摘要
提出了一种将径向基神经网络作为代理模型用于贝叶斯框架的损伤识别方法。首先采用拉丁超立方抽样技术,选取一定数量的结构输入输出样本,训练出一个径向基神经网络。然后将其用于基于马尔科夫链蒙特卡洛抽样的贝叶斯损伤识别方法。其中抽样方法采用吉布斯抽样。数值算例显示,在考虑测量误差的情况下,提出的方法能准确识别出简支梁的损伤,有效避免了损伤识别反问题的不适定性。其计算效率较传统的方法提高了数十倍,是一种很有潜力的损伤识别方法。
A method was proposed to use radial basis function neural networks as surrogate models for damage identification within the Bayesian framework.Initially,Latin hypercube sampling was employed to select a specific number of structural input-output samples,leading to the training of a radial basis function neural network.Subsequently,this network was applied to a Bayesian damage identification method based on Markov chain Monte Carlo sampling.Gibbs sampling was utilized as the sampling method.Numerical examples demonstrated that,considering measurement errors,the proposed method accurately identified damage in simply supported beams,effectively avoiding the ill-posed nature of the inverse problem in damage identification.The computational efficiency of this method was improved by several orders of magnitude compared to traditional approaches,making it a highly promising damage identification method.
作者
卢小丽
文韬
郭丽丽
LU Xiao-li;WEN Tao;GUO Li-li(School of Machinery and Engineering,Wuhan University of Engineering Science,Wuhan 430200,China)
出处
《建材世界》
2024年第2期110-114,共5页
The World of Building Materials
基金
2022湖北省教育厅科学技术研究计划指导性项目(B2022387)。
作者简介
卢小丽(1981-),副教授.E-mail:luxiaoli1981whues@163.com;通讯作者:郭丽丽(1985-),讲师.E-mail:Guolili1985Whues@163.com。
