Structural Damage Identifacation Using An Optimization Technique Based On Generalized Flexibility Matrix
Structural Damage Identifacation Using An Optimization Technique Based On Generalized Flexibility Matrix
Structural Damage Identifacation Using An Optimization Technique Based On Generalized Flexibility Matrix
https://doi.org/10.1093/jom/ufad047
Advance access publication 22 December 2023
Regular Article
A B ST R A C T
A generalized flexibility matrix-based objective function utilized for structure damage identification is firstly constructed. After transforming the
damage identification into a constrained nonlinear least squares optimization problem, the trust-region algorithm is applied to find the solution
of the inverse problem in multiple damage cases. Thereinto, the sensitivity analysis of the objective function with respect to the design variables
is derived using the Nelson’s method. At last, two numerical examples with several damage cases are investigated, including a steel truss bridge
model as well as a drilling rig derrick model. Based on the computational results, it is evident that the presented approach provides excellent
validity and reliability for the large and complicated engineering structures.
KEY WOR DS: generalized flexibility matrix, damage identification, constrained nonlinear least squares, trust-region algorithm
attention. Masoumi et al. presented an objective function on the eigenvectors, and satisfy the mass-normalized condition, such
basis of GFM for solving a constrained optimization problem that
in damage detection procedure via imperialist competitive algo-
T K = (3)
rithm [21]. Considering the non-negativity of the damage in-
dex, an improved GFM approach without and with noises was and
proposed by Liu et al. [22]. Later, the improved GFM is applied
to deal with incomplete mode shape data of structural damage T M = I , (4)
problem [23]. A damage identification process for a jacket-type where I is the identity matrix with dimension n × n. As the flex-
offshore platform structure was presented by using GFM and ibility matrix F is the inverse matrix of K in Eq. (3), F can be
optimal GA by Aghaeidoost et al. [24]. expressed by
The article is organized as follows. Primarily, the damage iden-
tification problem is constructed and the GFM is reviewed in
n
ϕ j ϕTj
−1 −1
F =K = = .
where Kd and Ku represent overall stiffness matrix of the dam- 3. THE PROPOSED M ETHOD
aged prediction model with dimension n × n and the one of 3.1 Objective function
the undamaged structure, respectively. K eu j describes the stiff- On the basis of GFM, an optimization model is employed as a
ness matrix of the jth element in an undamaged structure with damage detection problem for searching a series of damage ex-
expended dimension n × n. α j (0 ≤ α j ≤ 1.0) denotes the di- tents 0 ≤ α j ≤ 1.0 (j = 1, 2, …, Ne ), i.e.
mensionless damage index of corresponding jth element stiff-
g
ness matrix, and α j = 0 indicates that the corresponding element f (α) = F gexp − F d (α), (8)
has not been harmed. α j can be interpreted as any geometric or g g
physical parameter of the prediction model, such as moment of where F exp is the n × n experimental measured GFM, i.e. F exp =
inertia, stiffness, boundary condition and so on. Ne is the count exp −2exp exp , exp and exp are eigenvector matrices and
T
of finite elements. Therefore, the damage identification problem eigenvalue matrices of the actual damaged structure, respec-
g
becomes the problem of finding a set of values α j (j = 1, 2, …, tively. F d (α) is the analytical GFM which corresponds to global
Ne ). stiffness matrix Kd of the damaged prediction model, expressed
by
2.2 The generalized flexibility matrix
F d (α) = d (α)−2
g T
The GFM is reviewed in this section. First, consider the free vi- d (α)d (α) , (9)
bration of structural eigenproblem with the expression where d (α) and d (α) are matrices of eigenvalues and the
corresponding eigenvectors for the damage prediction model.
K = M, (2)
The issue is to find a set of values 0 ≤ α j ≤ 1.0 (j = 1, 2,
where K and M are structural stiffness and mass matrix …, Ne ) through minimizing the discrepancy between data from
with dimension n × n. = diag(λ1 , λ2 , . . . , λn ) and = experimental measurement and data from the analytical dam-
[ϕ1 ,ϕ2 ,…, ϕn ] are matrices of eigenvalues and corresponding aged model. When 2-norm is applied to analyze the discrepancy
Structural damage identification using an optimization technique • 13
3.2 Optimization algorithm Significantly, the TR algorithm requires the Jacobian matrix of
The optimization model above is actually a nonlinear least the objective function with related design variable α j (j = 1,
squares problem (NLS) with constraint conditions and can be 2,…,Ne ), i.e.
solved by TR algorithm. For this propose, the TR algorithm is
firstly reviewed [25, 26]. The general formulation of the ques- J (α) = [J (α1 ),J (α2 ), . . . ,J (αNe )] (18)
tion is expressed by and
2 2
min G(x)2 = y − A(x)2 ∂f
, (13) J (α j ) = vec
x
∂α j
s.t. li ≤ xi ≤ ui , i = 1, 2, . . . m T
∂ −2 T −3 ∂ T −2 ∂
where A ∈ Rm , x ∈ Rm and y ∈ Rm . The TR algorithm provides the = vec − + 2 −
∂α j ∂α j ∂α j
principle idea of transforming Eq. (13) into the kth step iterative
TR sub-problem j = 1, 2, . . . ,Ne , (19)
2
min k (sk ) = G(xk ) + J (xk )sk where J(α) is the Jacobian matrix with size n × Ne . The sen-
2
x 2
sitivities of eigenvalues and eigenvectors to the model updating
, (14)
sk 2 ≤ k parameter α j (j = 1, 2,…,Ne ) will be deduced below by Nelson’s
s.t.
li ≤ xi ≤ ui , i = 1, 2, . . . m method [27].
where xk represents the current point, sk denotes the solution 3.3 Sensitivity analysis
of Eq. (14), κ represents trust-region area at kth step. J(xk ) is In this paper, it is supposed that the itheigenvalue λi is simple so
the Jacobian matrix of A(x). Then the actual reduction is defined that the corresponding eigenvector ϕi is unique. Directly differ-
as entiation Eq. (2) with respect to α j yields
Are sk = G(xk ) − G(xk + sk ) (15)
∂ϕi ∂K ∂M ∂λi
( K − λi M ) + − λi − M ϕi = 0
∂α j ∂α j ∂α j ∂α j
the predicted reduction is described as
j = 1, 2, . . . ,Ne . (20)
Pre sk = (xk ) − (xk + sk ). (16)
Premultiplying Eq. (20) by ϕTi and combining with Eq. (2), the
and the ratio of actual reduction to predicted one is defined as sensitivity of the ith eigenvalue λi can be achieved
rk =
Are sk
, ∂λi ∂K ∂M
(17) = ϕi T
− λi ϕi , j = 1, 2, . . . ,Ne . (21)
Pre sk ∂α j ∂α j ∂α j
14 • Journal of Mechanics, 2024, Vol. 40
Figure 2 Damage presets of the steel truss bridge structure for all scenarios in Table 1. (a) Scenario Ⅰ, (b) scenario Ⅱ and (c) scenario Ⅲ.
Structural damage identification using an optimization technique • 15
4. NUM ER IC A L E X A M PLES
In this part, two simulation models are illustrated to show the
validity of the proposed method. Each of the two examples takes
account of three damage scenarios. All the simulated damage sce-
narios only result in reducing stiffness of specified elements. If
the calculated extent of damage is less than 5%, the related el-
ement is considered to be undamaged [19]. All NLS with con-
straint conditions problems included in the examples are imple-
mented by the trnlspbc command in Intel® oneAPI Math Kernel
Library.
Example 1: A steel truss bridge is presented in Fig. 1. The
bridge is 12 m wide, 10.5 m high and 90 m long. The material
properties are listed below: elasticity modulus E = 210 Gpa,
mass density ρ = 7800 kg/m3 and Poisson’s ratio ν = 0.31. The
structural finite element model has 94 elements and 40 nodes.
The boundary conditions are defined as all displacement con-
straints at two points on the left side and two horizontal displace-
Figure 3 Simulation results of the steel truss bridge structure for all
ment constraints at two points on the right side at the bottom.
scenarios in Table 1. (a) Scenario Ⅰ, (b) scenario Ⅱ and (c) scenario
Ⅲ. Therefore, the number of total DOFs is 230.
In this example, damage condition can be ascertained sim-
ply by the use of the first frequency and the related vibration
shape. The damage positions for the scenarios given in Table 1
are labeled in Fig. 2. The corresponding computational results
16 • Journal of Mechanics, 2024, Vol. 40
Figure 6 Simulation results of the drilling rig derrick structure for all
scenarios in Table 2. (a) Scenario Ⅰ, (b) scenario Ⅱ and (c) scenario
Figure 5 Damage presets of the drilling rig derrick structure for all Ⅲ.
scenarios in Table 2. (a) Scenario Ⅰ, (b) scenario Ⅱ and (c) scenario
Ⅲ.
Structural damage identification using an optimization technique • 17
using the presented approach are illustrated in Fig. 3. For dam- FUNDING
age scenario Ⅰ, the predefined damage positions are mapped ex- This work was supported by the National Natural Science Foun-
actly on element 8 with damage extents 0.1831, and element 9 dation of China under Grant No. 42372356 and 41972323.
with stiffness reduction 0.2321. In scenario Ⅱ, the damage ex-
tents and locations calculated are 0.1960, 0.2388 and 0.1901 for
elements No. 9, No. 35 and No. 36, respectively. From the cal- CONFLICT OF IN TER E ST
culation results of the damage scenario Ⅲ shown in Fig. 3c, it
is observed that damage locations occurred in element 15, el- The authors declare that they have no known competing finan-
ement 34, element 61 and element 83, and the values of cor- cial interests or personal relationships that could have appeared
responding reduction in stiffness are 0.1565, 0.1931, 0.2543 to influence the work reported in this paper.
and 0.3250, respectively. All these results indicate that this ap-
proach has the capability of precise identification of the damaged
data and using mode shape curvature estimation. Applied mathemat- 22. Liu H, Li Z. An improved generalized flexibility matrix approach for
ical modelling 2021;91:670–694. structural damage detection. Inverse Problems in Science and Engineer-
17. Najafabadi AA, Daneshjoo F, Ahmadi HR. Multiple damage detec- ing 2020;28(6):877–893.
tion in complex bridges based on strain energy extracted from sin- 23. Liu H, Wu B, Li Z. The generalized flexibility matrix method for
gle point measurement. Frontiers of Structural and Civil Engineering structural damage detection with incomplete mode shape data.
2020;14(3):722–730. Inverse Problems in Science and Engineering 2021;29(12):2019–
18. Liu G, Zhai Y, Leng D, Tian X, Mu W. Research on structural dam- 2039.
age detection of offshore platforms based on grouping modal strain 24. Aghaeidoost V, Afshar S, Tajaddod NZ, Asgarian B, Shokrgozar
energy. Ocean Engineering 2017;8:43–49. HR. Damage detection in jacket-type offshore platforms via general-
19. Li J, Wu B, Zeng QC, Lim CW. A generalized flexibility matrix based ized flexibility matrix and optimal genetic algorithm (GFM-OGA).
approach for structural damage detection. Journal of Sound and Vibra- Ocean Engineering 2023;281:114841.
tion 2010;329(22):4583–4587. 25. Zhang J, Wang Y. A new trust region method for nonlinear equations.
20. Yang QW, Liu JK. Damage identification by the eigenparameter de- Mathematical Methods of Operations Research 2003;58(2):283–
composition of structural flexibility change. International Journal for 298.