Search

CN-121980775-A - Test design method based on non-convex inequality constraint

CN121980775ACN 121980775 ACN121980775 ACN 121980775ACN-121980775-A

Abstract

The invention discloses a test design method based on non-convex inequality constraint, which comprises the following steps of S1, defining design variables and constraint conditions, S2, mapping the design variables to a multidimensional hypercube space, simultaneously recording the original physical upper and lower bounds of each variable, generating an initial population containing a plurality of samples by Latin hypercube sampling, converting infeasible samples in the population into a non-linear programming problem with constraint, calling a non-linear solver IPOPT to solve, obtaining a sufficient feasible solution, finally sampling by adopting standard normal distribution to generate a plurality of reference vectors, S3, generating a child sample set by using an updating strategy iteration, updating the population, S4, screening out sample points which meet the constraint and ensure uniformity from a father population and the child population, constructing an elite population, S5, mapping the sample points in the final elite population back to the physical space by a decoder, and outputting a test design scheme meeting the constraint.

Inventors

  • LIU FEI
  • ZHAO MINGYU
  • ZHOU KAI

Assignees

  • 华南理工大学

Dates

Publication Date
20260505
Application Date
20251231

Claims (10)

  1. 1. The experimental design method based on the non-convex inequality constraint is characterized by comprising the following steps of: S1, defining design variables and constraint conditions; S2, mapping design variables to a multidimensional hypercube space, simultaneously recording the original physical upper and lower bounds of each variable, generating an initial population containing a plurality of samples by Latin hypercube sampling, converting infeasible samples in the population into a nonlinear programming problem with constraint, calling a nonlinear solver IPOPT to solve, obtaining a sufficient quantity of feasible solutions, and finally sampling by adopting standard normal distribution to generate a plurality of reference vectors; s3, iteratively generating a child sample set by using an updating strategy, and updating the population; s4, screening sample points which meet constraint and ensure uniformity from parent population and offspring population to construct elite population; S5, mapping sample points in the final elite population back to a physical space through a decoder, and outputting a test design scheme meeting the constraint.
  2. 2. The method of claim 1, wherein the design variables are crack characteristics of the wing surface including coordinates of the crack center in the machine span direction, coordinates of the crack center in the wing chord direction, radius of the crack, and direction angle of the crack.
  3. 3. The method of claim 1, wherein the set of constraints is safely constructed around the location and structure of the aircraft wing crack requiring a crack midpoint Endpoint(s) And end point The crack is required to completely fall inside the wing defined by six boundary lines, and the sum of the minimum distances from two endpoints of the crack to any two different boundaries is not less than 4mm, wherein the center axis of the wing is not contained, and the crack distribution is ensured to meet the engineering physical boundary and structural safety requirements.
  4. 4. The test design method based on the non-convex inequality constraint according to claim 1, wherein in step S3, the following steps S31-S33 are repeatedly performed on each sample point in the current population until all samples are traversed, thereby constructing a complete offspring population for subsequent screening; S31, determining an evolution direction vector of each sample in the current population by adopting an evolution direction guiding method based on local energy gradient rejection; s32, shifting each sample point in the population based on a pre-evolution position bias strategy of neighborhood rejection to obtain an adjusted sample point; And S33, updating the sample points according to the evolution direction vector and the adjusted sample points, and adding the updated sample points into the offspring population set.
  5. 5. The method of claim 4, wherein step S32 includes the steps of first calculating a strong repulsive force vector based on distance, and then fine-tuning the sample position according to the strong repulsive force vector and the current attenuation coefficient to obtain an adjusted sample point.
  6. 6. The method of claim 4, wherein the step S33 comprises the steps of Updating sample points by adopting a constraint boundary control strategy based on a tangent space mapping and a nonlinear solver, comprising the following steps of: Firstly, judging whether the adjusted sample points meet all constraint conditions or not to construct a set of sample points conforming to the constraint And a set of sample points that violate a constraint ; The following different mechanisms for updating sample point locations are adopted for sample points in a set of sample points that meet a constraint and a set of sample points that violate a constraint: (1) For sample points in the set of sample points conforming to the constraint, calculating the distance from the sample point to the nearest constraint boundary according to the current spatial position of the sample point Dynamically adjusting a step length factor by using a hyperbolic tangent function, calculating a moving step length, and updating a sample point, wherein the updating calculation formula of the sample point is as follows: Wherein the method comprises the steps of Representing the updated sample points; a vector representing the direction of evolution of the sample; Representing the adjusted sample points; Representing a movement step; (2) When a sample point is close to a constraint boundary, constructing a tangential path by utilizing zero space projection, calculating a jacobian matrix for activating the constraint, solving a zero space base matrix of the jacobian matrix, carrying out tangential projection, projecting an original target evolution direction to a zero space to obtain a tangential evolution vector, and updating a position along the tangential vector, wherein an updating formula of the sample point is as follows: Wherein the method comprises the steps of The step size of the movement in the tangential direction is indicated, Representing a tangential evolution vector; (3) For infeasible sample points in the set of sample points that violate the constraint, the update formula for the sample points is: Wherein the method comprises the steps of Representing the sample points after the adjustment, The adaptive step size is indicated as such, Representing the generation of the final evolutionary direction vector from the repair vector and the exploration vector.
  7. 7. The method of claim 6, wherein the calculation of the final evolutionary direction vector comprises: Firstly, a mixed repair strategy is adopted to change the mixed repair strategy into a feasible sample point, so that probability is realized Constructing a nonlinear programming model based on minimum distance projection, taking a current infeasible sample point as a reference target, taking a repaired viable sample point as a decision variable, establishing an optimization model taking the square of Euclidean distance between the minimum two as an objective function and taking all inequality constraints of an original problem and upper and lower boundaries of variables as constraint conditions, then calling a nonlinear solver IPOPT to solve the problem, projecting the current infeasible sample point back to the vicinity of a constraint manifold surface, updating the current infeasible sample point to be a new sample point, and calculating the constraint violation degree of the new sample point, wherein the optimization model is a model based on the probability Directly calculating the constraint violation degree of the current infeasible sample point, calculating the sum of negative gradients of all violated constraint after the constraint violation degree is calculated, taking the sum as a gradient repair vector, calculating a vector pointing to a target reference point as an exploration vector, then constructing a dynamic mixing weight by using a hyperbolic tangent function, and finally generating a final evolution direction vector according to the repair vector and the exploration vector.
  8. 8. A non-based on the method of claim 1 a test design method of convex inequality constraint, the method is characterized in that the screening in the step S4 comprises the following steps: in the first iteration process, selecting from the population Individual feasible solutions to form elite populations if the feasible solutions are insufficient Selecting the solution with the minimum constraint violation degree to form elite population, and in the subsequent iteration process, if the current population has feasible solution number More than And then from And selecting a feasible solution from the feasible solutions to randomly replace the solutions in the elite population, and replacing the solutions in the elite population with the selected solutions if the uniformity of the elite population is improved, so as to ensure the uniformity of the solutions in the elite population.
  9. 9. A computer device comprising a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, wherein the computer program, when executed by the processor, causes the processor to implement the method as claimed in any one of claims 1 to 8.
  10. 10. A computer readable storage medium storing a computer program, wherein the computer program is executed by a processor, the processor implementing the method according to any one of claims 1 to 8.

Description

Test design method based on non-convex inequality constraint Technical Field The invention belongs to the field of experimental design in simulation experiments, and particularly relates to a non-convex inequality constraint-based experimental design method. Background At present, test design under complex constraint space still faces serious technical challenges. In engineering scenarios involving high-dimensional continuous variables and complex nonlinear constraints, efficient sampling is extremely difficult due to the too strict constraints and extremely irregular shape of the feasible region. Particularly, when facing non-convex or non-connected complex constraint, the existing method lacks effective utilization of constraint boundary information and a clear search guiding mechanism, and blindness and inefficiency exist in the calculation process. As Pang Y et al propose a latin supersvolume design method （Pang Y, Yang L, Wang Y, et al. A Latin hypervolume design for irregular sampling spaces and its application in the analysis of cracks[J]. Engineering with Computers, 2023, 39(5): 3509-3526.）, for irregular sampling space, which uses monte carlo sampling to approximate the calculation of supersvolume and performs division of design points based on the volume, so as to solve the problem that the conventional latin hypercube design is difficult to maintain space filling and non-collapsing in constrained space. However, this method is computationally expensive when dealing with high dimensional space and extremely irregular constraints, and adopts a simulated annealing algorithm in the process of finding a feasible solution, and its search efficiency remains to be improved for extremely narrow or broken feasible regions. You Yang et al propose a space filling test design method (You Yang, golden, pan Zhengjiang, etc.) based on random coordinate exchange, which constrains space to be approximately orthogonal [ J ]. System engineering and electronics, 2021,43 (07): 1831-1837.) the method constructs criteria based on the distance between design points and correlation coefficients, and is applicable to both convex constraint and non-convex constraint spaces. However, when facing complex non-convex constraint boundaries, the coordinate exchange-based algorithm has difficulty in controlling the distribution of sample points near the boundaries, easily causing excessive aggregation or trapping of sample points in local optimum at the boundaries, and difficulty in ensuring uniformity while meeting strict constraints. Disclosure of Invention The invention aims to solve the technical problem that the feasibility and the distribution uniformity of sample points are difficult to consider when the existing test design method faces the constraint of complex non-convex inequality. Therefore, the invention provides a test design method based on non-convex inequality constraint, which solves the problem of random sampling of limited continuous variables in a complex constraint space by introducing an evolutionary search and constraint repair mechanism. The above object of the present invention is achieved by at least one of the following technical means. A test design method based on non-convex inequality constraint comprises the following steps: S1, defining design variables and constraint conditions; S2, mapping design variables to a multidimensional hypercube space, simultaneously recording the original physical upper and lower bounds of each variable, generating an initial population containing a plurality of samples by Latin hypercube sampling, converting infeasible samples in the population into a nonlinear programming problem with constraint, calling a nonlinear solver IPOPT to solve, obtaining a sufficient quantity of feasible solutions, and finally sampling by adopting standard normal distribution to generate a plurality of reference vectors; s3, iteratively generating a child sample set by using an updating strategy, and updating the population; s4, screening sample points which meet constraint and ensure uniformity from parent population and offspring population to construct elite population; S5, mapping sample points in the final elite population back to a physical space through a decoder, and outputting a test design scheme meeting the constraint. Further, the design variables are crack characteristics of the wing surface, including coordinates of a crack center in a wing span direction, coordinates of a crack center in a wing chord direction, a radius of the crack, and a direction angle of the crack. Further, the position and structure safety construction of the constraint condition set around the crack of the airplane wing requires that the midpoint of the crackEndpoint(s)And end pointThe crack is required to completely fall inside the wing defined by six boundary lines, and the sum of the minimum distances from two endpoints of the crack to any two different boundaries is not less than 4mm, wherein the ce