摘要
针对飞机结构健康管理中高精度实时寿命预测与智能运维的迫切需求,提出了一种基于飞参-载荷-寿命数字孪生模型的飞机结构健康管理方法。首先,通过飞行参数和结构关键部位载荷实测数据,训练基于极致梯度提升(XGBoost)增量学习的飞参-载荷数字孪生模型,实现关键部位载荷的高精度动态映射。其次,通过参数化模型仿真数据,训练基于非侵入式降阶的载荷-应力场数字孪生模型,实现关键部位应力场的高精度动态重构。进而,基于细节疲劳额定值(DFR)法构建疲劳寿命评估模型,通过上述数字孪生模型预测结果计算疲劳寿命评估模型参数,预测寿命损耗,实现应力场-剩余寿命的高精度动态预测,形成飞参-载荷-寿命数字孪生模型。在此基础上,构建了机队维修及飞行任务多目标规划模型,并通过智能优化算法生成历史规划问题解集。然后,基于聚类中心距离和决定系数指标,量化历史规划(源域)问题和新规划(目标域)问题在解集分布和特征空间上的相似程度。最后,将高相似源域问题解迁移为目标域问题的初始种群,形成基于历史信息迁移学习的机队维修及飞行任务智能规划方法,实现机队寿命高效管理。经典型飞行试验数据集验证,本方法能够动态预测结构关键部位载荷及剩余寿命,载荷预测误差为5.30%,寿命预测误差为-7.19%。同时,针对考虑20架飞机的复杂维修和飞行任务规划问题,本方法相比直接优化方法效率分别提升33.9%和14.5%,验证了提出方法的有效性。
Aimed at the urgent needs of high-precision real-time life prediction and intelligent operation and mainte-nance in aircraft structural health management,this paper proposes an aircraft structural health management approach based on flight parameter-load-life digital twin models.First,measured data of flight parameters and loads in structural key parts are used to train flight parameter-load digital twin models based on the incremental learning eXtreme Gradi-ent Boosting(XGBoost),so as to realize high-precision dynamic mapping of the loads of the key parts.Second,simulated data of parametric models are used to train load-stress field digital twin models based on the non-intrusive reduced-order technique to realize high-precision dynamic reconstruction of the stress field of the key parts.Further-more,the fatigue life estimation model is constructed based on the Detail Fatigue Rating(DFR)method.The param-eters of the fatigue life estimation model are calculated from the prediction results of the above digital twin models,and the life consumption is estimated to realize the high-accuracy dynamic prediction of the remaining life,forming a flight parameter-load-life digital twin model.On this basis,fleet maintenance and flight task multi-objective planning models are constructed respectively,and solution sets of historical planning problems are formed by intelligent optimization al-gorithm.Then,based on clustering center distance and coefficient of determination metrics,similarity between histori-cal planning(source-domain)problems and new planning(target-domain)problem in terms of solution set distribution and feature space is quantified.Finally,the highly similar source-domain problem solutions are transferred to the initial population of the target-domain problems to form an intelligent planning method for fleet maintenance and flight task based on transfer learning of historical information,which realizes the intelligent and efficient management of fleet life.The results of a typical flight-testing example show that this method can dynamically predict the loads and remaining life of the structural key parts with a load prediction error of 5.30%and a life prediction error of-7.19%.Meanwhile,for the complex maintenance and flight task planning problems of 20 aircraft,the efficiency of the proposed method is im-proved by 33.9%and 14.5%,respectively,compared with the direct optimization method,which verifies the effective-ness of the proposed method.
作者
黄蕾
郭聪
张小波
孙良臣
左迎荟
王博
田阔
HUANG Lei;GUO Cong;ZHANG Xiaobo;SUN Liangchen;ZUO Yinghui;WANG Bo;TIAN Kuo(Department of Engineering Mechanics,Dalian University of Technology,Dalian 116024,China;China Southern Technic Shenyang Base,Shenyang 110169,China;Liaoning Provincial Key Laboratory of Digital Twins for Structural Strength of Aircraft,Shenyang 110035,China)
出处
《航空学报》
北大核心
2025年第21期143-161,共19页
Acta Aeronautica et Astronautica Sinica
基金
国家自然科学基金(124B2037,12572391)
辽宁省人工智能领域科技创新项目(2023JH26/10100007)
辽宁省优秀青年基金(2024JH3/10200003)
陕西省自然科学基础研究计划(2025SYS-SYSZD-102)。
关键词
结构健康管理
数字孪生
降阶模型
疲劳寿命预测
飞行和维修规划
structural health management
digital twin
reduced-order model
fatigue life prediction
flight and maintenance planning
