Mechanical Systems and Signal Processing 163 (2022) 108166

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Substructural damage detection using frequency response function

based inverse dynamic substructuring
Mohammad Hadi Jalali *, D. Geoff Rideout
Faculty of Engineering and Applied Science, Memorial University, St. John’s, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: In this paper, a substructural damage detection approach is presented using frequency response
Structural health monitoring function (FRF)-based inverse dynamic substructuring and FRF-based model updating. In practice,
Substructural damage detection often only one subsystem of the global structure is critical and susceptible to damage and
Inverse dynamic substructuring
therefore needs monitoring. In the proposed method, one subsystem is the “main” subsystem and
Frequency-based decoupling
Finite element model updating
is susceptible to damage and the “residual” subsystem(s) are considered undamaged and the
damage identification is only applied to the main subsystem. The FRF matrix of the main sub­
system is obtained as a “standalone” component that is completely decoupled from the residual
subsystem(s) without the need of interface virtual support or identifying interface forces. The FRF
measurement is performed on the global damaged structure but the FRFs of the main substructure
(damaged) is obtained by “decoupling” the known FRF(s) of the residual subsystem from the
global system FRFs and the damage detection is performed only on the main substructure. An
FRF-based model updating method using numerical sensitivities is then used for damage identi­
fication of the main subsystem. The FRF obtained from the finite element model of the main
substructure is updated using the FRF of the damaged main substructure (obtained from
decoupling) and the location and quantity of the damage is identified. Numerical and experi­
mental examples are presented to illustrate the damage detection procedure and the damage in
the main substructure is detected, located and quantified with good accuracy.

1. Introduction

Structural health monitoring mainly aims at evaluating the symptoms of deterioration or damage that may affect the operation of
the structure. Vibration-based methods are popularly used for the non-destructive damage identification of large-scale structures. The
fundamental idea for the vibration-based damage identification is that the damage-induced changes in the physical properties (mass,
damping, and stiffness) will cause detectable changes in modal information such as natural frequencies, modal damping, mode shapes
and frequency response function (FRF). The substructural system identification approach is receiving increasing attention due to the
advantage of a reduction in the number of unknown parameters to be identified, and a reduction in the system DoFs involved in the
computation, with improvement in the convergence and computational efficiency [1].
In substructural damage identification methods, the global structure is usually divided into substructures and damage identifi­
cation is applied locally only on the desired substructure rather than on the global structure [2]. This reduces the computational time
and increases the accuracy since the modal parameters of the global structure have small sensitivity to the local damage [3]. Zhang

* Corresponding author.
E-mail address: [email protected] (M.H. Jalali).

https://doi.org/10.1016/j.ymssp.2021.108166
Received 27 January 2021; Received in revised form 3 May 2021; Accepted 18 June 2021
Available online 2 July 2021
0888-3270/© 2021 Elsevier Ltd. All rights reserved.
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

et al. [4] proposed a substructure identification procedure for large shear structures. In their method, an inductive identification
procedure was proposed to estimate the structural story parameters from top to bottom recursively. A numerical example of a five-
story structure was presented to illustrate the efficacy of the proposed substructure identification method. Zhang et al. [5] pro­
posed a substructure damage identification method for shear structures utilizing the dynamic equilibrium of a one-floor substructure
and using the cross-power spectral densities between structural floor accelerations and a reference response. A ten-story shear
structure was used to illustrate the effectiveness of the proposed substructure method. Xing et al. [6] developed a substructual damage
identification method by adopting the well-known ARMAX model to extract the modal information of each substructure in shear
structures. The extent of damage was measured by using the squared original frequency and the squared damaged frequency. Many
other studies have been performed on substructural damage identification for shear structures using different damage indices and
different substructure identification [7–10].
Hou et al. [11] proposed a substructure isolation approach for local health monitoring. Their proposed approach was based on the
virtual distortion method (VDM) and used force distortions to model fixed supports in the boundary nodes to isolate the considered
substructure from influences of the rest of the structure. The isolation method was based on applying virtual numerical fixed support
on all the boundary DoFs of the target substructure. In another study, Hou et al. [12] extended the substructure isolation method by
proposing a free virtual support and investigated the method experimentally. Model updating based on natural frequencies was then
used for damage identification of the isolated substructure. Hou et al. [13] applied the substructure isolation method in the frequency
domain and concluded that the frequency-domain method is computationally faster than the time-domain approach. Law et al. [14]
and Liu et al [15] proposed a substructure damage detection method based on identification of interface and external forces acting on
the target substructure and using time domain information. There have been also some studies on using the substructural flexibility
matrix or eigen solutions of the substructures to detect damage in the substructure [16–18]. In these studies, the global structure is
divided into many manageable substructures and changes in eigen solution or flexibility matrix of the target substructure are used for
damage identification.
Reviewing the literature reveals that there have been many studies published on development of substructural damage identifi­
cation methods specifically for shear structures such as multi-story buildings and multi-story frames. These methods are restricted to
shear structures and mostly use time response of different stories in the damage detection procedure. In other papers, either interface
forces between the substructures are required or a numerical virtual support is needed in the target substructure boundary to isolate
the target substructure from the rest of the structure.
This paper presents a combination of two well-known approaches in structural dynamics, namely an FRF-based substructuring and
an FRF-based finite element model updating, to overcome these issues. FRF-based substructuring is a well-known method for coupling
and decoupling of the substructures [19]. The dual form of substructuring was developed by Voormeeren. et al. [20] and it presented a
simple formulation to be used in experiments. FRF-based substructuring has been used in many different applications mostly for
dynamic identification purposes [21–24] but not for damage detection. FRF-based finite element model updating uses the finite
element model of the structure to detect, localize and quantify the damage by comparing the FRF obtained from the FE model to the
measured FRF of the damaged structure over many updating iterations [25,26]. FRF-based model updating differs from the common
modal-based model updating [27] because it directly uses the FRF data for the damage detection and does not need the extraction of
the modal parameters [28].
In this paper, substructure decoupling is used to decouple the target substructure from the assembled structure using FRFs of the
assembled structure and FRFs of the residual subsystem. The proposed method can be used for any civil or mechanical structure. In this
method, one subsystem is considered the “main” subsystem and is susceptible to damage and the “residual” subsystem(s) are
considered unchanged and the damage identification is only applied to the main subsystem. The residual substructure(s) are
considered unchanged which could be due to the fact that the residual substructure(s) are either not operating or their material or
loading is such that they are not susceptible to dynamical change. The FRF matrix of the “main” subsystem is obtained as a “stand­
alone” component that is completely decoupled from the residual subsystem(s) without the need of interface virtual support or
identifying interface forces. Then an FRF-based finite element model updating is used to detect, localize and quantify the damage in the
main substructure. This FRF-based model updating differs from the usual model updating methods by using numerical sensitivities to
solve the inverse problem instead of analytical sensitivities [26]. In analytical sensitivity-based methods, direct access to mass and/or
stiffness matrices is required, and a sensitivity matrix should be obtained [25,29]. In this paper, magnitudes of experimental and
numerical FRFs are directly used for model updating with no need for a sensitivity matrix. The FRF-based model updating is used since
the substructure decoupling technique is based on FRF measurement and calculation. Also, FRF-based model updating reveals more
information than model updating methods based on modal characteristic properties [30]. In addition, curve fitting on the measured
data is needed for extracting the modal properties, causing errors in the estimation of modal parameters, whereas FRF data do not
require curve fitting and hence will not be contaminated with numerical errors and loss of information [31,32]. The FRF matrix of the
residual substructure and the FRF matrix of the damaged global system are experimentally obtained and the main substructure’s FRFs
(damaged) are obtained in the decoupling process. Then the FE model of the main substructure is only needed for model updating and
the damage is detected by iteratively comparing the numerically obtained FRF with the decoupled experimental FRF of the main
substructure. The damage is simulated with a stiffness reduction in the main subsystem. The FE model of the main subsystem is divided
into different sections. In each iteration, a stiffness reduction is tried in each section and the FRF is obtained. Finally, the damage
location is the section for which stiffness reduction leads to the damaged FRF. The quantity of the damage is also detected using the
value of the stiffness reduction. In order to reduce the computation time in the model updating of the damaged structure and as a
verification for decoupling analysis of the structure under study, decoupling can be first performed on the undamaged structure (before
assembly of the substructures) and the undamaged main substructure FRF can be obtained. The accuracy of decoupling in different

2
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

frequency ranges can be verified by comparing the decoupled FRF of the main substructure and the directly measured one. The ac­
curacy of decoupling in different frequency ranges in undamaged condition would likely be the same in the damaged condition as well.
Therefore, we will disregard the frequency ranges for which decoupling has significant error, if those regions of the FRF are not near
peaks. We must consider the frequency ranges in the neighborhood of the peaks since damage (stiffness reduction) would likely shift
the peak(s) location.
Two numerical examples and an experimental example are presented in order to illustrate the procedure. In the experimental
example, the computation time reduction technique is implemented. Section 2 describes the substructure decoupling method and
Section 3 describes the substructural damage identification procedure. Numerical and experimental examples are presented in Section
4 and 5, respectively.

2. Substructure decoupling

The purpose of substructure decoupling is to find the dynamics of subsystem A as a “standalone” component that is completely
decoupled from the rest of the structure (AB and B). Without limitation, only two subsystems are considered here to illustrate the
procedure, but the method works for any number of subsystems. The FRF matrix of the assembled structure AB and the residual
substructure B are known and the FRF matrix of Substructure A is to be obtained.
In Fig. 1, system AB consists of degrees of freedom “internal to substructure B”, ub AB , “internal to substructure A”, ua AB and the
coupling DoF, ucou AB . The subsystems A and B also have their internal DoFs, ua A and ub B , respectively and the coupling DoFs, ucou A and
ucou B , respectively. Assuming the dynamics of system AB and B are known (measured), the following is a dynamic stiffness repre­
sentation of the assembled system AB in compact form [33]:

ZAB uAB = f AB + gAB

⎡ ⎤
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ZAB ZAB 0
⎢ AB bb bcou
⎥ ub fb 0
⎢Z
⎣ coub ZAB
coucou ZAB
coua
⎥⎣ ucou ⎦ = ⎣ f cou ⎦ + ⎣ gcou ⎦
⎦ (1)
0 ZAB ZAB ua fa 0
acou aa

And subsystem B:

ZB uB = f B − gB
[ ][ ] [ ] [ ]
ZBbb ZBbcou uBb fb 0
= − (2)
ZBcoub ZBcoucou uBcou f cou gBcou
⎤
⎡ [ ]
AB
ub
B uBb
where u = ucou is the vector of DoF of system AB (Fig. 1) (the superscripts AB are omitted for brevity), u =
⎣ ⎦ is the vector
ua uBcou
of DoF of substructure B, ZAB and ZB are the stiffness matrices of system AB and B, respectively, vectors f * are the external force vectors
and vectors g* represent the additional disconnection forces (with non-zero entries only at the interface DoF) felt from the decoupling
of the neighboring components [33].
The displacement compatibility condition, Eq. (3) and force equilibrium condition, Eq. (4) between the two (sub)structures are as
follows, respectively:

uBcou = ucou (3)

Fig. 1. Substructure decoupling schematic.

3
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

gcou + gBcou = 0 (4)

A more systematic description of the problem can be written by introducing the Boolean matrices B and L:
[ ]
[ ] uAB
Bu = BAB BB B = uBcou − ucou = 0 (5)
u
⎡ ⎤
⎢ 0 ⎥
[ ] ⎢ ⎥
[ ] gAB ⎢ 0 ⎥
T
L g= L ABT
L BT =⎢
⎢ 0
⎥=0
⎥ (6)
gB ⎢ ⎥
⎣ ⎦
gcou + gBcou

Eq. (5) presents the compatibility equation and Eq. (6) presents the equilibrium equation and matrices B and L are:
[ ]
B = BAB BB = [ 0 I 0 ‖ 0 − I ] (7)
⎡ ⃦ ⎤
I 0 0 ⃦ 0 0
[ ] ⃦
⎢0 0 I ⃦ 0⎥
0 (8)
T T
LT = LAB LB =⎣ ⃦ I 0⎦
0 0 0 ⃦
0 I 0 ⃦ 0 I

Therefore, the decoupling problem can be described in the following equations:

[ AB ][ AB ] [ AB ] [ AB ]
Z 0 u f g
= +
0 Z B
uB fB − gB
[ ]
[ ] uAB
BAB BB =0
uB
[ ]
[ ] gAB
(9)
T T
LAB LB B =0
g

Employing the dual formulation for decoupling [34], the interface forces are satisfied a priori by choosing interface forces of the
form of:

g = − BT λ (10)

where λ are Lagrange multipliers, corresponding physically to the interface force intensities. The equilibrium condition in Eq. (6) thus
becomes:

LT g = − L T B T λ = 0 (11)
But this condition is always satisfied [19]. The decoupling problem can therefore be formulated in a dual way as:
⎡ AB T ⎤⎡ AB ⎤ ⎡ AB ⎤
Z 0 BAB u f
⎣ 0 B ⎦⎣ u ⎦ = ⎣ 0 ⎦ (12)
B B T B
− Z
BAB BB 0 λ 0

where we assumed f B = 0, since subsystem B is not our particular interest here [33]. In a more general form, different Boolean matrices
for the compatibility and equilibrium conditions can be taken [33]:
⎡ AB T ⎤⎡ AB ⎤ ⎡ AB ⎤
Z 0 EAB u f
⎣ 0 E ⎦⎣ u ⎦ = ⎣ 0 ⎦ (13)
B B T B
− Z
CAB CB 0 λ 0

where E* are the Boolean matrices defining the location of the uncoupling forces while C* are the matrices enforcing compatibility.
After eliminating Lagrange multipliers λ in Eq. (13) and considering H = Z− 1 :
[( )+ ] [ ( )+ ]
Zu + ET CZ− 1 ET CZ− 1 f = f ⇒ u = H − HET CHET CH f (14)

[ ] [ AB ] [ AB ] [ ]
ZAB 0 u f ABT [ ]
where Z = B u = B , f = , ET = E BT and C = CAB CB and + denotes the (Moore–Penrose) pseudo-
0 − Z u 0 E
inverse since the matrix in the parenthesis is not necessarily a square matrix. The reason for this is that a different number of
compatibility and equilibrium conditions (in C and E matrices, respectively) are often adopted in experimental practice to better deal

4
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

with measurement error and/or when some DoFs cannot be properly accessed/measured. Therefore, the decoupled FRF matrix in a
general form can be found:
[ AB ] [ AB ][ ]( [ ][ ] )+ [ ]
H 0 H 0 EAB
T
[ AB ] HAB 0 EAB
T
[ AB ] HAB 0
HA = B − B BT C CB B BT C CB B (15)
0 − H 0 − H E 0 − H E 0 − H

Matrix HA in Eq. (15) is called the decoupled FRF matrix of substructure A but also contains extra rows and columns corresponding
to subsystem B. Also, those rows and the columns of HA corresponding to compatibility and equilibrium DoFs appear twice, both in the
rows and columns in − HB and HAB [35]. Therefore, only the rows and columns corresponding to the subsystem A are extracted from
matrix HA and are considered decoupled FRFs of subsystem A.
Based on choosing different DoFs for compatibility (C) and equilibrium (E), various collocated and non-collocated decoupling
approaches can be used. A good discussion of this categorization is presented in [33].

3. Substructural damage identification procedure

In the proposed damage identification procedure, the FRF measurement is performed on the global damaged structure and on the
residual substructure (before assembly) and the damaged main substructure’s FRFs are obtained using Eq. (15). Then the decoupled
FRF (of the damaged main substructure) is used for FRF-based model updating and damage detection. Fig. 2 depicts the schematic
diagram of the damage identification procedure.
In Fig. 2 Hd A is the FRF matrix of the damaged substructure A obtained from decoupling, Han A is analytical FRF obtained from the
FE model of substructure A and Hd A is one (or a few) FRF in the FRF matrix Hd A to be used for model updating. The damage is simulated
with the decrease in the stiffness at some locations in the main substructure. In each iteration of model updating, the stiffness in
different locations in the substructure A FE model is changed and the FRF obtained from the FE model is compared to the decoupled
experimental FRF Hd A in the frequency range of interest. This process continues until the analytical and decoupled FRF are matched
with the desired accuracy. The FRF-based model updating procedure is presented in the following section.

3.1. FRF-based model updating

The accelerance FRF Ha,b (ω) is the acceleration response at node a, Aa (ω), divided by the force at node b, Fb (ω), both in the fre­
quency domain:

Aa (ω) ∑ m ̂ ai ϕ
− ω2 ϕ ̂ bi
Ha,b (ω) = ≅ 2 + 2iωΩ ξ + Ω2
(16)
Fb (ω) i=1 − ω i i i

where m is the number of modes used for constructing the FRF, ω is frequency, Ωi is ith natural frequency of the system, ξi is ith damping
ratio corresponding to ith mode and ϕ̂ ai and ϕ
̂ bi are the scalar values of mode shape i at response node a and excitation node b,
respectively [36]. The form of FRF used for model updating is as follows, with the values of the FRF magnitude in dB:

Fig. 2. Substructural damage identification procedure diagram.

5
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

(⃒ ⃒)
H a,b (ω) = 20log10 ⃒Ha,b (ω) ⃒ (17)

The goal of the FE model updating is to change the properties of a FEM such that the analytical response of the model matches that
of measured data. After developing the FE model, Eqs. (16) and (17) are used to obtain the FRF at the response and excitation nodes
corresponding to the response and excitation locations considered in the experiment (decoupled FRF). Then the difference between
analytical and measured FRFs, the error or residual, is found and a scalar objective error function value is calculated [26]. With the
scalar objective error function value, numerical optimization techniques are used to minimize the difference between the analytical
and measured FRFs by modifying the parameters of the FEM [26]. Substituting Eqs. (16) and (17) and using modal responses as
functions of the unknown updating parameters p results in Eq. (18) [26]:
(⃒ ⃒)
⃒∑ m ̂ ai (p) ϕ
− ω2 ϕ ̂ bi (p) ⃒
⃒ ⃒
H a,b (ω, p) = 20log10 ⃒ 2 ⃒ (18)
⃒ i=1 − ω2 + 2iωΩi (p)ξi (p) + Ω i (p) ⃒

The error function, or residual e, is defined as:

(19)
a m
e(ω, p) = H ω, p) − H
a,b ( ω, p)
a,b (

a m
where H a,b (ω, p) and H a,b (ω, p) are analytical and measured FRFs (decoupled), respectively. Then, a scalar objective error function J
is created and minimized through bounded numerical optimization:

J(p) = eT (ω, p)e(ω, p) (20)

The purpose of optimization is to minimize J(ω, p) by considering the upper and lower bounds for p [28]. The scalar objective error
function is normalized by the initial value of the scalar objective error function J(p)1 so that the optimization starts at a value of 1 and
the perfect match would be a value of 0.
J(p)i
J(p)i = (21)
J(p)1

The parameter p is a unitless modifier that modifies stiffness parameter.

4. Numerical examples

In this section, two numerical examples are presented to illustrate the substructural damage identification procedure.

4.1. Numerical case study 1

The simple problem used here is shown in Fig. 3 and consists of two lightly damped mass-spring-damper systems. Subsystem A has
5 degrees of freedom and is the main subsystem under damage, subsystem B possesses 3 DoF and it is not susceptible to damage. The
two subsystems are coupled at a 2 DoF interface (mA4 , mB1 and mA5 , mB2 ). The system properties are given in Table 1. Two damage
scenarios are considered. The first scenario D1 is a single-damage scenario and the damage in subsystem A is simulated by 20%
reduction in the value of stiffness parameter kA2 and the second scenario D2 is a multi-damage case and the damage is simulated by
10% stiffness reduction in kA3 and 5% reduction in kA5 . The first step in the procedure is obtaining the decoupled FRFs of subsystem A
from the assembled structure.
Fig. 4 shows the FRF H34 A , one of the FRFs in subsystem A, and the similar FRF obtained from decoupling in the undamaged and D1

Fig. 3. Lumped parameter system: Case study 1.

6
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Table 1
Lumped parameter system parameters.
Subsystem A Subsystem B

mA1 = 4,mA2 = 3,mA3 = 3,mA4 = 6,mA5 = 2kgkA1 = 500,kA2 = 2000,kA3 = 2000,kA4 = 1000, mB1 = 2, mB2 = 4, mB3 = 8kgkB1 = 1000, kB2 = 1000, kB3 =
N Ns N Ns
kA5 = 500 cA1 = 0.2, cA2 = 0.5, cA3 = 0.1, cA4 = 0.1, cA5 = 0.1 2000 cB1 = 0.4, cB2 = 0.1, cB3 = 0.3
m m m m

Fig. 4. Subsystem A FRF H34 A and the decoupled FRF obtained from decoupling in damaged D1 and undamaged conditions.

damaged conditions. As can be seen, the FRFs completely match in both the conditions. This result is a motivation to use the decoupling
method to obtain the main subsystem FRF when the global system is under operation and it is impossible to separate the subsystems.
After obtaining the main subsystem FRF, the FRF-based model updating is used only on the main subsystem to detect the damage.
The decoupled FRF in the damaged condition in Fig. 4 is used in the FE model updating using the subsystem A model to detect the
location and quantity of stiffness reduction. Fig. 5 shows the undamaged and damaged FRF (obtained from decoupling) and the
updated FRF after model updating. The initial guess in the updating process is the undamaged FRF and the final result is the updated
FRF. The optimization variables are the five stiffness parameter modifiers between 0 and 1 that is multiplied by each stiffness
parameter kA1 ,kA2 ,kA3 ,kA4 ,kA5 . In each iteration, these five parameters are changed and the obtained FRF is compared to the damaged
one until the two FRFs match with the desired accuracy. Fig. 6 shows the stiffness parameter multiplier values after the updating
process. The second stiffness parameter kA2 has 20% reduction and it is detected in the updating process. Stiffness parameters 1–5 are
corresponding to kA1 , kA2 , kA3 , kA4 , kA5 , respectively.
Fig. 7-a shows the FRF H34 A in D2 scenario and the similar decoupled FRF. The decoupled FRF matches the true one in the multi-
damage case as well. Fig. 7-b shows the updated FRF after updating process similar to D1 case. The stiffness parameter modifiers in D2

Fig. 5. Subsystem A FRF H34 A in model updating: D1 scenario.

7
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 6. Stiffness parameter modifiers in subsystem A: D1 scenario.

Fig. 7. Subsystem A FRF H34 A and the decoupled FRF obtained from decoupling in damaged D2 and undamaged conditions.

scenario are shown in Fig. 8. The location and quantity of stiffness reduction is detected accurately.

4.2. Numerical case study 2

The global system in this numerical example is shown in Fig. 9 and it consists of 34 nodes and 33 three-dimensional beam elements.

8
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 8. Stiffness parameter modifiers in subsystem A: D2 scenario.

Fig. 9. Numerical case study 2 FE model.

The main subsystem (A) and the residual subsystem (B) are shown. The material for the main subsystem has Young’s modulus of E =
90GPa and the density of 500 mkg3 and the residual subsystem has Young’s modulus of E = 220GPa and density of 3000 mkg3 . Three points
P1, P2 and P3 are used for obtaining the FRFs of the assembled system (in X direction) and the residual subsystem and the FRFs of the
main subsystem are obtained through the decoupling process. Damage detection is performed for the five sections shown in the figure.
Each section is 1 m length in the model. Two damage scenarios D1 and D2 are considered. In the first scenario D1, the damage is
simulated by 20% reduction in the Young’s modulus value in section 3 and scenario D2 is a multi-damage scenario and is simulated by
20% reduction in young modulus of section 2, 20% reduction in section 4 and 10% reduction in section 5. The FE model is developed

9
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

using Abaqus software and FE updating is performed with the use of Abaqus2Matlab software [37] and MATLAB’s fmincon nonlinear
constrained optimization routine. The optimization variables are five stiffness parameter modifiers between 0 and 1 that are multiplied
by the Young’s modulus of sections 1–5.
FRFs corresponding to points P1, P2 and P3 in the assembled structure are obtained by applying a unit 1 N load at each of the points
and obtaining the acceleration at the other points in the frequency domain. Then the FRF matrix HAB is obtained by inserting all the
FRFs in a matrix. Only one of the two equal symmetric array elements needs to be calculated. The FRF matrix of the residual subsystem
contains only one FRF H11 B since only point P1 is needed in the decoupling process. The FRF matrices are then inserted in Eq. (15) to
obtain the decoupled FRFs of the main subsystem. Fig. 10-a shows the FRF corresponding to point P3 in subsystem A, H33 A , in the
undamaged and damaged D1 condition and the similar FRF obtained from decoupling (Eq.(15)). The FRF obtained from the main
subsystem-only model and the one obtained from decoupling (Damaged Main Subs – Decoupled) match very well. There is a slight
difference around the first mode which could be due to numerical singularity in the decoupling. Fig. 10-b shows the damaged and the
updated FRF of the main subsystem obtained from model updating. Fig. 11 shows the stiffness parameter modifiers of sections 1–5
obtained after model updating in D1 scenario. The location of the damage (stiffness reduction) is detected very well but the quantity of
it (20%) has a slight error.
Figs. 12 and 13 show the same result as Figs. 10 and 11, respectively for the scenario D2. The locations of the damage is detected
again but the quantity has a small error (Fig. 13).

5. Experimental case study

The experimental example is a simple structure consisting of two subsystems that are bolted together. The structure is shown in
Fig. 14 with the dimensions shown. Two aluminum beams are bolted together at point P1 (Coupling point) and the main (A) and

Fig. 10. a) FRF H33 A of the main subsystem and the decoupled FRF H33 A in damaged D1 and undamaged conditions b) Damaged, undamaged and
updated FRF in D1 scenario.

10
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 11. Stiffness parameter modifiers in D1 scenario.

Fig. 12. a) FRF H33 A of the main subsystem and the decoupled FRF H33 A in damaged D2 and undamaged conditions b) Damaged, undamaged and
updated FRF in D2 scenario.

residual (B) substructures are shown in Fig. 14. The cross section of the beams are 20 × 2mm2 with a Young’s modulus of E = 70GPa
and density of 2700 mkg3 . The structure was first tested in the undamaged state. Afterwards, one cut with the dimensions shown in
Fig. 14-b was introduced in the main substructure. The width of the cut is 4 mm on each side with 8 cm length, therefore the width of

11
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 13. Stiffness parameter modifiers in D2 scenario.

Fig. 14. a) Schematic of the experimental case study b) The cut in the main subsystem to introduce damage.

the beam in the cut section is 12 mm, simulating damage as 40% stiffness reduction in that section. Five sections, sections 1–5 as shown
in Fig. 14-b, are used in the model updating of the main subsystem and the location of the damage (section 1) is identified in the
damage detection. Each section is 8 cm length, the same length as the cut. Before model updating, experimental vibration measure­
ment was performed on the global structure and the residual subsystem. The FRFs of the global structure in both the undamaged and
damaged states were obtained at points P1 and P2 and the FRF of the residual substructure was obtained at P1. The residual subsystem
was the same before and after introducing the damage since we consider this subsystem not susceptible to damage. Fig. 15 shows the

12
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 15. Experimental case study a) Damaged global structure b) The residual subsystem (Free-free boundary conditions) c) Coupling point
accelerometer d) The cut in the main substructure.

damaged assembled structure, the cut in the main subsystem, the residual subsystem modal test setup and mounting the sensor on the
coupling point. The residual subsystem was tested in free-free condition and two thin elastic bands were used to simulate the free-free
boundary condition. Two B&K 4507-B-004 accelerometers were used at the two locations P1 and P2 to obtain the FRFs of the
assembled structure and one accelerometer is used at point P1 to measure the FRF of the residual subsystem. FRFs were obtained using
a B&K 8205-002 impact hammer, a National Instruments c-DAQ 9172 chassis and the software ModalView [38] was used to analyze
the data. In each measurement, the structure was impacted five times to average the signals using RMS averaging. A 2 × 2 matrix HAB
and a 1 × 1 matrix HB were measured and the measured FRFs are used in Eq. (15) to predict the FRF of the main subsystem, decoupled
from the rest of the structure. In all the measurements, only the out of plane motion (Z direction in Fig. 14) was excited and measured to
obtain the FRFs.
The FE model of the subsystem A (a cantilever beam) with 3D beam elements is developed using Abaqus software and FE updating
is performed with the use of Abaqus2Matlab software [37] and MATLAB’s fmincon nonlinear constrained optimization routine. Five

13
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

sections 1–5 are defined (Fig. 14) in the FE model and five modifiers between 0 and 1 are used to multiply to the cross-section width of
the sections in each iteration. The initial FRF obtained from the model was significantly different from the experimental FRF. The
thickness of the beam was changed to 3.5 mm in the FE model (instead of 2 mm) and the FRF result improved significantly. This
thickness change in the model does not affect the damage detection results since we need the stiffness reduction percentage (per­
centage of width reduction) and more importantly where the stiffness reduction exists in the structure. Fig. 16 shows the FRF H11 A
directly measured from subsystem A and obtained from decoupling in the undamaged state, as well as the FRFs obtained from the
initial FE model and the modified FE model (after thickness change). The decoupled and directly-measured FRFs match well in most of
the frequency ranges and the error could be due to the error in simulating the free-free condition of the residual subsystem and bolted
joint nonlinearity. The bolted joint nonlinear effects on the substructuring are investigated in [39,40]. A frequency range of 30–60 Hz
is used for model updating of the damaged state since there is a good match between FRFs in the undamaged state and it is in the
neighborhood of the peak. As can be seen, the numerical FRF (obtained from FEM) has a better match with the experimental one after
the modification.
Fig. 17 shows the FRF H12 A , obtained from the FE model before and after updating, the experimentally obtained FRF in the un­
damaged state (directly measured from the cantilever beam) and the FRF obtained from decoupling of the damaged main subsystem.
These FRFs are obtained (in the model and test) by the excitation of the out of plane vibration of point P1 and obtaining the out of plane
acceleration of point P2. The “initial” FRF in this figure is the FRF obtained after modification of the model. The damaged FRF shows a
shift to the left which is a sign of stiffness reduction compared to the undamaged FRF and the updated FRF match with the damaged
FRF. The stiffness modifiers are shown in Fig. 18. The location of the damage is successfully detected (section 1) with only a slight error
in magnitude.
Future research efforts to enrich the experimental results could include using rotational accelerometers [41] and/or using more
sensors (on the residual subsystem) to capture the bending and torsional coupled vibration of the system. The accuracy of decoupling,
as one step in this damage detection method, can be improved by using a higher number of sensors and using an algorithm to choose
the optimum sensor locations. The decoupling and model updating in this paper assumed linearity. The bolted connection could be a
source of nonlinearity, likely proportional to tightening torque and excitation amplitude, the quantification of which would lead to a
better understanding of the range of operating and test conditions for which the damage detection method is unaffected.

6. Conclusion

This paper proposes a substructural damage detection method using frequency response functions. In the proposed method, one
substructure is considered the main substructure (susceptible to damage) and the rest of the substructure(s) (“residual”) are considered
undamaged or not susceptible to damage, a situation which often happens in practice. In the proposed method, unlike other sub­
structural damage detection methods, there is no need for placing a numerical virtual support in the target substructure boundary or
obtaining the interface forces. The frequency response functions of the assembled damaged structure and the frequency response
function(s) of the residual substructure(s) are needed to obtain the FRFs of the main substructure, completely decoupled from the rest
of the structure. An FRF-based model updating is then performed using only the FE model of the main substructure to detect, localize
and quantify the damage in the main substructure. In order to reduce the computation time in the model updating of the damaged
structure, a smaller frequency range in the FRF can be chosen based on the accuracy of decoupling results in the undamaged structure.
Two numerical examples and an experimental case study are presented to illustrate the procedure. The damage in the main sub­
structure is detected and quantified with a good accuracy in the numerical and experimental case studies.

Fig. 16. FRF of the main subsystem,H11 A , obtained from direct measurement, decoupling, initial FE model and the modified FE model.

14
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

Fig. 17. FRF updating of the main subsystem.

Fig. 18. Stiffness parameter modifiers in experimental case study.

Credit Author Statement

Mohammad Hadi Jalali analyzed the literature, performed the experiments, collected the data, developed the codes and prepared
the manuscript. D. Geoff Rideout edited and helped organize the manuscript.

Funding

The authors would like to thank Newfoundland and Labrador Hydro for their financial and in-kind support, and the Natural
Sciences and Engineering Research Council of Canada (NSERC) for their support under the Collaborative Research and Development
(CRD) program.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

References

[1] J. Li, S.S. Law, Y. Ding, Substructure damage identification based on response reconstruction in frequency domain and model updating, Eng. Struct. 41 (2012)
270–284.
[2] S. Weng, H. Zhu, Y. Xia, J. Li, W. Tian, A review on dynamic substructuring methods for model updating and damage detection of large-scale structures, Adv.
Struct. Eng. 23 (3) (2020) 584–600.
[3] H. Zhu, J. Li, W. Tian, S. Weng, Y. Peng, Z. Zhang, Z. Chen, An enhanced substructure-based response sensitivity method for finite element model updating of
large-scale structures, Mech. Syst. Signal Process. 154 (2021), 107359.

15
M.H. Jalali and D.G. Rideout Mechanical Systems and Signal Processing 163 (2022) 108166

[4] D. Zhang, E.A. Johnson, Substructure identification for shear structures I: Substructure identification method, Struct. Control Heal. Monit. 20 (5) (2013)
804–820.
[5] D. Zhang, E.A. Johnson, Substructure identification for shear structures: cross-power spectral density method, Smart Mater. Struct. 21 (5) (2012) 055006,
https://doi.org/10.1088/0964-1726/21/5/055006.
[6] Z. Xing, A. Mita, A substructure approach to local damage detection of shear structure, Struct. Control Heal. Monit. 19 (2) (2012) 309–318.
[7] D. Zhang, S. Li, H. Li, Adaptive substructure identification for shear structures with virtual control system, Mech. Syst. Signal Process. 121 (2019) 426–440.
[8] C. DeVore, Z. Jiang, R.E. Christenson, G. Stromquist-LeVoir, E.A. Johnson, Experimental verification of substructure identification for damage detection in shear
buildings, J. Eng. Mech. 142 (2016) 4015060.
[9] L. Mei, H. Li, Y. Zhou, W. Wang, F. Xing, Substructural damage detection in shear structures via ARMAX model and optimal subpattern assignment distance,
Eng. Struct. 191 (2019) 625–639.
[10] R. Hou, Y. Xia, Review on the new development of vibration-based damage identification for civil engineering structures: 2010–2019, J. Sound Vib. 491 (2021)
115741, https://doi.org/10.1016/j.jsv.2020.115741.
[11] J. Hou, Ł. Jankowski, J. Ou, A substructure isolation method for local structural health monitoring, Struct. Control Heal. Monit. 18 (6) (2011) 601–618.
[12] J. Hou, Ł. Jankowski, J. Ou, Experimental study of the substructure isolation method for local health monitoring, Struct. Control Heal. Monit. 19 (4) (2012)
491–510.
[13] J. Hou, Ł. Jankowski, J. Ou, Frequency-domain substructure isolation for local damage identification, Adv. Struct. Eng. 18 (1) (2015) 137–153.
[14] S.S. Law, D. Yong, Substructure methods for structural condition assessment, J. Sound Vib. 330 (15) (2011) 3606–3619.
[15] K. Liu, S.-S. Law, X.-Q. Zhu, Substructural condition assessment based on force identification and interface force sensitivity, Int. J. Struct. Stab. Dyn. 15 (2015)
1450046.
[16] S. Weng, Y. Xia, Y.-L. Xu, H.-P. Zhu, An iterative substructuring approach to the calculation of eigensolution and eigensensitivity, J. Sound Vib. 330 (14) (2011)
3368–3380.
[17] S. Weng, H.-P. Zhu, Y. Xia, L. Mao, Damage detection using the eigenparameter decomposition of substructural flexibility matrix, Mech. Syst. Signal Process. 34
(1-2) (2013) 19–38.
[18] S. Weng, H.-P. Zhu, P.-H. Li, Y. Xia, L. Ye, Construction of orthogonal projector for the damage identification by measured substructural flexibility, Measurement
88 (2016) 441–455.
[19] D. de Klerk, D.J. Rixen, S.N. Voormeeren, General framework for dynamic substructuring: History, review and classification of techniques, AIAA J. 46 (2008)
1169–1181.
[20] S.N. Voormeeren, D.J. Rixen, A dual approach to substructure decoupling techniques, in: Proc. IMAC-XXVIII, Jacksonville, Florida, USA, 2010.
[21] M.V Van der Seijs, Experimental dynamic substructuring: Analysis and design strategies for vehicle development, 2016.
[22] M.H. Jalali, D.G. Rideout, Frequency-based decoupling and finite element model updating in vibration of cable–beam systems, J. Vib. Control.
1077546321996936 (2021).
[23] S. Rahimi, D. de Klerk, D.J. Rixen, The Ampair 600 wind turbine benchmark: results from the frequency based substructuring applied to the rotor assembly, in:
Top. Exp. Dyn. Substruct. Vol. 2, Springer, 2014: pp. 179–192.
[24] Mohit Law, Hendrik Rentzsch, Steffen Ihlenfeldt, Predicting mobile machine tool dynamics by experimental dynamic substructuring, Int. J. Mach. Tools Manuf.
108 (2016) 127–134.
[25] John E. Mottershead, Michael Link, Michael I. Friswell, The sensitivity method in finite element model updating: A tutorial, Mech. Syst. Signal Process. 25 (7)
(2011) 2275–2296.
[26] J.D. Sipple, M. Sanayei, Finite element model updating using frequency response functions and numerical sensitivities, Struct. Control Heal. Monit. 21 (2014)
784–802.
[27] J.J. Moughty, J.R. Casas, A state of the art review of modal-based damage detection in bridges: Development, challenges, and solutions, Appl. Sci. 7 (2017) 510.
[28] Jesse D. Sipple, Masoud Sanayei, Finite element model updating of the UCF grid benchmark using measured frequency response functions, Mech. Syst. Signal
Process. 46 (1) (2014) 179–190.
[29] Chao-Dong Zhang, You-Lin Xu, Multi-level damage identification with response reconstruction, Mech. Syst. Signal Process. 95 (2017) 42–57.
[30] Z. Niu, Two-step structural damage detection method for shear frame structures using FRF and Neumann series expansion, Mech. Syst. Signal Process. 149
(2021), 107185.
[31] Akbar Esfandiari, Mansureh-Sadat Nabiyan, Fayaz R. Rofooei, Structural damage detection using principal component analysis of frequency response function
data, Struct. Control Heal. Monit. 27 (7) (2020), https://doi.org/10.1002/stc.v27.710.1002/stc.2550.
[32] M.H. Jalali, G. Rideout, Development and Validation of a Numerical Model for Vibration of Power Lines, in: CSME Int. Congr. Toronto, 2018, CSME-SCGM,
Toronto, 2018.
[33] S.N. Voormeeren, D.J. Rixen, A family of substructure decoupling techniques based on a dual assembly approach, Mech. Syst. Signal Process. 27 (2012)
379–396.
[34] D. de Klerk, D.J. Rixen, J. de Jong, The frequency based substructuring (FBS) method reformulated according to the dual domain decomposition method, in:
24th Int. Modal Anal. Conf. St. Louis, MO, 2006.
[35] Walter D’Ambrogio, Annalisa Fregolent, Substructure decoupling as the identification of a set of disconnection forces, Meccanica. 52 (13) (2017) 3117–3129.
[36] Mohammad Hadi Jalali, Geoff Rideout, Analytical and experimental investigation of cable–beam system dynamics, J. Vib. Control. 25 (19-20) (2019)
2678–2691.
[37] G. Papazafeiropoulos, M. Muñiz-Calvente, E. Martínez-Pañeda, Abaqus2Matlab: a suitable tool for finite element post-processing, Adv. Eng. Softw. 105 (2017)
9–16.
[38] ABSignal, Getting started with ModalView, 2012.
[39] Şerife Tol, H. Nevzat O¨zgu¨ven, Dynamic characterization of bolted joints using FRF decoupling and optimization, Mech. Syst. Signal Process. 54-55 (2015)
124–138.
[40] Taner Kalaycıoğlu, H. Nevzat Özgüven, FRF decoupling of nonlinear systems, Mech. Syst. Signal Process. 102 (2018) 230–244.
[41] Armin Drozg, Gregor Čepon, Miha Boltežar, Full-degrees-of-freedom frequency based substructuring, Mech. Syst. Signal Process. 98 (2018) 570–579.

16