基于高斯过程响应面的结构有限元模型修正方法

Structural finite element model updating based on Gaussian process response surface methodology

作为一种基于贝叶斯原理的非参数模型,高斯过程模型近年来得到研究人员的广泛关注。高斯过程具有灵活的协方差函数形式、预测精度高和量化预测不确定性等优点。基于此,提出一种基于高斯过程响应面的有限元模型修正方法,并介绍了高斯过程响应面方法的基本理论。算例结果表明,相对于传统参数型响应面方法,高斯过程响应面方法应用于有限元模型修正更具优势。

As Bayesian framework-based and nonparametric models, Gaussian process （GP） models draw a considerable attention of researchers in recent years. GP models have many advantages, including the flexibility of the forms of covariance functions, high accuracy of prediction, quantifying the prediction quality with uncertainties and so on. Here, a novel model updating technique based on GP response surface methodology （RSM） was proposed, and the relevant basic principle was elaborated. Results of numerical examples showed that GP response surface is superior to the traditional response surfaces, i. e. , the polynomial models in structural finite element （FE） model updating.