Proc. of the Intl. Conf.

on Advances In Engineering And Technology - ICAET-2014

Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

Health Monitoring and Damage Assessment

by Parameter Estimation of Framed
Structures From Static Responses
Jaydeep Chowdhury1 and Priyabrata Guha2

Abstract—Structural damage assessment of framed structure overcome this type of difficulties parameter estimation
utilizing a sensitivity based parameter identification method is process is used, in which the major differences between
presented in this paper. A subset of applied static forces and subset estimated and expected parameters are classified as damage.
of measured strains are used to identify the elemental stiffness In all cases, damage can severely affect the safety and
parameters of all or a portion of a finite element model of the serviceability of the structure. Hence an early detection of
structure. Here we developed a method for parameter estimation of
linear elastic structures using static strain measurements, preserving
any damage is necessary. Owing to its practical importance,
structural connectivity and determines the changes in cross-sectional damage identification in structures, especially the monitoring
properties, including large changes or elemental failure for stable of structural health has been the subject of extensive
structures. To linearize the associated non linear problem a first investigations over the last two decades. As a result, a great
order Taylor series expansion is used as an iterative scheme. The deal of research work on using either the static or the
algorithm automatically adjusts the structural element stiffness dynamic response of structures for damage detection has been
parameters in order to improve the comparison between a measured carried out.
and theoretical response in an optimal way. In this study we have Displacement measurements from applied loads are used for
artificially generated the required measured input. The identified static finite element method. It can be a difficult task to make
cross-sectional element properties can be used for damage
assessment of any structure. This procedure also identifies the
displacement measurement on full-scale structure; for this a
selection of limited number of degrees of freedom required to frame of reference must be established. There is having a
perform successful parameter identification, as well as reduces the number of non-destructive tests to assess the material
impact of measurement errors of the identified parameters. Two strengths of the structure without damaging the structure.
numerical examples, including two-dimensional (2D) and three- However, all these techniques normally assess the damage
dimensional (3D) frame structures are presented and the element qualitatively. System identification techniques, based on
stiffness are successfully and accurately evaluated, and to analyze dynamic data, have been developed extensively, compared to
any frame structure with this numeric process easily, a program is static data. But, the static responses are more locally sensitive
implemented in MATLAB programming according to the explained than the frequency in structural damage detection.
process, so that we can easily change the input data for analyzing
different types of frame structures.
The model-based approach is implemented by using
Keywords—Finite element model; Static strain Measurement; computer model of the structure of interest, such as Finite-
Damage detection;Error elimination; Element Method (FEM), to identify structural parameters
based on the measured test data. Though non-contact
displacement processes are available, but, they are costly and
Introduction difficult. Use of strain gauges to measure the static strain can
In recent years data-driven structural health monitoring overcome those limitations, but some small errors can occur,
(SHM) has been actively investigated by the civil engineering which we have to deal with or control. In this paper for
community as an important tool for future infrastructure structural health monitoring and damage assessment of
maintenance. To properly manage civil infrastructure, its structure, static strain measurement process is used. Static
condition, or serviceability must be assessed. Measurements strain measurements are definitely more useful than
and proper data processing are expected to give a reasonable displacement measurements, as it reduces the use of costly
assessment of serviceability that can be improved based on equipments and also avoids the difficulties of displacement
the assessment. A static condensation method is proposed to measurements on full scale structures.
adjust the parameters of the finite element model (FEM), by Implementing a damage identification strategy for aerospace,
minimizing the difference between the analytical data and the civil, and mechanical engineering infrastructures is referred
nondestructive test (NDT) data. The parameter to as Structural Health Monitoring (SHM). There are two
adjustmentsbased on trial and error is not appropriate for main approaches to SHM: (a) non-model-based approach and
structures with large number of parameter values. To (b) model-based approach. Both approaches have been
successfully used for damage detection in structures. The
1. Jaydeep Chowdhury. non-model-based approach relies on the signal processing of
Narula Institute of Technology (West Bengal University of Technology). experimental data, while the model-based approach relies on
India. mathematical descriptions of structural systems. The
jaydeepchowdhury18@gmail.com alternatives to the non-model-based method include: modal
2. Priyabrata Guha. analysis, dynamic flexibility measurements, matrix update
Narula Institute of Technology (West Bengal University of Technology).
methods and wavelet transform technique, which are used to
India.
priyabrata.guha@gmail.com determine changes in structures to identify damage. In model-

276
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

base approach uses computer model, such as Finite-Element elemental strains and aligning the system degree of freedoms
Method (FEM), to identify structural parameter based on horizontally with the system [M] matrix, we get,
measured test data.
The focus of the present work is the sensitivity based damage ᶯ
{ } = [M] {U} (6)
detection method. In this paper a method of estimating the
parameter of linear-elastic structures using static strain Now, by assembling{ Ɛ} = element strain vector of size
measurements, preserving structural connectivity is NEL×1; {U}= global displacement vector; we got the [M]
presented. Sensitivity based update methods are based on the matrix of size NEL×NDOF.
first order Taylor series, which minimizes the error of non- Creation of mapping vector :
linear problem occurrence. This process allows single or For linear elastic behavior of frame element, the actions of
several static forces to be applied at a subset of degree of axial deformation and bending both causes axial strains on
freedoms (DOF), and the strains to be measured at a subset of the element surface at a known distance from the neutral axis
structural components. It is also possible to identify all or a are superimposed. These strains are measured length wise,
portion of structural cross-sectional properties, including which are in direction.
element failure. The proposed method has numerically
demonstrated in this paper for two-dimensional (2D) and ᶯn = -ȳ (7)
three-dimensional (3D) frame structures, and the cross-
sectional parameters of those structures are successfully where, ȳ= distance from the neutral axis to strain
identified. measurement surface and = translation in ȳ direction. A
natural coordinate system ξ is induced for finite element
analysis, ξ can be defined as
Estimation of Parameter:
A sensitivity based parameter estimation process is presented ξ= -1 (8)
in this paper to modify the parameters of a finite element
model with simulated static strain measurements. Forces are ξ ranges from -1 to +1, for the substituted xvalue, which
induced at a subset of degrees of freedoms and with respect ranges from 0 to L.
to that strains are measured on a limited number of structural We can express mapping vector {Mn} as
elements. {Mn} = { } [Tn] (9)
Damage Assessment Using Finite Element Where, { } = mapping vector of an element n; The size of
Model: { } will be 1×6 for 2D frame and 1×12 for 3D frame, and
The static finite element equation for frame structure can be they are represented as
expressed as
[K]{U} = {F} (1) { } for 2D frame =
Where, [K] is the global stiffness matrix and {F} and {U} { (6ξ - 2) (6ξ + 2) }n;
are force and displacement vector, respectively. Finite
element model is based on the stiffness relationship between { } for 3D frame =
forces and displacements. The relationship will be created in { 0 (6ξ- 2) (6ξ - 2) 0
the form of an element mapping vector {M n} in global
coordinates.
First step is to create {Mn} in the local coordinates such that (6ξ +2) (6ξ + 2) }n;
ᶯ
the following relation is satisfied for an elemental strain ( n)
and, [Tn] = transformation matrix of size 6×6 for 2D frame
and 12×12 for 3D frame, and represented as
and displacement {Ūn}. The relation is,  l xx l xy 0 0 0 0
 
ᶯn= { }{ } (2) [Tn] for 2d frame =
l xx
 0
l xy

0
0

1
0

0
0

0
0

0


 
then{Un} is transformed from the local coordinates to the  0 0 0 l xx l xy 0
global coordinates  0 0 0 l xx l xy 0
 
 U n  = [Tn] {Un} (3)
 0 0 0 0 0 1
n
Where, [Tn] = mapping matrix; and {Un} = element nodal
displacement in the global coordinates.
and,
{Mn} = { } [Tn] (4)
By substituting the value of  U  and {
n
} from (3) and (4)
in (2), we can get the global mapping relation for one
structural element. That is

ᶯn={Mn}{Un} (5)
From equation (5), we got the strain-displacement relation for
system of n elements. Now by vertically augmenting the

277
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

and, [Tn] for 3D frame = function arrives. To linearize [e(p)], the error function matrix
is vectorized by concatenating all the columns vertically. This
 l xx l xy l xz 0 0 0 0 0 0 0 0 0  produces an {e(p)} of size NM × 1, where, the number of
  measurements (NM) will be the product value of NMS ×
l yx l yy l yz 0 0 0 0 0 0 0 0 0
 
NSF. To linearize this {e(p)}, a first-order Taylor series
 l zx l zy l zz 0 0 0 0 0 0 0 0 0 
  expansion is used as
 0 0 0 l xx l xy l xz 0 0 0 0 0 0 
 0 0 0 l yx l yy l yz 0 0 0 0 0 0 
{e(p + Δp)} = {e(p)} + [C] {Δp} (13)
 
 0 0 0 l zx l zy l zz 0 0 0 0 0 0  By differentiating [e(p)] one time with respect to p we got the
 0 0 0 0 0 0 l xx l xy l xz 0 0 0  sensitivity matrix [C] of size NM × NUP and Δp is the
 
 0 0 0 0 0 0 l yx l yy l yz 0 0 0 
change in parameter of size NUP×1.
  A scalar performance error function R(p) is came form the
0 0 0 0 0 0 l zx l zy l zz 0 0 0
  reduction of error function {e(p)}, and it can be expressed as
 0 0 0 0 0 0 0 0 0 l xx l xy l xz 

0 0 0 0 0 0 0 0 0 l l l yz
 R(p) = {e(p + Δp)}T {e(p + Δp)} (14)
 
yx yy

 0 0 0 0 0 0 0 0 0 l zx l zy l z z 
n To minimize R(p) with respect to {p}, subjected to pi ≥ 0 for
i = 1 to NUP, the gradient of R(p) with respect to {Δp}T is set
to zero, which results a linearized system of equation as,
where,
=cos(αx), =cos(αy), =cos(αz); [C]T [C] {Δp} = -[C]-1 {e(p)} (15)
If [C] is a square matrix then direct inversion may be used
=cos(αz), =cos(αx), =cos(αy);
and {Δp} will be
=cos(αy), =cos(αz), =cos(αx); {Δp} = -[C]T {e(p)} (16)
By substituting (1) in (6) it becomes but, when [C] will not be square (NM > NUP), the least-
{e} = [M] [K(p)] {F} -1
(9) square method results {Δp} as
{Δp} = -([C]T [C])T [C]T {e(p)} (17)
ᶯ
Where, { }= NEL ×1; [M]= NEL ×NDOF; [K(p)]= NDOF ×
NDOF; and {F}=NDOF ×1. Examples for Parameter Estimation of
Frame Structures:
The element stiffness parameters of the stiffness matrix may
be a cross-sectional area A, or a moment of inertia Ixx, Iyy, 2D Frame Example:
Izzand stored in the vector of parameters {p}. Number of set The first example consist of a 2D frame as shown in“Fig. 1”.
of force(NSF) load cases are used in parameter identification As it is a frame structure, it is capable of bending and axial
method for solving the unknown parameters. By arranging deformation, so, it will have two parameters as: cross-
the test data horizontally in {e} and {F}, the equation sectional area (A) and moment of inertia (Iz). This frame is
becomes having 18 elements or 18 members, 15 number of nodes. As
shown in “Fig. 1”, node 11 to 15 is fixed.so, there can be a
ᶯ
[ ] = [M] [K(p)]-1[F] (10) maximum of 36 unknown parameters, total number of DOF
where, [e] = NEL×NSF; and [F] = NDOF×NSF. will be 30. Modulus of elasticity (E) for all elements of the
To eliminate the less important unmeasured strain, [e] is frame is taken as 280.6 GPa. and cross-sectional properties
are taken as specified in“Table-1”. There are 9 cases for the
partitioned as [ ], correspond to this [M] is also partitioned parameter identification of the 2D frame example. In case
1,3,4,5,6,7,8,9 applied forces are 500 N or 200 N.m. and for
ᶯ
as [ ], where [ b] is the unmeasured strain. After cases 11,12,13,15 applied forces are 600 N or N.m,800 N or
elimination of the unmeasured strain the equation becomes N.m,700 N or N.m,900 N or N.m respectively, for case-14
forces are 800 N or 400 N.m.
ᶯ
[ a] = [Ma] [K(p)]-1[F] (11) Table-1. Cross-sectional Properties of 2D Frame

ᶯ
where, [ a] = NMS × NSF; [Ma] = NMS × NDOF; and NMS
= number of measured strain.
Estimation and Minimization of Error
Function:
The error function is defined as the difference between the
analytical strain value [ea] and the measured strain value
ᶯ
[ a]m. It can be expressed as

ᶯ
[ (p)] = [ a] – [ a]m ᶯ ᶯ (12)
The error function will be of size NMS × NSF, and in
equation (11) due to the inversion of [K(p)], a nonlinear
278
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

Table -2. Different Applied Load Cases for Analysis:

Case FDOF
ᶯ Load Iteratio
Parameters Initial True

(N or N.m) ns A1 to A4
and
1 16 to 30 1 to 18 500 N or 200 N.m 3
Area A10to A13 900 × 102 mm2 855 × 102 mm2
2,5,8,11,14,17, 4,7,9,12,
2 500 N or 200 N.m 4
20,23,26,29 15,18
A5to A9
4,5,8,9,
3 1,6,12,17 500 N or 200 N.m 3 and
15,21,25
1,2,12,15,16 A14to A18 1600 × 102 mm2 1760 × 102 mm2
4 6,8,11,16 500 N or 200 N.m 2
17,27,30
10,11,16, Iz 1 to Iz 4
5 6,9,12,16,21,27 600 N or 600 N.m 6
17 and
6 19,20,27 2,9,10,18 800 N or 800 N.m 3 Moment of Iz10 to Iz13 67500 × 104 mm4 81000 × 104 mm4
7 6,12,21,30 6,13 700 N or 700 N.m Singular Inertia
8 2,6,18,30 5,9,15 800 N or 400 N.m 5 Iz 5 to Iz 9
9 11,12,19,21 3,6,8,11 900 N or 900 N.m 3 and
Iz14 to Iz18 213333 × 104 mm4 170666.4 × 104 mm4

Fig. 1 – 4 Bay 2 Storied 2D Frame

Bar chart-1: Error in Cross-sectional Area (2D Frame):

Bar chart-2: Error in Moment of Inertia (2D Frame):

279
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

Load case analysis of case-9 of “Table-2” is done with the

help of MATLAB and in this case we have detected some
errors in moment of inertia which is very less and there is Parameters Initial True
having no errors in cross-sectional area of all members
A1 to A2
they as shown in the “Bar chart-1” and “Bar chart-2”.
and
Area A6to A10 900 × 102 mm2 855 × 102 mm2
3D Frame Example:
“Fig. 2” shows the example of a 3D frame structure. As it A3to A5
is also a frame structure, it will also be capable of bending and
and axial deformation and parameters will also be similar A11to A13 1600 × 102 mm2 1760 × 102 mm2

1 4 to 12 2,4,8,9, 200 N or 4
49 to 54 11,13 100 N.m

2 8,9,44,45, 3,5,7, 200 N Singular

50,51 11,13

Bar chart-3: Error in Cross-sectional Area (3D Frame):

Fig. 2 – 2 Bay 1 Storied 3D Frame

to the previous frame example.This frame is having 13
elements or 13 members, 12 number of nodes. As shown in
“Fig. 2”, node 4 to 6 and 10 to 12 is fixed.so, there can be
a maximum of 13 unknown parameters, total number of
NDOF will be 36, but in “Fig. 2” the DOf s are represented
as their global number. Modulus of elasticity (E) for all
elements of the frame is taken as 280.6 GPa. torsional
constant (J) is taken as 27795 GPa and cross-sectional Load case analysis of case-1 of “Table-4” is done with the
properties are taken as specified in “Table-3(a)” and help of MATLAB and in this case we are not be able to
“Table-3(b)”. There are 2 cases for the parameter identify the moment of inertia properly, there are having so
identification of the 3D frame example. The load cases are many errors which consist of large magnitude, but in case
specified in “Table-4”. of identification of cross-sectional area, this process has
successfully identified a part of the members, more
Table-3 (a). Cross-sectional Properties of 3D Frame
specifically the beam elements have been identified very
closely, as they are having very less magnitude of error. As
Table-3 (b). Cross-sectional Properties of 3D Frame
shown in “Bar chart-3” and “Fig. 2”, member 1, 2, 6, 7, 8,
9, 10 are identified with a very less error.
Parameters Initial True

Iz 1 to Iz 4 Conclusions
and This method has successfully identified the parameters at
67500 × 104 mm4 81000 × 104 mm4
Iz10 to
Moment Iz13 the element level by applying forces at some limited
of numbers of degrees of freedoms and also by taking strain
Inertia Iz 5 to Iz 9 measurements at some selective locations for linear elastic
and structures. The difference between analytical strain and
213333 × 104 mm4 170666.4 × 104 mm4
Iz14 to
Iz18 measured strain forms the performance error function. The
cross-sectional element properties for frame such as, areas
and/or moments of inertias are identified by minimizing
that performance error function.
Table -4. Different Applied Load Cases for Analysis: Two examples are presented here and assessment has
been done successfully to some extent. It is also identified
Case FDOF
ᶯ Load Iterations that if NM ≥ NUP, it is possible to identify the parameters.
(N or N.m) So, this method can successfully identify the parameters of
2D structures. In case of 3D structures the capability of
280
 Proc. of the Intl. Conf. on Advances In Engineering And Technology - ICAET-2014
Copyright © Institute of Research Engineers and Doctors. All rights reserved.
ISBN: 978-1-63248-028-6 doi: 10.15224/ 978-1-63248-028-6-03-55

simultaneous identification parameters are reduced. Either

subset of measurements has to change to avoid large
changes in identified parameters or only a partial set of
parameters can be identified.
Future work can include some improvement in this method
to overcome the limitations of identifying all the
parameters properly and the damage assessment of frame
structure for dynamic responses can also be done. So that,
we will be able to identify the parameters of any type of
frame structures for both static and dynamic responses.

References
[1] William McGuire, Richard H. Gallagher, Ronald D Zeimian.
“Matrix Structural Analysis” (Second Edition). Chapter- 4
and 5.
[2] P.Seshu. “Textbook of Finite Element Analysis” , Chapter-5.
[3] Masoud Sanayei, Michael J. Saletnik.(1996) “Parameter
Eastimation of Structures From Static Strain Measurements.
I:Formulation” .
[4] G. Aditya and S. Chakraborty.(2008) “Sensitivity Based
Health Monitoring of Structures With Static Response”.
[5] Masoud Sanayei, Oladipo Onipede.(1991) “Damage
Assessment of Structures Using Static Test Data”.

281