A solidification distortion optimization method based on adaptive enumeration-annealing

Abstract

The application provides a curing deformation optimization method based on adaptive enumeration-annealing, comprising the following steps: S1: obtaining the semi-paving layer number N of a laminated plate and the discrete angle category number K, and calculating the full arrangement scale K^N; if K^N is less than a preset enumeration threshold T, it is determined as being enumerable, and S2 enumeration optimization is entered; otherwise, it is determined as being unenumerable, and S3 annealing optimization is entered; S2: all semi-paving layer sequences meeting process constraints are generated through systematization, and a finite element simulation model and a simulation history database are comprehensively utilized to obtain a curing deformation calculation result, so that a global optimal layer scheme is ensured to be obtained; S3: a simulated annealing algorithm is used to efficiently search in a huge solution space, and a process constraint discriminator and a simulation history database are utilized to completely eliminate invalid and repeated deformation evaluations.

Description

Technical Field

[0001] This invention belongs to the field of composite material structure optimization design technology, and relates to a curing deformation optimization method based on adaptive enumeration-simulated annealing. Background Technology

[0002] During the curing process, composite laminates often exhibit warping and springback deformations after demolding due to the anisotropic chemical shrinkage of the resin and the thermal expansion of the fiber-matrix, severely affecting assembly accuracy and structural performance. Controlling curing deformation by optimizing the layup angle and stacking sequence is a core technical requirement in the manufacturing of composite components.

[0003] Ply optimization is a high-dimensional discrete combinatorial problem. Taking the commonly used angles of 0°, +45°, -45°, and 90° as examples, when the number of half-ply layers in a symmetrical laminate is N, the theoretical total number of ply sequences is 4^N. When N=12, the scale reaches 1.6×10^7, and when N=24, it exceeds 2.8×10^14, with the solution space expanding dramatically. Meanwhile, engineering specifications impose several strict constraints on ply sequences: the number of ply layers at ±45° angles must be equal (balance), and the number of consecutive layers at the same angle cannot exceed 4, etc. Finding the ply sequence that satisfies all constraints and minimizes curing deformation in such a huge discrete space is extremely challenging.

[0004] Existing optimization methods mostly rely on finite element simulation, which takes several minutes or even tens of minutes to evaluate the solidification deformation of each scheme, and the optimization cycle often reaches tens of hours. The main problems are as follows: a large number of sequences that do not meet the process constraints are subject to invalid prediction; repeated sequences are repeatedly evaluated, resulting in computational waste; the optimization algorithms themselves (such as genetic algorithms) require population maintenance and complex operations, parameter tuning is cumbersome, and they cannot guarantee exhaustive search for the global optimum for small-scale problems. Summary of the Invention

[0005] This invention provides a curing deformation optimization method based on adaptive enumeration-simulated annealing, which can solve the technical problems of high computational cost and low optimization efficiency of existing methods.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A curing deformation optimization method based on adaptive enumeration-simulated annealing, characterized by comprising: S1: Obtain the number of half-layout layers N and the number of discrete angle types K of the laminate, and calculate the permutation scale K^N; if K^N≤ preset enumeration threshold T, it is determined to be enumerable, and proceed to S2 enumeration optimization; otherwise, it is determined to be unenumerable, and proceed to S3 simulated annealing optimization. S2: Generate all half-layout sequences that satisfy the constraints through a generation algorithm with process constraint pruning to form a feasible sequence set; for each sequence in the sequence set, query the simulation history database. If a match is found, directly read the maximum curing deformation amount; otherwise, call the finite element simulation model to directly calculate the maximum curing deformation amount and store it in the simulation history database; after traversing all sequences, output the sequence with the smallest deformation as the optimal layup scheme. S3: Initialize a feasible sequence as the current solution and the global optimal solution, and set the initial temperature, termination temperature, cooling factor, and number of isothermal iterations; perform multiple neighborhood searches at each temperature, and after generating a new sequence, check its compliance with the process constraint discriminator. If it is not feasible, set a maximum value for the maximum curing deformation. If it is feasible, query the simulation history database through a unique identifier to obtain the deformation value; use the Metropolis criterion to determine whether to accept the new solution and update the global optimal solution; decrease the temperature according to the cooling factor until the termination condition is met, and output the global optimal layup scheme.

[0007] In summary, this application provides a curing deformation optimization method based on adaptive enumeration-simulated annealing, with the following technical effects: Adaptive strategies guarantee efficiency and optimality: for small-scale problems, exhaustive search is performed to obtain the true global optimum; for large-scale problems, simulated annealing is used for single-point iteration, which has fewer parameters, is easy to implement, and can probabilistically escape local minima.

[0008] Computational overhead is minimized: the pre-process constraint discriminator filters out all infeasible sequences before calculating deformation, and the simulation history database eliminates finite element simulation calculations for duplicate sequences. In the 48-layer board optimization test, about 60% of the candidate sequences were intercepted by the constraint discriminator, and the remaining sequences were further reduced by more than 70% due to historical reuse, resulting in a reduction of more than an order of magnitude in the total number of simulation calls.

[0009] Superior convergence performance: Simulated annealing combined with neighborhood perturbation and the Metropolis criterion steadily approaches the global optimum within the feasible region, avoiding premature convergence and parameter tuning burden in population algorithms.

[0010] It has extremely strong engineering applicability: the entire process is fully automated, requiring no manual intervention, and can be directly embedded into the digital design platform for composite material components. Attached Figure Description

[0011] Figure 1 The above is a model diagram of an embodiment provided in this application.

[0012] Figure 2 The optimization iteration curve in Embodiment 1 provided in this application.

[0013] Figure 3 The optimization iteration curve in Embodiment 2 provided in this application. Detailed Implementation

[0014] This application provides a curing deformation optimization method based on adaptive enumeration-simulated annealing. This method adaptively switches between exhaustive and heuristic search based on the problem size, and thoroughly eliminates redundant computations through strict pre-constraint filtering and historical simulation reuse. Furthermore, simulated annealing, as a simple, theoretically guaranteed global convergence single-point search method, has fewer parameters and is easier to implement than genetic algorithms. If it can be organically integrated with adaptive strategies and computational reuse mechanisms, it will provide a more efficient solution for ply curing deformation optimization.

[0015] Example 1 This application provides a curing deformation optimization method based on adaptive enumeration-simulated annealing, including: S1: Obtain the number of half-layout layers N and the number of discrete angle types K of the laminate, and calculate the permutation scale K^N; if K^N≤ preset enumeration threshold T, it is determined to be enumerable, and proceed to S2 enumeration optimization; otherwise, it is determined to be unenumerable, and proceed to S3 simulated annealing optimization. Specifically, S1 includes: S1-1: Determine the dimension N of the ply design variables and the number of discrete angle types K; For symmetrical laminates, N is the number of half-lays; for asymmetrical laminates, N is the total number of layups. K=4 corresponds to 0°, +45°, -45°, and 90°.

[0016] S1-2: Calculate the theoretical size of all permutations S=K^N.

[0017] S1-3: Set the enumeration feasibility threshold T, which is adjusted according to hardware performance and acceptable waiting time.

[0018] The typical value is T=10^6.

[0019] S1-4: If S≤T, the problem is determined to be enumerable, and the process proceeds to step S2 to perform enumeration optimization; otherwise, the problem is determined to be non-enumerable, and the process proceeds to step S3 to perform simulated annealing optimization.

[0020] S2: Generate all half-layout sequences that satisfy the constraints through a generation algorithm with process constraint pruning to form a feasible sequence set; for each sequence in the sequence set, query the simulation history database. If a match is found, directly read the maximum curing deformation amount; otherwise, call the finite element simulation model to calculate the maximum curing deformation amount and store it in the simulation history database; after traversing all sequences, output the sequence with the smallest maximum curing deformation amount as the optimal layup scheme. The generation algorithm for the process-constrained pruning is a depth-first recursive search. When adding angles layer by layer, it checks the difference between adjacent angles and the number of consecutive layers in the same direction in real time, and performs balance prediction pruning.

[0021] Specifically, S2 includes: S2-1: Construct a sequence generator with constraint pruning, and verify the angle in real time as each layer is added during the recursive or depth-first search process; The preferred real-time verification method is: (a) Adjacent angle difference constraint: If the absolute value of the angle difference between the new layer and the previous layer is greater than 45°, then backtracking is performed.

[0022] (b) Maximum number of consecutive layers in the same direction constraint: Maintain the current consecutive same angle count. If the new layer has the same angle as the previous consecutive layers and causes the count to exceed the preset upper limit (e.g., 4), then backtrack.

[0023] (c) Balance constraint prediction: Count the number of +45° and -45° generated so far, and combine the remaining number of layers to predict whether it is possible to achieve the same value after the sequence is completed; if not, prune in advance.

[0024] S2-2: A sequence that satisfies all constraints and has a length of N is output as a feasible sequence, which ultimately forms the feasible sequence set F.

[0025] S2-3: For each sequence in F, its maximum curing deformation is obtained through the following sub-process: Preferably, S2-3 includes: S2-3-1: Generate a unique identifier. Convert the sequence into a one-hot encoded matrix or an integer index vector, apply a secure hash algorithm (such as SHA-256), and truncate a fixed-length string as the key.

[0026] S2-3-2: Perform a key-value query in the simulation history database.

[0027] The simulation history database is a key-value pair storage structure, where the key is an identifier and the value is the corresponding solidification deformation amount.

[0028] S2-3-3: If the target is hit, the deformation value is read directly; if the target is not hit, the pre-established finite element simulation model is called to calculate the maximum solidification deformation, and the key-value pair is stored in the simulation history database.

[0029] When the laminate is an asymmetric structure, N is the total number of layers, K^N is the total permutation size, and the other steps remain unchanged.

[0030] S2-4: After traversing all sequences in F, compare the maximum curing deformation and output the sequence with the minimum maximum curing deformation as the optimal layup scheme, and the optimization ends.

[0031] S3: Initialize a feasible sequence as the current solution and the global optimal solution, and set the initial temperature, termination temperature, cooling factor, and number of isothermal iterations; perform multiple neighborhood searches at each temperature, and after generating a new sequence, check its compliance with the process constraint discriminator. If it is not feasible, assign a maximum deformation penalty value; if it is feasible, query the simulation history database through a unique identifier to obtain the deformation value; use the Metropolis criterion to determine whether to accept the new solution and update the global optimal solution; decrease the temperature according to the cooling factor until the termination condition is met, and output the global optimal layup scheme.

[0032] Specifically, S3 includes: S3-1: Simulated annealing initialization.

[0033] Specifically, S3-1 includes: S3-1-1: Generate an initial solution. A feasible half-layout sequence x0 is generated by randomly generating the sequence and immediately performing constraint checks; if the randomly generated sequence is not feasible, it is regenerated until it is feasible.

[0034] S3-1-2: Set the initial temperature T0, the termination temperature Tf, the cooling factor α (0 < α < 1), and the number of iterations L at each temperature.

[0035] The initial temperature T0 and cooling factor α can be determined through a small number of experimental runs to balance the globality of the search with the convergence speed.

[0036] The cooling factor α ranges from 0.8 to 0.99, and the termination temperature Tf is set to 10^-4 to 10^-3 times the initial temperature.

[0037] S3-1-3: Current solution x=x0, current optimal solution x_best=x0. Call the deformation acquisition sub-process (S2-3) to calculate the solidified deformation value f(x0) of x0, which is used as the current optimal value f_best.

[0038] S3-2: Construct a process constraint discriminator to perform sequence checks on any complete sequence; Specifically, the process constraint discriminator checks in the following order: (1) Balance: Count the number of +45° and -45°. If they are not equal, it is directly determined to be infeasible.

[0039] (2) Adjacent angle difference: Traverse adjacent layer pairs. If there is an absolute value of angle difference > 45°, it is determined to be infeasible.

[0040] (3) Maximum number of consecutive layers in the same direction: If there are consecutive layers with the same angle in the scan sequence, it is determined to be infeasible.

[0041] If all passes, it is deemed feasible.

[0042] S3-3: Simulated annealing iteration main loop, at each temperature, perform L inner loops; Specifically, S3-3 includes: S3-3-1: Generate a neighborhood solution x_new. Apply a neighborhood perturbation operation to the current solution x, which can be selected from one or a combination of the following: randomly select a ply position, and change its angle to a random allowable angle different from the current angle; randomly select two positions and swap their angles; reverse the angle order of a certain subsequence. The neighborhood operation must ensure that the generated sequence is still a valid sequence of length N.

[0043] As a preference, the neighborhood perturbation operation in S3-3-1 is specifically: randomly select one of single-point angle mutation, two-point swap or subsequence reversal with a predetermined probability; when performing single-point angle mutation, the new angle is selected with equal probability from the set {0°, +45°, -45°, 90°} after excluding the current angle.

[0044] S3-3-2: Send x_new to the process constraint discriminator. If it is determined to be infeasible, set its deformation value f(x_new)=P, where P is a very large deformation penalty value; if it is determined to be feasible, enter the simulation reuse step — generate a unique identifier for x_new and query the simulation history database: if there is a hit, directly read the deformation value f(x_new), otherwise call the finite element simulation model for calculation and store it in the simulation history database.

[0045] Wherein, the very large deformation penalty value P is far greater than any actual deformation magnitude, and is set to more than 1000 times the actual possible maximum deformation, for example, P=10^6mm, so that the infeasible sequence is accepted with an almost zero probability under the simulated annealing acceptance criterion.

[0046] S3-3-3: Calculate ΔE=f(x_new)–f(x).

[0047] S3-3-4: If ΔE≤0, accept the new solution, set x=x_new, and if f(x_new)<f_best, update x_best and f_best.

[0048] S3-3-5: If ΔE>0, accept the new solution with the probability of exp(-ΔE / T); if accepted, set x=x_new, but do not update the optimal solution.

[0049] S3-4: Current temperature T=α×T, if T<Tf, the termination condition is satisfied, go to S3-5; otherwise, return to S3-3 to continue the search at the next temperature.

[0050] S3-5: Output the global optimal solution x_best and its corresponding curing deformation value f_best, as the optimal ply scheme.

[0051] It should be noted that the simulation history database is shared between the enumeration branch and the simulated annealing branch, which allows the deformation values ​​of the same layup sequence to be reused even in problems of different sizes or multiple optimization runs, further expanding the benefits of computational reuse.

[0052] Specifically, the neighborhood search described in S3 generates a new sequence through one or more combinations of the following perturbation operations: randomly selecting a ply position and changing its angle to another allowed angle; randomly swapping the angles of two ply positions; and reversing the angle order of a subsequence.

[0053] Specifically, the unique identifier is generated by encoding the layup sequence into a one-hot matrix or an integer vector and then performing a hash operation to generate a fixed-length string; the simulation history database adopts a hash table structure, with the identifier as the key and the solidified deformation amount as the value.

[0054] Example 1: Enumeration optimization of 10-layer symmetrical laminate This embodiment uses a carbon fiber / bismaleimide resin U-shaped laminate as an example; the geometric dimensions are shown below. Figure 1 The total number of layers is 10, with N=5 half-lay layers and a single layer thickness of 0.15mm. The layup angle set is {0°, +45°, -45°, 90°}. Process constraints: uniformity (equal number of ±45° layers), adjacent angle difference ≤45°, number of consecutive layers in the same direction ≤4, and the layup ratio of 0°, ±45°, and 90° is constrained to be 2:2:1, with the outermost layup being +45°.

[0055] S1 size prediction: 4^5 = 1024 < 10^6, so it is determined to be enumerable and enters the enumeration branch.

[0056] S2 Enumeration Optimization: A recursive algorithm with constrained pruning is used to generate all 10 feasible sequences. All feasible sequences are as follows: (1) [45 / -45 / 0 / 90 / 0 / 0 / 90 / 0 / -45 / 45] (2) [45 / -45 / 0 / 0 / 90 / 90 / 0 / 0 / -45 / 45] (3) [45 / 90 / 0 / -45 / 0 / 0 / -45 / 0 / 90 / 45] (4) [45 / 90 / 0 / 0 / -45 / -45 / 0 / 0 / 90 / 45] (5) [45 / 0 / -45 / 90 / 0 / 0 / 90 / -45 / 0 / 45] (6) [45 / 0 / -45 / 0 / 90 / 90 / 0 / -45 / 0 / 45] (7) [45 / 0 / 90 / -45 / 0 / 0 / -45 / 90 / 0 / 45] (8) [45 / 0 / 90 / 0 / -45 / -45 / 0 / 90 / 0 / 45] (9) [45 / 0 / 0 / -45 / 90 / 90 / -45 / 0 / 0 / 45] (10)[45 / 0 / 0 / 90 / -45 / -45 / 90 / 0 / 0 / 45] The simulation history database is implemented using a hash table, with keys directly generated from the integer indices of the layup sequences. All feasible sequences are traversed, and for each sequence, the simulation history database is queried. If a match is found, the value is directly retrieved; otherwise, the finite element simulation model is called to calculate the maximum curing deformation, which is then stored in the simulation history database. After 10 finite element simulations, the maximum curing deformation corresponding to all feasible sequences is obtained. The maximum curing deformation corresponding to all feasible sequences is shown below. Figure 2 The dashed lines in the diagram represent the sequence number on the x-axis and the maximum curing deformation on the y-axis. The solid lines represent the minimum maximum curing deformation after each deformation evaluation. The sequence with the minimum maximum curing deformation was selected as the second sequence, namely [45 / -45 / 0 / 0 / 90 / 90 / 0 / 0 / -45 / 45], with a maximum curing deformation of 1.863 mm.

[0057] Example 2: Simulated Annealing Optimization of 20-Layer Symmetrical Laminate This embodiment applies to the same material system and the same U-shaped structure; dimensions are shown below. Figure 1 The total number of laminate layers is 20, with 10 symmetrical half-lay layers. The layup angle and constraints are the same as above.

[0058] S1 size prediction: 4^24=1048576 is greater than 10^6, so it is determined to be unenumerable, and enter the simulated annealing branch.

[0059] S3 Simulated Annealing Optimization: Initialization: A feasible sequence x0 is randomly generated, with the corresponding layup sequence [45 / -45 / 90 / 0 / 0 / 0 / 0 / 90 / -45 / 45 / 45 / -45 / 90 / 0 / 0 / 0 / 0 / 90 / -45 / 45]. The maximum curing deformation is calculated using a finite element simulation model to be f0 = 0.549 mm, and these are designated as the current solution x and the optimal solution x_best. The initial temperature T0 = 1.0, the termination temperature Tf = 1 × 10^-4, the cooling factor α = 0.95, and the iterations L = 200 times at each temperature. The penalty value P = 10^6 mm.

[0060] Neighborhood operations: perform a single-point angle mutation with a probability of 0.6, perform a random two-point swap with a probability of 0.2, and perform a subsequence reversal with a probability of 0.2.

[0061] Inner loop: Repeat L times at the current temperature. Each time, a neighborhood solution x_new is generated and checked by a constraint discriminator. If it is infeasible, assign f(x_new)=P; if it is feasible, generate a unique identifier to query the simulation history database, and call the finite element simulation model if there is no hit. Accept or reject according to the Metropolis criterion. If accepted and f(x_new) is less than the current f_best, update the optimal solution.

[0062] Temperature update: After every L iterations are completed, T=T×0.95. Stop when T<Tf. During the optimization process, the simulation history database is continuously expanded. The iteration curve of the maximum curing deformation during the optimization process is shown in Figure 3 , and the optimal layup sequence is finally obtained, which is [45 / -45 / 0 / 90 / 0 / -45 / 90 / 0 / 0 / 45 / 45 / 0 / 0 / 90 / -45 / 0 / 90 / 0 / -45 / 45], with the corresponding maximum curing deformation of 0.497mm.

[0063] It can be seen from the embodiment that the simulated annealing branch also achieves remarkable optimization effects, with intuitive algorithm parameter settings and stable convergence.

[0064] In summary, the present application provides a curing deformation optimization method based on adaptive enumeration - simulated annealing, which is a layup sequence optimization method that combines exhaustive search and simulated annealing algorithm, and eliminates redundant deformation calculation through pre-process constraint discrimination and a simulation history database. It is specifically used to efficiently obtain the layup scheme with the minimum curing deformation under layup process constraints such as balance, adjacent angle difference, and maximum number of consecutive layers in the same direction. Exhaustive search or simulated annealing optimization is automatically selected according to the layup scale, and with the help of the process constraint discriminator and the simulation history database, deformation evaluation of any constraint-violating sequences and duplicate sequences is avoided throughout the optimization process, so that the layup scheme that meets all process constraints and has the minimum curing deformation is obtained at an extremely low computational cost.

Claims

1. A curing deformation optimization method based on adaptive enumeration-simulated annealing, characterized in that, include: S1: Obtain the number of half-layout layers N and the number of discrete angle types K of the laminate, and calculate the permutation scale K^N; if K^N≤ preset enumeration threshold T, it is determined to be enumerable, and proceed to S2 enumeration optimization; otherwise, it is determined to be unenumerable, and proceed to S3 simulated annealing optimization. S2: Generate all half-layout sequences that satisfy the constraints through a generation algorithm with process constraint pruning to form a feasible sequence set; for each sequence in the sequence set, query the simulation history database. If a match is found, directly read the maximum curing deformation amount; otherwise, call the finite element simulation model to directly calculate the maximum curing deformation amount and store it in the simulation history database; after traversing all sequences, output the sequence with the smallest deformation as the optimal layup scheme. S3: Initialize a feasible sequence as the current solution and the global optimal solution, and set the initial temperature, termination temperature, cooling factor, and number of isothermal iterations; perform multiple neighborhood searches at each temperature, and after generating a new sequence, check its compliance with the process constraint discriminator. If it is not feasible, set a maximum value for the maximum curing deformation; if it is feasible, query the simulation history database using a unique identifier to obtain the deformation value; use the Metropolis criterion to determine whether to accept the new solution and update the global optimal solution. The temperature decreases according to the cooling factor until the termination condition is met, and the globally optimal layup scheme is output.

2. The method according to claim 1, characterized in that, S1 includes: S1-1: Determine the dimension N of the ply design variables and the number of discrete angle types K; S1-2: Calculate the theoretical size of all permutations, S=K^N; S1-3: Set the enumeration feasibility threshold T, which is adjusted according to hardware performance and acceptable waiting time; S1-4: If S≤T, the problem is determined to be enumerable, and the process proceeds to step S2 to perform enumeration optimization; otherwise, the problem is determined to be non-enumerable, and the process proceeds to step S3 to perform simulated annealing optimization.

3. The method according to claim 1, characterized in that, S2 include: S2-1: Construct a sequence generator with constraint pruning. During the recursive or depth-first search process, constraint and repeatability need to be checked in real time for each layer. S2-2: A sequence that satisfies all process constraints and has a length of N is output as a feasible sequence, which ultimately constitutes the feasible sequence set F; S2-3: For each sequence in F, its maximum curing deformation is obtained through the following sub-process: S2-4: After traversing all sequences in F, compare the maximum curing deformation and output the sequence with the minimum maximum curing deformation as the optimal layup scheme, and the optimization ends.

4. The method according to claim 3, characterized in that, The real-time verification method is as follows: Maximum number of consecutive layers in the same direction constraint: Maintain the count of the current consecutive layers with the same angle. If a new layer has the same angle as the previous consecutive layers and causes the count to exceed the preset upper limit, backtrack. Balance constraint prediction: Count the number of +45° and -45° that have been generated so far, and combine them with the number of remaining layers to predict whether it is possible to achieve the same result after the sequence is completed; if not, prune in advance.

5. The method according to claim 1, characterized in that, S2-3 includes: S2-3-1: Convert the sequence into an integer index vector, apply a secure hash algorithm, and extract a fixed-length string as the key; S2-3-2: Perform a key-value query in the simulation history database; S2-3-3: If the maximum curing deformation can be queried, it is directly read; if not, the pre-established finite element model is invoked to calculate the maximum curing deformation, and the key-value pair corresponding to the simulation result and the current sequence is stored in the simulation history database.

6. The method according to claim 1, characterized in that, S3 comprising: S3-1: Perform simulated annealing initialization; S3-2: Construct a process constraint discriminator to perform sequence check on any complete sequence; S3-3: Execute the main loop of simulated annealing iteration, and perform L inner loops at each temperature; S3-4: Set the current temperature T=α×T, if T<Tf, the termination condition is satisfied, and the process proceeds to S3-5; otherwise, return to S3-3 to continue the search at the next temperature; S3-5: Output the global optimal solution x_best and its corresponding curing deformation value f_best, which are used as the optimal layup scheme.

7. The method according to claim 6, characterized in that, S3-1 comprises: S3-1-1: Generate a feasible semi-layup sequence x0 by randomly generating and immediately performing process constraint check; if the randomly generated sequence is not feasible, regenerate it until it is feasible; S3-1-2: Set the initial temperature T0, the termination temperature Tf, the cooling factor α (0<α<1) and the number of iterations L at each temperature; S3-1-3: Set the current solution x=x0, and the current optimal solution x_best=x0; invoke the maximum curing deformation acquisition sub-process S2-3 to calculate the curing deformation value f(x0) of x0, which is used as the current optimal value f_best.

8. The method according to claim 6, characterized in that, The process constraint discriminator performs checks in the following order: Balance property: Count the number of +45° and -45°, if they are not equal, the sequence is directly determined to be infeasible; Maximum number of consecutive layers in the same direction: Scan the sequence, if the number of consecutive layers with the same angle is greater than the upper limit, the sequence is determined to be infeasible; If all checks are passed, the sequence is determined to be feasible.

9. The method according to claim 6, characterized in that, S3-3 comprises: S3-3-1: Generate a neighborhood solution x_new; S3-3-2: Send x_new to the process constraint discriminator; if it is determined to be infeasible, set its deformation value f(x_new)=P, where P is a very large deformation penalty value; if it is determined to be feasible, enter the simulation reuse step — generate the unique identifier of x_new and query the simulation history database: if there is a hit, directly read the deformation value f(x_new), otherwise invoke the finite element simulation model for calculation and store the result in the simulation history database; S3-3-3: Calculate ΔE=f(x_new)–f(x); S3-3-4: If ΔE≤0, accept the new solution, set x=x_new, and if f(x_new)<f_best, update x_best and f_best; S3-3-5: If ΔE>0, accept the new solution with the probability of exp(-ΔE / T); if accepted, set x=x_new without updating the optimal solution.

Images