A SEQUENTIAL LOCAL ENUMERATION-BASED IMPROVED LATIN HYPERCUBE SAMPLING METHOD FOR BALLISTIC CONSTRAINT DESIGN SPACE

The design and optimization of complex aircraft usually requires thousands to tens of thousands of numerical simulations to complete, a costly process. Surrogate models have been widely used in engineering problems due to their feature of replacing complex and time-consuming numerical simulations. Design of experiment (DoE) is an important and indispensable part of the process of establishing a surrogate model, and its sampling results have a crucial impact on the accuracy of the model and the subsequent optimal design. Existing DoE methods such as Latin hypercube sampling (LHS) are mainly targeted at unconstrained design space. But in actual engineering problems, such as the ballistic constrained design space in aircraft design is often subject to multiple and strong constraints on the ballistic trajectory, and then the unconstrained DoE method is no longer applicable. The constrained DoE method: sequential local enumeration-based Latin hypercube sampling for constrained design space (SLE-CLHS) for the constraint space is unable to establish the constraints of the ballistic constraint design space of the whole vehicle. Therefore, based on SLE-CLHS, this paper proposes a sequential local enumeration-based improved Latin hypercube sampling method for ballistic constraint design space (BCDS-ILHS), which ensures better exploration of the design space by establishing the constraints of the ballistic trajectory, improves the correctness and accuracy of the sampling, and establishes a higher accuracy surrogate model with as few samples as possible. In order to verify the feasibility of the proposed BCDS-ILHS in the ballistic constraint design space, we apply it to a 3D numerical case and a 3D engineering case. The results show that BCDS-ILHS is able to generate samples that satisfy the constraints compared to LHS. And in most cases, with the same prediction accuracy of the surrogate model, BCDS-ILHS requires fewer samples.