CAJ | 학술논문

结构系统的可靠性评估是结构设计的一个重要研究内容,而极限状态函数的建立是进行可靠性评估的基础.但是,大型结构系统的极限状态函数极为复杂,响应面法用简单的多项式进行模拟的精度较低,导致误差较大.文章提出用神经网络替代多项式来拟合复杂的极限状态函数,形成所谓的神经网络响应面.然后,基于塑性极限理论,文中提出了不依赖于失效模式的极限状态函数表达形式及采用ICP对该极限状态函数进行计算的方法.最后,依照拟合得到的神经网络响应面,给出了大型结构系统失效概率的方法.通过两个算例计算并和其它方法进行比较,表明该方法的计算精度较高,而计算时间大大降低.
결구계통적가고성평고시결구설계적일개중요연구내용,이겁한상태함수적건립시진행가고성평고적기출.단시,대형결구계통적겁한상태함수겁위복잡,향응면법용간단적다항식진행모의적정도교저,도치오차교대.문장제출용신경망락체대다항식래의합복잡적겁한상태함수,형성소위적신경망락향응면.연후,기우소성겁한이론,문중제출료불의뢰우실효모식적겁한상태함수표체형식급채용ICP대해겁한상태함수진행계산적방법.최후,의조의합득도적신경망락향응면,급출료대형결구계통실효개솔적방법.통과량개산례계산병화기타방법진행비교,표명해방법적계산정도교고,이계산시간대대강저.
The evaluation of the failure probability of structural system is of extreme importance in structural design and the limit state function (LSF) is the key to do it. As the basis function of Response Surface (RS), the quadratic polynomial is too weak to approximate the real LSF for largescale structure, which is very complex. In this paper by substituting the Artificial Neural Network (ANN) for the quadratic polynomial, a Neural Network Response Surface (NNRS) was built to approximate the very complex LSF of large-scale structure accurately. Based on ductile limit analysis theory,a system LSF independent of structural system failure modes and a method to compute it with Improved Compact Procedure (ICP)were presented. By approximating this system LSF with NNRS.Finally,an evaluation method of system failure probability of large-scale structural was presented. Compared with FORM,the accuracy and efficiency of this method was demonstrated using two numerical examples.