simple configurations through machine learning, and lack efficient solution methods for complex configurations. This invention introduces near-orthogonal Latin hypercube sampling to extract feature parameter vectors for complex configurations, and builds an efficient surrogate model for complex configurations, providing an important reference for the optimization of complex topology structures under impact loads.

[0025] Figure 1A A schematic flowchart illustrating an optimization method for a transient impact structure provided in this application embodiment. Figure 1B A schematic flowchart illustrating another method for optimizing a transient impact structure provided in this application embodiment is shown as an example and not a limitation. See also Figure 1A and Figure 1B The method includes: Step 101: Based on the pre-set overall structural cell outer dimensions of the transient impact structure, and combined with the pre-set curve, model the transient impact structure to obtain the cell model of the transient impact structure.

[0026] Among them, the overall structural cell outer dimension is used to represent the uniform size of each cell structure in the transient impact structure.

[0027] In the process of optimizing a transient impact structure, we can first model each cell structure of the transient impact structure, so that we can optimize the transient impact structure according to the state of different cell structures when they are impacted, based on the cell structure with the best state.

[0028] Optionally, the overall structural cell outer dimensions of the transient impact structure can be obtained first, and the boundary curve of the cell structure can be generated based on the overall structural cell outer dimensions and the pre-set spline curve. Then, the cell model can be obtained by modeling based on the boundary curve and the pre-set cell height.

[0029] Specifically, based on the parameters set by the user, the overall structural cell outer dimensions of the transient impact structure can be obtained, and the cell structure contour can be extracted. Then, combined with the overall structural cell outer dimensions, the boundary curve of the cell structure can be calculated based on the pre-set spline curve (such as a B-spline curve).

[0030] Then, based on the pre-set cell height, the boundary curve can be stretched, thereby establishing the cell model of the transient impact structure based on the coordinates of the boundary curve and the cell height.

[0031] For example, the outer dimensions of the overall structural cell can be ,in, For long, For width, It is high. Furthermore, the overall external load on the transient impact structure can be set as a transient impact, with the load form being... The impact waveform is a rectangular wave.

[0032] Correspondingly, such as Figure 2A and Figure 2B As shown, see Figure 2ABased on Bézier curves, B-spline curves can be used to describe the topological structure of cell structures. Based on the second-order basis functions of B-spline curves, the boundary curves of cell structures can be calculated by selecting control points.

[0033] in, , This represents the coordinates of the i-th control point. This represents the x-coordinate of the i-th control point. Let represent the ordinate of the i-th control point, which can be in angular coordinate form, i.e.:

[0034]

[0035]

[0036] in, and For the center point coordinates of each B-spline curve, Where n is the radial length and n is the number of basic control points. It is the angle of each control point relative to the x-coordinate, for example, such as Figure 2B As shown, Only 0 degrees, 45 degrees, 90 degrees, 135 degrees, 180 degrees, 225 degrees, 270 degrees, and 315 degrees can be selected. Furthermore, as... Figure 2C As shown, considering the case of multiple B-spline curves, Boolean operations can be used for processing.

[0037] Moreover, it can limit Within the outer dimensions of the cell structure, i.e. Furthermore, considering the smooth connection performance of the curves:

[0038] in, For derived control points, These are the 1st and nth control points.

[0039] Furthermore, the i-th basis function of the B-spline curve can be calculated. And obtain the coordinates of the curve points: when hour,

[0040] when hour,

[0041] in, Let be a node vector, which is a monotonically non-decreasing sequence of real numbers; and These are the th elements in a monotonically non-decreasing sequence. Point, number Point and the The sequence value of the point.

[0042] and, for and The intermediate value of W is the W-th B-spline basis function.

[0043] Correspondingly, the coordinates of the corresponding points on the curve It can be obtained from equation (7), and the calculation formula is as follows:

[0044] Step 102: Simulate the cell model to obtain the feature dataset.

[0045] After modeling the cell structure and obtaining the cell model, the cell model can be simulated to obtain the characteristic data of the cell structure. In subsequent steps, the cell structure can be simulated based on the characteristic data.

[0046] Optionally, the topological description parameters of the transient impact structure can be obtained first based on the orthogonality and space-filling properties of the cell description parameters. Then, the cell model can be simulated according to the contact mode corresponding to the pre-set load curve and boundary curve to obtain the transient data of the cell model. Based on the transient data, the time-domain strain information of the transient impact structure can be obtained. Thus, a feature dataset can be obtained by combining the topological description feature parameters and the time-domain strain information.

[0047] Among them, the cell description parameters of the cell structure can be R={ , Where R represents the topology description parameter, and For the center point coordinates of each B-spline curve, This is the radial length.

[0048] Specifically, the orthogonality of the cell description parameters can be calculated first, and then the space-filling property of the cell description parameters can be calculated. Thus, the orthogonality and space-filling property can be used as the objective function for near-orthogonal Latin hypercube sampling, and the structural cell topological description parameters of the transient impact structure can be calculated.

[0049] Subsequently, based on the topological description characteristic parameters and combined with the pre-set load form and contact method, a load corresponding to the load form can be applied to the transient impact structure in a rigid contact manner, thereby simulating the cell model and obtaining the transient data of each cell model in the transient impact structure.

[0050] Correspondingly, the transient data corresponding to the reference point of a part of the transient impact structure can be extracted to obtain the time-domain strain information of the reference point for incident, reflection and transmission. Then, the time-domain strain information is normalized to obtain the time-domain strain information of the transient impact structure. Finally, the topological description feature parameters and time-domain strain information of the transient impact structure are combined to form a feature dataset.

[0051] For example, the orthogonality of the cell topological description parameters can be calculated using equation (8), and the spatial filling property of the cell topological description parameters can be calculated using equation (9). Orthogonality and spatial filling property are used as objective functions for near-orthogonal Latin hypercube sampling, and the desired topological description feature parameters are returned. .

[0052] Calculate orthogonality:

[0053] The condition number can be obtained by singular value decomposition (SVD), where X is the orthogonal design parameter. and They are The maximum and minimum SV; Formula (8) is used to calculate orthogonality; the condition number of the near-orthogonal form should be greater than 1.

[0054] Equation (9) is shown below and is used to calculate space filling:

[0055] Where q is the number of variables, and num is the number of runs. Design parameters for filling space.

[0056] Correspondingly, topological description feature parameters It can be represented as:

[0057] in, Here, D is a near-orthogonal Latin hypercube sampling function, and D is the number of cell topological description parameters. and For near-orthogonal Latin hypercube sampling, there are two objective functions: orthogonality and space-filling property.

[0058] Furthermore, if data R1 is generated using feature configuration-derived data, then the structural cell topology description parameter R can be expressed as:

[0059]

[0060] Where Rand is a random number function, It is a small amount.

[0061] For example, the parameter selection range is shown in Table 1 below, and the initial configuration is as follows: Figure 3A As shown, the corresponding feature configurations are as follows: Figure 3B Demonstration. To obtain a sufficient number of samples, data was generated in the form of feature configuration-derived data. Based on the parameters of each feature configuration, noise was added to increase the amount of data, and 40 new data points were derived for each feature parameter.

[0062] Table 1

[0063] Correspondingly, in the process of establishing a cell model to simulate transient impact structures, the given load curve can be used as a basis. Given a rigid contact boundary condition, a load is applied to the end of the incident rod of the transient impact structure to simulate the actual experimental load. The transient data is then obtained by calculating and solving the problem using the explicit dynamics module of third-party software. Based on this transient data, a dataset can be set up at a reference point in the middle of the incident and transmission rods to extract the incident, reflection, and transmission time-domain strain information at the reference point, followed by data normalization.

[0064] Specifically as follows:

[0065] in, It is the z-th value in the time-domain parameters. and These are the maximum and minimum values ​​of the parameters, respectively. The range is [0, 1].

[0066] Accordingly, based on the structural cell topology description parameter R and the normalized temporal strain information... We can create a feature dataset Z={R,yz}.

[0067] For example, regarding parameter settings in a co-simulation program, typical calculation cases include: Figure 5 As shown, the rod has a diameter of 50mm, is made of stainless steel, and has a density of 7.85×10⁻⁶. 3 kilograms per cubic meter (kg / m³) 3 The elastic modulus is 210 gigapascals (GPa), and the Poisson's ratio is 0.3. C3D8R was selected as the element type in the simulation. The sample material is Al6061 with a density of 2.7 × 10⁻⁶. 3kg / m 3 The elastic modulus is 69 GPa, and the Poisson's ratio is 0.33. In the simulation, C3D6 was selected as the element type, and the fragment shape is a rectangular block of 30*30*20mm. The Johnson-Cook constitutive model is used, as shown below: (11) In the formula: A = 278.93 MPa is the dynamic yield stress; B = 294.89 MPa is the static yield stress of the material at the reference strain rate and reference temperature; and B = 294.89 MPa is the strain hardening parameter of the material. =E is the equivalent plastic strain; n=0.58 is the strain hardening exponent of the material; C=0.069 is the strain rate sensitivity coefficient; m=0.7 is the temperature softening exponent. For strain rate (dimensionless). =0.001 is the reference strain rate. For dimensionless temperature parameters, =293k is room temperature =873K is the melting point of the material.

[0068] Regarding the construction of experimental platforms, such as Figure 5 As shown, the air gun pressure was set to 0.3 MPa, and the initial projectile velocity, as tested, ranged from 23.207 to 32.883 m / s. In the simulation, 25.51 m / s was taken as the projectile's incident velocity. The peak pressure load was approximately 200 MPa. The dynamic testing system used was the LK2109A dynamic testing system, with BE120-3AA strain gauges and a sensitivity of 2.22%. The bridge voltage is 2 volts (V), the gain is 1000, and the sampling frequency is 10 MHz. For example... Figure 4A and Figure 4B As shown, it is a comparison of simulation and experiment for two configurations, configuration one and configuration two. Through simulation calculation, the stress and strain response of the configurations were obtained, and the experimental results were compared. The error was within 5%.

[0069] It should be noted that in practical applications, steps 101 and 102 above provide sufficient feature data for step 103. By introducing nearly orthogonal Latin hypercube sampling (NOLHS), feature parameters are effectively extracted to provide feature data vectors under multiple design parameters, reducing the high computational cost caused by parameter selection. This successfully achieves feature configuration sample extraction for complex topological configurations, i.e. B-spline configurations, reducing the possibility of local overfitting in step 103 and providing necessary data support for subsequent optimization.

[0070] Step 103: Train the pre-set initial curve model based on the feature dataset to obtain the target curve model.

[0071] After obtaining the feature dataset, the initial curve model can be trained based on the data included in the feature dataset to obtain the target curve model. In subsequent steps, the transient impact structure can be optimized using the target curve model.

[0072] Optionally, an initial curve model can be constructed based on a pre-set input layer, output layer, and at least one hidden layer. The training set in the dataset can then be input into the initial curve model to obtain a temporary response curve. Based on the temporary response curve and the loss function, the initial curve model can be iteratively trained to obtain the target curve model.

[0073] Specifically, the feature dataset can be first divided into a training set, a test set, and a validation set. The training and test sets are used to train the initial curve model; for example, the training set can be used for each iteration, and the test set can be used for each performance test. The validation set consists of randomly selected data samples used to verify the accuracy of the training on real samples after completion.

[0074] Correspondingly, an overall network framework can be built first based on the input layer, hidden layer, and output layer. The mapping relationship from input to output can be established through the inter-layer connections of multiple neurons, and the weights and bias matrices can be iterated to establish a nonlinear mapping between the input layer and the output layer, thereby obtaining the target curve model.

[0075] During training, the structural cell topology description parameter R from the feature dataset can be input into the input layer of the initial curve model. The structural cell topology description parameter R is then processed and transformed through the hidden layer to obtain the temporary response curve. Finally, the temporary response curve in the time domain is output through the output layer.

[0076] Subsequently, the temporal strain information corresponding to the topological description parameter R of the structural cell can be obtained from the temporary response curve and the feature dataset. The pre-set loss function adjusts the weights and bias matrices of the hidden layers in the initial curve model, and the output temporary response curve and time-domain strain information are then compared. If the difference between them meets the pre-set difference threshold, or if the number of training iterations of the initial curve model reaches the pre-set threshold, training can be stopped to obtain the target curve model, thereby enabling the explicit solution of the target curve model for complex topological configurations under transient impact.

[0077] Among these methods, mean squared error (MSE) can be selected as the loss function, and the neural network can be optimized using a stochastic gradient descent (SGD) optimizer to reduce MSE, thereby generating the optimal target curve model.

[0078] For example, the MSE formula is shown below:

[0079] Where nt is the number of samples. For predicted values, This is the actual value.

[0080] Furthermore, the target curve model can consist of a weight matrix, a bias matrix, and a nonlinear activation function. For a feedforward neural network with hidden layers, the parameter mapping relationship is set as follows:

[0081]

[0082]

[0083] in, and These are the input and output parameters, respectively. It is the hidden layer, and K is the number of neurons in the hidden layer. It is the first The number of neurons in each hidden layer It is the activation function of the s-th layer. and These are the hidden layer, weights, and bias matrix of the s-th layer, respectively. , It is the activation function of the input layer. matrix, The paranoia matrix, where r is the number of hidden layers. The activation function of the output layer. These are the weight matrix and bias matrix of the output layer, respectively. Let r be the hidden layer of the r-th layer.

[0084] For example, see Figure 6 The neural network was configured with two hidden layers, trained 5000 times, and the batch size was set to 256. After normalizing the target parameters, all values ​​were adjusted to be greater than 0 by increasing the weights. ReLU, MSE, and SGD were chosen as the activation function, loss function, and optimization function, respectively. The validation rate was 0.2, the learning rate was 0.5, and an optimized implementation of a third-party software library was used to train the MLP architecture. Figure 7A and Figure 7B As shown, for both random mechanism types, the surrogate model's predicted values ​​are almost equal to the simulated values, verifying the accuracy of the surrogate model.

[0085] Step 104: Input the randomly generated cell structure into the target curve model to obtain the simulated response curve of the cell structure.

[0086] After obtaining the target curve model, multiple cell structures can be randomly generated to form a cell population. The cell population is then input into the target curve model to obtain the simulated response curves of each cell structure. In subsequent steps, the cell population can be evolved based on the simulated response curves, and the transient impact structure can be optimized based on the evolved cell population.

[0087] In the process of randomly generating multiple cell structures, random noise can be introduced based on the feature dataset. This allows for the random generation of multiple cell topology description parameters by combining the topology description feature parameters and time-domain strain information in the feature dataset with the random noise. In this way, multiple cell structures can be constructed through Boolean operations.

[0088] Optionally, multiple initial cell structures can be randomly generated based on the cells corresponding to the feature dataset, and an initial cell population can be formed based on the multiple initial cell structures. Then, each initial cell structure in the initial cell population can be input into the target curve model to obtain the simulated response curve corresponding to each initial cell structure.

[0089] Specifically, for each initial cell structure in the initial cell population, the structural cell topology description parameters corresponding to the initial cell structure can be input into the target curve model. The target curve model performs feature extraction and analysis on the structural cell topology description parameters and outputs the simulated response curve corresponding to the initial cell structure.

[0090] Step 105: Based on the simulated response curve and the pre-set optimization objective, iteratively evolve the cell structure until the updated cell structure satisfies the iteration conditions or optimization objective.

[0091] Corresponding to step 104, after obtaining the simulated response curves corresponding to each initial cell structure, the initial cell structure can be selected from multiple simulated response curves by calculating the crowding degree and sorting the non-dominated structures according to the determined structural optimization objective, and then iteratively evolved based on the selected initial cell structure.

[0092] Here, crowding degree is used to represent the arrangement density of the initial cell structure. The higher the crowding degree, the sparser and more diverse the distribution of the solution region. Within the same front layer, in order to avoid excessive concentration of solutions, they are sorted in ascending order according to the objective function values, and the perimeter of the enclosing rectangle (i.e., the crowding distance) between adjacent points of each solution is calculated, where the crowding degree of the boundary solutions is set to infinity.

[0093] Furthermore, non-dominated sorting is used to classify all solutions that are not dominated by other individuals into the first frontier layer (i.e., Pareto optimal) by comparing the objective function values ​​among individuals. After removing individuals from this layer, the next frontier layer is selected, and so on, to stratify the population and ensure that the algorithm converges to the Pareto frontier.

[0094] For example, when making selections, individuals with higher frontier levels can be prioritized; while within the same level, individuals with greater crowding can be prioritized to balance convergence and diversity.

[0095] Correspondingly, in the iterative evolution process, parameter ranges, boundary constraints, and optimization objectives can be defined first. For the cell topology description parameters of the input initial cell structure, before each iteration, the target value is calculated to give each design parameter a corresponding fitness value. The topology description parameters with better fitness values ​​are retained, and vice versa. After reaching the maximum number of iterations, the result that satisfies the constraints is obtained.

[0096] Optionally, the structural optimization objective can be determined first. Then, based on the structural optimization objective and the simulated response curves corresponding to each initial cell structure in the initial cell population, multiple candidate cell structures can be determined. Based on the multiple candidate cell structures, multiple random cell structures can be randomly generated to evolve into a temporary cell population. The multiple random cell structures in the temporary cell population can then be input into the target curve model to obtain the simulated response curve corresponding to each random cell structure. This process continues until the simulated response curves meet the optimization objective or the number of iterations meets the iteration conditions.

[0097] Moreover, in determining the structural optimization target, the dynamic mechanical parameters of the cell structure can be calculated first based on the cell structure, and then the structural optimization target can be obtained based on the dynamic mechanical parameters and the transmission peak load of the cell structure.

[0098] For example, the dynamic mechanical parameters of the cell structure can be calculated using equations (18-19) based on the data of the incident rod and the transmission rod of the transient impact structure.

[0099]

[0100]

[0101] in, , and These represent strain, stress, and the elastic wave velocity of the rod, respectively. The length of the sample and These are the cross-sectional areas of the rod and the sample, respectively. and Let Eb be the reflected wave and the transmitted wave, respectively, and Eb be the Young's modulus of the incident and transmitted rod materials. Figure 7D The strain rate-time plot is shown below. Figure 7A and Figure 7B The reflection load curves are consistent.

[0102] like Figure 7C As shown, peak transmission load and dynamic stiffness are used as structural optimization objectives, where dynamic stiffness is represented by the stress-to-strain ratio at the nr-th point. The calculation formula for the structural optimization objectives is as follows:

[0103] Step 106: Optimize the transient impact structure based on the strain response curve corresponding to the cell structure.

[0104] When the number of iterations of the initial cell population reaches a preset threshold, or when the simulated response curve corresponding to any random cell structure in the temporary cell population satisfies the structure optimization objective, the transient impact structure can be optimized based on the structural cell topology description parameters corresponding to the cell structure.

[0105] Optionally, based on the structural optimization objective, strain response curves can be selected from the simulated response curves corresponding to multiple random cell structures, and the corresponding target cell structure can be determined based on the strain response curves. Then, the transient impact structure can be optimized based on the parameters corresponding to the target cell structure.

[0106] Based on the structural optimization objective, crowding degree calculation and non-dominated sorting are performed. The topological description parameters of randomly given cell structures are iteratively optimized by combining selection, crossover and mutation operators to obtain Pareto front data, which is used as the optimization result. Finally, the structure cell diagram is obtained through visualization processing.

[0107] The specific optimizations are listed below: The variables are:

[0108] Minimize object:

[0109] Constraints:

[0110]

[0111]

[0112] in, For peak transmission load, The area of ​​the structural holes, and Let x and y be the upper and lower limits of the center coordinates of the i-th (i=1,2) B-spline curve. Let be the radial length of the j-th (j=1,2...8) control point of the i-th B-spline.

[0113] The final result is as follows Figures 8A to 8D As shown, with the increase of the number of iterations, the overall performance gradually approaches the origin of the coordinate system. Among these, Figure 8A To optimize the results for the dual objectives, Figure 8B To obtain the dual-objective optimization results under the same hole area, i.e., 0.24 of the total area is used as a constraint, the structural transmission time-domain load curve of each generation is obtained. The peak load is calculated to be 116.17 MPa, while the low peak load is 80.7 MPa. Under the constraint of consistent hole area, the peak load is reduced by 30.53%. Figure 8C To obtain the dual-objective optimization results under the same load reduction performance, the perforation area ratios were 0.34, 0.42, and 0.5, respectively, indicating that a larger perforation area does not necessarily result in higher load reduction performance. Figure 8D As a result of the dual-objective optimization with consistent overall performance, the predicted values ​​of the surrogate model are generally close to the explicit solutions of ABAQUS, which verifies the accuracy of the proposed method.

[0114] In summary, the optimization method for transient impact structures disclosed in this application, considering the generalization of structural parametric design, differs from traditional simple structural design methods by employing MMV (Mean Multivariable Vector Model), i.e., B-spline curve, for structural parametric description. B-spline curves are a type of Bézier curve, constructed through polynomial expansion and defined by control points and node vectors. With their unique local control and smoothing properties, B-spline curves allow shape manipulation through node selection, thereby facilitating a wide range of structural designs with minimal control parameters. This inherent characteristic ensures the surrogate model's versatility and generalization ability. Furthermore, to reduce the impact of the number of parameters on the neural network, near-orthogonal Latin hypercube sampling is used for feature configuration extraction, improving space-filling performance while ensuring parameter orthogonality. By introducing NOLHS (Normally Orthogonal Parameters), feature parameters are effectively extracted to provide feature data vectors under multiple design parameters, reducing the high computational cost caused by parameter selection and successfully achieving feature configuration sample extraction for complex topological configurations.

[0115] Furthermore, considering that explicit finite element method (FEM) calculations increase exponentially with the number of iterations and the time required for each calculation, this invention, to reduce the high computational cost, constructs a mapping network from structural dimensional parameters to the structural temporal response based on a multilayer perceptron (MLP). Based on a three-part network—input layer, hidden layer, and output layer—with interconnected internal neurons, the network parameters are iteratively updated using weight and bias matrices. A pre-trained surrogate model replaces the FEM, thereby reducing the high computational cost. The introduction of NOLHS ensures high feature diversity while maintaining a low data volume, giving the transient impact optimization method provided in this patent high generalization ability.

[0116] In addition, the evolutionary algorithm NSGA-II solves the strong nonlinearity problem caused by instantaneous and high loads under transient impact, avoiding the difficulty of sensitivity calculation in traditional gradient algorithms, as well as the optimization problem that can only be applied to low-parameter configurations in transient problems. The optimization method provided by this invention can solve the problem of difficult optimization of protective structure parameters under strong impact for complex multi-parameter configurations.

[0117] In addition, based on the optimization results of transient impact protection structures, specific impact protection application scenarios can be considered, such as aerospace, shipbuilding and marine, railway transportation, mining, individual soldier protection and other equipment and personnel protection equipment, to provide high-lightweight load-bearing impact-resistant structural dimensional parameters, providing a reference for the design and improvement of protection structures.

[0118] If the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0119] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0120] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0121] In the embodiments provided in this application, it should be understood that the disclosed apparatus / devices and methods can be implemented in other ways. For example, the apparatus / device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0122] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0123] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0124] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0125] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0126] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An optimization method for a transient impact structure, characterized in that, The method includes: Based on the pre-set overall structural cell outer dimensions of the transient impact structure, and combined with the pre-set curve, a cell model of the transient impact structure is obtained. The cell model was simulated to obtain a feature dataset; Based on the feature dataset, the pre-set initial curve model is trained to obtain the target curve model; The randomly generated cell structure is input into the target curve model to obtain the simulated response curve of the cell structure; Based on the simulated response curve and the pre-set optimization objective, the cell structure is iteratively evolved until the updated cell structure satisfies the iteration conditions or the optimization objective. The transient impact structure is optimized based on the strain response curve corresponding to the cell structure.

2. The method according to claim 1, characterized in that, The process of modeling the transient impact structure by combining the pre-set overall structural cell outer dimensions with a pre-set curve to obtain the cell model of the transient impact structure includes: Obtain the overall structural cell outer dimensions of the transient impact structure; Based on the overall structural cell outer dimensions, and combined with the pre-set spline curves, the boundary curves of the cell structure are generated. Based on the boundary curve and the pre-set cell height, the cell model is obtained by modeling.

3. The method according to claim 1, characterized in that, The simulation of the cell model yields a feature dataset, including: Based on the orthogonality and space-filling property of the cell description parameters of the transient impact structure, the topological description feature parameters of the transient impact structure are obtained; The cell model is simulated according to the pre-set load curve and the contact mode corresponding to the boundary curve to obtain the transient data of the cell model; Based on the transient data, the time-domain strain information of the transient impact structure is obtained; The feature dataset is obtained by combining the topological description feature parameters and the time-domain strain information.

4. The method according to claim 1, characterized in that, The step of training a pre-set initial curve model based on the feature dataset to obtain a target curve model includes: The initial curve model is constructed based on a pre-defined input layer, output layer, and at least one hidden layer; The training set from the dataset is input into the initial curve model to obtain a temporary response curve; Based on the temporary response curve and the loss function, the initial curve model is iteratively trained to obtain the target curve model.

5. The method according to claim 1, characterized in that, The step of inputting the randomly generated cell structure into the target curve model to obtain the simulated response curve of the cell structure includes: Based on the cells corresponding to the feature dataset, multiple initial cell structures are randomly generated; An initial cell population is formed based on multiple initial cell structures; Each initial cell structure in the initial cell population is input into the target curve model to obtain the simulated response curve corresponding to each initial cell structure.

6. The method according to claim 1, characterized in that, The step of iteratively evolving the cell structure based on the simulated response curve and a pre-set optimization objective until the updated cell structure satisfies the iteration conditions or the optimization objective includes: Define the structural optimization objective; Based on the structure optimization objective, and combined with the simulated response curves corresponding to each initial cell structure in the initial cell population, multiple candidate cell structures are determined. Based on the multiple candidate cell structures, multiple random cell structures are randomly generated, and a temporary cell population is obtained through evolution. The random cell structures in the temporary cell population are input into the target curve model again to obtain the simulated response curve corresponding to each random cell structure, until the simulated response curve satisfies the optimization objective, or the number of iterations satisfies the iteration condition.

7. The method according to claim 6, characterized in that, The determination of the structural optimization objective includes: Based on the cell structure, calculate the dynamic mechanical parameters of the cell structure; Based on the dynamic mechanical parameters and the transmission peak load of the cell structure, the structural optimization target is obtained.

8. The method according to claim 7, characterized in that, The optimization of the transient impact structure based on the strain response curve corresponding to the cell structure includes: According to the structural optimization objective, the strain response curve is selected from the simulated response curves corresponding to the multiple random cell structures; Based on the strain response curve, determine the corresponding target cell structure; The transient impact structure is optimized based on the parameters corresponding to the target cell structure.

9. An electronic device, characterized in that, The device includes a processor, a memory, and a communication bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory through the communication bus, and the processor executes the machine-readable instructions to implement the method according to any one of claims 1 to 8.

10. A storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the method according to any one of claims 1 to 8.

Images