基于Bayes数据融合和改进鲸鱼算法的可展开天线结构损伤识别研究

金路 ,  刘安清 ,  田大可 ,  赵丙峰

工程科学与技术 ›› 2026, Vol. 58 ›› Issue (03) : 48 -58.

PDF (2425KB)
工程科学与技术 ›› 2026, Vol. 58 ›› Issue (03) : 48 -58. DOI: 10.12454/j.jsuese.202500063
可展开结构及其在航天工程中的应用

基于Bayes数据融合和改进鲸鱼算法的可展开天线结构损伤识别研究

作者信息 +

Research on Damage Identification of Deployable Antenna Structure Based on Bayes Data Fusion and Improved Whale Optimization Algorithm

Author information +
文章历史 +
PDF (2482K)

摘要

为了提高大型复杂工程结构损伤识别精度,本文引入两阶段损伤识别策略,结合Bayes数据融合和改进鲸鱼优化算法,提出一种改进的两阶段结构损伤识别方法。首先,通过Bayes数据融合将跨模型模态应变能指标和跨模型模态应变能变化率指标融合得到新的损伤定位指标,进而定位复杂结构损伤的具体位置。其次,针对鲸鱼优化算法存在容易陷入局部最优、全局搜索能力不足等缺陷,采用Sobol序列初始化种群、双种群并行搜索和交流等4种策略加以改进,提出改进的鲸鱼优化算法。最后,采用改进的鲸鱼优化算法修正先前的损伤定位结果,并确定结构杆件实际损伤程度。对含有415个杆件的大口径空间可展开天线结构开展损伤识别研究,研究表明:在考虑噪声的随机干扰下,通过Bayes数据融合方法融合跨模型模态应变能指标和跨模型模态应变能变化率指标进行损伤定位,能有效降低非损伤杆件被误判为疑似损伤杆件的可能,相比仅采用跨模型模态应变能指标,可得到更准确的定位结果;在0.2%噪声干扰下,采用改进的鲸鱼优化算法在第一阶段损伤定位基础上准确识别损伤实际发生位置和具体程度,抗噪性良好。该两阶段结构损伤识别方法可为大型复杂工程结构损伤识别提供参考。

Abstract

Objective With increasingly stringent requirements for satellite precision in fields such as deep space exploration, remote sensing and navigation, and military reconnaissance, large-aperture modular deployable antennas are becoming a key development trend in astronautical deployable antenna systems. However, due to factors such as the disintegration or explosion of defunct spacecraft, the number of small space debris objects continues to increase year by year, posing a significant threat to their on-orbit service. There is currently limited research on damage identification for large and complex structures, such as deployable antennas. In such cases, accurately identifying damage presents significant challenges, often leading to the misclassification of undamaged components and resulting in reduced identification accuracy. In this paper, an improved two-stage structural damage identification method is proposed by combining Bayesian data fusion and an improved whale optimization algorithm. Numerical simulations of damage identification are carried out on a large-aperture space deployable antenna support structure with 415 rods to verify the identification accuracy of the two-stage method under a certain level of noise interference. Methods First, six damage conditions are designed based on three types of damage: single damage, double damage, and multiple damage. The Bayesian data fusion method is adopted to integrate the cross-model modal strain energy index (CMSEI) and the cross-model modal strain energy change rate index (CMSECR). A damage threshold is defined, and rods with values exceeding this threshold are identified as suspected damaged rods, thereby enabling localization of structural damage. The localization accuracy of Bayesian data fusion is then compared with that of using only CMSEI under noise levels of 0, 0.1%, and 0.2%. Second, to overcome the limitations of the whale optimization algorithm (WOA), such as its tendency to fall into local optima and its insufficient global search capability, a hybrid multi-strategy whale optimization algorithm (HMWOA) is proposed. This approach incorporates multiple improvement strategies, including Sobol sequence-based population initialization and parallel search with dual populations and communication mechanisms. The performance of the proposed algorithm is compared with that of WOA and five other optimization algorithms using six benchmark test functions. Finally, after Bayesian data fusion is used to identify the suspected damaged rods of the deployable antenna support structure in the first stage, the HMWOA algorithm is applied in the second stage to quantify the damage. This process further eliminates non-damaged rods and determines the damage severity of the actual damaged rods. Results and Discussion From the damage localization results of single-damage scenario 1, it can be seen that both methods, CMSEI and Bayesian data fusion, can clearly identify the actual damaged rods at noise levels of 0, 0.1%, and 0.2%. However, Bayesian data fusion achieves more accurate damage localization and reduces the possibility of misclassification, especially at a noise level of 0.2%. From the damage localization results of multi-damage scenarios 4 to 6, it can be observed that the sensitivity of different rod types to structural damage varies. The central vertical rods are the least sensitive to damage, followed by the edge vertical rods, while the chord and diagonal web rods are more sensitive. Therefore, it is necessary to define different damage thresholds for different types of rods. Based on the results of the three working conditions, the damage thresholds are set to 0.001 for central vertical rods, 0.03 for edge vertical rods, and 0.2 for chord and diagonal web rods. The effect of noise on multi-damage scenarios is more significant than that on single-damage scenarios, increasing the number of non-damaged rods misclassified by both methods for damage localization. However, the use of Bayesian data fusion effectively reduces the effect of noise. For damage scenario 4, the damage identification results using CMSEI show that, under 0.2% noise, 31 rods are identified as suspected damaged rods, compared to 8 rods under noise-free conditions. In contrast, using Bayesian data fusion under 0.2% noise results in 11 suspected damaged rods, while only 3 rods are identified in the noise-free case. To verify the solution accuracy and convergence speed of HMWOA, we compare the results of WOA, an improved whale optimization algorithm (MSWOA), the sailfish optimization algorithm (SFO), the sparrow search algorithm (SSA), and the zebra optimization algorithm (ZOA). Six benchmark test functions from CEC2005 are selected to evaluate the performance of the different algorithms. To reduce the influence of randomness on the experimental results, we conduct 30 independent runs and calculate the mean and standard deviation of the solutions. The results indicate that the HMWOA algorithm demonstrates superior accuracy and stability in optimizing functions F1 to F6. In particular, for functions F3, F5, and F6, the advantage of the HMWOA algorithm is most pronounced. Compared with other algorithms, HMWOA also maintains the fastest optimization speed across all six benchmark test functions, significantly reducing computation time. In the damage quantification stage, due to the high accuracy of damage localization using Bayesian data fusion at noise levels of 0 and 0.1%, only the case with a noise level of 0.2% is examined. From the three scenarios of single-damage scenario 1, double-damage scenario 3, and multi-damage scenario 4, it can be seen that, after using Bayesian data fusion in the first stage to identify suspected damaged rods, the HMWOA algorithm in the second stage can further identify the actual damaged rods and accurately determine the damage severity. The damage quantification error is kept within 5%, and no non-damaged rods are misclassified. Conclusion The results show that using Bayesian data fusion to integrate CMSEI and CMSECR for damage localization yields more accurate identification results. The four improvement strategies effectively enhance the optimization accuracy and convergence speed of the WOA. For complex structures with a large number of rods, the two-stage damage identification method based on Bayesian data fusion and the improved WOA proposed in this paper can accurately identify structural damage.

Graphical abstract

关键词

结构损伤识别 / Bayes数据融合 / 改进鲸鱼算法 / 跨模型模态应变能 / 可展开天线结构

Key words

structural damage identification / Bayes data fusion / improved whale optimization algorithm / cross-model modal strain energy / deployable antenna structure

引用本文

引用格式 ▾
金路,刘安清,田大可,赵丙峰. 基于Bayes数据融合和改进鲸鱼算法的可展开天线结构损伤识别研究[J]. 工程科学与技术, 2026, 58(03): 48-58 DOI:10.12454/j.jsuese.202500063

登录浏览全文

4963

注册一个新账户 忘记密码

本刊网刊
在长期服役过程中,工程结构长期受荷载作用、材料老化及腐蚀等诸多因素影响,会因累积损伤导致结构承载能力下降,从而影响结构的安全性能[1]。有效的损伤识别方法能够准确反映结构损伤现状,从而降低风险,保障安全。目前,通过模态参数[2]定位分析结构损伤是结构损伤识别的常用方法。史治宇等[3]提出结构单元模态应变能的理念,并采用单元模态应变能变化率指标对一个二维结构进行损伤位置识别;刘晖等[4]考虑单元的模态应变能耗散率会引起单元模态应变能的变化,提出了基于应变能耗散率理论的识别方法;Seyedpoor[5]提出了模态应变能基指标,可提高损伤定位的精度;郭惠勇等[6]提出了模态应变能等效指标,并比较了不同损伤指标的识别能力;伍晓顺等[7]提出了跨模型模态应变能指标,通过包含392根杆的网架结构仿真算例验证了其有效性。
基于模态应变能的损伤识别方法对结构损伤敏感,但应用于大型复杂结构时,单元数目制约了结构损伤识别精度及效率。由此提出减少待识别变量数目这一思路,引入两阶段损伤识别理念。在损伤定位阶段,由于单一指标的局限性,诸多学者采用主成分分析[89]、Bayes理论[1011]、D‒S证据理论[1213]和卡尔曼滤波[14]等方法,融合多个指标进行损伤定位,减少单一指标对结果产生误判的现象;在损伤量化阶段,将结构损伤识别问题视为数学上的优化问题,设定目标函数,优化求解确定结构单元损伤程度。随着计算机和优化算法的发展,粒子群算法[15]、向日葵优化算法[16]、灰狼优化算法[17]、蚁狮算法[18]等智能优化算法已取得较好效果。
在轨服役期间,星载可展开天线长期暴露于极端热交变、空间碎片超高速撞击及强空间辐照等复杂空间环境载荷作用下,易导致其支承结构发生损伤,影响结构在轨运行性能。大口径模块化的可展开天线结构复杂、杆件数量众多,对其损伤识别的精度和计算效率带来了巨大挑战。为实现精准确定结构杆件损伤位置和程度,本文针对星载大口径可展开天线在恶劣空间环境下的在轨运行需求,引入两阶段损伤识别策略,提出了基于Bayes数据融合和改进鲸鱼优化算法的两阶段识别方法,并考虑噪声干扰对该方法损伤识别精度的影响,对天线支承结构展开损伤识别研究。

1 基于Bayes数据融合的结构损伤识别

1.1 损伤定位指标

结构损伤前、后,第ii=1,2,,mm为模态总阶数)阶模态下任一单元jj=1,2,,NN为结构单元总数)的模态应变能[3]UijŨij分别为:

Uij=ΦiTKjΦi
Ũij=Φ̃iTKjΦ̃i

式(1)、(2)中:Κj为单元j在整体坐标系下的单元刚度矩阵,扩阶到与整体刚度矩阵同阶;ΦiΦ̃i分别为未损结构和损伤结构第i阶模态下的振型向量。

伍晓顺等[7]在此基础上提出了跨模型模态应变能的概念。定义单元j在结构损伤后对应模态i的跨模型模态应变能Ũijc和跨模型模态应变能增量Eij的表达式分别为:

Ũijc=ΦiTKjΦ̃i
Eij=Ũijc-Uij

跨模型模态应变能指标(cross-model modal strain energy index,CMSEI,记为ICMSE)被定义为:

ICMSE,j= max0,1mi=1mEijEi,max

式中,Ei,max=max(Eij)

类比于模态应变能变化率指标[3]的构建方法,构建跨模型模态应变能变化率指标(cross-model modal strain energy change rate,CMSECR,记为ICMSECR),其表达式为:

ICMSECR,j=max0,1mi=1mFijFi,max

式中:Fij为中间变量,Fij=Eij/UijFi,max=max(Fij)

1.2 Bayes数据融合

在随机试验中,有n个相互独立事件A1,A2,,Ak,,An,有且只有一个事件Ak发生的概率用P(Ak)表示,即k=1nP(Ak)=1。假定B为任意一个事件,根据条件概率的定义及全概率的公式,Bayes的后验概率P(AkB)表达式为:

P(AkB)=P(AkB)P(A)=P(BAk)P(Ak)k=1nP(BAk)P(Ak)

将Bayes数据融合理论引入结构健康监测时可以表示为:假定A1,A2,,Ak,,Ann个待识别的相互独立目标(或n个单元),S1,S2,,Sll个传感器(或l个评价指标),确定待识别目标的先验概率P(Ak)后,目标Ak的后验概率为:

P(AkS1,S2,,Sl)=P(SAk)P(Ak)p=1nP(SAp)P(Ap)=P(S1Ak)P(S2Ak)P(SlAk)P(Ak)p=1nP(S1Ap)P(S2Ap)P(SlAp)P(Ap)

式中,P为待识别相互独立目标下标,P=1,2,,n

本文采用两种评价指标,故l=2。考虑损伤产生位置的不可预知,假设各单元的损伤先验概率相等,即P(Ak)=1/nk=1,2,,n。则目标Ak 的后验概率为:

P(Ak|S1,S2)=S1(k, 1) S2(k, 1)p=1nS1(p, 1) S2(p, 1)

2 混合多策略改进的鲸鱼优化算法

2.1 改进的鲸鱼优化算法

鲸鱼优化算法(WOA)是Mirjalili等[19]将座头鲸捕食方式转化成数学模型而提出的智能优化算法,具有调节参数少、结构简便等优点,但其存在容易陷入局部最优、全局搜索能力不足等缺陷。本文在基本WOA算法的基础上,采用以下4种改进策略,提出了改进的鲸鱼优化算法(hybrid multi-strategy whale optimization algorithm,HMWOA)。

2.1.1 Sobol序列初始化种群

采用Sobol序列对种群进行初始化,将尽可能均匀的点填充至搜索空间中,对于大部分问题有利于提高算法性能。在[-1,1]范围内产生规模为500的种群,随机生成和Sobol序列生成种群在二维空间的分布如图1所示。从图1可以看出,采用Sobol序列所得到的种群分布均匀,且对解空间的覆盖更完整,遍历性更广。

2.1.2 双种群并行搜索和交流机制

基本鲸鱼优化算法采用单种群搜索。为降低其搜索陷入局部最优的可能性,同时加大全局搜索范围,采用双种群并行搜索的搜索机制,确定最优位置。每次迭代完成后两个种群的最优个体进行交流,确认新的最优个体。

定义两个子种群分别为P1P2,每次迭代后搜索到各自的最优个体分别为O1*O2*,针对两个子种群之间提出的一种新的交流机制为:

G(g)=G(g-1)+r1(F(g)-M(g))+            r2(F(g)-O*(g-1))+            r3(O*(g-1)-M(g)),g2
Oc*(g)=r4O*(g-1)+G(g)

式中:Oc*(g)为第g次迭代后经过两个子种群交流后得到的个体;O*(g-1)为第g-1次迭代中全局最优个体,是第g-1次迭代中O1*O2*O*(g-1)三者最优的个体;G(g)为第g次迭代的变化量;F(g)M(g)分别表示第g次迭代时O1*O2*比较下较优和较次的个体;r1r2r3r4为范围在[0,1]之间的随机数。定义O*(1)=F(1),即在第1次迭代中,双种群最优个体中较好的个体被视为全局最优个体。从第2次迭代开始,全局最优个体由上一次全局最优个体和当前的变化量确定。

2.1.3 变异反向学习

对两个子种群P1P2分别引入不同的反向学习策略,并采用柯西变异加以扰动,提高种群的多样性,降低陷入局部极值的可能性。

1)透镜成像反向学习

图2为透镜成像反向学习示意图。假设某一空间中,Da是将一个高度为h的物体Wa投影到x轴上得到的点,坐标为Xx轴的上下限为lbub,在原点o上放一个焦距为f的凸透镜,原点o的坐标为(lb+ub)/2,通过凸透镜成像可以得到一个高度为h* 的物体Wb,此时物体Wbx轴上投影得到的点为Db,坐标为X*uv分别为点Da、Db到原点o的距离。

根据凸透镜成像原理可以得出:

(lb+ub)/2-XX*-(lb+ub)/2=hh*

h/h*=λ,通过变换得到X*的表达式:

X*=(lb+ub)2+(lb+ub)2λ-Xλ

式(13)可知,当前解与反向解之间的距离由λ决定。当λ=1时,为反向学习策略。本文取λ=1+r5r5为[0,1]中的随机数。

2)动态反向学习

在反向学习的基础上引入随机数提高反向学习的随机性,计算式如下:

X*=X+αr6(lb+ub-X)
α=1,r7>0.5;-1,r70.5

式(14)、(15)中,α为中间变量,r6r7为[0,1]中的随机数。

3)柯西变异

鲸鱼个体位置的迭代更新由最优个体引导,在原有算法中加入柯西变异扰动算子,对两个子种群最优个体分别进行柯西扰动,扩大算法的搜索空间,增加种群的多样性,其模型为:

Xnew=Xbest(1+cauchy(0, 1))

式中:Xnew为柯西变异后的个体;Xbest为子种群的最优位置;cauchy(0,1)为柯西算子,即服从柯西分布的随机数。

为了保证变异之后的解优于原始解,对比扰动前后解的适应度值,择优保留。

2.1.4 非线性时变因子

参数A是WOA算法中调节全局勘探和局部开发的重要参数,主要受收敛因子a的影响。WOA算法中收敛因子a随迭代次数的增加而线性下降,使得算法前期搜索猎物不彻底,且后期围猎过程收敛速度较慢。为了改善收敛因子a对算法的影响,提出一种分段非线性收敛因子:

a=-4tT2+2,t0.5T;1+cosπtT,t>0.5T

式中,T为最大迭代次数,t为迭代次数。

图3为收敛因子a与参数A在最大迭代次数为100时分布。

图3可知,本文提出的改进收敛因子和原始收敛因子相比在迭代早期较大,后期较小,从而在没有改变全局勘探和局部开发的分割点的同时,增强了算法前期全局勘探和后期局部开发能力。

2.2 算法性能评价

将提出的HMWOA算法与WOA算法、基于混合策略改进的鲸鱼优化算法(MSWOA)[20]、旗鱼优化算法(SFO)[21]、麻雀搜索算法(SSA)[22]以及斑马优化算法(ZOA)[23]在6个基准测试函数中进行比较,函数见表1

为公平比较,对每个测试函数,各算法均独立运行30次,并记录每次寻得的最小值,进而计算各算法所得最优值的平均值(Mean)与标准差(Std)。最大迭代次数设置为1 000次,种群大小P设置为30,HMWOA算法的两个子种群的大小为P1=P2=15,6种算法在基准测试函数的优化结果见表2。从表2可知,在函数F1F6的寻优中,HMWOA算法寻优结果的准确度和稳定性均最高,尤其针对函数F3F5、F6,HMWOA算法的优势最为明显。

图4为各算法的收敛曲线。由图4可知,本文提出的HMWOA算法相比其他5种算法,在6个基准函数优化过程中均保持最快寻优速度,可有效节省寻优时间。

3 可展开天线支承结构损伤识别

3.1 有限元模型及工况设定

基于Jin[24]和田大可[25]等提出的模块化可展开天线支承结构开展损伤识别研究[2425],模块化可展开天线支承结构由19个单模块通过标准连接件组合而成,支承结构基本单元如图5所示。标准单模块支承单元主要由12根交叉布置的张紧索及6个放射状分布的肋单元组成,肋单元为具有展开锁定功能的多连杆机构,通过节点处的铰链实现天线结构的展开与锁定,各杆件均为刚性杆,不可伸缩。采用软件ANSYS Mechanical建立结构的有限元模型(图6),结构杆件采用铝合金2A12,张紧索仅受拉且施加200 N预紧力,采用不锈钢。弦杆、斜腹杆及边缘竖杆直径为10 mm,中心竖杆直径为12 mm,厚度均为1 mm,均采用BEAM188单元进行模拟。张紧索直径为1.2 mm,采用LINK180单元进行模拟。将弹簧驱动模块及连接件等部件等效成点质量施加于中心竖杆两端节点,采用MASS 21单元进行模拟。为真实反映天线结构和卫星连接方式,在结构边缘竖杆处施加约束模拟伸展臂与天线连接支点。

支承结构共415根杆件,鉴于其杆件数量众多、拓扑关系复杂的特点,参考文献[2627]中损伤识别的情况,从结构整体出发,以确定具体损伤杆件为目标,未考虑杆件的局部损伤或结构节点损伤,将单根杆件划分为一个单元,因此,共146个节点,杆件单元总数为415。

在损伤力学中,基于应变等效假定,弹塑性材料的损伤程度可通过其有效弹性模量来表征。一般来说,实际构件受载时,内部应力状态通常不均匀,为简化分析,假定各点应力状态相同,则可通过折减整个构件的弹性模量来表示其整体损伤程度,因此,本文通过弹性模量折减的方法模拟结构损伤[4]。考虑单损伤、两损伤以及多损伤3种损伤类型并划分损伤工况,对不同程度的结构损伤进行度量,本文工况的损伤程度处于10%~50%之间,具体损伤工况见表3

3.2 结构损伤定位

ICMSE指标和ICMSECR指标通过Bayes数据融合方法构建新的损伤定位指标,记为IDF,分别采用ICMSEIDF指标对3类工况损伤定位,取结构的前10阶模态信息,同时考察噪声对定位效果的影响,对振型添加噪声:

Φ̃ikS=Φ̃ikM(1+γnrandΦ̃max,iM)

式中:Φ̃ikSΦ̃ikM分别为损伤结构第i阶模态在第k个自由度的含噪、无噪振型数据;γ为噪声水平,本文取0、0.1%、0.2%;nrand为均值为0、方差为1的标准正态分布随机数;Φ̃max,iM为无噪声影响下第i阶模态中节点位移最大绝对值。

由于噪声在每一次测量过程中具有随机性,本文将多次计算,最终取代表性结果。对计算结果归一化处理,初始将0.2倍所有杆件结果的最大值设置为阈值[28],即损伤阈值为0.2,计算结果大于阈值的杆件判定为疑似损伤杆件。以单损伤工况1和多损伤工况4~6为例,识别结果如图79所示。

图7为损伤工况1在不同程度噪声干扰下分别采用ICMSEIDF指标进行损伤定位的结果。由图7可以看出,在噪声水平为0、0.1%和0.2%的3种情况下,两种方法都能够明确指示出损伤杆件96,噪声水平越大,未损伤杆件对结果的干扰也越大。在噪声水平为0和0.1%时,采用ICMSE指标损伤定位时会误判少量非损伤杆件,而IDF指标未出现误判。当噪声水平升至0.2%时,ICMSE指标误判的非损伤杆件数达13,而IDF指标误判数为2。由此表明,针对单损伤工况,相比于ICMSE指标,采用IDF指标能够更精准地实现损伤定位,减少杆件误判的可能。

图8为多损伤3种工况(工况4~6)在无噪声干扰下分别采用ICMSEIDF指标进行损伤定位的结果。由图8可知,3种工况下采用IDF指标均比采用ICMSE指标定位更加精准,但将所有杆件损伤阈值都设置为0.2时,损伤工况4中实际损伤杆件25、损伤工况5中实际损伤杆件39及损伤工况6中实际损伤杆件15和54都将会被漏判,可见不同杆件类型对于结构损伤的敏感程度有所不同,支承结构中心竖杆对于损伤最不敏感,边缘竖杆次之,而弦杆和斜腹杆对损伤较为敏感。经试算,为不漏判损伤杆件,需要对不同类型杆件设置不同的损伤阈值,综合以上3种工况的损伤识别结果,建议确定中心竖杆的损伤阈值为0.001,边缘竖杆的损伤阈值为0.030,弦杆及斜腹杆的损伤阈值为0.200为宜。

图9为多损伤3种工况在0.2%噪声干扰下分别采用ICMSEIDF指标进行损伤定位的结果。由于存在0.2%水平的噪声干扰,两种方式判别的疑似损伤杆件数目都有所增加,但采用IDF指标时,非损伤杆件被误判的可能性更小。例如损伤工况4在0.2%噪声干扰下,采用ICMSE指标进行损伤识别得到的疑似损伤杆件数目为31,而无噪声下疑似损伤杆件数目为8;在0.2%噪声干扰下,采用IDF指标进行损伤识别得到的疑似损伤杆件数目为11,而无噪声下疑似损伤杆件数目为3。

3.3 结构损伤量化

3.3.1 目标函数

结构损伤会引起模态特征参数(如频率、振型等)的改变,通过模态特征构建目标函数,求解目标函数,迭代寻优,修正损伤识别因子,使得有限元模型与模态实测值之间的差异最小化。然而,仅以最小化模型计算与实际测量特征误差为目标,在噪声及不完备测量等影响下,容易出现识别结果精度不高的问题。考虑到实际损伤具有稀疏分布特征[29],本文采用L1正则化的方式施加稀疏性约束,提高结构损伤识别精度。

由于大口径可展开天线支承结构杆件数目繁多,少量杆件损伤对结构频率的影响较小,故而本文仅在以振型为特征的基础上,引入L1正则化,目标函数αopt如下:

αopt=argα minϕ(α)+βα1=          argα minϕ(α)+βj=1Nαj

式中:ϕα为中间变量,ϕ(α)=i=1mΦ˜iTΦ˜i-ΦiTΦiFΦiTΦiFα 为结构损伤识别因子向量;F为F范数;αj为第j个单元的损伤程度,0αj0.99β为正则化参数,表示系数约束的参与比重,通常由试算确定,本文β=1×10-41为1范数。

3.3.2 损伤量化

当噪声水平为0或0.1%时,采用IDF指标损伤定位较为精准,判定为疑似损伤杆件的数目较少,因此本文在损伤量化阶段仅考察噪声水平为0.2%的情况。以单损伤工况1、两损伤工况3及多损伤工况4为例,对振型添加0.2%的噪声,采用IDF指标判定的疑似损伤杆件如表4所示。

采用提出的HMWOA算法对疑似损伤杆件进一步识别,HMWOA算法的两个子种群的数量都为80,最大迭代次数设置为100,3个工况的损伤量化结果如表5所示。

表5可以看出,在0.2%噪声干扰下,在第1阶段采用IDF指标进行损伤定位确定疑似损伤杆件之后,第2阶段采用HMWOA算法能够在疑似损伤杆件中进一步确定实际损伤杆件并较为准确地识别其损伤程度,损伤量化误差保持在5.00%以内,且没有非损伤杆件的误判。

4 结 论

本文提出了一种基于Bayes数据融合和改进鲸鱼算法的两阶段结构损伤识别方法,通过对415杆的可展开天线支承结构数值模型单损伤、两损伤、多损伤数值模拟,验证了本文所提识别方法的有效性和可行性。具体结论如下:

1)选取6个基准测试函数进行优化实验,通过Sobol序列初始化种群,采用双种群并行搜索和交流机制,对种群施加变异反向学习及提出改进的非线性时变因子这4种改进策略能较好地提高鲸鱼优化算法的寻优精度和收敛速度,解决了其容易陷入局部最优、全局搜索能力不足的问题。

2)采用Bayes数据融合综合跨模型模态应变能指标和跨模型模态应变能变化率指标的判断结果,相比于单独采用跨模型模态应变能指标,在噪声水平为0~0.2%下,损伤定位更加精准,有效减少对非损伤杆件的误判。

3)可展开天线结构损伤识别表明,在0.2%噪声干扰下,经过Bayes数据融合进行损伤定位得到疑似损伤杆件后,采用改进的鲸鱼优化算法能够精准确定实际损伤杆件,且损伤程度识别误差小于5%。针对复杂、杆件繁多的结构,本文提出的基于Bayes数据融合和改进鲸鱼算法的两阶段损伤识别方法能够精准识别结构损伤,为可展开天线在轨维护提供参考。

经试算,该两阶段识别方法不仅适用于本文的19模块的模块化可展开天线支承结构,对于杆件规模相似的其他杆系结构,例如三折叠树状可展开天线、辐射肋式天线[30]等,同样能够实现良好且可靠的损伤识别效果,展现了该方法在该类结构损伤识别研究的潜在应用价值。目前,受限于实物样机在轨部署环境难以模拟,获取真实在轨工程环境下的结构响应数据存在较大挑战,未来将进一步研究缩尺模型的在轨环境构建及地面实验,并考虑如何利用稀疏的、在轨可行的传感器布局和航天器固有的环境激励,结合先进的信号处理与特征提取技术,实现基于有限响应的损伤定位。

参考文献

[1]

Mishra M, Lourenço P B, Ramana G V.Structural health monitoring of civil engineering structures by using the internet of things: A review[J].Journal of Building Engineering,2022,48:103954. doi:10.1016/j.jobe.2021.103954

[2]

Bao Longsheng, Cao Yue, Zhao Ning,et al.Application of BP neural network and curvature mode theory in bridge damage identification[J].Journal of Shenyang Jianzhu University(Natural Science),2021,37(2):296‒302.

[3]

包龙生,曹悦,赵宁,.BP神经网络和曲率模态理论在桥梁损伤识别中的应用[J].沈阳建筑大学学报(自然科学版),2021,37(2):296‒302.

[4]

Shi Zhiyu, Law S S, Zhang Lingmi.Determination of structural damage location based on elemental modal strain energy change[J].Journal of Vibration Engineering,1998(3):109‒113. doi:10.1088/0256-307X/16/9/020

[5]

史治宇,罗绍湘,张令弥.结构破损定位的单元模态应变能变化率法[J].振动工程学报,1998(3):109‒113. doi:10.1088/0256-307X/16/9/020

[6]

Liu Hui, Qu Weilian, Yuan Runzhang.Structural damage detection method based on the theory of dissipationratio of modal strain energy[J].Journal of Vibration and Shock,2004(2):120‒123. doi:10.4203/ccp.83.294

[7]

刘晖,瞿伟廉,袁润章.基于模态应变能耗散率理论的结构损伤识别方法[J].振动与冲击,2004(2):120‒123. doi:10.4203/ccp.83.294

[8]

Seyedpoor S M.A two stage method for structural damage detection using a modal strain energy based index and particle swarm optimization[J].International Journal of Non-Linear Mechanics,2012,47(1):1‒8. doi:10.1016/j.ijnonlinmec.2011.07.011

[9]

Guo Huiyong, Sheng Mao.Comparision of different damage indices based on modal strain energy[J].Journal of Hohai University(Natural Sciences),2014,42(5):444‒450. doi:10.3876/j.issn.1000-1980.2014.05.013

[10]

郭惠勇,盛懋.基于模态应变能的不同损伤指标对比[J].河海大学学报(自然科学版),2014,42(5):444‒450. doi:10.3876/j.issn.1000-1980.2014.05.013

[11]

Wu Xiaoshun, Xia Juwei, Hu Yuefang.Damage localization of space trusses based on indicators expressed by cross-model modal strain energy[J].Journal of Zhejiang University(Engineering Science),2020,54(2):248‒256. doi:10.3785/j.issn.1008-973X.2020.02.005

[12]

伍晓顺,夏巨伟,胡岳芳.基于跨模型模态应变能指标的网架结构损伤定位[J].浙江大学学报(工学版),2020,54(2):248‒256. doi:10.3785/j.issn.1008-973X.2020.02.005

[13]

Liu Yuchi, Jiang Yufeng, Wang Shuqing,et al.Data fusion and residual convolutional auto-encoder based structural damage identification[J].Journal of Vibration and Shock,2023,42(4):194‒203.

[14]

刘玉驰,蒋玉峰,王树青,.基于数据融合及残差卷积自编码器的结构损伤识别方法[J].振动与冲击,2023,42(4):194‒203.

[15]

Cao Shancheng, Guo Ning, Xu Chao.Robust damage localization in plate-type structures by using an enhanced robust principal component analysis and data fusion technique[J].Mechanical Systems and Signal Processing,2022,162:108091. doi:10.1016/j.ymssp.2021.108091

[16]

Li Xueyan, Ye Xinquan, Law S S.Damage identification with fusion of estimates from covariance of IRF in different frequency bands[J].Mechanical Systems and Signal Processing,2019,134:106327. doi:10.1016/j.ymssp.2019.106327

[17]

Barman S K, Mishra M, Maiti D K,et al.Vibration-based damage detection of structures employing Bayesian data fusion coupled with TLBO optimization algorithm[J].Structural and Multidisciplinary Optimization,2021,64(4):2243‒2266. doi:10.1007/s00158-021-02980-6

[18]

Wang Xiaojuan, Lan Xiangyong, Zhou Hongyuan,et al.A two-stage damage recognition method based on data fusion[J].Journal of Vibration and Shock,2024,43(17):132‒144. doi:10.13465/j.cnki.jvs.2024.17.015

[19]

王小娟,兰祥勇,周宏元,.基于数据融合的两阶段损伤识别方法[J].振动与冲击,2024,43(17):132‒144. doi:10.13465/j.cnki.jvs.2024.17.015

[20]

Wang Fengdan, Li Rongpeng, Xiao Yuzhu,et al.A strain modal flexibility method to multiple slight damage localization combined with a data fusion technique[J].Measurement,2021,182:109647. doi:10.1016/j.measurement.2021.109647

[21]

Wang Yishou, He Mengyue, Sun Lei,et al.Weighted adaptive Kalman filtering-based diverse information fusion for hole edge crack monitoring[J].Mechanical Systems and Signal Processing,2022,167:108534. doi:10.1016/j.ymssp.2021.108534

[22]

Li Xiaolin, Serra R, Olivier J.A multi-component PSO algorithm with leader learning mechanism for structural damage detection[J].Applied Soft Computing,2022,116:108315. doi:10.1016/j.asoc.2021.108315

[23]

Gomes G F, Giovani R S.An efficient two-step damage identification method using sunflower optimization algorithm and mode shape curvature(MSDBI‒SFO)[J].Engineering with Computers,2022,38(2):1711‒1730. doi:10.1007/s00366-020-01128-2

[24]

Sang‒To T, Le‒Minh H, Mirjalili S,et al.A new movement strategy of grey wolf optimizer for optimization problems and structural damage identification[J].Advances in Engineering Software,2022,173:103276. doi:10.1016/j.advengsoft.2022.103276

[25]

Chen Chengbin, Yu Ling, Pan Chudong,et al.Structural damage detection based on an ant lion optimizer algorithm and trace sparse regularization[J].Journal of Vibration and Shock,2019,38(16):71‒76. doi:10.13465/j.cnki.jvs.2019.16.011

[26]

陈承滨,余岭,潘楚东,.基于蚁狮优化算法与迹稀疏正则化的结构损伤识别[J].振动与冲击,2019,38(16):71‒76. doi:10.13465/j.cnki.jvs.2019.16.011

[27]

Mirjalili S, Lewis A.The whale optimization algorithm[J].Advances in Engineering Software,2016,95:51‒67. doi:10.1016/j.advengsoft.2016.01.008

[28]

Qiu Xingguo, Wang Ruizhi, Zhang Weiguo,et al.Improved whale optimizer algorithm based on hybrid strategy[J].Computer Engineering and Applications,2022,58(1):70‒78. doi:10.3778/j.issn.1002-8331.2012-0316

[29]

秋兴国,王瑞知,张卫国,.基于混合策略改进的鲸鱼优化算法[J].计算机工程与应用,2022,58(1):70‒78. doi:10.3778/j.issn.1002-8331.2012-0316

[30]

Shadravan S, Naji H R, Bardsiri V K.The sailfish optimizer: A novel nature-inspired metaheuristic algorithm for solving constrained engineering optimization problems[J].Engineering Applications of Artificial Intelligence,2019,80:20‒34. doi:10.1016/j.engappai.2019.01.001

[31]

Xue Jiankai, Shen Bo.A novel swarm intelligence optimization approach: sparrow search algorithm[J].Systems Science & Control Engineering,2020,8(1):22‒34. doi:10.1080/21642583.2019.1708830

[32]

Trojovská E, Dehghani M, Trojovský P.Zebra optimization algorithm:A new bio-inspired optimization algorithm for solving optimization algorithm[J].Ieee Access,2022,10:49445‒49473. doi:10.1109/access.2022.3172789

[33]

Jin Lu, Li Boheng, Tian Dake,et al.Impact dynamic response of large aperture space deployable antenna supporting structures based on a dual-scale model[J].Thin‒Walled Structures,2024,195:111432. doi:10.1016/j.tws.2023.111432

[34]

Tian Dake, Guo Zhenwei, Jin Lu,et al.Deployment dynamics modeling and motion planning of modular deployable antenna mechanism[J].Journal of Mechanical Engineering,2023,59(17):56‒66. doi:10.3901/JME.2023.17.056

[35]

田大可,郭振伟,金路,.模块化可展天线机构展开动力学建模与运动规划[J].机械工程学报,2023,59(17):56‒66. doi:10.3901/JME.2023.17.056

[36]

Zeng Bin, Zhou Zhen, Zhang Qingfang,et al.Multi-position damage identification and anti-noise analysis of cable-stayed arch-truss based on data fusion[J].Journal of Building Structures,2020,41(S1):36‒43.

[37]

曾滨,周臻,张庆方,.基于数据融合的张弦桁架多位置损伤识别与抗噪性分析[J].建筑结构学报,2020,41(S1):36‒43.

[38]

Qiu Shumao, Yang Haifeng, Wu Ziyan,et al. Damage identification of truss structure based on strain mode and sparse Bayesian learning[J].Chinese Journal of Computational Mechanics,2022,39(1):63‒69. doi:10.7511/jslx20201112001

[39]

仇树茂,杨海峰,吴子燕,.基于应变模态和稀疏贝叶斯学习的网架结构损伤识别[J].计算力学学报,2022,39(1):63‒69. doi:10.7511/jslx20201112001

[40]

Miao Bingrong, Zhang Ying, Huang Zhong,et al.Structural damage identification optimization method using change rate of modal strain energy[J].Journal of Vibration Engineering,2023,36(2):477‒486.

[41]

缪炳荣,张盈,黄仲,.利用模态应变能变化率的结构损伤识别优化方法[J].振动工程学报,2023,36(2):477‒486.

[42]

Hou Rongrong, Wang Xiaoyou, Xia Yong.Sparse damage detection via the elastic net method using modal data[J].Structural Health Monitoring,2022,21(3):1076‒1092. doi:10.1177/14759217211021938

[43]

Tian Dake, Song Yuqun, Jin Lu,et al.Design and analysis of radiating rib space deployable antenna mechanism[J].Chinese Space Science and Technology,2024,44(2):40‒50.

[44]

田大可,宋玉群,金路,.辐射肋式空间可展开天线机构设计与分析[J].中国空间科学技术(中英文),2024,44(2):40‒50.

基金资助

辽宁省教育厅领军人才团队项目(U222410153096)

东北大学航空动力装备振动及控制教育部重点实验室研究基金项目(VCAME202207)

国家自然科学基金联合基金项目(U2341237)

AI Summary AI Mindmap
PDF (2425KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/