Multi-point approximation method used for structure optimization and based on approximate Hessian matrix
By adopting a multi-point approximation method based on the approximate Heisen matrix and the Euclidean distance point selection strategy in the optimization of spacecraft structure, the problem of low optimization efficiency caused by the large number of iterations in the traditional method is solved, and an efficient spacecraft structure design is achieved. Simplifies the analysis process and improves design efficiency.
Patent Information
- Application Number
- CN202410921156.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-10
AI Technical Summary
When traditional dimension optimization methods deal with large spacecraft structures, the number of iterations leads to too long structural analysis time and low optimization efficiency, which cannot meet the needs of modern aerospace missions.
Using a multi-point approximation method based on the approximation Heisen matrix, the Heisen matrix of the multi-point approximation function is calculated by the quasi-Newtonian method, and the neighborhood point selection strategy of Euclidean distance is used, and the optimization program is developed in combination with Nastran and Matlab to improve the complex structure. Dimensionally optimized design efficiency.
The number of structural analysis times during the spacecraft structure optimization process is significantly reduced, the optimization efficiency is improved, and the optimization results obtained are real solutions, which are suitable for constraints under static and dynamic responses, simplifying the design process.
Smart Images

Figure CN118965057A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a multi-point approximation method based on an approximate Hessian matrix for structural optimization. The method is aimed at complex spacecraft structures composed of rods, beams, plates, shells and combinations thereof, and belongs to the field of spacecraft structure and mechanism design. Background Art
[0002] Spacecraft structural design is a key link to ensure the success of space missions. It requires that the structure not only meet the requirements of the extreme space environment, but also be lightweight enough to reduce launch costs. In particular, in terms of lightweight and multi-environment adaptability, advanced optimization design technology is needed to improve the performance of spacecraft structures.
[0003] Structural optimization design usually involves complex nonlinear behavior, multidisciplinary coupling, and has a large number of design variables and complex constraints, making the solution of optimization problems challenging. With the development of technology and the improvement of spacecraft performance requirements, traditional size optimization design methods, especially the first-order methods integrated in commercial software, have gradually failed to meet the needs of modern space missions. For spacecraft structures with large finite element scales, optimization iterations may bring a large number of structural analysis times, which ultimately leads to excessive optimization time. Therefore, there is an urgent need for an efficient optimization design method (or optimization system) to improve the efficiency of spacecraft structure optimization.
[0004] The present invention proposes a multi-point approximation method based on approximate Hessian matrix for structural optimization. This method introduces approximate second-order information into the construction of multi-point approximate function for the first time. Specifically, the quasi-Newton method is used for the calculation of the Hessian matrix of the multi-point approximate function, thereby enhancing the approximation accuracy and improving the efficiency of the size optimization design of complex structures. Furthermore, it is proposed to adopt a neighborhood point selection strategy based on Euclidean distance to select known points for constructing the approximate function. Finally, based on the optimization strategy of sequential approximation, a set of engineering-usable spacecraft size optimization programs were jointly developed using Nastran and Matlab, which can be used to solve the size optimization design of spacecraft structures such as space trusses and satellites under typical design conditions. Summary of the invention
[0005] In order to solve the problem of long structural analysis time and low optimization efficiency caused by a large number of iterations in the optimization solution of large aerospace structures by traditional size optimization methods, the present invention proposes a new multi-point approximation method for the optimization design of spacecraft structures. It should be emphasized that the finite element model is a numerical representation established to accurately predict the physical behavior of the structure during the design and analysis process. The spacecraft structure finite element model described in the present invention is composed of rods, beams, plates, shells, and solid units. The optimization object is the properties of the rods, beams, plates, and shell units of the model, such as cross-sectional dimensions, plate thickness, node coordinates, etc.
[0006] The optimization method includes the following steps:
[0007] Step 1: Create a finite element model of the spacecraft based on the initial design of the spacecraft structure.
[0008] Step 2: Perform mechanical prediction of the spacecraft under typical working conditions. Analysis methods include static analysis, modal analysis, buckling analysis, transient response, etc.
[0009] The mechanical response value of the structure needs to be obtained after mechanical analysis, so as to clarify the current status of the structural design and establish the objective function or constraint conditions in the optimization problem. Static analysis refers to the calculation of the stress, strain and displacement of the spacecraft structure under static load. The basic finite element equation to be solved is:
[0010] KU=P
[0011] Among them, K is the global stiffness matrix, U is the node displacement vector, and P is the external load vector. The unit stress and other response values can be further obtained through the material stress-strain relationship and interpolation formula.
[0012] Modal analysis is used to determine the natural frequency and vibration mode of the structure, which is crucial to avoid resonance between the satellite and the rocket. The basic finite element equation to be solved is:
[0013] KΦ-ω 2 MΦ=0
[0014] Here, Φ is the matrix containing the mode shape vectors and ω is the eigenvalue (square root of the natural frequency).
[0015] Linear buckling analysis is a type of analysis used to evaluate the buckling behavior of a structure under static loads. The buckling load factors and mode shapes obtained from this analysis can be used to evaluate the deformation mode when the structure buckles. The basic finite element equations solved are:
[0016] KU-λK u U=0
[0017] Among them, K u is the geometric stiffness matrix of the structure under initial load, usually related to the node displacement vector U, and λ is the linear buckling load factor.
[0018] Transient response analysis is used to determine the dynamic response of a structure under time-varying loads. This analysis takes into account the time-varying loads and the inertia and damping characteristics of the structure, and the basic finite element equations to be solved are no longer given. Based on the above analysis, the mechanical response values of interest can be obtained to evaluate the rationality of the structural design and to establish subsequent optimization models.
[0019] Step 3: Establish a mathematical model for structural optimization, that is, clarify the optimization objectives, constraint functions and design variables.
[0020] The mathematical description of the original problem of spacecraft structure optimization with respect to continuous variables is:
[0021]
[0022] Where X is a continuous dimensional variable related to the structural member (such as the cross-sectional area of the member, the thickness of the plate, the coordinates of the node, etc.), k are the elements in it, and m is the number of continuous variables; are the upper and lower limits of the variable, respectively; f(X) is the structural quality and is the objective function of the optimization design; g j (X) is the jth structural property constraint, which is an implicit function of the design variables, such as stress constraint, displacement constraint, natural frequency constraint, and stability constraint, etc. 0 is the number of constraints. To eliminate the influence of the response size, the constraints corresponding to the structural response are normalized.
[0023] Step 4: Establish a sequence approximation problem that considers second-order information to approximate the original problem.
[0024] Use multi-point approximation function and select intermediate variables Establish a sequential explicit approximation problem for the original optimization problem. The approximation problem for the pth iteration is expressed as follows:
[0025]
[0026] In the formula, f (p) (X) and are the objective and constraint functions established by multi-point approximation technology at the pth iteration, and are unified by function w (p) (X) to indicate; J 1 is the number of critical constraints retained; and is x k The upper and lower limits at this time are and is x k The motion limit at the pth iteration, and the motion limit width of each design variable at the current point is generally taken as 20% of the variable interval width of each variable. Using the accumulated information of the known points, the multi-point approximation method considering the approximate second-order information proposed by the present invention is used to establish the following approximate function containing the second-order derivative information about the variable X:
[0027]
[0028] Where H is the number of known points, and its upper limit is H max, if the number of known points is greater than H max , then only the last H is retained max known points; X t is the tth known design point; intermediate variable Y t (X t ) is the value of the intermediate variable at the known point; w (p) (X) is the objective or constraint function established by the multi-point approximation technique at the pth iteration; w(X t ) is the function value of the objective or constraint function; is the Hessian matrix of the t-th known point at the p-th stage; h t (X) is the weight function; It is an approximate function that contains the first-order information of the known points. t (X) and The calculation method is as follows:
[0029]
[0030]
[0031] In the formula, w(X t )and They are the original function at the known point X t The function value and partial derivative at are obtained by semi-analytical method; X s And l is a substitution symbol in calculation and has no actual meaning; r t is an adaptive parameter that controls the nonlinearity of the approximate function and is obtained by solving the following least squares problem:
[0032]
[0033] In the formula, r t L ,r t U r t The upper and lower limits are generally r t0 =-1,r t L =-5,r t U =5.
[0034] As the structural optimization process proceeds, the adaptive parameter r needs to be corrected. t. In view of the fact that the least squares problem is more complicated after considering the second-order information, the interior point algorithm is used to solve it in the present invention. The interior point algorithm uses points inside the feasible domain to avoid dealing with boundary conditions and ensures the feasibility of the solution through the barrier function. The original optimization problem is made explicit into an approximate problem through the approximate function. The approximate problem is a nonlinear constrained programming problem. The interior point algorithm is still used to solve it in the present invention. The interior point algorithm is also more effective for solving problems with complex constraints and large-scale optimization problems.
[0035] Step 5: Use the quasi-Newton method to estimate the Hessian matrix of the multi-point approximation function.
[0036] Specifically, use To approximate Each iteration only needs to use the function value and first-order information to correct Initial Matrix In the absence of a better estimate, the identity matrix is usually taken. It should be noted that only the Hessian matrix of the current point (i.e., the Hth known design point) is corrected That is (When t=1,2,...,H-1 and H≥2). Based on the quasi-Newton condition, the BFGS correction formula for the Hth known point is derived as follows:
[0037]
[0038] In the formula, and are the approximate Hessian matrices of the Hth known design point at the p stage and the p-1 stage respectively; s (p -1) is the increment of the intermediate variable at adjacent iteration points; (p-1) is the gradient change of the intermediate variable at adjacent iteration points, which has the following form:
[0039] s (p-1) =Y H (X (p) )-Y H-1 (X (p-1) )
[0040]
[0041] Where, X (p) is the initial point of the pth iteration, i.e., the suboptimal point obtained in the p-1th iteration; r H and r H-1are the adaptive parameters of the Hth known point and the H-1th known point respectively; the conventions of other parameters are the same as above. Quasi-Newton correction is applicable to the approximate calculation of the second-order derivatives of frequency constraints (or eigenvalues) and node displacements. Since stress constraints are often accompanied by a large number of finite element stress response values, when the constraints are deleted, the finite element stress response values may be retained or deleted in adjacent iterations, so the stress constraint cannot use this scheme to approximate the Hessian matrix. In practice, the stress constraint can be is a zero matrix.
[0042] Step 6: Select known points using the neighborhood point selection strategy based on Euclidean distance.
[0043] As the iteration proceeds, the number of known points will gradually increase, and we need to consider which point information to select to construct the multi-point approximate function. First, we need to set the upper limit H of the number of known points to be selected. max This is because if too many known points are used, it will cause errors in the calculation of nonlinear adaptive parameters and the reference value of the information between design points that are relatively far away is not very significant. Therefore, this method proposes a neighborhood point selection strategy based on Euclidean distance. First, the concept of boundary circle is introduced. Figure 1 An example of the iterative process shown, X (9) is the latest suboptimal value. Take the latest H max =5 points to construct a new approximate function (i.e. select X (5) ~X (9) ), where the distance X (9) The farthest point is X (6) So build with X (9) As the center of the circle, from its (6) The Euclidean distance of the boundary circle is the radius. Obviously, the previous iteration point X (2) , X (4) Also within the boundary circle, a reasonable suggestion is to use X (2) Replace X (6) To construct a multi-point approximate function, this is because the points closer to the current optimal point are more valuable for reference.
[0044] For optimization problems involving 3D and higher dimensional design variables, the boundary circle can be extended to a 3D and higher dimensional spherical domain. Based on the above description and analysis, the method of selecting known points in the present invention is: H≤H max When , the known points to be selected are increased in sequence; H ≥ H max , follow these steps:
[0045] (1) Let the latest point be X (p) , select the latest 5 known points to establish the known point set A;
[0046] (2) Calculate the maximum envelope radius X t ∈A, and establish the bounding sphere and divide the distance X (p) The farthest point is denoted by X t,max (The point is on the bounding sphere);
[0047] (3) Determine whether there are other known points in the sphere that do not belong to A. If so, record the set of other known points as B and go to step (4); if not, go to step (5);
[0048] (4) Find the distance X in B (p) The nearest known point is added to the set A and X t,max Remove and go to step (2);
[0049] (5) After the known points are selected, output set A.
[0050] During the iteration of the sequence approximation problem (step 4), the quasi-Newton method is used to estimate the Hessian matrix of the multi-point approximation function (step 5) and the known points are selected based on the neighborhood point selection strategy based on Euclidean distance (step 6).
[0051] Step 7: Convergence judgment and establishment of automatic optimization system.
[0052] In order to improve the efficiency of spacecraft structural optimization design, the entire optimization iteration must be guaranteed to be automatic. When the difference / relative difference of the objective function of adjacent iterations is less than the tolerance value and the constraints are met, the iteration is considered to have converged. The operation logic of the optimization system is: complete the initial modeling of the structural finite element in Patran, call Nastran for analysis by submitting the bdf file; establish the main program in Matlab, including modules such as convergence judgment, establishment of approximate functions, optimizer optimization, and visualization of optimization results; in addition, external files are required to store the iteration results. Figure 2 The development framework of the optimization platform is given. Some core commands or functions in the program system are given below, and the input, output and calling relationship are explained in detail.
[0053] (1)[optx,fval]=fmincon(@objfun,x0,[],[],[],[],move_lower,move_upper,@constraint,options);
[0054] This function is used to optimize the approximate problem, and the interior point algorithm is used by default. The input items include the handle function @objfun and @constraint, which are the optimization objective and constraint functions respectively, established by the multi-point approximation method proposed in the present invention; x0 is the initial value of the design variable; move_lower is the lower limit vector in the variable motion limit; move_upper is the upper limit vector in the variable motion limit; option is a structure used to define the convergence tolerance or single-step display content. In the output items, optx is the optimal design variable of the current approximate problem, and fval is the optimal objective function value of the current approximate problem. In each iteration of the approximate problem, the motion limit value needs to be updated and this function needs to be called.
[0055] (2)[rsp(LOOP,:),drsp(LOOP,:,:)]=Sensitivity_anaysis(optx,lower_limit,upper_limit);
[0056] After obtaining the optimized value, this function needs to be called for structural analysis. In the input items, optx is the optimal design variable of the current approximate problem (same as above). In the output items, LOOP is the number of iterations; rsp(LOOP,:) is the actual structural response value obtained after the LOOPth approximate problem optimization iteration. Similarly, drsp(LOOP,:,:) is the sensitivity value. An example of a call statement for commercial software is system('D:\MSCsoftware\Nastran2018\bin\nastran.exe"',bdf_name,'.bdf'],'-echo'), where the variable bdf_name is the file name.
[0057] (3)[ap,ap_dg]=Approximate_Function(in_x,x,g,dg,r);
[0058] The core of this function is the multi-point approximation method based on the approximate Hessian matrix proposed by the present invention, which is used to establish the approximate function of the objective function and the constraint function. In the input item, in_x is the design variable in the internal optimization process of the approximate problem, x is a matrix containing the design variable values of known points; g is a vector containing the constraint function values at known points; dg is a matrix containing the sensitivity of the constraint function to the design variable at known points; r stores the nonlinear adaptive parameter values in the multi-point approximate function. In the output item, ap is the objective / constraint function value obtained based on the approximate function when the design variable takes the value of in_x; ap_dg is the approximate sensitivity value. By calling this function, it is avoided that the structural analysis program is called for each iteration of the optimization process, thereby reducing the entire optimization time and improving the design efficiency.
[0059] Step 8: Method comparison, analysis and iterative explicitness. An optimization example is set to illustrate the general process from spacecraft design to analysis and then to optimization. The effectiveness of the method proposed in the present invention is verified by comparing the results with the optimization methods of typical commercial software. Through the multi-point approximation method based on the approximate Hessian matrix proposed in the present invention, the optimal value of the objective function and the value of the constraint function can be obtained, and the iteration curve of the objective function can be drawn.
[0060] The advantages and beneficial effects of the present invention are as follows: the method / optimization strategy mentioned in the present invention can improve the approximate accuracy of the approximate function, improve the optimization efficiency, and is suitable for the mixed optimization of size and shape. Based on the method mentioned in the present invention, a practical optimization system is established after secondary development to ensure that the entire optimization iteration is automatic. In particular, for large-scale plates, beams and their combined structures in the aviation and aerospace fields, this method can significantly reduce the number of structural analyses in the optimization process, thereby reducing the time-consuming optimization design of engineering structures. This method can handle the constraints under static and dynamic responses, and the obtained optimization results are true solutions, which can be directly applied to actual structural design without particularly complex post-processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 An example of an iterative process for optimization.
[0062] Figure 2 A development framework for optimizing the platform.
[0063] Figure 3 It is a 72-rod truss structure.
[0064] Figure 4 Iteration history of the objective function in Example 1.
[0065] Figure 5 This is the initial structure of the 37-rod hybrid frame.
[0066] Figure 6 It is the optimal design of 37-rod planar hybrid frame.
[0067] Figure 7a , Figure 7b , Figure 7c , Figure 7d This is the optimized configuration mode obtained by the multi-point approximation method proposed in the present invention.
[0068] Figure 8 Iteration history of the objective function in Example 2.
[0069] Figure 9a , Figure 9b Satellite modal analysis results.
[0070] Fig.10Comparison of iterative curves for whole star mass.
[0071] Fig.11 Comparison of fundamental frequency iteration curves. DETAILED DESCRIPTION
[0072] The following is combined with Figure 1-11 The following table describes in detail the specific implementation of the multi-point approximation method based on the approximate Hessian matrix, and the optimization objects are typical truss structures and satellite structures. The following description includes 3 embodiments, which are only used to illustrate the present invention, but not to limit the scope of the present invention.
[0073] Example 1: Optimization of the size of a seventy-two-bar truss
[0074] Truss structure is also one of the classic forms of spacecraft structure. This example is a complex truss structure with 72 rods and a four-layer skeleton, which is used to further test the approximate effect and reliability of the multi-point approximation method based on the approximate Hessian matrix.
[0075] Step 1: Based on the initial design of the spacecraft structure, a finite element model of the spacecraft is established. The finite element model of the truss structure is established in the commercial software Patran and simulated using rod elements. The truss dimensions and rod numbers are as follows: Figure 3 As shown. The elastic modulus of the material is E = 68.971 GPa, and the density is ρ = 2767.991 kg / m 3 The structure is subjected to two loading conditions, as shown in Table 1.
[0076] Table 1 Load conditions of 72-bar truss
[0077]
[0078] Note: 22241.1N is 5 kips (kilopound force, imperial unit)
[0079] Step 2: Perform mechanical prediction of the spacecraft under typical working conditions. The analysis method in this example includes static analysis. The maximum response displacement is u under working condition 1. 1x =u 1y =9.774mm.
[0080] Step 3: Establish a mathematical model for structural optimization, that is, clarify the optimization objectives, constraint functions and design variables. The size optimization problem is described as:
[0081] Objective function: Minimize the total mass of the 72-bar truss
[0082] Design variables: Considering the chain relationship of variables, the truss section is divided into 16 groups, and the cross-sectional area of the rods in each group is the same. The chain relationship of variables is shown in Table 2. In the optimization design, the initial cross-sectional area is 322.58mm 2 And Ai ≥64.516mm 2 ,i=1,2,…,16;
[0083] Constraints: The x- and y-direction displacements of the topmost nodes 1, 2, 3, and 4 of the structure are all less than ±6.35 mm; the allowable tensile and compressive stresses of all rods are 172.4 MPa;
[0084] Step 4: Establish a sequence approximation problem that considers second-order information to approximate the original problem. Take the upper limit H of the known points max =4, using the accumulated information of known points, the multi-point approximation method proposed by the present invention considering approximate second-order information is used to establish an approximate function containing approximate second-order derivative information about the variable X, and the approximate function is an approximation for the node displacement. At the known point X t The function value w(X t ) and partial derivative values Obtained by Nastran analysis.
[0085] Step 5: Estimate the Hessian matrix of the multi-point approximate function using the quasi-Newton method. Use the BFGS correction formula of the Hth known point given in the present invention to approximately calculate the second-order derivative of the node displacement.
[0086] Step 6: Select known points using the neighborhood point selection strategy based on Euclidean distance. It should be noted that only the Hessian matrix of the current point (i.e., the Hth known design point) is corrected. That is (t=1,2,…,H-1 and H≥2). In the iterative process of the sequence approximation problem (step 4), it is necessary to use the quasi-Newton method to estimate the Hessian matrix of the multi-point approximation function (step 5), and select known points based on the neighborhood point selection strategy based on Euclidean distance (step 6), and the maximum number of known points selected is H max .
[0087] Step 7: Convergence determination and establishment of automatic optimization system. Set the maximum number of iterations to 20 and use Figure 2 The optimization platform given is used to automatically update the design variables.
[0088] Step 8: Method comparison, analysis and iterative explicitness. The mass of the truss at the initial optimization point is 193.48 kg. The initial point violates all displacement constraints under condition 1, and the maximum response displacement u 1x =u 1y=9.774mm, so the initial point is an infeasible point. The quasi-Newton method for correcting the Hessian array is compared with Ray optimization, Adaptation of the method of inscribed hypersphere (AIP), and Nastran built-in optimization algorithm.
[0089] Table 2 Variable chaining of 72-bar trusses
[0090]
[0091] Figure 4 is the iteration history of the objective function, and Table 3 is the statistics of the optimal design values of this truss under different optimization methods. From the optimization results of the objective function, the method proposed in the present invention has higher calculation efficiency, the least number of structural analysis and the lightest result.
[0092] Table 3 Statistics of optimal design values of seventy-two-bar trusses
[0093]
[0094] Note: Design variable A in the table i Unit: mm 2 ; No.S represents the number of structural analyses; The displacement constraint results of Nastran's built-in optimization are slightly violated.
[0095] Example 2: Shape and size hybrid optimization of a planar hybrid structure
[0096] The 37-bar planar hybrid frame is a standard problem for many scholars to study shape optimization. Structural optimization with frequency constraints is a challenging problem because it involves a highly nonlinear, discontinuous and non-convex search space consisting of multiple local optima. This example studies the performance of the multi-point approximation method based on the quasi-Newton corrected Hessian matrix in shape and size hybrid optimization.
[0097] Step 1: Establish a finite element model of the spacecraft based on the initial design of the spacecraft structure. The geometry and support conditions of the 37-rod planar hybrid frame are as follows: Figure 5 As shown in the figure, node 1 is a hinged support and node 20 is a sliding hinged support. The lower chord member is modeled with a rectangular cross-section beam element with B = 8 cm and H = 5 cm, and 10 kg of non-structural mass is added at each node; the upper chord member is modeled with a simple bar element, and the initial cross-sectional area is A = 1 cm 2 . Material elastic modulus E = 201GPa, material density ρ = 7800kg / m 3 , Poisson's ratio ν = 0.3.
[0098] Step 2: Perform mechanical prediction of the spacecraft under typical working conditions. The analysis method in this example is modal analysis.
[0099] Step 3: Establish a mathematical model for structural optimization, that is, clarify the optimization objectives, constraint functions and design variables. The size optimization problem is described as:
[0100] The objective function is to minimize the mass of the hybrid frame, and the initial mass is 336.3 kg;
[0101] The design variables include 14 size variables, 5 shape variables and symmetry. The size variable is the cross-sectional area of the upper chord, with a value range of 1×10 -4 ~10×10 -4 m 2 And satisfy A 1 =A 27 , A 2 =A 26 , A 3 =A 24 , A 4 =A 25 , A 5 =A 23 , A 6 =A 21 , A 7 =A 22 , A 8 =A 20 , A 9 =A 18 , A 10 =A 19 , A 11 =A 17 , A 12 =A 15 , A 13 =A 16 ; The shape variable is the y coordinate of the upper chord member, ranging from 1.0 to 3.0 m and satisfying y 3 =y 19 ,y 5 =y 17 ,y 7 =y 15 ,y 9 =y 13 ;
[0102] The constraints are the first three natural frequency constraints: f 1 ≥20Hz, f 2 ≥40Hz, f 3 ≥60Hz;
[0103] Step 4: Establish a sequence approximation problem that considers second-order information to approximate the original problem. Similarly, take the upper limit H of the known points max=4, using the accumulated information of the known points, the multi-point approximation method proposed by the present invention considering the approximate second-order information is used to establish an approximate function containing the approximate second-order derivative information about the variable X, and the approximate function is approximated for the first three frequencies. t The function value w(X t ) and partial derivative values Obtained by Nastran analysis.
[0104] Step 5: Estimate the Hessian matrix of the multi-point approximate function using the quasi-Newton method. Use the BFGS correction formula of the Hth known point given in the present invention to approximately calculate the second-order derivatives of each order frequency.
[0105] Step 6: Select known points using the neighborhood point selection strategy based on Euclidean distance. Similarly, only correct the Hessian matrix of the current point (i.e., the Hth known design point) That is (t=1,2,…,H-1 and H≥2). In the iterative process of the sequence approximation problem (step 4), it is necessary to use the quasi-Newton method to estimate the Hessian matrix of the multi-point approximation function (step 5), and select known points based on the neighborhood point selection strategy based on Euclidean distance (step 6), and the maximum number of known points selected is H max .
[0106] Step 7: Convergence determination and establishment of automatic optimization system. Set the maximum number of iterations to 40 and use Figure 2 The optimization platform given is used to automatically update the design variables.
[0107] Step 8: Method comparison, analysis and iterative explicitness. The optimization methods include HS (HarmonySearch algorithm based on meta-inspiration), HSCFA (hybrid sine-cosine firefly algorithm) and the multi-point approximation method based on quasi-Newton corrected Hessian matrix proposed in the present invention. The optimal design obtained by the method proposed in the present invention is as follows: Figure 6 As shown in the figure, the first four modes after optimization are as follows Figure 7a , Figure 7b , Figure 7c , Figure 7d As shown, Figure 7a In, f 1 =19.992Hz, Figure 7b In, f 2 =40.004Hz, Figure 7c In, f 3 =60.001Hz, Figure 7d In, f 4 =72.507Hz.
[0108] The various algorithms based on metaheuristics were calculated repeatedly 20 times, and the best values and average values were recorded in Table 4. It is not difficult to see that the metaheuristic algorithm requires thousands of structural analyses, which is time-consuming. The solution efficiency of the method proposed in the present invention is extremely high, and only 27 structural analyses are required to converge. The structural quality after optimization is significantly better than the calculation result of HS but slightly higher than the result of HSCFA. Table 5 shows the comparison of the natural frequencies of this planar hybrid frame after optimization by various optimization algorithms. Figure 8 The iteration history of the objective function of the method proposed in the present invention is given.
[0109] Table 4 Comparison of optimized designs of 37-rod plane hybrid frame
[0110]
[0111]
[0112] Note: The unit of A is cm 2 , the unit of y is m; N / A means no corresponding result; No.S represents the number of structural analysis.
[0113] Table 5 Comparison of natural frequencies of 37-rod plane hybrid frames
[0114]
[0115]
[0116] Example 3: Optimization design of box-plate satellite
[0117] The initial mass of a certain satellite is 173.2 kg. Its side panels, top panels, bottom panels and internal partitions all adopt honeycomb sandwich structures. The method proposed in the present invention is used to optimize the design of satellite size, which can reflect the engineering value of the method.
[0118] Step 1: Establish a finite element model of the spacecraft based on the initial design of the spacecraft structure. Use Patran to establish a finite element model of the entire spacecraft, which is mainly composed of plate and beam elements, and contains a total of 142,010 nodes and 141,430 elements.
[0119] Step 2: Perform mechanical prediction of the spacecraft under typical working conditions. The analysis method in this example is modal analysis. The modal analysis uses the boundary conditions of the fixed support of the middle node of the bottom adapter. The first-order frequency of the satellite is 49.779 (base frequency) and the second-order frequency is 54.615Hz. Figure 9a , Figure 9b Shown are the vibration shape results for the first two modes of the satellite.
[0120] Step 3: Establish a mathematical model for structural optimization, that is, clarify the optimization objectives, constraint functions and design variables. The description of the whole satellite size optimization problem is as follows:
[0121] Objective function: Minimize the mass of the entire star;
[0122] Design variables: Core thickness T of satellite side panels (4), top panel, bottom panel, internal X-direction partition, and internal Z-direction partition 2 , upper and lower panel thickness T 1 and T 3 , a total of 24 continuous size variables. The value range is 0.1≤T 1 ,T 3 ≤1.0mm, 10.0≤T 2 ≤40.0mm, variable names and initial optimization values are shown in Table 6;
[0123] Constraints: The fundamental frequency of the entire satellite is ≥ 50Hz;
[0124] Step 4: Establish a sequence approximation problem that considers second-order information to approximate the original problem. Similarly, take the upper limit H of the known points max =4, using the accumulated information of known points, the multi-point approximation method proposed by the present invention considering approximate second-order information is used to establish an approximate function containing approximate second-order derivative information about the variable X, and the approximate function is an approximation for the frequency of the satellite structure. t The function value w(X t ) and partial derivative values Obtained by Nastran analysis.
[0125] Step 5: Estimate the Hessian matrix of the multi-point approximate function using the quasi-Newton method. Use the BFGS correction formula of the Hth known point given in the present invention to approximately calculate the second-order derivative of the fundamental frequency of the entire satellite.
[0126] Step 6: Select known points using the neighborhood point selection strategy based on Euclidean distance. Similarly, only correct the Hessian matrix of the current point (i.e., the Hth known design point) That is (t=1,2,…,H-1 and H≥2). In the iterative process of the sequence approximation problem (step 4), it is necessary to use the quasi-Newton method to estimate the Hessian matrix of the multi-point approximation function (step 5), and select known points based on the neighborhood point selection strategy based on Euclidean distance (step 6), and the maximum number of known points selected is H max .
[0127] Step 7: Convergence determination and establishment of automatic optimization system. Set the maximum number of iterations to 50 and use Figure 2 The optimization platform given is used to automatically update the design variables.
[0128] Step 8: Method comparison, analysis and iterative explicitness. The above problem is solved by using the Nastran built-in optimization method and the multi-point approximation method based on the quasi-Newton corrected Hessian matrix proposed in the present invention. Fig.10 and Fig.11 They are the iterative curves of the whole star mass and fundamental frequency, and the optimization results are shown in Table 6. The Nastran solution converged after 40 iterations, the number of structural analysis was large and obvious sawtooth phenomenon appeared in the iteration process. The multi-point approximation method converged after 13 iterations, and the whole star mass was slightly lower, indicating that the convergence effect of the method proposed in the present invention is better.
[0129] The optimization results obtained by the multi-point approximation method based on the quasi-Newton corrected Hessian matrix proposed in the present invention are used to design and adjust the satellite structure. After the ground-based star system test, the satellite fundamental frequency only drifted slightly and the characteristic curves of typical positions on the main structure overlapped well, indicating that the mechanical properties of the satellite did not change significantly. The satellite has been successfully launched and has undergone the mechanical environment assessment of the active segment.
[0130] Table 6 Physical meaning, initial values and optimization results of optimization variables
[0131]
[0132]
Claims
1. A multi-point approximation method based on approximate Hessian matrix for structural optimization, characterized by: The steps include: Step 1: Establish a finite element model of the spacecraft based on the initial design of the spacecraft structure; Step 2: Perform mechanical prediction of the spacecraft under typical working conditions, using analysis methods including static analysis, modal analysis, buckling analysis, and transient response; Step 3: Establish a mathematical model for structural optimization, that is, clarify the optimization objectives, constraint functions and design variables; Step 4: Establish a sequence approximation problem that considers second-order information to approximate the original problem; Step 5: Use the quasi-Newton method to estimate the Hessian matrix of the multi-point approximate function; Step 6: Select known points using the neighborhood point selection strategy based on Euclidean distance; Step 7: Convergence determination and establishment of automatic optimization system; Step 8: Method comparison, analysis and iterative explicitness; An optimization example is set up to illustrate the process from spacecraft design to analysis and then to optimization, and the effectiveness is verified by comparing the results with those of commercial software optimization methods.
2. A multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 2, static analysis refers to the calculation of the stress, strain and displacement of the spacecraft structure under static load. The basic finite element equation to be solved is: KU=P Among them, K is the global stiffness matrix, U is the node displacement vector, and P is the external load vector; the unit stress response value is obtained through the material stress-strain relationship and interpolation formula; Modal analysis is used to determine the natural frequencies and vibration modes of a structure. The basic finite element equation to be solved is: KF-oh 2 MΦ=0 Where Φ is the matrix containing the vibration mode vectors and ω is the eigenvalue; Buckling analysis is used to evaluate the buckling behavior of a structure under static loads. The buckling load factor and mode shape obtained are used to evaluate the deformation mode of the structure when it buckles. The basic finite element equation to be solved is: KU-λK u U=0 Among them, K u is the geometric stiffness matrix of the structure under initial load, which is related to the node displacement vector U, and λ is the linear buckling load factor.
3. A multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 3, the mathematical description of the original problem of spacecraft structure optimization with respect to continuous variables is: In the formula, X is a continuous dimensional variable related to the structural part, x k are the elements in it, and m is the number of continuous variables; are the upper and lower limits of the variable, respectively; f(X) is the structural quality and is the objective function of the optimization design; g j (X) is the jth structural property constraint, and J0 is the number of constraints.
4. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 4, a multi-point approximation function is used and the intermediate variables are selected Establish a sequential explicit approximation problem for the original optimization problem; the approximation problem for the pth iteration is expressed as follows: In the formula, f (p) (X) and are the objective and constraint functions established by the multi-point approximation method at the pth iteration, and are unified by the function w (p) (X) represents; J1 is the number of critical constraints retained; and is x k The upper and lower limits at this time are and is x k The motion limit at the pth iteration, and the motion limit width of each design variable at the current point is taken as 20% of the variable interval width of each variable.
5. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 4, characterized in that: Using the accumulated information of known points, the following approximate function containing second-order derivative information about the variable X is established: Where H is the number of known points, and its upper limit is H max , if the number of known points is greater than H max , then only the last H is retained max known points; X t is the tth known design point; intermediate variable Y t (X t ) is the value of the intermediate variable at the known point; w (p) (X) is the objective or constraint function established by the multi-point approximation method at the pth iteration; w(X t ) is the function value of the objective or constraint function; is the Hessian matrix of the t-th known point at the p-th stage; h t (X) is the weight function; It is an approximate function containing the first-order information of the known points; h t (X) and The calculation method is as follows: In the formula, w(X t )and They are the original function at the known point X t The function value and partial derivative at are obtained by semi-analytical method; X s And l is a substitution symbol in calculation and has no actual meaning.
6. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 5, characterized in that: r t is an adaptive parameter that controls the nonlinearity of the approximate function and is obtained by solving the following least squares problem: In the formula, r t The upper and lower limits of r t0 =-1, As the structural optimization process proceeds, the adaptive parameter r needs to be corrected. t ; Since the least squares problem is more complicated after considering the second-order information, the interior point algorithm is used to solve it.
7. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 5, the Hessian matrix of the multi-point approximation function is estimated using the quasi-Newton method; use To approximate Each iteration only needs to use the function value and first-order information to correct Initial Matrix Correct only the Hessian matrix of the current point That is When t=1,2,...,H-1 and H≥2; the BFGS correction formula for the Hth known point is derived based on the quasi-Newton condition: In the formula, and are the approximate Hessian matrices of the Hth known design point at the p stage and the p-1 stage respectively; s (p-1) is the increment of the intermediate variable at adjacent iteration points; (p-1) is the gradient change of the intermediate variable at adjacent iteration points, which has the following form: s (p-1) =And H (X (p) )-AND H-1 (X (p-1) ) In the formula, X (p) is the initial point of the pth iteration, i.e., the suboptimal point obtained in the p-1th iteration; r H and r H-1 are the adaptive parameters of the Hth known point and the H-1th known point respectively.
8. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 6, the known points are selected using the neighborhood point selection strategy based on Euclidean distance; Set the upper limit H of the number of known points to be selected max , X (9) is the latest suboptimal value; take the latest H max =5 points to construct a new approximate function, that is, select X (5) ~X (9) , where the distance X (9) The farthest point is X (6) So build with X (9) As the center of the circle, from its (6) The Euclidean distance of the boundary circle is the radius; the previous iteration point X (2) , X (4) Also within the bounding circle, use X (2) Instead of X (6) To construct a multi-point approximation function.
9. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 8, characterized in that: For optimization problems involving 3D and higher dimensional design variables, the method for selecting known points is: H≤H max When , the known points to be selected are increased in sequence; H ≥ H max The steps are as follows: (1) Let the latest point be X (p) , select the latest 5 known points to establish the known point set A; (2) Calculate the maximum envelope radius X t ∈A, and establish the bounding sphere and divide the distance X (p) The farthest point is denoted by X t,max ; (3) Determine whether there are other known points in the sphere that do not belong to A; if so, record the set of other known points as B and go to step (4); if not, go to step (5); (4) Find the distance X in B (p) The nearest known point is added to the set A and X t,max Remove and go to step (2); (5) After the known points are selected, output set A.
10. The multi-point approximation method based on approximate Hessian matrix for structural optimization according to claim 1, characterized in that: In step 7, the initial modeling of the structural finite element is completed in Patran, and Nastran is called for analysis by submitting the bdf file; the main program is established in Matlab, including convergence judgment, establishment of approximate functions, optimizer optimization, and optimization result visualization modules; in addition, external files are required to store iterative results; some core commands or functions in the program system are given below, and the input, output and calling relationships are explained in detail; (1)[optx,fval]=fmincon(@objfun,x0,[],[],[],[],move_lower,move_upper,@constraint,options); This function is used to optimize the approximate problem. The default is to use the interior point algorithm. The input items include the handle function @objfun and @constraint, which are the optimization objective and constraint function respectively. x0 is the initial value of the design variable. move_lower is the lower limit vector in the variable motion limit. move_upper is the upper limit vector in the variable motion limit. option is a structure used to define the convergence tolerance or single-step display content. In the output items, optx is the optimal design variable of the current approximate problem, and fval is the optimal objective function value of the current approximate problem. In each iteration of the approximate problem, the motion limit value needs to be updated and this function needs to be called. (2) [rsp(LOOP,:), drsp(LOOP,:,:)] = Sensitivity_anaysis(optx, lower_limit, upper_limit). After obtaining the optimized value, this function needs to be called for structural analysis; in the input item, optx is the optimal design variable of the current approximate problem; in the output item, LOOP is the number of iterations; rsp(LOOP,:) is the actual structural response value obtained after the LOOP-th approximate problem optimization iteration, and drsp(LOOP,:,:) is the sensitivity value. The example of the call statement for commercial software is system('D:\MSCsoftware\Nastran2018\bin\nastran.exe"',bdf_name,'.bdf'],'-echo'), where the variable bdf_name is the file name; (3)[ap,ap_dg]=Approximate_Function(in_x,x,g,dg,r); Used to establish approximate functions of objective functions and constraint functions; in the input item, in_x is the design variable in the internal optimization process of the approximate problem, x is a matrix containing the design variable values at known points; g is a vector containing the constraint function values at known points; dg is a matrix containing the sensitivity of the constraint function to the design variables at known points; r stores the nonlinear adaptive parameter values in the multi-point approximate function; in the output item, ap is the objective / constraint function value obtained based on the approximate function when the design variable takes the value of in_x; ap_dg is the approximate sensitivity value.
Citation Information
Patent Citations
Spacecraft transient thermal analysis model inversion correction method
CN105930676A
Apparatus and method for design by using sequential approximate optimization(SAO) and record media recorded program for realizing the same
KR1020080082285A
Integration method for accurate modeling and analysis and reliability-based design optimization of variable stiffness composite plate and shell structures
US20190080040A1
High-dimensional data analysis
WO2011159255A2
AU2020103808A4
