结构应力逆向推演模型及其不确定度定量分析

目的 针对现有结构安全在线监测与评估方法存在的短板，结合人工智能方法探索新的解决方案。提出一种基于有限测点应力逆向推演整体结构应力分布的方法，以数据驱动的模式，基于神经网络技术搭建算法模型。方法 以结构有限元仿真数据为基础，运用相关性分析方法，获取代表结构响应特征的有限个测点，利用神经网络方法构建基于有限测点应力推演结构全场应力分布的算法模型。结果 以科学试验平台的连接器结构为对象，开展该算法模型的应用研究，并且对该应用实例下的算法模型开展不确定度分析，推演结果的相对不确定度u95rel为8.6%。结论 该算法模型的推演结果正确反映了结构总体响应特征。从建模过程角度分析，该算法模型的不确定度来源主要包括相关性分析方法、神经网络建模以及模型收敛条件3个方面。

Reverse Deduction Model of Structural Stress and Quantitative Analysis of Its Uncertainty

The work aims to explore an artificial intelligence method for solving the shortcomings of structural safety online monitoring and evaluation and propose a method for reversely deducing the stress distribution of the whole structure based on limited measuring points, so as to build a data-driven algorithm model based on neural network technology. Firstly, based on the finite element simulation data of the structure, a limited number of measuring points representing the structural response characteristics were obtained by the correlation analysis method. Then, neural network method was adopted to build the algorithm model of deducing stress distribution in the whole field of structure based on limited measuring points. With the connector structure of Scientific Research & Demonstration Platform (SRDP) as the object, the application of the algorithm model was studied, and the uncertainty analysis of the algorithm model in this application case was carried out. Relative uncertainty u95rel of the deduced results was 8.6%. The deduced results of this algorithm model correctly reflect the overall response characteristics of the structure. According to the modeling process, the uncertainty mainly results from correlation analysis method, neural network modeling and model convergence condition.