Damage Detection for Space Truss Structures Based on Strain Mode under Ambient Excitation Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. Zhao-Dong Xu1 and Ke-Yi Wu2 Abstract: Safety assurance and detection of potential damage for space truss structures have been challenging topics. The two most critical problems are considered in this paper. One is to develop an effective damage detection method based on strain data under ambient excitation, and the other is then to optimize the installment of strain sensors owing to numerous structural members in the space truss structures. A method of damage detection for space truss structures, called the environmental excitation incomplete strain mode (EEISM) method, is proposed. Four steps are taken in the EEISM method. First, strain mode parameter identification is carried out based on the cross-correlation function of the strain responses through a combination of the empirical mode decomposition method and the peak amplitude series method. Second, the strain sensors are located optimally in the space truss structures through sensitive analysis of the strain mode perturbation matrix, which are obtained by perturbation theory. Third, the modal assurance criterion (MAC) value is applied to locate the damages; that is, the members with the larger MAC values are defined as the damaged members. Finally, a damage index obtained by solving the perturbation equation is used for damage quantification. Numerical analysis of a long-span space truss structure including damage location and quantification for single-member and multimember damages, detection of the various severities of damage, and the effect of the number of sensors is performed to verify the effectiveness of the proposed EEISM method. It is shown from the analysis results that the EEISM method is effective in the location and quantification of damages for single-member and multimember damages. The quantity of the strain sensors has an effect on the damage location and has no remarkable effect on the damage quantification for the determined damage members. DOI: 10.1061/(ASCE)EM.1943-7889.0000426. © 2012 American Society of Civil Engineers. CE Database subject headings: Damage; Strain; Trusses; Excitation. Author keywords: Damage detection; Strain mode; Space truss structure; Ambient excitation; Incomplete strain mode method. Introduction Around the world, long-span space truss structures are often designed for large-scale public buildings, such as stadiums, gymnasiums, and exhibition halls. To ensure their healthy operational condition and safety, early identification of damage caused by strong wind or earthquakes in these structures is of great importance. Researchers have been committed to finding a reliable damage detection strategy for long-span space truss structures to address problems associated with inspection difficulties and the importance of considering cost. The displacement mode shape is usually adopted in structural damage detection (Pandey et al. 1991; Lam et al. 1998; Kaouk et al. 2000). However, the accuracy of damage detection tests for space truss structures with a large number of members and degrees of freedom using the complete displacement mode shape is a difficult task. Kim and Membertkowicz (2001) proposed a two-step damage detection method for large-scale structures based on an 1 Professor, Key Laboratory of C&PC Structures of the Ministry of Education, Southeast Univ., Nanjing 210096, China (corresponding author). E-mail: xuzhdgyq@seu.edu.cn 2 Doctoral Student, Key Laboratory of C&PC Structures of the Ministry of Education, Southeast Univ., Nanjing 210096, China. E-mail: keyi.wuu@ gmail.com Note. This manuscript was submitted on June 9, 2009; approved on February 23, 2012; published online on September 14, 2012. Discussion period open until March 1, 2013; separate discussions must be submitted for individual papers. This paper is part of the Journal of Engineering Mechanics, Vol. 138, No. 10, October 1, 2012. ©ASCE, ISSN 07339399/2012/10-1215–1223/$25.00. incomplete displacement mode shape. The first step was to determine the approximate damage area using the best matrix correction method, and the second step was to determine damaged members in the detected damage area using a sensitivity analysis method. The displacement mode shape would be identified from the acceleration responses by modal analysis in the time or frequency domain. In a real application, the strain sensors are also commonly applied as accelerators. Moreover, previous research has proven that strain responses are more sensitive to local damage. Yao et al. (1992) observed the strain mode changed during the damage process of a 5-story frame structure and noticed that the strain mode was superior to the displacement mode shape in detecting local damage. Unger et al. (2005) applied the strain mode in the damage location of a prestressed concrete beam and concluded that the mode shapes were sensitive to the local change of stiffness nearby the sensor locations. The objective of this paper is to develop a damage detection strategy for long-span space truss structures based on the strain mode. However, two critical problems should also be taken into consideration when the strain mode is used for damage detection of space truss structures. First, because space truss structures are usually large-scale buildings exposed to environmental conditions, the excitation input to the structures can be naturally considered to be subjected to these environments. Second, the strain sensors should be installed optimally because of the large quantity of members in the space truss structures. Therefore, a damage detection method for space truss structures, named the environmental excitation incomplete strain mode (EEISM) method, is proposed. The EEISM method consists of four steps. The first step is identification of the strain mode parameters. In this step, JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 / 1215 J. Eng. Mech., 2012, 138(10): 1215-1223 Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. a cross-correlation function of strain responses under ambient excitation is derived. It can be proven that the cross-correlation function of the strain responses under ambient excitation has a similar equation as the strain impulse response. Then, the strain mode parameters are determined by combining the empirical mode decomposition method (Huang et al. 1998) and the peak amplitude series method (Li et al. 2001) based on the cross-correlation function of the strain responses. Second, based on the perturbation theory, the optimal location of the strain sensors can be obtained through sensitivity analysis of the strain mode perturbation matrix (Chen and Garbat 1980). Third, the modal assurance criterion (MAC) value is applied to the damage location, and members with larger MAC values are defined as the damage members. Finally, the damage index obtained by solving the perturbation equation is used for damage quantification. Numerical analysis of a long-span space truss structure, including the damage location and quantification for single-member and multimember damages and the effect of the number of sensors, is performed to verify the effectiveness of the proposed method. Strain Mode Parameter Identification the strain responses can be expressed similarly to Eq. (1) as follows (Tsang 1990): «ðtÞ ¼ Cross-Correlation Function of the Strain Response Based on the vibration theory (Clough and Penzien 1993), the displacement responses of the structure can be expressed as follows as the linear combination of all mode shapes: xðtÞ ¼ n P fr qr ðtÞ ð1Þ n cp fk P r r × expð2zr vnr tÞ sinðvdr tÞ r¼1 mr vdr «pk ðtÞ ¼ xi ðtÞ ¼ n P fir fkr r¼1 mr vdr × expð2zr vnr tÞ sinðvdr tÞ 2x where fk ðuÞ 5 excitation imposed on point k, and gr ð × Þ can be defined as 1 × expð2zr vnr tÞ sinðvdr tÞ ð7Þ gr ðtÞ ¼ mr vdr If stationary white noise is imposed on point k, the crosscorrelation function between member p and member q can be expressed as (Clough and Penzien 1993) ð8Þ Rpqk ðtÞ ¼ E «pk ðt þ tÞ«qk ðtÞ It is known that there is no ideal stationary white noise in nature and normal ambient excitation can usually be regarded as one kind of wide-band random process. Research (Clough and Penzien 1993) shows that Eq. (8) is adaptive to the wide-band random process. Substituting Eq. (6) into Eq. (8), the cross-correlation function Rpqk ðtÞ can be rewritten as n P n P cpr cqs fkr fks Rpqk ðtÞ ¼ r¼1 s¼1 ðt 2x ð t1t 2x gr ðt þ t 2 sÞ gs ðt 2 uÞ E½fk ðsÞfk ðuÞdsdu ð9Þ In accordance with the theory of dynamics (Clough and Penzien 1993), the characteristic of white noise can be expressed as E½ fk ðsÞfk ðuÞ ¼ 2pS0 dðs 2 uÞ mass of where vnr and mr 5 rth natural frequency and the rth modal pffiffiffiffiffiffiffiffiffi the system without damping, respectively; vdr 5 vnr 1 2 z2r , and zr 5 rth damping ratio. Substituting Eq. (2) into Eq. (1), the displacement responses of point i under impulse excitation on point k can be expressed as ð10Þ where dð × Þ 5 Dirac function and S0 5 spectral density of white noise. Substituting Eq. (10) into Eq. (9), the cross-correlation function Rpqk ðtÞ can be written as Rpqk ðtÞ ¼ n P n P 2pS0 cpr cqs fkr fks r¼1 s¼1 ðt ð3Þ where mode shape fir 5 specific vibration equilibrium state, and the strain mode is the corresponding strain distribution state. Therefore, ð5Þ where vnr , mr , and cPr 5 rth natural frequency, rth modal mass, and rth strain mode of member p without system damping, repffiffiffiffiffiffiffiffiffi spectively; vdr 5 vnr 1 2 z2r ; and zr 5 rth damping ratio. When excitation is imposed on point k, based on the Duhamel integration, the strain response of member p can be expressed as follows: ðt n P p k cr fr × fk ðuÞgr ðt 2 uÞdu ð6Þ «pk ðtÞ ¼ where xðtÞ 5 displacement responses, fr 5 rth displacement mode shape, and qr ðtÞ 5 coordinates associated with the rth mode. If impulse excitation is imposed on point k, then qr(t) can be written as ð2Þ ð4Þ where cr 5 rth strain mode of the space truss structures. The strain impulse response of member p under impulse excitation on point k can be written as (Jones 2005) r¼1 fkr qr ðtÞ ¼ × expð2zr vnr tÞ sinðvdr tÞ mr vdr cr qr ðtÞ r¼1 r¼1 The typical method, the natural excitation technique (NExT) (James et al. 1995) was developed to identify the mode parameters under ambient excitation based on displacement responses. The basic idea of NExT can be expressed as follows. The cross-correlation function of the displacement responses under ambient excitation has an equation similar to the displacement impulse responses. Thus, the mode parameters can be identified from the cross-correlation function based on the classical mode parameter identification methods in the time domain. The strain mode parameter identification technique is also developed in this paper based on the NExT idea; i.e., to find a relationship between the cross-correlation function of the strain responses and the strain impulse response. n P 2x gr ðt þ t 2 uÞ gs ðt 2 uÞdu Assuming l 5 t 2 u, Eq. (11) can be simplified as 1216 / JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 J. Eng. Mech., 2012, 138(10): 1215-1223 ð11Þ Rpqk ðtÞ ¼ n P n P 2pS0 cpr cqs fkr fks r¼1 s¼1 ðx gr ðl þ tÞ gs ðlÞdl 0 ð12Þ In accordance with Eq. (7), gr ðl 1 tÞ can be expressed as intrinsic mode function rpqk ðtÞ corresponding to the first natural frequency (Huang et al. 1998), and the peak amplitude series method is used to obtain amplitude of rpqk ðtÞ (Li et al. 2001). This can be described as rpqk ðtÞ ¼ Apq expð2j1 vn1 tÞ sinðvd1 t þ gÞ expð2jr vnr lÞ sinðvdr lÞ mr vr expð2jr vnr lÞ cosðvdr lÞ þ ½expð2jr vnr tÞ sinðvdr tÞ mr vr ð13Þ Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. gr ðl þ tÞ ¼ ½expð2jr vnr tÞ cosðvdr tÞ Because gs ðlÞ is independent of t, by substituting Eq. (13) into Eq. (12), the cross-correlation function Rpqk ðtÞ can be written as Rpqk ðtÞ ¼ n h P Arpqk expð2jr vnr tÞ cosðvdr tÞ r ¼1 þ Arpqk Brpqk Brpqk i expð2jr vnr tÞ sinðvdr tÞ 9 n 2pS0 cp cq fk fk > P Irs > r s r s > ¼ > = 2 2 m m v J þ I r s dr rs rs s¼1 n 2pS0 cp cq fk fk > P Jrs > r s r s > ¼ > ; 2 2 m m v J þ I r s dr rs rs s¼1 ð14Þ ð15Þ Arpqk ¼ Brpqk qffiffiffiffiffiffiffiffiffiffiffiffiffi 1 2 j21 vn1 ð19Þ where Td 5 time interval between the adjacent positive (or negative) amplitude of rpqk ðtÞ. Considering the relationship vn1 5 2pf1 and Eq. (19), f1 can be written as f1 ¼ 1 qffiffiffiffiffiffiffiffiffiffiffiffi ffi Td 1 2 j21 ð20Þ assuming the time series corresponding to the positive amplitude of rpqk ðtÞ is Tð1Þ; Tð2Þ; Tð3Þ; . . . ; TðSÞ. Because S is the quantity of the positive amplitude, Td can be expressed theoretically as follows: Td ¼ Tð j þ 1Þ 2 Tð jÞ ð j ¼ 1; 2; :::; S 2 1Þ where Irs 5 2vdr ðjr vnr 1 js vns Þ and Jrs 5 ðv2ds 2 v2dr Þ 1 ðjr vnr 1js vns Þ2 . Assuming tanðgn Þ 5 In =Jn , Eq. (15) can be rewritten as 9 n 2 cpr P > 2 21=2 > brs þ I sinðg Þ J rs > rs = mr vdr s¼1 qk rs p n cr P > 2 21=2 > ¼ brs J 2 þ Irs cosðgrs Þ > ; mr vdr s¼1 qk rs vd1 ¼ 2p ¼ Td ð16Þ ð18Þ ð21Þ Usually, S cycles of the peak value attenuation curves should be applied to calculate Td to reduce error Td ¼ TðSÞ 2 Tð1Þ S21 ð22Þ Assuming the positive amplitudes of rpqk ðtÞ are rð1Þ; rð2Þ; . . . ; rðSÞ, the following equation can be obtained: rð1Þ ¼ j1 vn1 ½TðSÞ 2 Tð1Þ ð23Þ In rðSÞ Based on Eqs. (19), (22), and (23), j1 can be expressed as q k k where brs qk 5 2pS0 cs fr fs =ms . By substituting Eq. (16) into Eq. (14), the cross-correlation function of the strain responses can be expressed as p n n 2 P cr P 2 21=2 brs Rpqk ðtÞ ¼ qk Jrs þ Irs m v r dr s¼1 r¼1 expð2jr vnr tÞ sinðvdr t þ grs Þ ð17Þ Comparing Eq. (5) with Eq. (17), it can be found that the crosscorrelation function of the strain responses under white noise excitation has a similar equation to the strain impulse response. Therefore, the cross-correlation function of the strain responses can be expressed as a linear combination of a series of free attenuation responses. In practical applications, the measured noise data between two different test channels have no correlation. That is, the cross-correlation function of the measured noise data between two different test channels equals zero (Halvorsen and Brown 1977); i.e., when the cross-correlation function of the strain responses is calculated, the measurement noise can be neglected in theory. rð1Þ In rðSÞ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi j1 ¼ 2ffi rð1Þ 1 In 2pðS 2 1Þ 1 þ 2pðS 2 1Þ rðSÞ ð24Þ Substituting Eq. (22) into Eq. (20), f1 can be rewritten as f1 ¼ S 2 1qffiffiffiffiffiffiffiffiffiffiffiffiffi ½TðSÞ 2 Tð1Þ 1 2 j21 ð25Þ The following equation is tenable at peak values points of rpqk ðtÞ: drpqk tj ¼ 0 ½ti ¼ TðjÞ; j ¼ 1; 2; . . . ; S ð26Þ dtj Substituting Eq. (26) in Eq. (18), the following relationship can be obtained: qffiffiffiffiffiffiffiffiffiffiffiffiffi sin vd1 tj þ g ¼ 1 2 j21 tj ¼ TðjÞ; Strain Mode Parameter Identification j ¼ 1; 2; . . . ; S ð27Þ Based on the cross-correlation function Rpqk ðtÞ given by Eq. (17), the empirical mode decomposition method is applied to extract Consequently, rðjÞ can be rewritten as JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 / 1217 J. Eng. Mech., 2012, 138(10): 1215-1223 qffiffiffiffiffiffiffiffiffiffiffiffiffi rðjÞ ¼ Apq exp½2j1 vn1 TðjÞ 1 2 j21 ðj ¼ 1; 2; . . . ; SÞ In theory, Apq is independent of j. To improve precision, Eq. (28) should be rewritten as Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. Apq ¼ S S P 1 P 1 Aj ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffi rð jÞ exp½j1 vn1 Tð jÞ S j¼1 pq 2 j¼1 S 1 2 j1 Pr1 ¼ ð28Þ ð29Þ n P s¼1;sr lr where rð jÞ qffiffiffiffiffiffiffiffiffiffiffiffiffi exp½2j1 vn1 Tð jÞ 1 2 j21 Considering the strain mode of the member is one kind of ratio, the first strain mode of member p can be expressed as cp1 5 Apq =Aqq if member q is regarded as the reference member when calculating the strain mode, and the strain mode of member q is determined as cq1 5 1. A comparison analysis of the strain mode parameter identification will be introduced in detail subsequently. ð33Þ where n 5 quantity of members, s 5 sth strain mode; and lr and ls 5 rth and sth eigenvalues of the stiffness matrix in Eq. (32). In Eq. (33), K1« 5 small change of K« when the structure is damaged, and K1« can be expressed as K1« ¼ Ajpq ¼ 1 cT K1 c c 2 ls s « r s m P cj K « ð34Þ j¼1 where cj ð0 # cj # 1Þ 5 reduction coefficient for the strain mode stiffness of the jth damaged member, which is defined as the damage index. Substituting Eq. (34) into Eq. (33), the rth strain mode change can be expressed as Dcr ¼ ¼ n P m 1 cT P cj K « c r c s s s¼1;sr lr 2 ls j¼1 m P j¼1 cj n P 1 cT K« cr cs l 2 ls s r s¼1;sr ð35Þ Optimization of the Sensor Location Eq. (35) can be rewritten in the form of a matrix The strain modes of the structure can be obtained by the previously mentioned technique. However, it is unlikely that a strain sensor will be installed on every member because of the large number of members and economic considerations in practical applications. Here, sensor locations will be determined optimally based on the sensitivity analysis of the strain mode perturbation matrix, which can be obtained by the perturbation theory. Application of the Perturbation Theory in the Strain Mode The relationship between the rth strain mode cr and the rth displacement mode Fr can be expressed as (Yam et al. 1996) cr ¼ TFr ð30Þ where cr 5 ½ c1 c2 ⋯ cm T ; Fr 5 ½ f1 f2 ⋯ fn T ; T 5 transformation matrix from the displacement mode to the strain mode, which only depends on the geometric conditions of the structure; n 5 quantity of members; and m 5 number of degrees of freedom. The rth mode potential energy Ep can be expressed in the form of the strain mode as follows: Ep ¼ FTr KFr ¼ cTr K« cr ð31Þ where K« 5 axial stiffness of the members, which can be expressed in the form of a matrix as follows: 3 2 ðEA=lÞ1 0 7 6 ... 7 6 7 6 ðEA=lÞj K« ¼ 6 ð32Þ 7 7 6 5 4 ... 0 ðEA=lÞm where E, A, and l 5 elastic modules, cross area, and length of the member, respectively, and (EA/l)j 5 strain mode stiffness of member j. The first-order perturbation matrix for the rth strain mode can be expressed as (Chen et al. 1993) Dcr ¼ Pr1 a ð36Þ and Eq. (36) is defined as a perturbation equation, where Pr1 5 firstorder perturbation matrix of the rth strain mode, and a 5 damage index vector, a 5 ½ c1 c2 ⋯ cm T . The jth column of Pr1 represents the strain mode change of all members when unit damage ðcj 5 1Þ occurs to the jth member. Eq. (36) shows that the rth strain mode change Dcr is the linear combination of column vectors in Pr1 when unit damage occurs in various members, and the damage index cj can be considered as the combination coefficient. Optimization of the Sensor Location An analysis method on the optimization of sensor locations is introduced based on the strain modes of space truss structures and the perturbation theory. The perturbation equation [Eq. (36)] is used to estimate the quantity of damaged members, and a can be written as a ¼ ½PTr1 Pr1 21 PTr1 Dcr ð37Þ where P0 5 PTr1 Pr1 , which is defined as the Fisher information matrix. Imamovic and Ewins (1997) applied the Fisher information matrix to define sensitive members based on the displacement modes. Here, the Fisher information matrix is applied to define the sensitive members based on the strain modes. The Fisher information matrix P0 contains the perturbation information of all members. The best unbiased estimation for a can be obtained when P0 takes the largest trace value. Various members have various contributions to their trace values. Therefore, it should be ensured that the trace values of P0 are as large as possible when the sensor quantity is determined in order to obtain the best estimation of a. That is, the sensors on the members with a greater contribution to their trace values should be retained, while the sensors on the members with less of a contribution to their trace values should be discarded; thus, reasonable sensor placement can be obtained. This is the so-called sensitivity analysis for the strain mode perturbation matrix. In practice, P0 can be decomposed in accordance with each test channel 1218 / JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 J. Eng. Mech., 2012, 138(10): 1215-1223 P0 ¼ m P j¼1 T Pjr1 Pjr1 ¼ m P j¼1 Pj0 Calculating the trace value of P0 , the following equation can be obtained: tr P0 ¼ m P Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. j¼1 tr Pj0 ð39Þ Here, the contribution ratio of the damage index (CRDI) is defined as CRDI ¼ trPj0 =trP0 MACj $ MAC þ gsMAC ð38Þ ð40Þ Therefore, the sensors should be placed on the members with larger CRDI values to get the maximum trace value of P0 . An example analysis of the optimal placement of the sensors will be introduced in detail subsequently. Damage Location and Quantification When the optimal placement of sensors is determined, the incomplete strain mode can be obtained by the mode parameter identification step of the EEISM method described previously. To date, a limited amount of research has been conducted on the damage location and quantification of space truss structures using the incomplete strain mode. Here, the EEISM method is proposed to identify the damage location and damage quantification of space truss structures based on the incomplete strain mode. ð42Þ where MAC and sMAC 5 average value and the standard deviation of all MACj , respectively; g 5 guarantee coefficient; and the smaller g 5 higher identification probability for the damaged members. Certainly, some members whose MAC values are situated in the defined scope shown in Eq. (42) are not damaged. These misjudged members will be eliminated through the damage quantification index. Damage Quantification The perturbation equation expressed in Eq. (36) can be used for estimation of the damage quantity, and the first mode is usually adopted in calculating damage quantification index a, which is presented in the following equation: P11 a ¼ Dc1 ð43Þ where P11 5 first-order perturbation matrix of the first strain mode and Dc1 5 change of the first strain mode when the structure is damaged. Eq. (43) is an overdetermined equation if the number of the damaged members is less than the number of sensors. The damage quantification index a can be solved by the least-squares method. Numerical Analysis To verify the effectiveness of the proposed EEISM method, a numerical analysis of a long-span space truss structure was carried out. A finite-element model of the space truss structure with a span of 9 3 9 m, 61 nodes, and 200 members shown in Fig. 1. All members with Damage Location When the sensor locations are determined, perturbation matrix P11 should be modified such that the rows corresponding to the locations without sensors are set as zero. Meanwhile, the first strain mode change vector Dc1 should also be modified; i.e., the values corresponding to the elements without sensors should be set as zero. A comparison of the degree of similarity between Dc1 and each P11 column (i.e., Pj11 ) is carried out in accordance with the assurance criterion (Ting et al. 1993), which can be expressed as MACj ¼ ðDcT1 × Pj11 Þ2 j ðDcT1 × Dc1 Þ × ðPjT 11 × P11 Þ ð41Þ where MACj 5 MAC value of the jth member. The members with large MAC values will be defined as the damaged members. Actually, the MAC value can be regarded as the similarity degree between the tested strain mode change and the theoretical strain mode change when a special member is damaged; i.e., a larger MAC value represents a higher degree of similarity, and the corresponding member can be considered as the damaged member. When several members are damaged at the same time, the MAC values between Dc1 and Pj11 that participate in the combination of Dc1 will be large, while the MAC values between Dc1 and Pj11 that does not participate in the combination of Dc1 will be small. The damaged members usually exhibit larger MAC values; however, they do not always have the largest MAC values. Consequently, a larger MAC value scope must be determined first, as shown in Eq. (42) (Gonçalves et al. 2007), and the damaged members are certainly among the members whose MAC values meet Eq. (42) Fig. 1. Space truss structure in the numerical analysis JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 / 1219 J. Eng. Mech., 2012, 138(10): 1215-1223 Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. the same cross section were assumed to be axial tensioncompression members. The stress was considered as fully axial and uniform along the member, which is a basic assumption for space truss structures. Although bending deformation actually existed, it was neglected compared with the axial deformation. In a real application, bending deformation can also be eliminated by a compensation strain sensor that can be placed on the symmetric position of the member. The bearings were set on the four corner bottom nodes, which are hinge supports in the X, Y, and Z dimensions. In the finite-element model, there were 40 bottom chord members with the number ranging from 1 to 40, 60 top chord members with the number ranging from 41 to 100, and 100 web members with the number ranging from 101 to 200. The loading was applied on the intermediate top chord point. In the numerical analysis, a comparison analysis of the strain mode identification results was first introduced to observe the efficiency of the strain mode parameter identification of the EEISM method. Then, the damage location and quantification analysis followed. Finally, the damage detection analysis under various numbers of sensors was carried out. To realize the aforementioned research contents, 12 damage cases were adopted in the numerical analysis (listed in Table 1). Damage Cases 1–4 were used to verify the damage location capability when various members were damaged. Member No. 7 (bottom chord), Member No. 63 (top chord), and Member No. 108 (web member) with a stiffness reduction of 20% were selected to simulate single-member damage. Member No. 11 (bottom chord) with a stiffness reduction of 15% and Member No. 83 (top chord) with a stiffness reduction of 20% were supposed to simulate multimember damage. Cases 5–8 were used to verify the damage Table 1. Damage Cases in the Numerical Analysis Damage case Damaged member Damage quantity (%) Sensor quantity Case Case Case Case Case Case Case Case Case Case Case Case 1 2 3 4 5 6 7 8 9 10 11 12 7 63 108 11 and 83 53 53 53 53 34 34 34 34 20 20 20 15 and 20 10 15 20 40 20 20 20 20 50 50 50 50 50 50 50 50 25 50 75 100 quantification ability when various stiffness reductions occurred to a member; i.e., Member No. 53 with stiffness reductions of 10, 15, 20, and 40%, respectively. Cases 9–12 were used in the damage detection analysis when various sensor quantities were applied to the damage detection; i.e., when the sensor quantities were 25, 50, 75, and 100, and a 20% stiffness reduction occurred to Member No. 34. The previously defined damaged members are shown as bold lines in Fig. 1. It must be noted that the structure still remained linear after slight damage, and the damage was usually simulated by stiffness deterioration of the members rather than by integral deterioration. Strain Mode Identification Analysis Stationary filtering white noise was employed to simulate the ambient excitation imposed on the space truss structure show in Fig. 1. The strain responses were obtained by finite-element dynamic analysis, and the strain modes of the members were identified based on the mode parameter identification step of the EEISM method described previously. Fig. 2 shows a comparison between the first strain mode values obtained by the EEISM method and the first strain mode values obtained by the finite-element mode analysis. In Fig. 2, the horizontal axis is the member number, and the vertical axis is the corresponding strain mode value. Here, Member No. 1 is used as the reference member whose strain mode value is c11 5 1, and the strain mode values of the other members are the ratios to the strain of the reference member. As seen in Fig. 2, the first strain mode values obtained by the EEISM method agree well with the first strain mode values obtained by the finite-element analysis, and the maximum error is around 10%. This demonstrates that the EEISM method has an excellent ability of strain mode identification. Based on Fig. 2, about 30 members have no obvious amplitude response for the first strain mode relative to Member No. 1. Here, two aspects should be noticed. First, the strain mode of the sensitive member changes greatly. When damage occurs on a member of the structure, it does not matter if this member has large or small strain responses. That is, the structure damage will lead to a change of the strain mode of the sensitive members first. At the same time, the amplitudes of the strain responses are not necessarily connected to the sensitive members. In practical applications, it is beneficial to install the sensors on sensitive members with relatively large strain responses in order to decrease the relative error in measurement. Second, sensors are not set on the member with large strain response intentionally. Based on the damage detection method proposed in this paper, more attention is paid to the strain mode change of the sensitive member rather than the strain mode change of the damage member when some members are damaged. Fig. 2. Comparison of the strain mode of the space truss structure 1220 / JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 J. Eng. Mech., 2012, 138(10): 1215-1223 Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. Damage Location Analysis from Fig. 6 that Member Nos. 11 and 83 have peak MAC values of 0.246 and 0.593, and they can obviously be identified as the damaged members. It can be concluded from the previous analysis that the EEISM method has excellent damage location ability for single damaged members of bottom chord members, top chord members, and web members. At the same time, the EEISM method also has excellent damage location ability for multidamaged members. As described previously, the strain mode parameters can be identified precisely. Then, the change of the incomplete strain mode between the damaged and initial structures (i.e., Dc1 ) should be obtained with relatively good accuracy. The MAC values that indicate the degree of similarity between Dc1 and each column of perturbation matrix Pj11 are calculated by Eq. (41), and then the damages can be located by Eq. (42). In Damage Case 1, a stiffness reduction of 20% occurring in the bottom chord of Member No. 7 was assumed. The MAC values of all members can be calculated by Eq. (41). Fig. 3 shows the MAC values of all members when Member No. 7 is damaged. It can be seen that Member No. 7, which has the largest MAC value, can be easily determined to be a damaged member. In Damage Case 2, a stiffness reduction of 20% occurring in the bottom chord of Member No. 63 was assumed. Fig. 4 shows the MAC values of all members when Member No. 63 is damaged. It can be seen that Member No. 63 has the largest MAC value of 0.933. Member No. 63 can be determined to be a damaged member. Damage may also occur in web members, and in Damage Case 3 the stiffness reduction of 20% occurring in the web Member No. 108 was assumed. Fig. 5 shows the MAC values of all members when Member No. 108 is damaged. It can be seen that Member No. 108 has the largest MAC value of 0.892, and it can be determined to be a damaged member. The aforementioned cases all were used in the damage location analysis of single damaged members. However, damage may also occur in multimembers. In Damage Case 4 it was assumed that a stiffness reduction of 15% occurred in Member No. 11 (bottom chord) and a stiffness reduction of 20% occurred in Member No. 83 (top chord) at the same time. Fig. 6 shows the MAC values of all members when Member Nos. 11 and 83 are damaged. It can be seen To observe the damage quantification ability of the EEISM method, a damage quantification analysis was carried out by assuming that Member No. 53 had various amounts of damage severity with 10, 15, 20, and 40% stiffness reductions. The MAC values of all members can be calculated by Eq. (41), and the damage member can be located by Eq. (42); then, damage index a can be estimated by Eq. (43). In Damage Cases 5–8, stiffness reductions of 10, 15, 20, and 40%, respectively, occurring to Member No. 53 were assumed. Fig. 7 shows the MAC values of all members in Damage Cases 5–8. As seen, Member No. 53 has the largest MAC value in every damage case, and it can easily be determined to be the damaged member. With an increase of the degree of damage, the MAC value of Member No. 53 also increased. The MAC values of Member No. 53 were 0.609, 0.642, 0.692, and 0.848 under damage quantities of 10, 15, 20, and 40%, respectively. The damage quantity can be estimated by Eq. (43). Fig. 8 and Table 2 show comparisons between the real damage quantity and the detected damage quantity. The detected damage quantities were 10.16, 15.98, 22.15, and 52.63% under real damage quantities of 10, 15, 20, and 40%, respectively. The relative error for estimation of the damage quantity increased with the increase of the real damage quantity. The reason is that the Fig. 3. MAC values of Damage Case 1 Fig. 5. MAC values of Damage Case 3 Fig. 4. MAC values of Damage Case 2 Fig. 6. MAC values of Damage Case 4 Damage Quantification Analysis JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 / 1221 J. Eng. Mech., 2012, 138(10): 1215-1223 first-order perturbation matrix Pr1 ðr 5 1Þ is adopted in Eq. (33), where the higher-order terms that have more of an effect on severe damage are neglected. Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. Damage Detection Analysis under Various Sensor Quantities Sensors with larger contribution ratios of damage index CRDI calculated by Eq. (40) were chosen as the optimal placement of the sensors. The number of sensors is usually determined in accordance with economic considerations and damage identification accuracy before optimization analysis of sensor placement. The CRDI values under various numbers of sensor quantities are shown in Fig. 9. As seen, the CRDI value increases with the increase of sensor quantities; however, the increasing trend becomes slower with the increase of the sensor quantity. This means that when the sensor quantity exceeds a given value, installing more sensors only slightly benefits the damage information; thus, increasing the sensor quantity contributes less to the damage detection results when the sensor quantity exceeds a given value. To analyze the effect of the sensor quantity on the damaged location and quantification abilities of the EEISM method, damage cases were set as 9–12, in which it was assumed that sensor quantities of 25, 50, 75, and 100 corresponding to Cases 9–12 and Member No. 34 were damaged with a stiffness reduction of 20%, as shown in Table 3. The MAC values under all cases are shown in Fig. 10. When the sensors quantities were 25 and 50, the MAC values of Member No. 34 were 0.666 and 0.752, and the MAC values of Member No. 35 were 0.585 and 0.403, respectively. As seen, Member Nos. 34 and 35 have larger MAC values in Cases 9 and 10, and the two members can be detected to be damaged members in accordance with Eq. (42). Therefore, Member No. 35 was misdiagnosed as the damaged member when the sensor quantity was not enough. When the sensor quantity was increased to 75 and 100, the MAC values of Member No. 34 were 0.834 and 0.863, respectively, while the MAC values of Member No. 35 were 0.221 and 0.097, which means Member No. 34 had the largest MAC value in Cases 11 and 12, Member No. 34 was detected as the damaged member, and Member No. 35 was not misdiagnosed. Fig. 11 shows the damage quantification result of Member No. 34 estimated by Eq. (43) under Fig. 9. Total CRDI value under a special sensor quantity Fig. 7. MAC values of Member No. 53 Table 3. Damage Detection for Member No. 34 MAC value Real damage quantity (%) Detected damage quantity (%) Damage case No. 34 No. 35 No. 34 No. 35 No. 34 No. 35 No. 34 Case Case Case Case 23.00 22.60 22.35 22.23 9.15 8.45 — — 15.00 13.00 11.75 11.15 9 10 11 12 0.666 0.752 0.834 0.863 0.609 0.642 0.692 0.848 20 20 20 20 0 0 0 0 Fig. 8. Damage quantity detection of Member No. 53 Table 2. Damage Detection for Member No. 53 Damage case Case Case Case Case 5 6 7 8 MAC value Real damage quantity (%) Detected damage quantity (%) Relative error (%) 0.609 0.642 0.692 0.848 10 15 20 40 10.16 15.98 22.15 52.63 1.60 6.53 10.75 31.58 Relative error (%) Fig. 10. MAC values of Member Nos. 34 and 35 1222 / JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 J. Eng. Mech., 2012, 138(10): 1215-1223 will only slightly improve the accuracy of the damage quantification. Downloaded from ascelibrary.org by "National Institute of Technology, Silchar" on 07/24/23. Copyright ASCE. For personal use only; all rights reserved. Acknowledgments This research was supported by the Momentous Research Plan in the National Natural Science Foundation of China (90915004), Key Projects in the National Science & Technology Brace Program of China (2011BAK02B03), and the 333 High-level Talent Project in Jiangsu Province, which the writers gratefully acknowledge. References Fig. 11. Damage quantity detection of Member No. 34 Damage Cases 9–12. As listed in Table 3, the real damage quantity is 20%, and the detected damage quantities are 23.00, 22.60, 22.35, and 22.23% when the sensors quantities are 25, 50, 75, and 100, respectively. The relative error between the real damage quantity and the detected damage quantity is 15.00, 13.00, 11.75, and 11.15%. This shows the EEISM method has good damage quantification ability. It can be concluded from the previous analysis that the damaged location ability of the EEISM method obviously increases with the increase of the sensor quantity; however, the accuracy of the damage quantification improves slightly. Especially when the sensor quantity is large, the sensor quantity has no obvious effect on the improvement of the accuracy of the damage quantification. Therefore, determining the sensor quantity is required in accordance with economic considerations and damage identification accuracy. It can be suggested from the research that the sensor quantities can be determined as 1/8–3/8 of the total member quantity, which is not described in detail here for the sake of brevity. Obviously, the proposed method could detect damage by the incomplete strain mode, which can save cost and reduce data processing work. In this paper, the damage detection problem for three or more members is not taken into consideration, and this problem will be studied in the future. Conclusions A new damage detection method for space truss structures, named the EEISM method, is proposed in this paper. Numerical analysis of a long-span space truss structure was performed to verify the effectiveness of the EEISM method. The following conclusions may be drawn from this study. 1. A similar equation exists between the cross-correlation function of the strain response under environmental excitation and the strain impulse response. Consequently, the strain mode can be obtained based on the cross-correlation function of the strain responses. 2. The EEISM method has good ability in damage location for single damaged members of bottom chord members, top chord members, and web members, as well as in damage location for multidamaged members. 3. The damage index a in the EEISM method can be used to estimate the damage quantity with good accuracy. 4. The capability of locating damage using the EEISM method will be improved by increasing the sensor quantity but Chen, J. C., and Garbat, J. A. (1980). “Analytical model improvement using modal test results.” AIAA J., 18(6), 684–690. Chen, S.-H., Liu, Z.-S., Shao, C.-S., and Zhao, Y.-Q. (1993). “Perturbation analysis of vibration modes with close frequencies.” Int. J. Numer. Methods Eng., 9(5), 427–438. Clough, R. W., and Penzien, J. (1993). Dynamics of structures, McGrawHill, New York. Gonçalves, M. H., Cabral, M. S., Ruiz de Villa, M. C., Escrich, E., and Solanas, M. (2007). “Likelihood approach for count data in longitudinal experiments.” Comput. Stat. Data Anal., 51(12), 6511–6520. Halvorsen, W. G., and Brown, D. L. (1977). “Impulse technique for structural frequency response testing.” Sound Vib., 11(11), 8–21. Huang, N. E., et al. (1998). “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis.” Proc. R. Soc. London Ser. A, 454, 903–995. Imamovic, N., and Ewins, D. J. (1997). “Optimization of excitation DOF selection for modal tests.” Proc., 15th Int. Modal Analysis Conf., SEM, Orlando, FL, 1945–1951. James, G. H., III, Carne, T. G., and Lauffer, J. P. (1995). “The natural excitation technique (NExT) for modal parameter extraction from operating structures.” Int. J. Anal. Exp. Modal Anal., 10(4), 260–277. Jones, I. S. (2005). “Impulse response model of thermal striping for hollow cylindrical geometries.” Theor. Appl. Fract. Mech., 43(1), 77–88. Kaouk, M., Zimmerman, D. C., and Simmermacher, T. W. (2000). “Assessment of damage affecting all structural properties using experimental modal parameters.” Trans. ASME, J. Vib. Acoust., 122(4), 456–462. Kim, H. M., and Membertkowicz, T. J. (2001). “An experimental study for damage detection using a hexagonal truss.” Comput. Struct., 79(2), 173–182. Lam, H. F., Ko, J. M., and Wong, C. W. (1998). “Localization of damaged structural connections based on experimental modal and sensitive analysis.” J. Sound Vib., 210(1), 91–115. Li, Z. F., Hua, H. X., Shi, Y., Chen, Z. Y. (2001). “On line modal parameter identification technique.” Proc., 19th Int. Modal Analysis Conf., SEM, Kissimmee, FL, 1518–1523. Pandey, A. K., Biswas, M., and Samman, M. M. (1991). “Damage detection from changes in curvature mode shapes.” J. Sound Vib., 145(2), 321–332. Ting, T., Chen, T. L. C., and Twomey, W. (1993). “Correlating mode shapes based on the modal assurance criterion.” Finite Elem. Anal. Design, 14(4), 353–360. Tsang, W. F. (1990). “Use of dynamic strain measurement for the modeling of structures.” Proc., 8th Int. Modal Analysis Conf., SEM, Kissimmee, FL, 1246–1251. Unger, J. F., Teughels, A., and De, R. G. (2005). “Damage detection of a prestressed concrete beam using modal strains.” J. Struct. Eng., 131(9), 1456–1463. Yam, L. H., Leung, T. P., Li, D. B., and Xue, K. Z. (1996). “Theoretical and experimental study of modal strain analysis.” J. Sound Vib., 191(2), 251–260. Yao, G. C., Chang, K. C., and Lee, G. C. (1992). “Damage diagnosis of steel frames using vibrational signature analysis.” J. Eng. Mech., 118(9), 1949–1961. JOURNAL OF ENGINEERING MECHANICS © ASCE / OCTOBER 2012 / 1223 J. Eng. Mech., 2012, 138(10): 1215-1223
0
You can add this document to your study collection(s)
Sign in Available only to authorized usersYou can add this document to your saved list
Sign in Available only to authorized users(For complaints, use another form )