A method for rapidly calculating implosion load field in asymmetric complex multi-chamber structure

Abstract

The application discloses a kind of fast calculation methods of asymmetric complex multi-chamber structure implosion load field, specifically related to the field of explosion mechanics and structure impact response prediction, establish asymmetric complex multi-chamber structure numerical model, determine the sampleable area of explosion center position;Multiple load response measuring points are arranged in each chamber to obtain explosion impact load response parameters;With explosion center coordinates, equivalent as variable, generate explosion point sample through Latin hypercube sampling;Combined with multi-chamber model, carry out spatial constraint verification, eliminate and supplement invalid sample;Uniform measuring point completes explosion load simulation and data extraction, constructs parameter mapping dataset;Build multi-chamber implosion fluid-solid coupling numerical simulation model to reproduce implosion process;Based on Latin hypercube sampling, generate explosion working condition sample, construct full working condition simulation calculation and explosion load dataset, as the basis of proxy model training;Two-stage modeling is used to construct load prediction proxy model, to realize global load fast calculation.

Description

Technical Field

[0001] This invention relates to the fields of explosion mechanics and structural impact response prediction, specifically to a rapid calculation method for implosion load fields in asymmetric complex multi-compartment structures. Background Technology

[0002] Currently, methods for predicting explosive shock loads mainly include empirical formula methods, numerical simulation methods, and data-driven methods. Empirical formula methods based on explosion similarity theory estimate load parameters such as peak overpressure and impulse using parameters like proportional distance, offering advantages such as computational simplicity and high efficiency. However, these methods are typically applicable to open spaces or simple structural conditions, and struggle to accurately describe the complex phenomena of shock wave propagation, reflection, and coupling between compartments in asymmetric, complex multi-compartment structures, leading to insufficient prediction accuracy. High-precision numerical simulation methods can realistically simulate the propagation of explosive shock waves and their interaction with the structure, obtaining complete spatiotemporal load distribution information. However, these methods are computationally expensive and time-consuming, making them unsuitable for rapid assessment and engineering applications in multi-compartment complex structures undergoing multi-condition analysis. In recent years, machine learning methods have been used for explosive shock load prediction, achieving rapid prediction by establishing a mapping relationship between input parameters and load response. However, these methods rely on large amounts of high-quality training data. For multi-compartment structures, the high cost of high-precision simulation, limited data sample size, and the complexity and variability of the structure and explosion scenarios result in insufficient model generalization ability. In addition, existing methods are mostly aimed at predicting single-point or local loads, and lack the ability to effectively reconstruct the load distribution across the entire field in complex structures.

[0003] Therefore, existing technologies struggle to achieve rapid, full-field, and efficient prediction of implosion load fields in asymmetric, complex, multi-compartment structures while ensuring prediction accuracy. Summary of the Invention

[0004] The present invention aims to solve the problems of low calculation efficiency and difficulty in rapid reconstruction of full-field loads in asymmetric complex multi-compartment structures in the prior art for predicting implosion load fields. Therefore, it provides a method for rapid calculation of implosion load fields in asymmetric complex multi-compartment structures to solve this problem.

[0005] To achieve the above objectives, this invention provides the following technical solution: a rapid calculation method for implosion load fields in asymmetric complex multi-compartment structures, combining high-precision numerical simulation and data-driven proxy models to achieve rapid prediction of explosion impact load fields, applicable to blast-resistant design and rapid load assessment of ship structures, comprising the following steps:

[0006] A numerical model of an asymmetric complex multi-compartment structure was established to clarify the spatial topological relationship, geometric parameters, and air domain boundary constraints of each compartment, which serve as the range of values ​​for the explosion point parameters and determine the sampleable area at the explosion center. Multiple load response measurement points were arranged in each compartment to obtain the explosion impact load response parameters.

[0007] Using the explosion center coordinates and equivalent quantity as variables, explosion point samples are generated through Latin hypercube sampling; spatial constraint verification is carried out in combination with a multi-compartment model, invalid samples are eliminated and supplemented; then, explosion load simulation and data extraction are completed through unified measurement points, and a parameter mapping dataset is constructed for subsequent model training.

[0008] A fluid-structure interaction (ALE) algorithm was used to build a multi-compartment internal explosion fluid-structure interaction numerical simulation model. Different solution methods, material models and state equations were selected for different regions to fully reproduce the implosion process. Explosion condition samples were generated based on Latin hypercube sampling (LHS) to construct a full-condition simulation calculation and explosion load dataset as the basis for training the surrogate model.

[0009] Based on the Kriging surrogate model, a two-stage modeling approach is adopted to construct a load prediction surrogate model. Combined with Bayesian optimization and K-fold cross-validation, training and accuracy optimization are completed to achieve fast calculation of the global load.

[0010] Preferably, a three-dimensional structural model is constructed based on multiple compartments formed by the compartmentalized structure. Each compartment is spatially separated by longitudinal bulkheads, transverse bulkheads and decks to form a spatial topology, so as to realize the construction of a numerical model of an asymmetric complex multi-compartment structure.

[0011] The main structure of the compartment structure is modeled using shell elements, and includes longitudinal bulkheads, transverse bulkheads, and reinforcing structures. The air domain is modeled using solid meshes, and a 3D gradient mesh is applied to the air domain. The specific geometric boundaries are as follows:

[0012] A layered grid region is established based on the explosion center, and the grid size is set to... It expands outward layer by layer according to a geometric progression, and its grid size satisfies:

[0013] ;

[0014] in, For the first Layer area grid size; The grid size is for the central area; This represents the grid growth coefficient.

[0015] Mesh convergence analysis shows that when the air mesh size converges to the order of 15 mm, the calculation error of the peak shock wave pressure tends to stabilize, with the average error controlled within 7%. Therefore, the mesh size of the central region is adopted as 15.625 mm, and the mesh growth factor is 2. The mesh sizes of each layer are 15.625 mm, 31.25 mm, 62.5 mm and 125 mm respectively, to ensure the effectiveness of mesh division.

[0016] Preferably, the three-dimensional coordinates of the explosion center and the explosive yield are used as sampling variables; each sampling variable is divided into several equally probable intervals, and the Latin hypercube sampling method is used to randomly extract sample values ​​from each interval and combine them to generate multiple sets of explosion point samples covering the entire compartment space, ensuring that each sampling variable is evenly distributed within its value range and that the samples cover the entire compartment space.

[0017] After generating explosion point samples using Latin hypercube sampling, the validity of each explosion point sample is determined as follows:

[0018] Based on the established numerical model of the asymmetric complex multi-compartment structure, it is determined whether the explosion center is located within the effective space of the compartment, and the minimum distance between the explosion center and the classic structural parts is calculated. The classic structural parts include longitudinal bulkheads, transverse bulkheads, decks and their corresponding structural boundaries. When the explosion center is located inside the structural entity, or the minimum distance between the explosion center and the classic structural parts is less than the preset safety threshold, the sample is determined to be an invalid sample and is removed.

[0019] The preset safety threshold is 1%-5% of the characteristic length of the compartment, in order to avoid the explosion center being too close to the bulkhead, deck or compartment structure, which would cause the explosion point to fall into the structural solid area or form an atypical near-wall condition.

[0020] For invalid samples that are removed, a supplementary sampling method is used to regenerate new explosion point samples within the spatial range that meets the constraints, and the screening process is repeated until an explosion condition sample set that meets the spatial constraints and sample spacing requirements is obtained.

[0021] For each set of detonation point samples, explosion load simulation calculations are performed in the established asymmetric complex multi-compartment structure numerical model, and the corresponding explosion impact load response parameters are extracted according to the pre-arranged load response measurement points. The load response measurement points cover each compartment and typical structural parts to ensure that response data covering the entire structural range can be obtained for each set of explosion conditions. The positions of the load response measurement points remain consistent during the calculation of each explosion condition. Each set of detonation point samples corresponds to the same set of load response measurement points to establish the mapping relationship between the detonation point parameters and the impact load response parameters of each measurement point, providing training samples for the subsequent construction of a rapid calculation model of the shock wave force field.

[0022] The preferred method for building a multi-compartment internal explosion fluid-structure interaction numerical simulation model is as follows:

[0023] The air domain was solved using the Euler description method, and the multi-compartment main structure was solved using the Lagrange description method. Fluid-structure interaction calculations were performed using the ALE algorithm. Based on the mechanical behavior of the compartment materials (907A steel was used for longitudinal and transverse bulkheads and decks, and 921A steel was used for reinforcing structures), the Johnson–Cook constitutive model was used for description.

[0024] ;

[0025] In the formula, The yield stress; It is the strain hardening constant; Equivalent plastic strain; The strain hardening index; The strain rate correlation coefficient; This represents the ratio of equivalent plastic strain rates. Relative temperature; Temperature-related index;

[0026] The air material is described using the *MAT_NULL model, combined with a linear polynomial equation of state:

[0027] ;

[0028] In the formula: For pressure; Internal energy per unit volume; For relative volume, For air materials constants;

[0029] The explosive medium is described by combining the high-energy combustion model *MAT_HIGH_EXPLOSIVE_BURN with the JWL equation of state.

[0030] The high-energy combustion model is used to define the basic parameters of the explosive, and the combustion fraction F is used to simulate the initiation and detonation transition process of the explosive; the pressure-volume-energy relationship of the detonation products is described by the JWL equation of state, the expression of which is:

[0031] ;

[0032] In the formula: For the pressure of detonation products, For relative volume, The internal energy per unit initial volume, , , , , These are the material constants for TNT;

[0033] The high-energy combustion model is coupled with the JWL equation of state through the combustion fraction F, thereby simulating the release process of explosive chemical energy into detonation product pressure energy; the actual pressure of the explosive unit is expressed as:

[0034] ;

[0035] in, The combustion fraction, ranging from 0 to 1; when This indicates that the explosives have not yet detonated. The time indicates that the explosive has completely detonated;

[0036] The explosive charge area is defined by the keyword INITIAL_VOLUME_FRACTION_GEOMETRY and placed at the predetermined explosion location inside the compartment; the detonation point and detonation time are defined by INITIAL_DETONATION, thereby realizing the numerical simulation of the explosion process inside the complex multi-compartment structure.

[0037] Explosion case samples were generated based on Latin hypercube sampling (LHS), as follows:

[0038] At the center of the explosion As sampling variables, Latin hypercube sampling is used to generate valid burst point conditions, forming a complete set of conditions:

[0039] ;

[0040] in, They represent the first A sample of the trigger points in Coordinate values ​​in three spatial directions; These represent the sampleable areas of the explosion point. Minimum boundary coordinates in the direction; These represent the sampleable areas of the explosion point. Maximum boundary coordinates in the direction; This represents the total number of samples generated by Latin hypercube sampling; They represent the first time. The random perturbation generated within each interval has a value range of [0,1), and is used to randomly determine the sample location within the corresponding interval. Indicates the first Interval numbering in each dimension A random permutation function is used to ensure the uniform distribution of sampling points in each dimension and avoid the correlation of sample points in multidimensional space;

[0041] For each set of explosion conditions, high-precision numerical simulation calculations were performed, and shock wave load parameters, including peak pressure, were extracted at multiple pre-arranged pressure measurement points. Time of arrival of the shock wave duration of positive phase Specific impulse and quasi-static pressure This allows us to construct an explosive load dataset.

[0042] Preferably, the load prediction proxy model is constructed as follows:

[0043] In the first stage, a load prediction model for each measuring point is constructed, establishing a mapping relationship between the explosion center location parameters and the load parameters at each measuring point; let the first stage be... The coordinates of the measuring points are The corresponding load response parameters are Then the mapping relationship is expressed as:

[0044] ;

[0045] in, This represents a nonlinear mapping function established based on data. This indicates the load parameters at the corresponding measuring point;

[0046] The mapping function is constructed using a Kriging surrogate model. Both stages of the load prediction surrogate model employ a second-order polynomial regression function to describe the global trend term and a Gaussian correlation function to characterize local smooth changes. Its basic form is as follows:

[0047] ;

[0048] in, It is a second-order polynomial regression function. For a zero-mean Gaussian random process, its covariance is determined by the correlation function;

[0049] The hyperparameters in the correlation function are identified through maximum likelihood estimation and efficiently solved using Bayesian optimization. The likelihood function is expressed as:

[0050] ;

[0051] in, Represents the log-likelihood function value; For the sample size, This is the response value vector at the sample point. The regression matrix corresponds to the second-order polynomial basis functions. For the regression coefficient vector, This is the correlation matrix between sample points; The inverse matrix of the correlation matrix; This represents the determinant value of matrix R, used to calculate the volume effect term resulting from the correlation matrix; Represents the transpose of a matrix / vector;

[0052] To improve the generalization ability of the model, a K-fold cross-validation method is introduced during the training process of the load prediction surrogate model. The sample data is divided into K subsets, and a subset is selected alternately as the validation set, while the remainder is used as the training set. The model's prediction error is evaluated using the root mean square error (RMSE) as the evaluation metric, and a coefficient of determination is also introduced. The fit of the load prediction surrogate model was evaluated:

[0053] ;

[0054] in, The load parameters are the true values, i.e., the explosive loads obtained from the numerical simulation of the asymmetric complex multi-compartment structure; These are the predicted values ​​of the load parameters, i.e., the loads output by the load prediction surrogate model. This represents the sample mean of the load parameters;

[0055] In the second stage, after obtaining the load prediction results for each measuring point, the load prediction results are then calculated at any spatial location. As input, combined with the known coordinates of the measuring points and their corresponding load parameters, a spatial continuous field reconstruction model is constructed, and its mapping relationship is expressed as follows:

[0056] ;

[0057] in, Representative proxy modeling function The input is the coordinates of the measuring points in the first-stage measuring point load prediction model. and the set of spatial location and load data of all measuring points The output is the predicted load parameter value at that location. ;

[0058] The spatial reconstruction process is implemented based on the Kriging proxy model for the target point. Its predicted response value and prediction variance Represented as:

[0059] ;

[0060] in, It is a vector of regression coefficients. It is the correlation vector between the target point and the training samples, where the target point is... Including the first phase of measurement points and any point in space ;and , This indicates the correction term for the target point; It is transpose.

[0061] The present invention has the following advantages:

[0062] This invention proposes a rapid prediction method for implosion loads in asymmetric complex multi-compartment structures. Based on a combination of high-precision numerical simulation and a data-driven surrogate model, it significantly improves computational efficiency while maintaining prediction accuracy. By introducing the LHS sampling method to construct a representative set of explosion conditions and combining it with a two-stage Kriging surrogate model, it achieves efficient mapping between explosion center parameters and measurement point load responses, as well as rapid reconstruction of the overall load distribution. This enables accurate prediction of impact load distributions in complex multi-compartment structures under limited sample conditions. Compared with traditional high-precision numerical simulation methods, this invention reduces computation time from a large amount of kernel time to the second level, significantly lowering computational costs. It also overcomes the problem that empirical formulas are insufficient to describe the shock wave propagation and coupling effects in complex compartment structures, demonstrating good engineering applicability and widespread value. Attached Figure Description

[0063] Figure 1 Flowchart provided for this invention;

[0064] Figure 2 A perspective view of the three-dimensional numerical model of the asymmetric complex multi-compartment structure provided by the present invention;

[0065] Figure 3 This is a schematic diagram of the explosion / impact simulation mesh generation provided by the present invention;

[0066] Figure 4 A schematic diagram of load response measurement points for a three-dimensional numerical model of an asymmetric complex multi-compartment structure provided by the present invention;

[0067] Figure 5 A diagram of the two-stage load prediction model provided by this invention;

[0068] Figure 6 This is a standard flowchart for K-fold cross-validation provided by the present invention. Detailed Implementation

[0069] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] like Figure 1 As shown, the overall process of the fast calculation method for implosion load field of asymmetric complex multi-compartment structure proposed in this invention includes the following steps: structural modeling, implosion point sampling, high-precision numerical simulation, surrogate model construction, and fast load prediction and full-field reconstruction.

[0071] In one embodiment, based on the explicit dynamic finite element software LS-DYNA, an ALE method is used to establish a numerical simulation model of an implosion in an asymmetric complex multi-compartment structure, and a data-driven method is combined to construct a rapid prediction model of the explosion load. Specifically, the following steps are included:

[0072] 1. Establish a three-dimensional numerical model of an asymmetric complex multi-compartment structure. For example... Figure 2 As shown, the structure as a whole consists of multiple compartments, which are spatially separated by longitudinal bulkheads, transverse bulkheads, and decks, forming a complex topology. The bulkheads and reinforcing structures are modeled using shell elements, while the air domain is modeled using solid meshes, with a three-dimensional gradient mesh generation applied to the air domain. Figure 3 As shown, starting from the explosion center, the mesh size gradually increases from the inside out, with the central region having a mesh size of 15.625 mm, gradually transitioning to 31.25 mm, 62.5 mm, and 125 mm. This significantly reduces the computational scale while ensuring the accuracy of shock wave propagation. Mesh convergence analysis shows that when the air mesh size converges to the order of 15 mm, the calculation error of the peak shock wave pressure tends to stabilize, with the average error controlled within approximately 7%, thus verifying the effectiveness of the proposed mesh generation method.

[0073] Using the explosion center coordinates and equivalent quantity as variables, explosion point samples are generated through Latin hypercube sampling; spatial constraint verification is carried out in combination with a multi-compartment model, invalid samples are eliminated and supplemented; then, explosion load simulation and data extraction are completed through unified measurement points, and a parameter mapping dataset is constructed for subsequent model training.

[0074] Specifically, the three-dimensional coordinates of the explosion center and the explosive yield are used as sampling variables. Each sampling variable is divided into several equally probable intervals. The Latin hypercube sampling method is used to randomly extract sample values ​​from each interval and combine them to generate multiple sets of explosion point samples covering the entire compartment space, ensuring that each sampling variable is evenly distributed within its value range and that the samples cover the entire compartment space.

[0075] After generating explosion point samples using Latin hypercube sampling, the validity of each explosion point sample is determined as follows:

[0076] Based on the established numerical model of the asymmetric complex multi-compartment structure, it is determined whether the explosion center is located within the effective space of the compartment, and the minimum distance between the explosion center and the classic structural parts is calculated. The classic structural parts include longitudinal bulkheads, transverse bulkheads, decks and their corresponding structural boundaries. When the explosion center is located inside the structural entity, or the minimum distance between the explosion center and the classic structural parts is less than the preset safety threshold, the sample is determined to be an invalid sample and is removed.

[0077] The preset safety threshold is 1%-5% of the characteristic length of the compartment, in order to avoid the explosion center being too close to the bulkhead, deck or compartment structure, which would cause the explosion point to fall into the structural solid area or form an atypical near-wall condition.

[0078] Specifically, in this embodiment, a minimum distance of 0.4 m is used to calculate the Euclidean distance between the explosion point samples after spatial constraint screening. When the distance between any two explosion point samples is less than the preset minimum sample spacing, one of the samples is retained and the rest are removed. The minimum sample spacing is preferably 0.4 m to avoid excessive aggregation of samples in local space and to improve the uniformity and representativeness of sample coverage in the entire cabin space.

[0079] For invalid samples that are rejected, a supplementary sampling method is used to regenerate new explosion point samples within the spatial range that meets the constraints. The screening process is repeated until an explosion condition sample set that meets the spatial constraints and sample spacing requirements is obtained, so as to ensure the physical rationality of the samples and the effectiveness of subsequent numerical calculations, thereby forming a representative explosion condition sample set that covers the entire compartment space.

[0080] For each set of detonation point samples, explosion load simulation calculations are performed in the established asymmetric complex multi-compartment structure numerical model, and the corresponding explosion impact load response parameters are extracted according to the pre-arranged load response measurement points. The load response measurement points cover each compartment and typical structural parts to ensure that response data covering the entire structural range can be obtained for each set of explosion conditions. The positions of the load response measurement points remain consistent during the calculation of each explosion condition. Each set of detonation point samples corresponds to the same set of load response measurement points to establish the mapping relationship between the detonation point parameters and the impact load response parameters of each measurement point, providing training samples for the subsequent construction of a rapid calculation model of the shock wave force field.

[0081] 2. Construct a fluid-structure interaction (FSI) calculation model. The air domain is described using the Euler method, and the structure using the Lagrange method. The ALE algorithm is used to perform FSI calculations. Based on the mechanical behavior of the compartment materials (907A steel is used for longitudinal and transverse bulkheads and decks, and 921A steel is used for reinforcing structures), the Johnson-Cook constitutive model is used for description.

[0082]

[0083] In the formula: The yield stress; It is the strain hardening constant; The strain rate correlation coefficient; The strain hardening index; Temperature-related index; Equivalent plastic strain; This represents the ratio of equivalent plastic strain rates. This refers to relative temperature.

[0084] The air material is described using the *MAT_NULL model, combined with a linear polynomial equation of state:

[0085] ;

[0086] In the formula: For pressure; Internal energy per unit volume; For relative volume, is a material constant.

[0087] The explosive is described using a high-energy combustion model (*MAT_HIGH_EXPLOSIVE_BURN) and the JWL equation of state, defined by the keyword *INITIAL_VOLUME_FRACTION_GEOMETRY, and placed at a predetermined detonation location inside the compartment. The JWL equation of state, derived from isentropic curves, is expressed as follows:

[0088] ;

[0089] In the formula: For pressure, For relative volume, Internal energy per unit volume , , , , is a material constant.

[0090] 3. Based on the above numerical model, multiple sets of explosion center location samples were generated within the compartment space using LHS. This embodiment conducted 150 operational scenarios to form an explosion scenario set. Let the explosion center location be... Then we have:

[0091] ;

[0092] in, Indicates the first Interval numbering in each dimension A random permutation function is used to ensure the uniform distribution of sampling points in each dimension and to avoid the correlation of sample points in multidimensional space.

[0093] For each set of explosion conditions, high-precision numerical simulation calculations are performed. For example... Figure 4 As shown, shock wave load parameters, including peak pressure, are extracted at multiple pre-arranged pressure measurement points. Arrival time duration of positive phase Specific impulse and quasi-static pressure This allows us to construct an explosive load dataset.

[0094] 4. Construct a two-stage load prediction model. For example... Figure 5 As shown, in the first stage, the mapping relationship between the explosion center location parameters and the load parameters at each measuring point is established. Let the first stage be... The coordinates of the measuring points are The corresponding load response parameters are Then the mapping relationship can be expressed as:

[0095] ;

[0096] in, This represents a data-driven nonlinear mapping function. The inputs are the coordinates of the explosion center and the spatial coordinates of the set measurement points. The output is the load parameters (including peak pressure) at the corresponding measurement points. Time of arrival of the shock wave duration of positive phase Specific impulse and quasi-static pressure wait).

[0097] Preferably, the mapping function is constructed using a Kriging surrogate model. Both stages of the Kriging surrogate model use a second-order polynomial regression function to describe the global trend term and a Gaussian correlation function to characterize local smooth changes. Its basic form is as follows:

[0098]

[0099] in, It is a second-order polynomial regression function. It is a zero-mean Gaussian random process, and its covariance is determined by the correlation function.

[0100] The hyperparameters in the correlation function are identified through maximum likelihood estimation and efficiently solved using Bayesian optimization methods. The likelihood function can be expressed as:

[0101] ;

[0102] in, For the sample size, This is the response value vector at the sample point. The regression matrix corresponds to the second-order polynomial basis functions. For the regression coefficient vector, This is the correlation matrix between sample points.

[0103] Furthermore, to improve the model's generalization ability, a K-fold cross-validation method is introduced during model training. For example... Figure 6 The sample data is divided into K subsets. A subset is selected alternately as the validation set, and the remainder as the training set. The model's prediction error is evaluated using the root mean square error (RMSE) as the evaluation metric. A coefficient of determination is also introduced. The model fit was evaluated:

[0104] ;

[0105] in, These are the actual values ​​of the load parameters. These are the predicted values ​​of the load parameters. This represents the sample mean of the load parameters.

[0106] By employing a Bayesian optimization method with cross-validation error as the optimization objective, and training the sample data, it is possible to quickly predict the load parameters of each measuring point under arbitrary explosion location conditions.

[0107] In the second stage, after obtaining the load prediction results for each measuring point, the load prediction results are then calculated at any spatial location. Using the known coordinates of the measuring points and their corresponding load parameters as input, a spatial continuous field reconstruction model is constructed, and its mapping relationship can be expressed as:

[0108]

[0109] in, This represents a spatial interpolation or surrogate modeling function. The input is the coordinates of the target point and the set of spatial locations and load data of all measuring points. The output is the predicted value of the load parameters at that location.

[0110] Preferably, the spatial reconstruction process is also based on the Kriging model, for the target point Its predicted response value and prediction variance It can be represented as:

[0111]

[0112] in It is a vector of regression coefficients. It is the correlation vector between the target point and the training samples. It is the response vector, and .

[0113] Similarly, in the second-stage model construction process, K-fold cross-validation and Bayesian optimization methods are combined to optimize the model hyperparameters, thereby improving the accuracy and stability of load field reconstruction in asymmetric complex multi-compartment structures. By constructing a spatial correlation function, discrete measurement point data are processed into continuous data, thus obtaining the explosion load distribution at any location within the asymmetric complex multi-compartment structure and enabling rapid calculation of the explosion force field.

[0114] 5. Given any explosion location as input, the constructed proxy model can quickly output the shock wave load field parameter distribution in an asymmetric complex multi-compartment structure without the need for finite element simulation, realizing an integrated and rapid calculation process from "explosion point input - measurement point prediction - full field reconstruction".

[0115] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A rapid calculation method for implosion load field in an asymmetric complex multi-compartment structure, characterized in that: Includes the following steps: A numerical model of an asymmetric complex multi-compartment structure is established to clarify the spatial topological relationship, geometric parameters, and air domain boundary constraints of each compartment, which serve as the range of values ​​for the detonation point parameters and determine the sampleable area at the explosion center location. Multiple load response measurement points were arranged in each compartment to obtain the explosion shock load response parameters. Detonation point samples are generated using Latin hypercube sampling, with the explosion center coordinates and equivalent quantity as variables. Spatial constraint verification was carried out using a multi-compartment model, invalid samples were removed and supplemented; then, explosion load simulation and data extraction were completed through unified measurement points, and a parameter mapping dataset was constructed for subsequent model training. Using the ALE fluid-structure interaction algorithm, a numerical simulation model of fluid-structure interaction in multi-compartment implosion was built. The solution method, material model and state equation were selected for different regions to completely reproduce the implosion process. Explosion condition samples were generated based on Latin hypercube sampling (LHS), and a full-condition simulation calculation and explosion load dataset was constructed as the basis for training the surrogate model. Based on the Kriging surrogate model, a two-stage modeling approach is adopted to construct a load prediction surrogate model. Combined with Bayesian optimization and K-fold cross-validation, training and accuracy optimization are completed to achieve fast calculation of the global load.

2. The method for rapid calculation of implosion load field in an asymmetric complex multi-compartment structure according to claim 1, characterized in that: A three-dimensional structural model is constructed based on multiple compartments formed by the compartmentalized structure. The compartments are spatially separated by longitudinal bulkheads, transverse bulkheads and decks to form a spatial topology, so as to realize the construction of a numerical model of an asymmetric complex multi-compartment structure. The main structure of the compartment structure is modeled using shell elements, and includes longitudinal bulkheads, transverse bulkheads, and reinforcing structures. The air domain is modeled using solid meshes, and a 3D gradient mesh is applied to the air domain. The specific geometric boundaries are as follows: A layered grid region is established based on the explosion center, and the grid size is set to... It expands outward layer by layer according to a geometric progression, and its grid size satisfies: ； in, For the first Layer area grid size; The grid size is for the central area; This is the grid growth factor; Mesh convergence analysis shows that when the air mesh size converges to the order of 15 mm, the calculation error of the peak shock wave pressure tends to stabilize, with the average error controlled within 7%. Therefore, the mesh size of the central region is adopted as 15.625 mm, and the mesh growth factor is 2. The mesh sizes of each layer are 15.625 mm, 31.25 mm, 62.5 mm and 125 mm respectively, to ensure the effectiveness of mesh division.

3. The method for rapid calculation of implosion load field in an asymmetric complex multi-compartment structure according to claim 2, characterized in that: The three-dimensional coordinates of the explosion center and the explosive yield were used as sampling variables. Each sampling variable was divided into several equally probable intervals. The Latin hypercube sampling method was used to randomly extract sample values ​​from each interval and combine them to generate multiple sets of explosion point samples covering the entire compartment space, ensuring that each sampling variable is evenly distributed within its value range and that the samples cover the entire compartment space. After generating explosion point samples using Latin hypercube sampling, the validity of each explosion point sample is determined as follows: Based on the established numerical model of the asymmetric complex multi-compartment structure, it is determined whether the explosion center is located within the effective space of the compartment, and the minimum distance between the explosion center and the classic structural parts is calculated. The classic structural parts include longitudinal bulkheads, transverse bulkheads, decks and their corresponding structural boundaries. When the explosion center is located inside the structural entity, or the minimum distance between the explosion center and the classic structural parts is less than the preset safety threshold, the sample is determined to be an invalid sample and is removed. The preset safety threshold is 1%-5% of the characteristic length of the compartment, in order to avoid the explosion center being too close to the bulkhead, deck or compartment structure, which would cause the explosion point to fall into the structural solid area or form an atypical near-wall condition. For invalid samples that are removed, a supplementary sampling method is used to regenerate new explosion point samples within the spatial range that meets the constraints, and the screening process is repeated until an explosion condition sample set that meets the spatial constraints and sample spacing requirements is obtained. For each set of detonation point samples, explosion load simulation calculations are performed in the established asymmetric complex multi-compartment structure numerical model, and the corresponding explosion impact load response parameters are extracted according to the pre-arranged load response measurement points. The load response measurement points cover each compartment and typical structural parts to ensure that response data covering the entire structural range can be obtained for each set of explosion conditions. The positions of the load response measurement points remain consistent during the calculation of each explosion condition. Each set of detonation point samples corresponds to the same set of load response measurement points to establish the mapping relationship between the detonation point parameters and the impact load response parameters of each measurement point, providing training samples for the subsequent construction of a rapid calculation model of the shock wave force field.

4. The method for rapid calculation of implosion load field in an asymmetric complex multi-compartment structure according to claim 1, characterized in that: The numerical simulation model of fluid-structure interaction in a multi-compartment chamber explosion was built, and the construction process is as follows: The air domain was solved using the Euler description method, and the multi-compartment main structure was solved using the Lagrange description method. Fluid-structure interaction calculations were performed using the ALE algorithm. Based on the mechanical behavior of the compartment materials, the Johnson–Cook constitutive model was used for description. ； In the formula, The yield stress; It is the strain hardening constant; Equivalent plastic strain; The strain hardening index; The strain rate correlation coefficient; This represents the ratio of equivalent plastic strain rates. Relative temperature; Temperature-related index; The air material is described using the *MAT_NULL model, combined with a linear polynomial equation of state: ； In the formula: For pressure; Internal energy per unit volume; Relative volume For air, the material constant is used. The explosive medium is described by combining the high-energy combustion model *MAT_HIGH_EXPLOSIVE_BURN with the JWL equation of state. The high-energy combustion model is used to define the basic parameters of the explosive, and the combustion fraction F is used to simulate the initiation and detonation transition process of the explosive; the pressure-volume-energy relationship of the detonation products is described by the JWL equation of state, the expression of which is: ； In the formula: For the pressure of detonation products, Relative volume The internal energy per unit initial volume, , , , , These are the material constants for TNT; The high-energy combustion model is coupled with the JWL equation of state through the combustion fraction F, thereby simulating the release process of explosive chemical energy into detonation product pressure energy; the actual pressure of the explosive unit is expressed as: ； in, The combustion fraction, ranging from 0 to 1; when This indicates that the explosives have not yet detonated. The time indicates that the explosive has completely detonated; The explosive charge area is defined by the keyword INITIAL_VOLUME_FRACTION_GEOMETRY and placed at the predetermined explosion location inside the compartment; the detonation point and detonation time are defined by INITIAL_DETONATION, thereby realizing the numerical simulation of the explosion process inside a complex multi-compartment structure. Explosion case samples were generated based on Latin hypercube sampling (LHS), as follows: At the center of the explosion As sampling variables, Latin hypercube sampling is used to generate valid burst point conditions, forming a complete set of conditions: ； in, They represent the first A sample of the trigger points in Coordinate values ​​in three spatial directions; These represent the sampleable areas of the explosion point. Minimum boundary coordinates in the direction; These represent the sampleable areas of the explosion point. Maximum boundary coordinates in the direction; This represents the total number of samples generated by Latin hypercube sampling; They represent the first time. The random perturbation generated within each interval has a value range of [0,1), and is used to randomly determine the sample location within the corresponding interval. Indicates the first Interval numbering in each dimension A random permutation function is used to ensure the uniform distribution of sampling points in each dimension and avoid the correlation of sample points in multidimensional space; For each set of explosion conditions, high-precision numerical simulation calculations were performed, and shock wave load parameters, including peak pressure, were extracted at multiple pre-arranged pressure measurement points. Time of arrival of the shock wave Duration of positive phase Specific impulse and quasi-static pressure This allows us to construct an explosive load dataset.

5. The method for rapid calculation of implosion load field in an asymmetric complex multi-compartment structure according to claim 4, characterized in that: The load prediction proxy model is constructed as follows: In the first stage, a load prediction model for each measuring point is constructed, establishing a mapping relationship between the explosion center location parameters and the load parameters at each measuring point; let the first stage be... The coordinates of the measuring points are The corresponding load response parameters are Then the mapping relationship is expressed as: ； in, This represents a nonlinear mapping function established based on data. This indicates the load parameters at the corresponding measuring point; The mapping function is constructed using a Kriging surrogate model. Both stages of the load prediction surrogate model employ a second-order polynomial regression function to describe the global trend term and a Gaussian correlation function to characterize local smooth changes. Its basic form is as follows: ； in, It is a second-order polynomial regression function. For a zero-mean Gaussian random process, its covariance is determined by the correlation function; The hyperparameters in the correlation function are identified through maximum likelihood estimation and efficiently solved using Bayesian optimization. The likelihood function is expressed as: ； in, This represents the log-likelihood function value; For the sample size, This is the response value vector at the sample point. The regression matrix corresponds to the second-order polynomial basis functions. For the regression coefficient vector, This is the correlation matrix between sample points; This represents the inverse of the correlation matrix; This represents the determinant value of matrix R, used to calculate the volume effect term resulting from the correlation matrix; Represents the transpose of a matrix / vector; In the training process of the load prediction proxy model, a K-fold cross-validation method is introduced. The sample data is divided into K subsets, and a subset is selected in turn as the validation set, while the rest are used as the training set. The prediction error of the model is evaluated, and the root mean square error (RMSE) is used as the evaluation metric. A coefficient of determination is also introduced. The fit of the load prediction surrogate model was evaluated: ； in, The load parameters are the true values, i.e., the explosive loads obtained from the numerical simulation of the asymmetric complex multi-compartment structure; These are the predicted values ​​of the load parameters, i.e., the loads output by the load prediction surrogate model. This represents the sample mean of the load parameters; In the second stage, after obtaining the load prediction results for each measuring point, the load prediction results are then calculated at any spatial location. As input, combined with the known coordinates of the measuring points and their corresponding load parameters, a spatial continuous field reconstruction model is constructed, and its mapping relationship is expressed as follows: ； in, Representative proxy modeling function The input is the coordinates of the measuring points in the first-stage measuring point load prediction model. and the set of spatial location and load data of all measuring points The output is the predicted load parameter value at that location. ; The spatial reconstruction process is implemented based on the Kriging proxy model for the target point. Its predicted response value and prediction variance Represented as: ； in, It is a vector of regression coefficients. It is the correlation vector between the target point and the training samples, where the target point is... Including the first phase of measurement points and any point in space ;and , This indicates the correction term for the target point; It is transpose.

Images