基于神经网络与遗传算法的潜艇舱壁结构优化

建立了某潜艇端部耐压平面舱壁结构优化设计的数学模型,以舱壁上加强桁(筋)结构的剖面尺寸为设计变量,结构强度、稳定性、工艺性要求为约束条件,以加强桁(筋)结构总体积为目标函数,用人工神经网络方法代替结构的有限元分析,结合遗传算法完成了舱壁的结构优化设计。在为网络生成学习样本时,自行设计了正交表,解决了因为设计变量多,无法选取合适正交表而随机取样的问题。数值仿真结果表明,优化设计方案质量较原始初步设计方案减少了18.3%。

Structural Optimization of a Submarine End Plane Transverse Bulkhead Based on Neural Networks and Genetic Algorithm

A mathematical model of optimization for the sectional dimensions of stiffeners on a submarine end plane bulkhead is described.The sectional dimensions of stiffeners are selected to be the design variables.Structural strength,stability and technics are included in constraints while the total volume of the stiffeners is defined as the objective function.In this paper,the structural reanalysis is executed with trained BP-neural networks instead of the finite element method and the genetic algorithm is employed to conduct the structural optimization with discrete design variables.In the study,a self-defined design orthogonal form is proposed for selecting samples efficiently in the training of neural network with large number of design variables.Numerical simulations show that the weight of the optimal design scheme can be decreased 18.3% compared to the weight of the original design scheme.