基于BP神经网络和分解技术的汽轮机叶片可靠性反求设计

汽轮机叶片可靠性反求设计旨在确定叶片未知概率设计参数以满足给定的可靠度要求.针对叶片功能函数为随机变量隐性函数的情况,提出了基于有限元、BP神经网络和分解技术的可靠性反求设计方法,该方法将有限元和BP神经网络相结合以构造功能函数与随机输入变量之间的近似解析表达式,运用分解技术,将求解随机设计参数的全局优化问题分解为主问题和子问题,通过子问题直接调用标准优化工具箱得到可靠性指标,并运用分解迭代技术对主问题求解,从而得到随机设计参数及目标可靠性指标对各随机变量的敏感性.以某实验台汽轮机等直叶片为例,阐述了该方法的具体实施过程.该方法数学描述简单,并可直接应用标准优化程序,成功地解决了隐性功能函数下叶片可靠性反求设计,具有较好的工程应用价值.

Design of Steam Turbine Blades Based on BP (Back Propagation) Neural Network and Decomposition Techniques

The reliability reverse-solution-seeking design of steam turbine blades aims at determining the design parameters of blades with unknown probability to meet a given reliability requirement.In the light of the blade function being a random variable implicit function,a reliability reverse-solution-seeking design method was presented based on finite element method,BP neural network and decomposition techniques.It combined the finite element method with BP neural network to establish an approximate analytic expression showing the relationship between the performance function and the random input variables.By employing the decomposition techniques,the overall optimization problem involving the solution-seeking of random design parameters was decomposed into a main problem and sub-problems.By way of the sub-problems,the standard optimization toolbox was used directly to obtain the reliability indexes,and the decomposition and iterative techniques were employed to seek solutions to the main problem,thus obtaining the sensitivity of the random design parameters and target reliability indexes to various random variables.With the equal and straight blades of a steam turbine on a test rig serving as an example,the concrete application process of the method was expounded.The method features a simple mathematical expression and can be directly used in standard optimization programs.It successfully solved the reliability reverse-solution-seeking design problem of blades under an implicit function,thus enjoying a relatively good application value for engineering projects.