An automotive collision analysis method based on an adaptive surrogate model

Abstract

This application relates to the technical field of electrical digital data processing, and particularly to a vehicle collision analysis method based on an adaptive surrogate model. The method includes the following steps: S100, sampling the design space of vehicle collision using the Latin hypercube method to obtain N0 initial sampling points; S200, building an initial Kriging model based on the initial sample points and obtaining the accuracy λ0 of the initial Kriging model; S300, obtaining N1 and #imgabs0#; S400, using the Latin hypercube method to obtain N 1 1 uniformly sampled point newly added in the first generation; S500, using the importance sampling method to obtain N 2 1 importance sampled point newly added in the first generation; S600, generating a first-generation Kriging model using the first-generation sampling points; S700, if the number Q1 of the first-generation sampling points is ≥ Z, output the first-generation Kriging model, otherwise enter S800. The present invention improves the sampling efficiency and accuracy of the surrogate model for vehicle collision.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular, to a vehicle collision analysis method based on an adaptive surrogate model. Background Art

[0002] The collision safety of vehicles is an essential link in the vehicle design and development process. The body structure is the basis of collision safety. Designing a safe body with good collision energy absorption performance is a major goal of vehicle design. The collision tests required by the China New Car Assessment Program (C-NCAP) include: a frontal 100% overlap rigid barrier collision at a collision speed of 50 km / h, a frontal 40% overlap deformable barrier collision at a collision speed of 64 km / h, a deformable barrier side collision at a collision speed of 50 km / h, and other items, such as rear collision and pedestrian protection, etc.

[0003] However, in the process of vehicle collision simulation calculation, the cost of performing collision simulation analysis and optimization is huge. For example, a single calculation of the frontal 100% overlap rigid barrier collision simulation analysis of a certain vehicle model at a collision speed of 50 km / h takes more than 20 hours. If 8 design variables are considered, then the simulation analysis and optimization require 400 - 800 calculations, and the time cost spent is 8000 hours - 16000 hours. Such a high time cost is basically infeasible.

[0004] A surrogate model is a replacement for a real-world physical problem or a high-precision simulation model, and is a black-box model that can quickly calculate the output after a given input, and is trained by selecting a finite number of sample points using a specific sampling criterion. In order to reduce the time cost spent in the vehicle collision simulation calculation process, the surrogate model can be applied to the vehicle collision simulation calculation process. However, the existing surrogate models are generally constructed with sampling points based on the Latin square algorithm, that is, sample points are evenly selected in the entire design space, resulting in the problem of over-sampling at some positions with relatively small gradient changes, which affects the sampling efficiency of the surrogate model; while there is a problem of insufficient sampling at some positions with relatively large gradient changes, which affects the accuracy of the surrogate model. How to balance the sampling efficiency and accuracy of the surrogate model is an urgent problem to be solved. Summary of the Invention

[0005] The object of the present invention is to provide a vehicle collision analysis method based on an adaptive surrogate model to balance the sampling efficiency and accuracy of the surrogate model.

[0006] According to the present invention, there is provided a vehicle collision analysis method based on an adaptive surrogate model, including the following steps:

[0007] S100. Use the Latin hypercube method to sample the design space of vehicle collisions to obtain N0 initial sampling points, where N0 is the preset number of initial sampling points.

[0008] S200. Obtain the accuracy λ0 of the initial Kriging model, where the initial Kriging model is obtained from the N0 initial sampling points.

[0009] S300. Obtain N1 = (2 - λ0) * m and N1 is the number of newly added sampling points in the first generation, m is the number of design variables in the preset design space of vehicle collisions, is the ratio of uniform sampling and importance sampling in the first generation, and k is a preset value, k > 0.

[0010] S400. Use the Latin hypercube method to obtain N 1 1 newly added uniform sampling point in the first generation,

[0011] S500. Use the importance sampling method to obtain N 2 1 newly added importance sampling point in the first generation,

[0012] S600. Use the first-generation sampling points to generate the first-generation Kriging model, where the first-generation sampling points include N0 initial sampling points, N 1 1 newly added uniform sampling point in the first generation and N 2 1 newly added importance sampling point in the first generation.

[0013] S700. If the number of first-generation sampling points Q1 ≥ Z, output the first-generation Kriging model and end the method; Z is the preset maximum number of sampling points, Q1 = N0 + N1; if Q1 < Z, obtain the number of second-generation sampling points Q2 = Q1 + N2 and enter S800, where N2 is the number of newly added sampling points in the second generation.

[0014] S800. If Q2 ≥ Z, output the second-generation Kriging model and end the method; if Q2 < Z, obtain the number of third-generation sampling points Q3 = Q2 + N3, and so on, until the number of nth-generation sampling points Q n ≥ Z, and output the nth-generation Kriging model and end the method; N3 is the number of newly added sampling points in the third generation, n ≥ 3.

[0015] Compared with the prior art, the present invention has obvious beneficial effects. By means of the above technical solutions, the vehicle collision analysis method based on the adaptive surrogate model provided by the present invention can achieve considerable technical progressiveness and practicality, and has wide industrial utilization value. It has at least the following beneficial effects:

[0016] The present invention obtains sampling points based on uniform sampling and importance sampling. Among them, the overall profile of the entire design space can be obtained based on uniform sampling, and the number of sampling points in the areas with relatively drastic changes in the design space can be increased and the number of sampling points in the areas with relatively gentle changes in the design space can be reduced based on importance sampling. The present invention takes into account the global exploration and important position exploration of the design space, and through an adaptive iterative method, realizes the output of the optimal surrogate model under the maximum sampling point number limit, improving the sampling efficiency and accuracy of the surrogate model for vehicle collision. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a flowchart of the vehicle collision analysis method based on an adaptive surrogate model provided by an embodiment of the present invention. Detailed Embodiments

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0020] According to the present invention, a vehicle collision analysis method based on an adaptive surrogate model is provided, as Figure 1 shown, including the following steps:

[0021] S100, sampling the design space of vehicle collision using the Latin hypercube method to obtain N0 initial sampling points, where N0 is the preset number of initial sampling points.

[0022] In the present invention, N0 is input by the user or N0 is set to the empirical value 2*m + 1, where m is the number of design variables in the design space of vehicle collision preset.

[0023] Sampling using the Latin hypercube method is a prior art. Those skilled in the art know that any technical solution for obtaining sampling points using the Latin hypercube method in the prior art falls within the scope of protection of the present invention.

[0024] S200, obtaining the accuracy λ0 of the initial Kriging model, where the initial Kriging model is obtained from the N0 initial sampling points.

[0025] According to the present invention, the Kriging model is a surrogate model constructed by using the Kriging method.

[0026] It should be noted that the number S of output variables of vehicle collision y may be greater than or equal to 1. When the number S of output variables of the vehicle y is equal to 1, only the primary Kriging model corresponding to this 1 output variable needs to be constructed based on N0 primary sampling points. Correspondingly, the accuracy λ0 of the primary Kriging model is the accuracy of the primary Kriging model corresponding to this 1 output variable. In this case, where y i is the simulation calculation value of this output variable at the i-th sampling point, and y pred i is the surrogate model prediction value of the i-th sampling point of the primary Kriging model corresponding to this output variable, is the average value of the simulation calculation values of this output variable at all sampling points.

[0027] When the number S of output variables of the vehicle y is greater than 1, the primary Kriging model corresponding to each output variable is constructed respectively based on N0 primary sampling points. Thus, the accuracy λ 0,1 of the primary Kriging model corresponding to the 1st output variable, the accuracy λ 0,2 of the primary Kriging model corresponding to the 2nd output variable, …, the accuracy y of the primary Kriging model corresponding to the S-th output variable Finally, the minimum value of the accuracies of the primary Kriging models corresponding to these output variables is taken as λ0. In this case, where y j,i is the simulation calculation value of the j-th output variable at the i-th sampling point, and y pred j,i is the surrogate model prediction value of the i-th sampling point of the primary Kriging model corresponding to the j-th output variable, is the average value of the simulation calculation values of the j-th output variable at all sampling points, and the value range of j is from 1 to S y . Then, min means taking the minimum value.

[0028] S300, obtain N1 = (2 - λ0) * m and N1 is the number of newly added sampling points in the first generation, m is the number of design variables in the preset design space of vehicle collision, is the ratio of uniform sampling and importance sampling in the first generation, and k is a preset value, k > 0.

[0029] According to the present invention, the number of samples in the first generation is related to the accuracy λ0 of the primary Kriging model. The lower λ0 is, the higher the number of newly added sampling points in the first generation. Since 0 < λ0 < 1, the upper limit value of N1 is 2*m, and the lower limit value of N1 is m. Thus, the present invention can take into account both the number of iterations of vehicle collision and the speed of vehicle collision iteration.

[0030] According to the present invention, the ratio of the first-generation uniform sampling to the importance sampling is determined by the accuracy λ0 of the primary Kriging model. The lower λ0 is, the smaller the proportion of the newly added uniform sampling points in the first generation, and the larger the proportion of the newly added importance sampling points in the first generation. Preferably, 1 ≤ k ≤ 5. Thus, the lower limit value of the ratio of the first-generation uniform sampling to the importance sampling is 1:5, which can ensure the degree of exploration of the design space; the upper limit value of the ratio of the first-generation uniform sampling to the importance sampling is 1:1, which can ensure that the proportion of the importance sampling points is not too low.

[0031] S400, obtaining N by using the Latin hypercube method 1 1 newly added uniform sampling point in the first generation,

[0032] According to the present invention, the larger N1 is, the larger, N 1 the larger N1 is; the smaller N1 is, the smaller, N 1 the smaller N1 is.

[0033] S500, obtaining N by using the importance sampling method 2 1 newly added importance sampling point in the first generation,

[0034] According to the present invention, the larger N1 is, the smaller, N 2 the larger N1 is; the smaller N1 is, the larger, N 1 the smaller N1 is.

[0035] Those skilled in the art know that the technical solutions for obtaining importance sampling points by using any importance sampling method in the prior art fall within the protection scope of the present invention. Optionally, the importance sampling method is the Monte Carlo importance sampling method.

[0036] However, the present invention also proposes an importance sampling method, including the following steps:

[0037] S510, obtaining the uniform distribution function within the design space D is the design space of vehicle collision.

[0038] S520, for the uniform distribution function u(x1, x2, …, x m)Perform uniform sampling to obtain a set of sampling points \(A1 = [X1(x1,x2,\ldots,x m ),X2(x1,x2,\ldots,x m ),\ldots,X h (x1,x2,\ldots,x m ),\ldots,X H (x1,x2,\ldots,x m )]\), where \(X h (x1,x2,\ldots,x m )\) is the \(h\)-th sampling point obtained by performing uniform sampling on the uniform distribution function \(u(x1,x2,\ldots,x m )\). The value range of \(h\) is from 1 to \(H\), and \(H\) is the number of sampling points obtained by performing uniform sampling on the uniform distribution function \(u(x1,x2,\ldots,x m )\).

[0039] It should be noted that in order to reflect the trend of the uniform distribution function as much as possible and make \(T > N\) in S540 2 1, the value of \(H\) should be set as large as possible. For example, set \(H = 100*m\).

[0040] S530, traverse \(A1\) to obtain the weight of the \(h\)-th sampling point where \(p1(X h )\) is the value of the probability density \(p1(x1,x2,\ldots,x m )\) function corresponding to the first output variable of vehicle collision at the \(h\)-th sampling point, and \(u(X h )\) is the value of the uniform distribution function \(u(x1,x2,\ldots,x m )\) at the \(h\)-th sampling point.

[0041] S540, traverse \(A1\) and uniformly sample a sampling point \(u in the interval h . If \(u h < p1(X h )\), then append \(X h \) to \(D\) to obtain \(D = [X′1(x1,x2,\ldots,x m ),X′2(x1,x2,\ldots,x m ),\ldots,X′ t (x1,x2,\ldots,x m ),……,X′ T (x1,x2,\ldots,x m )]\), where \(X′ t (x1,x2,\ldots,x m )\) is the \(t\)-th sampling point appended to \(D\). The value range of \(t\) is from 1 to \(T\), and \(T\) is the number of sample points appended to \(D\). The initial value of \(D\) is Null.

[0042] S550, randomly select N from D 2 1 sampling point to obtain the set of sampling points corresponding to the first output variable of vehicle collision X″1(x1, x2, …, x m )、X″2(x1, x2, …, x m ) and are respectively the 1st, 2nd, N 2 1 randomly selected sampling point

[0043] S560, if the number S of output variables of vehicle collision y = 1, then determine B1 as the newly added importance sampling point of the first generation; if the number S of output variables of vehicle collision y > 1, then obtain and randomly select N 2 1 sampling point from C as the newly added importance sampling point of the first generation, B j is the set of sampling points corresponding to the jth output variable of vehicle collision

[0044] It should be understood that when the number S of output variables of the vehicle is y greater than 1, the set of sampling points corresponding to each output variable can be obtained with reference to S510 - S550 respectively

[0045] Those skilled in the art know that the technical solutions for implementing importance sampling using any probability density function in the prior art fall within the protection scope of the present invention. However, the present invention also proposes a method for obtaining a probability density function, including the following steps:

[0046] S531, obtain the function f1(x1, x2, …, x m ) = f0 + E1(x1, x2, …, x m ) of the first output variable corresponding to the initial Kriging model; where x1, x2, …, x m are respectively the 1st, 2nd, …, mth design variables of vehicle collision, f0 is the mean value of the output variable values corresponding to the initial sampling points, and E1(x1, x2, …, x m ) is the error function corresponding to f1(x1, x2, …, x m )

[0047] S532, obtain the gradient function of f1(x1, x2, …, x m )

[0048] S533, obtain the probability density function of f1(x1, x2, …, x m )

[0049] It should be understood that when the number S of output variables of the vehicle y is greater than 1, the probability density functions corresponding to the respective output variables can be obtained with reference to S531 - S533.

[0050] Based on the above probability density function, the present invention ensures that the distribution of sample points is associated with the gradient. By combining the importance sampling method, it is ensured that there are more sample points at positions with larger gradients and fewer sample points at positions with smaller gradients, thereby improving the efficiency and accuracy of sampling.

[0051] S600, generating a first - generation Kriging model using the first - generation sampling points, where the first - generation sampling points include N0 initial sampling points, N 1 1 newly added uniform sampling point of the first generation and N 2 1 newly added importance sampling point of the first generation.

[0052] According to the present invention, the first - generation sampling points are obtained by adding N1 sampling points to the initial sampling points. Thus, the accuracy of the first - generation Kriging model generated using the first - generation sampling points is improved compared to the initial Kriging model.

[0053] S700, if the number Q1 of the first - generation sampling points ≥ Z, then output the first - generation Kriging model and end the method; Z is a preset maximum number of sampling points, Q1 = N0 + N1; if Q1 < Z, then obtain the number Q2 of the second - generation sampling points Q2 = Q1 + N2, and enter S800, where N2 is the number of newly added sampling points of the second generation.

[0054] According to the present invention, N2=(2 - λ1)*m, where λ1 is the accuracy of the first - generation Kriging model.

[0055] Preferably, Z≥5*m. Small - batch experiments show that when Z≥5*m, the number of iterations of vehicle collision and the number of sample points in each iteration can be guaranteed, improving the accuracy of the output Kriging model.

[0056] The termination condition of the present invention is that the number of sample points is greater than or equal to Z. If the number Q1 of the first - generation sampling points is already greater than or equal to Z, then output the first - generation Kriging model and end the method; if the number Q1 of the first - generation sampling points is less than Z, then S800 needs to be executed.

[0057] S800, if Q2≥Z, then output the second - generation Kriging model and end the method; if Q2 < Z, then obtain the number Q3 of the third - generation sampling points Q3 = Q2 + N3, and so on, until the number Q n of the nth - generation sampling points ≥ Z, and output the nth - generation Kriging model and end the method; N3 is the number of newly added sampling points of the third generation, n≥3.

[0058] According to the present invention, the method for obtaining the second - generation Kriging model includes:

[0059] S810, Obtain which is the ratio of the second-generation uniform sampling and importance sampling.

[0060] S820, Use the Latin hypercube method to obtain N 1 two newly added second-generation uniform sampling points,

[0061] S830, Use the importance sampling method to obtain N 2 two newly added second-generation importance sampling points,

[0062] S840, Generate the second-generation Kriging model using the second-generation sampling points, where the second-generation sampling points include the first-generation sampling points, N 1 two newly added second-generation uniform sampling points and N 2 two newly added second-generation importance sampling points.

[0063] According to the present invention, Q n = Q n-1 + N n N n =(2 - λ n-1 ) * m, where λ n-1 is the accuracy of the (n - 1)-th generation Kriging model, N n is the number of newly added sampling points in the n-th generation, and Q n-1 is the number of sampling points in the (n - 1)-th generation.

[0064] According to the present invention, the method for obtaining the n-th generation Kriging model includes:

[0065] S811, Obtain which is the ratio of the n-th generation uniform sampling and importance sampling.

[0066] S821, Use the Latin hypercube method to obtain N 1 n newly added n-th generation uniform sampling points,

[0067] S831, Use the importance sampling method to obtain N 2 n newly added n-th generation importance sampling points,

[0068] S841, Generate the n-th generation Kriging model using the n-th generation sampling points, where the n-th generation sampling points include the (n - 1)-th generation sampling points, N 1 n newly added n-th generation uniform sampling points and N 2n An important sampling point newly added in the nth generation.

[0069] It should be noted that the process of constructing the second-generation Kriging model and the nth-generation Kriging model of the present invention can refer to the process of constructing the first-generation Kriging model. The difference is that when constructing the first-generation Kriging model, the initial Kriging model is referred to, and the parameters of the first-generation Kriging model are obtained by using the corresponding parameters of the initial Kriging model; while when constructing the second-generation Kriging model and the nth-generation Kriging model, the first-generation Kriging model and the (n - 1)th-generation Kriging model need to be referred to respectively, and the parameters of the second-generation Kriging model and the nth-generation Kriging model are obtained by using the corresponding parameters of the first-generation Kriging model and the (n - 1)th-generation Kriging model respectively.

[0070] As a specific embodiment, in order to analyze the frontal 100% overlap rigid wall barrier collision with a collision speed of 50 km / h, there are 3 output variables constructed, which are: ab (i.e., the peak acceleration of the B-pillar of the vehicle), ae (i.e., the peak acceleration of the engine mount of the vehicle), and m (i.e., the weight of the vehicle); there are 8 design variables in the corresponding design space, which are: x1 is the thickness of the outer panel of the engine hood, and the value range is 0.733 ≤ x1 ≤ 0.991; x2 is the thickness of the outer panel of the fender, and the value range is 0.719 ≤ x2 ≤ 0.973; x3 is the thickness of the inner panel of the fender, and the value range is 1.295 ≤ x3 ≤ 1.753; x4 is the thickness of the outer panel of the front section of the front longitudinal beam, and the value range is 1.294 ≤ x4 ≤ 1.750; x5 is the thickness of the inner panel of the front section of the front longitudinal beam, and the value range is 1.611 ≤ x5 ≤ 2.179; x6 is the thickness of the energy absorber box, and the value range is 1.317 ≤ x6 ≤ 1.781; x7 is the thickness of the front upper anti-collision beam, and the value range is 1.663 ≤ x7 ≤ 2.249; x8 is the thickness of the front lower anti-collision beam, and the value range is 1.287 ≤ x8 ≤ 1.741.

[0071] Set the maximum number of sampling points Z = 80, the number of initial sampling points N0 = 17, and k = 5. Respectively construct the initial Kriging models corresponding to ab, ae, and m, where the accuracy of the initial Kriging model corresponding to ab The accuracy of the initial Kriging model corresponding to ae The accuracy of the initial Kriging model corresponding to m Thus, determine λ0 = min{0.101, 0.133, 0.835} = 0.101.

[0072] Based on λ0, N1 = (2 - λ0) * m = 15.192, rounded to 15; N 1 1 = 2.68, rounded to 3; N 2 1 = 12.32, rounded to 12. Use N0 = 17 initial sampling points, N1 1 = 3 newly added uniform sampling points of the first generation and N 2 1 = 12 newly added importance sampling points of the first generation to obtain Q1 = 32 sampling points of the first generation. Based on these 32 sampling points of the first generation, the first-generation Kriging model can be obtained, including the first-generation Kriging model corresponding to ab, the first-generation Kriging model corresponding to ae, and the first-generation Kriging model corresponding to m.

[0073] Since Q1 = 32 is less than the maximum number of sampling points Z = 80, it is necessary to continue the iteration.

[0074] Obtain the accuracy of the first-generation Kriging model corresponding to ab The accuracy of the first-generation Kriging model corresponding to ae The accuracy of the first-generation Kriging model corresponding to m Thus, determine λ1 = min{0.322, 0.298, 0.933} = 0.298.

[0075] Based on λ1, N2 = (2 - λ1)*m = 13.616, rounded to 14; N 1 2 = 2.91, rounded to 3; N 2 2 = 11.09, rounded to 11. Use Q1 = 32 sampling points of the first generation, N 1 2 = 3 newly added uniform sampling points of the second generation and N 2 2 = 11 newly added importance sampling points of the second generation to obtain Q2 = 46 sampling points of the second generation. Based on these 46 sampling points of the second generation, the second-generation Kriging model can be obtained, including the second-generation Kriging model corresponding to ab, the second-generation Kriging model corresponding to ae, and the second-generation Kriging model corresponding to m.

[0076] Since Q2 = 46 is less than the maximum number of sampling points Z = 80, it is necessary to continue the iteration.

[0077] Obtain the accuracy of the second-generation Kriging model corresponding to ab The accuracy of the second-generation Kriging model corresponding to ae The accuracy of the second-generation Kriging model corresponding to m Thus, determine λ2 = min{0.579, 0.483, 0.992} = 0.483.

[0078] Based on λ2, N3 = (2 - λ2)*m = 12.136, rounded to 12; N 1 3 = 2.95, rounded to 3; N 2 3 = 9.05, rounded to 9. Use Q2 = 46 sampling points of the second generation, N 13 = 3 newly added uniform sampling points of the third generation and N 2 3 = 9 newly added importance sampling points of the third generation to obtain Q3 = 58 sampling points of the third generation. Based on these 58 sampling points of the third generation, the third-generation Kriging model can be obtained, including the third-generation Kriging model corresponding to ab, the third-generation Kriging model corresponding to ae, and the third-generation Kriging model corresponding to m.

[0079] Since Q3 = 58 is less than the maximum number of sampling points Z = 80, therefore, iteration needs to continue.

[0080] Obtain the accuracy of the third-generation Kriging model corresponding to ab The accuracy of the third-generation Kriging model corresponding to ae The accuracy of the third-generation Kriging model corresponding to m Thus, determine λ3 = min{0.666, 0.673, 0.999} = 0.666.

[0081] Based on λ3, N4 = (2 - λ3) * m = 10.672, rounded to 11; N 1 4 = 3.3, rounded to 3; N 2 4 = 7.7, rounded to 8. Use Q3 = 58 sampling points of the third generation, N 1 4 = 3 newly added uniform sampling points of the fourth generation and N 2 4 = 8 newly added importance sampling points of the fourth generation to obtain Q4 = 69 sampling points of the fourth generation. Based on these 69 sampling points of the fourth generation, the fourth-generation Kriging model can be obtained, including the fourth-generation Kriging model corresponding to ab, the fourth-generation Kriging model corresponding to ae, and the fourth-generation Kriging model corresponding to m.

[0082] Since Q4 = 69 is less than the maximum number of sampling points Z = 80, therefore, iteration needs to continue.

[0083] Obtain the accuracy of the fourth-generation Kriging model corresponding to ab The accuracy of the fourth-generation Kriging model corresponding to ae The accuracy of the fourth-generation Kriging model corresponding to m Thus, determine λ4 = min{0.697, 0.712, 0.999} = 0.697.

[0084] Based on λ4, N5 = (2 - λ4) * m = 10.424, rounded to 10; N 1 5 = 3.11, rounded to 3; N 2 5 = 6.89, rounded to 7. Use Q4 = 69 sampling points of the fourth generation, N 1 5 = 3 newly added uniform sampling points of the fifth generation and N2 5 additional importance sampling points of the 5th generation are obtained to get Q5 = 79 sampling points of the 5th generation. Based on these 79 sampling points of the 5th generation, the Kriging model of the 5th generation can be obtained, including the Kriging model of the 5th generation corresponding to ab, the Kriging model of the 5th generation corresponding to ae, and the Kriging model of the 5th generation corresponding to m.

[0085] Since Q5 = 79 is less than the maximum number of sampling points Z = 80, iteration needs to continue.

[0086] Obtain the accuracy of the Kriging model of the 5th generation corresponding to ab The accuracy of the Kriging model of the 5th generation corresponding to ae The accuracy of the Kriging model of the 5th generation corresponding to m Thus, it is determined that λ5 = min{0.733, 0.748, 0.999} = 0.733.

[0087] Based on λ5, N6 = (2 - λ5) * m = 10.136, rounded to 10; N 1 6 = 3.26, rounded to 3; N 2 6 = 6.74, rounded to 7. Using Q5 = 79 sampling points of the 5th generation, N 1 6 = 3 uniformly sampled points newly added in the 6th generation and N 2 6 = 7 importance sampling points newly added in the 6th generation to get Q6 = 89 sampling points of the 6th generation. Based on these 89 sampling points of the 6th generation, the Kriging model of the 6th generation can be obtained, including the Kriging model of the 6th generation corresponding to ab, the Kriging model of the 6th generation corresponding to ae, and the Kriging model of the 6th generation corresponding to m.

[0088] Since Q6 = 89 is greater than the maximum number of sampling points Z = 80, iteration is terminated and the Kriging model of the 6th generation is output.

[0089] After 89 samplings and 6 iterations, the accuracy of the Kriging model corresponding to the peak acceleration of the B-pillar of the vehicle and the peak acceleration of the engine base of the vehicle output by the present invention reaches above 0.75 (in the Kriging model of the 6th generation, the accuracy of the Kriging model of the peak acceleration of the B-pillar is 0.78, and the accuracy of the Kriging model of the peak acceleration of the engine base is 0.754), which can more accurately reflect the change trend between the design variables and the output variables; the accuracy of the Kriging model corresponding to the weight of the vehicle reaches 0.999 (in the Kriging model of the 6th generation, the accuracy of the Kriging model of the vehicle weight is 0.999), which can replace the simulation process of weight calculation.

[0090] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are for illustrative purposes only and not for limiting the scope of the present invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. An automobile collision analysis method based on an adaptive surrogate model, characterized in that, It includes the following steps: S100, sampling the design space of vehicle collision by using the Latin hypercube method to obtain N0 initial sampling points, where N0 is the preset number of initial sampling points; S200, obtaining the accuracy λ0 of the initial Kriging model, where the initial Kriging model is obtained from the N0 initial sampling points; S300, obtain N1 = (2 - λ0) * m and N1 is the number of newly added sampling points in the first generation, m is the number of design variables in the design space of the preset vehicle collision, is the ratio of the first-generation uniform sampling and importance sampling, k is a preset value, k > 0; S400, obtaining N using the Latin hypercube method 1 1 newly added uniform sampling point of the first generation S500, obtaining N using the importance sampling method 2 1 newly added importance sampling point of the first generation S600, generate the first-generation Kriging model using the first-generation sampling points, where the first-generation sampling points include N0 initial sampling points, N 1 1 newly added uniform sampling point of the first generation and N 2 1 newly added importance sampling point of the first generation; S700, if the number of the first-generation sampling points Q1 ≥ Z, output the first-generation Kriging model and end the method; Z is the preset maximum number of sampling points, Q1 = N0 + N1; if Q1 < Z, obtain the number of the second-generation sampling points Q2 = Q1 + N2 and enter S800, where N2 is the number of newly added sampling points in the second generation; S800, if Q2 ≥ Z, output the second-generation Kriging model and end the method; if Q2 < Z, obtain the number of sampling points Q3 of the third generation = Q2 + N3, and so on until the number of sampling points Q of the nth generation n ≥ Z, output the nth-generation Kriging model and end the method; N3 is the number of newly added sampling points in the third generation, n ≥ 3.

2. The method according to claim 1, characterized in that, When the number S of output variables of the vehicle y equals 1, where y i is the simulation calculation value of the output variable at the i-th sampling point, and y pred i is the surrogate model prediction value of the i-th sampling point of the initial Kriging model corresponding to the output variable, and is the average value of the simulation calculation values of the output variable at all sampling points.

3. The method according to claim 1, wherein When the number S of output variables of the vehicle y is greater than 1, min is to take the minimum value, where y j,i is the simulation calculation value of the j-th output variable at the i-th sampling point, and y pred j,i is the surrogate model prediction value of the i-th sampling point of the initial Kriging model corresponding to the j-th output variable, is the average value of the simulation calculation values of the j-th output variable at all sampling points, and the value range of j is from 1 to S y .

4. The method according to claim 1, characterized in that In S100, N0 = 2*m + 1.

5. The method according to claim 1, characterized in that, In S300, 1 ≤ k ≤ 5.

6. The method according to claim 1, wherein In S700, Z ≥ 5*m.

7. The method according to claim 1, wherein In S800, Q n = Q n-1 + N n , N n = (2 - λ n-1 ) * m, λ n-1 is the accuracy of the (n - 1)-th generation Kriging model, N n is the number of newly added sampling points in the n-th generation, Q n-1 is the number of sampling points in the (n - 1)-th generation.

8. The method according to claim 7, wherein In S800, the method for obtaining the nth-generation Kriging model includes: S811, Obtain is the ratio of the n-th generation of uniform sampling and importance sampling; S821, obtaining N 1 n uniform sampling points newly added in the nth generation by using the Latin hypercube method, S831, obtaining N importance sampling points newly added in the n-th generation by using the importance sampling method 2 n of them, S841. Generate the n-th generation Kriging model using the n-th generation sampling points. The n-th generation sampling points include the (n - 1)-th generation sampling points, N 1 n evenly-spaced sampling points newly added in the n-th generation, and N 2 n importance sampling points newly added in the n-th generation.

Images