基于人工智能的三维结构全局应力求解方法研究  

Solving approach for global stress field of the3D structures based on artificial intelligence

作　　者：张涛[1,2] 胡嘉骏 汪雪良[1,2] 徐业峻 郑重[4] 王子渊 ZHANG Tao;HU Jia-jun;WANG Xue-liang;XU Ye-jun;ZHENG Zhong;WANG Zi-yuan(China Ship Scientific Research Center,Wuxi 214082,China;Taihu Laboratory of Deepsea Technological Science,Wuxi 214082,China;CNOOC Energy Technology&Service Oil Production Service Co.,Tianjin 300452,China;Dalian Shipbuilding Industry Equipment Manufacturing Co.,Ltd.,Dalian 116052,China)

机构地区：[1]中国船舶科学研究中心,江苏无锡214082 [2]深海技术科学太湖实验室,江苏无锡214082 [3]中海油能源发展股份有限公司采油服务分公司,天津300452 [4]大连船舶重工集团装备制造有限公司,辽宁大连116052

出　　处：《船舶力学》2023年第2期238-249,共12页Journal of Ship Mechanics

基　　金：国家留学基金项目(202208320260)。

摘　　要：针对结构健康监测系统(SHM)中有限数量的应力测点难以监测结构全局应力的问题,提出一种基于人工智能的三维结构全局应力求解方法,通过少量应力测点实现对三维结构物应力场的实时求解。首先,基于有限单元之间对载荷响应的关联性,运用相关性分析找出求解三维结构物应力场的特征单元;其次,利用神经网络建立特征单元与相关单元的求解关系,并将特征单元设为应力测点进而求解结构全局应力;然后,以某海洋平台连接器结构为应用主体,考虑测点数量、神经网络架构、收敛准则等因素对求解精度的影响,建立连接器结构应力场人工智能算法优化模型;最后,开展实尺度模型试验,经比较分析得出该算法模型应力求解精度高达93.6%,该技术可切实提升SHM系统的实用性,将传统的对“点”监测提升为对“场”监测。For the existing problem in the structural health monitoring(SHM) system that the global stress distribution could not be measured by a limited number of strain sensors, a solving approach for the global stress field of the 3D structures based on the AI was proposed, by which the stress field of marine structures can be solved in real-time with only a few selected sensors. Firstly, based on the relevance of the load response among the finite elements, several characteristic elements for solving the stress field were selected from the 3D structure as the stress-measured points by correlation analysis. Secondly, the solution relationship between characteristic elements and correlation ones was established using the artificial neural network.Then, the proposed method was applied to a physical structure, which is part of a real offshore platform. Considering the influence of the number of strain sensors, neural networks framework, convergence criteria and other factors on the solution accuracy, an optimal neural network with comprehensive functions for the stress field of a connector structure was established. The real scale model test shows that the accuracy of the AI model reaches 93.6%. This technology effectively improves the practicability of the SHM system and upgrades the traditional‘point monitoring’to‘field monitoring’.

关 键 词：三维结构 船舶与海洋工程 全局应力 人工智能 结构健康监测 

分 类 号：U661.4[交通运输工程—船舶及航道工程]