基于卷积神经网络的海洋平台损伤识别实验系统开发  

Development of An Experimental System for the Damage Localization of Offshore Platform Based on CNN

作　　者：包兴先[1] 范同轩 张敬 张亚 BAO Xingxian;FAN Tongxuan;ZHANG Jing;ZHANG Ya(College of Petroleum Engineering,China University of Petroleum(East China),Qingdao 266580,Shandong,China)

机构地区：[1]中国石油大学(华东)石油工程学院,山东青岛266580

出　　处：《实验室研究与探索》2021年第1期49-52,80,共5页Research and Exploration In Laboratory

基　　金：国家重点研发计划项目(2016YFC0303800);国家自然科学基金项目(51979283);山东省自然科学基金(ZR2018MEE053)。

摘　　要：将卷积神经网络(CNN)引入到海洋平台损伤识别应用中,开发了基于一维CNN的导管架式海洋平台损伤识别实验系统。构建一维CNN结构损伤识别模型,基于振动台规则波激励获取导管架式海洋平台模型的应变响应实测数据,将实测数据分为训练集和测试集,进行CNN网络的训练和测试。在此过程中,考虑了完好工况、单损伤、多损伤等不同损伤位置共7个模拟工况。实验结果表明,一维CNN可以有效地识别海洋平台结构损伤位置,开发的实验系统可进一步扩展其功能。The convolutional neural network(CNN)is introduced into the damage localization for the offshore platform,and an experimental system based on one-dimensional CNN for jacket-type offshore platform is developed.In this developed experimental system,aone-dimensional CNN model is constructed,and the strain response signals are obtained from the platform model under the regular wave excitation produced by the shaking table,and then the data which are divided into the training set data and testing set data,are input into the CNN model for training and testing.During the experiments process,7 simulated cases including the intact,single damage location and multiple damage locations are considered.Results show that the one-dimensional CNN can identify the damage locations of the platform model effectively.

关 键 词：卷积神经网络 海洋平台 损伤识别 实验系统 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TU317[自动化与计算机技术—控制科学与工程]