Engineering Structures 28 (2006) 453–465

www.elsevier.com/locate/engstruct

A modal parameter based technique to inspect welded reinforcement splices

Sooyong Park a , Sanghyun Choi b,∗ , Norris Stubbs c , Robert Bolton d , Angie H. Price d ,
Charles Sikorsky e
a School of Architecture, Youngsan University, Junam-ri, Ungsang-up, Yangsan-si, Kyungnam 626-847, Republic of Korea
b Structural Systems and Site Evaluation Department, Korea Institute of Nuclear Safety, Yoosung PO Box 114, Guseong-dong, Yuseong-ku, Daejeon 305-600,
Republic of Korea
c Department of Civil Engineering, Texas A&M University, College Station, TX 77843, USA
d Department of Engineering Technology and Industrial Distribution, Texas A&M University, College Station, TX 77843, USA
e State of California Department of Transportation, Sacramento, CA 95816, USA

Received 3 June 2004; received in revised form 22 August 2005; accepted 1 September 2005
Available online 10 October 2005

Abstract

The objective of this study is to investigate the technical feasibility of using a modal parameter based global nondestructive damage evaluation
method to evaluate, quantitatively, welded reinforcement splices during the construction process. For the purpose of this study, an expression
which predicts the strength of the weld, in terms of the yield strength of the steel and the effective stiffness change in the weld, is developed.
Then, an experiment to nondestructively measure the fractional stiffness changes, using changes in modal parameters and a system identification
technique, in the welded region of 27 welded reinforcing bars is performed. For comparative purposes, a conventional nondestructive testing (NDT)
method, radiographic inspection, and destructive tensile tests are performed to evaluate the accuracy of the nondestructive strength prediction
procedure. The predicted strengths of the weld systems, obtained from the strength–damage models and the nondestructively measured fractional
stiffness change, are compared with those from the destructive tensile test. Also, areas for potential improvement for field implementation of the
methodology are investigated.
c 2005 Elsevier Ltd. All rights reserved.

Keywords: Nondestructive damage evaluation; Welded splices; Modal parameters; Strength prediction

1. Introduction splice as a function of the maximum dimension of any

Quality assurance of the various types of welded splices is
accomplished using nondestructive testing (NDT) techniques,
pre-approved installation procedures, and visual inspection.
For example, visual inspection of lap splices is sufficient for
evaluation, since the strength of the connection is a function of
the length of the lap splice. Mechanical couplers are typically
pre-approved on the basis of laboratory testing prior to their
installation in the field. While inspection of welded splices
could be accomplished via a visual inspection during the actual
welding of the splice, radiographs of the welded joints are
usually required, as described in the standard specifications.
The structural welding code [1] defines acceptance of a welded
∗ Corresponding author. Tel.: +82 42 868 0666; fax: +82 42 868 0523.
E-mail address: schoi@kins.re.kr (S. Choi).
single porosity or fusion-type discontinuity. Procedures used to
determine the acceptance of a welded reinforcement splice are
based on a statistical analysis of the imperfections in welded
splices collected from welds fabricated in the laboratory. The
detection of a given defect is dependent on such factors
as location, orientation and shape of the defect, the weld
material, the inspector, and the inspection environment [2].
Yet, to date, potential weaknesses exist in current testing and
acceptance procedures for welded reinforcement splices. Local
NDT methods such as acoustic or ultrasonic methods, magnetic
field methods, radiographs, eddy-current methods, and thermal
field methods [3] yield only a qualitative description of surface
and internal flaws in a weld system. The quantification of the
impact of the identified defects on the performance of the
weld system is not usually performed. Even if the geometry
of a flaw were known, Fracture Mechanics or other strength

0141-0296/$ - see front matter 

c 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2005.09.002
454 S. Park et al. / Engineering Structures 28 (2006) 453–465
predicting methods must be utilized to estimate the strength or
the remaining useful life of a structural element. In addition,
if the geometry of the defect were complex, such as in the
case of a defective weld, the application of Fracture Mechanics,
which applies mostly to isolated flaws with simple geometric
configurations, is limited. Given the catastrophic consequences
of a possible failure of weld system with an undetected
defect, a more reasonable and quantitative nondestructive field
inspection method for welded splices in steel reinforcement
bars is needed.
Another approach to NDT comprises the so-called global
methods. The term “global” refers to the fact that the integrity
of a structure is interrogated using properties that belong to the
entire structure (e.g., resonant frequencies and mode shapes).
Methods that utilize changes in the vibrational characteristics
of the structure to detect defects and estimate the severity
of the defects have recently received much attention [4,5].
The objective of this study is to investigate the technical
feasibility of using a vibration-based method to evaluate,
quantitatively, the quality of welded reinforcement splices
during the construction process. One of the premier vibration-
based global methods, the damage index method [6], has been
selected for this study. This method has been corroborated
using (1) numerically simulated data for various structural types
[7–9], (2) experimental modal data generated in a laboratory
environment [10,11], and (3) field data measured on bridge
structures [12,13]. Notably, a recent study on the performance
of various damage detection schemes found in the literature
[13] confirmed the efficacy of the damage index method to
locate and estimate the severity of damage.
The overall approach utilized to meet the stated objective
is driven by the following logic. First, the model used to
estimate the stiffness change at the location of the weld
is described. Second, a model which predicts the strength
of the weld in terms of the stiffness change and the yield
strength of the steel is developed. Third, an experiment to
nondestructively measure the fractional stiffness changes, using
changes in modal parameters and system identification, in
the weld region of 27 welded reinforcing bars is performed.
Fourth, a conventional NDT evaluation technique, radiographic
inspection, is performed independently to evaluate the quality
of the welds of the same specimens. Fifth, destructive tests
to measure the true weld capacities of the specimens are
performed. Finally, a comparison of the results using the three
different methods, the strength–damage model, radiographic
inspection, and the destructive tests is made.
2. Damage severity estimation scheme
The damage index method evolved from a sensitivity
approach that relates changes in modal responses, specifically
resonant frequencies, to changes in the mass, damping, and
stiffness of a structure [14]. The damage index method utilizes
the change in mode shapes of the pre-damage and post-damage
structure to detect and locate damage in a structure [6]. Some
advantages of the method include: it can detect, locate, and
quantify damage in a structure; it only requires a limited
Fig. 1. A rod element subjected to axial loading. (a) original (baseline)
element; (b) element with discontinuities; (c) element with effective uniform
stiffness.
number of mode shapes; and it can yield element level damage
information with statistically based confidence levels.
The damage index, β, which represents the ratio of the pre-
damaged and post-damaged stiffness parameters of an element
in a system, for the element j and NM eigenmodes is defined
as [15]:


N M  
φi∗T C j φi∗ + φi∗T K φi∗ φiT K φi
kj i=1
βj = ≈ (1)
k ∗j M
N

φiT C j φi + φiT K φi φi∗T K φi∗
i=1
where k j is the stiffness parameter of the j th element; φi is the
i th eigenvector; K is the system stiffness matrix; the superscript
T indicates the transpose of the matrix; the superscript asterisk
represents the parameters for the damaged structure; and the
matrix C j only involves geometric quantities and possibly
Poisson’s ratio. For the problem in hand, the damage index
represents the ratio of the effective stiffness between the
baseline analytical model and the weldment of the reinforcing
bar. The concept of the effective stiffness can be demonstrated
as follows [16].
Consider a prismatic bar element with the original (baseline)
stiffness k as shown in Fig. 1(a). Also consider a corresponding
welded part of a reinforcing bar with discontinuities as shown
in Fig. 1(b). Assuming that both the bar element and the
welded part are subjected to the same axial force, P, then the
corresponding elongations would be different as ∆ and ∆∗
due to the discontinuities. For this case, if an element with
uniform stiffness, k ∗ , as shown in Fig. 1(c) yields the same
elongation, ∆∗ , for the same axial force, P, then the effective
stiffness for the welded part would be k ∗ . Thus, the effective
reduction in the stiffness of the element due to the introduction
of discontinuities is the difference between k and k ∗ .
Using the damage index, the difference of the stiffness
between the analytical baseline model and the welded part can
be effectively expressed as follows [15]:
k ∗j − k j 1
αj = = − 1. (2)
kj βj
The severity index, α, represents the fractional change in
stiffness parameter of an element. Theoretically, α ranges from
−1, total loss of stiffness, to +∞, infinite increase in stiffness.
Note that if α < 0, the interpretation is stiffness decrease, and
if α > 0, stiffening has occurred.
 S. Park et al. / Engineering Structures 28 (2006) 453–465
3. Quantitative evaluation of weld capacity
In the discipline of Continuum Damage Mechanics, damage
and damage growth are related to the initial state of damage in
a material and the loading environment to which the material is
subjected. However, in that field, the consensus is that damage
is treated as an unobservable internal thermodynamic variable.
Physically, damage has been related to the fraction of voids
in a plane intersecting a material and the reduction in the
elastic modulus of the material. More specifically, Kachanov
[17] considered the damage variable as a surface density of
intersections of cracks and cavities. For quantifying a weld
capacity, a relationship between the damage in a material and
the capacity of the material is developed. Here, the capacity
is measured in terms of the yield strength of the system,
because the yield strength is being utilized as the most common
indicator of the strength of reinforcing steel bars in structural
design and analysis.
A rod containing a weld and subjected to an axial force P
is shown in Fig. 2. Let the weld system consist of the filler
material and the interface between the filler material and the
base metal. Restricting this analysis to failure in the filler
material, let the yield strength and cross-sectional area of the
weld system without any defects be σ y and AU , respectively.
Then, the load, PU y , to cause yield in the weld system without
defects is given by:
PU y = σ y AU . (3)
Assume that defects (in the form of cracks, voids, porosity,
incomplete fusion, etc.) are introduced into the weld system
such that the cross-sectional area is reduced to A D < AU . Then
the load, PDy , to cause yield in the weld system with defects
becomes:
PDy = σ y A D . (4)
Dividing Eq. (4) by Eq. (3) yields:
PDy AD E AD E ID
= = ≈ (5)
PU y AU E AU E IU
where E and I are, respectively, Young’s modulus and the
second moment of inertia of the cross-section. Here it is
assumed that the damaged region is small compared to
the cross-sectional area so that the ratio A D /AU can be
approximated to I D /IU . For a circular section, 2% deduction in
the cross-sectional area causes 1%–3% decrease (4% at most)
in the second moment of inertia of the cross-section. According
to Eq. (5), the ratio of loads to cause yield equals the ratio
of the bending stiffnesses of the defect-free and defected weld
systems. Note that the ratio EI D /EI U is the reciprocal of the
damage index, β, defined in Eq. (1). That is
E ID k∗ 1
= = = (1 + α). (6)
E IU k β
Combining Eqs. (5) and (6),
PDy = PU y (1 + α). (7)
455
Fig. 2. A rod containing weld subjected to an axial force.
Dividing both sides of Eq. (7) by AU yields:
σ Dy = σ y (1 + α). (8)
From Eqs. (1) and (2), α can be determined nondestructively by
utilizing the modal parameters of the structure containing the
weld. Since the intent here is to inspect the weld by measuring
α after welding, the analysis ignores damage growth subsequent
to loading. This situation is analogous to a rod with a circular
hole that is loaded to yield. In such a case, a strength of
materials analysis indicates no change in the geometry of the
hole at full yield. Furthermore, the yield strength, σ y , can
be associated with the nominal properties of the weld. Thus,
Eqs. (7) and (8) predict the load and strength at yield for a
defective weld, given the damage and the nominal properties
of a defect-free weld system.
4. Measurement of fractional stiffness changes in weld-
ments
A total of thirty Grade 60 reinforcing bar specimens
fabricated by State of California Department of Transportation
(CALTRANS) were evaluated. The length of each specimen
is 122 cm (48.03 in.). An overview of the damage evaluation
methodology used here to identify the fractional stiffness
changes in the welded bar specimens is shown in Fig. 3. Modal
tests are performed on the test specimens and, from the analysis
of the measured responses, modal parameters are extracted.
In order to perform the identification of the test specimen,
modal parameters of the test specimen as well as the modal
parameters of the baseline model are needed. The baseline
modal parameters were estimated using measured frequencies
for the test specimen as well as the finite element (FE) model of
the test specimen presented in the following subsection. Once
a corresponding set of measured modal parameters for the test
specimen and baseline modal parameters was made available,
Eqs. (1) and (2) was used to estimate the fractional stiffness
change in the weldment.
4.1. Methodology to identify baseline structure
The following system identification methodology was
utilized to identify baseline modal responses of a structure [8].
Consider a linear skeletal structure with NE members and N
nodes. Suppose k ∗j is the unknown stiffness of the j th member
of the structure for which M eigenvalues are known. Also,
suppose k j is a known stiffness of the j th member of an
FE model for which the corresponding set of M eigenvalues
are known. Then, relative to the FE model, the fractional
stiffness change of the j th member of the structure, α j , and
the stiffnesses are related according to the following equation:
k ∗j = k j (1 + α j ). (9)
456 S. Park et al. / Engineering Structures 28 (2006) 453–465

Fig. 3. Schematic for estimating stiffness change in weldments.

The fractional stiffness change of NE members may be obtained Table 1

using the following equation [8]: Welded specimen designation codes

Size #8 #11 #14

Z = Fα (10)
BP1 BP1 BP1
where α is an NE × 1 matrix containing the fractional changes BP2 BP2 BP2
in stiffness between the FE model and the structure, Z is an BP3 BP3 BP3
M × 1 matrix containing the fractional changes in eigenvalues PH1 PH2 PH3
S1 S1 S1
between the two systems, and F is an M × NE stiffness Designation code
S2 S2 S2
sensitivity matrix relating the fractional changes in stiffnesses S3 S3 S3
to the fractional changes in eigenvalues. W1 W1 W1
The M × NE, F matrix can be determined as follows: first, W2 W2 W2
M eigenvalues are numerically generated from the initial FE Control Control Control
model; second, the stiffness of the first member of the FE model
is modified by a known amount; third, the corresponding set
of M eigenvalues is numerically generated for the modified
FE model; fourth, the fractional changes between the M initial
eigenvalues and M eigenvalues of the modified structure are
computed; fifth, each component of the first column of the F
matrix (i.e., the M × 1, F matrix) is computed by dividing
the fractional changes in each eigenvalue by the magnitude of
the modification at member one; and finally, the M × NE, F
matrix is generated by repeating the entire procedures for all
NE members.
Using the above rationale as a basis, the following 6-step Fig. 4. A typical welded bar specimen.
algorithm is proposed to identify a given structure.
Once the baseline model is identified, its modal parameters
1. Select a target structure (e.g., a post-damage state of the can be numerically generated (e.g., using commercial software
structure) for which sufficient frequencies that can be used [18]).
to identify the baseline structure are available. (Note that the
mode shapes of the damaged structure in defining the target 4.2. Description of welded bar specimens
structure are ignored.)
2. Select an initial FE model of the structure, utilizing all The thirty specimens consisted of three, ten-specimen
possible knowledge about the design and construction of the subsets, fabricated from #8, #11, and #14 reinforcing bars. Each
structure. ten-specimen subset included one unwelded bar that served as a
3. As outlined above, compute the sensitivity matrix of the FE control. Table 1 summarizes the coding system used to identify
model. each bar in the ten-specimen subsets. Visual inspection of the
4. As outlined above, compute the fractional changes in welded specimens indicated that each bar was originally saw-
eigenvalues between the FE model and the target structure. cut to a 122 cm (48.03 in.) length and then beveled-cut again
5. Fine-tune the FE model by first solving Eq. (10) to estimate at the center of the bar to form the groove geometry for a butt
stiffness changes (i.e., to compute the NE × 1, α matrix) and weld. The final lengths of the specimens after welding were
next solving Eq. (9) to update the stiffness parameters of the approximately 1.27 cm (0.5 in.) to 2.54 cm (1.0 in.) longer than
FE model. their original lengths. A typical welded test specimen is shown
6. Repeat steps 4–5 until Z ≈ 0 or α ≈ 0 (i.e., as they approach in Fig. 4. The backing plate attached to each specimen during
zero) when the parameters of the FE model are identified. the welding process is also noted in the figure.
The converged FE model is the baseline model. It has the All specimens were tested and evaluated with backing
frequencies of the damaged (i.e., target) structure but none of plates attached. The reinforcing bars were welded with the
its members are damaged. Furthermore, the mode shapes of SMAW process using 0.3175 cm (0.125 in.) diameter E8018-
the baseline model differ from those of the damaged structure. C3 electrodes. The joint geometry was a single V with a 45◦
 S. Park et al. / Engineering Structures 28 (2006) 453–465 457

Fig. 5. Experiment model setup with free–free boundary. Fig. 6. Instrumentation for experimental modal testing.

included angle and 0.635 cm (0.25 in.) root. Stringer beads Table 2
were applied with cleaning after each bead using chipping and Modal test parameters for the test
brushing only; no grinding was performed on the specimens.
Parameter Setting Notes/Units
Short arc lengths were maintained and generator CC welding
machines set at 125 A were used. Accelerometer channels – PCB 352A10
Impact head – PCB 208A10
4.3. Experimental determination of modal parameters of bars Analyzer – DSP Technology 20-42
Sample frequency 5000 Hz
Sample length 8192 Samples per channel
The experimental setup used in the modal testing of the Spectral resolution 1.5625 Hz
reinforcing bars is shown in Fig. 5. A fixed response-roving Number of repetitions 10 Linear average
input test method was used to collect the response data. The Channel gain Varied Set to maximize resolution
test specimen shown in Fig. 5 was suspended with two flexible Trigger method +18% hammer FS Pre-trigger save all channels
cables to simulate a free–free boundary configuration in the Accelerometer window Exponential 99% down at end
Hammer window Rectangular 10% Window width
horizontal plane. A response accelerometer was attached to the
back of the specimen slightly off center and in the horizontal
plane. The impact head/slider assembly noted in Fig. 5 was directly from the FRF data because of the low damping, widely
aligned along the horizontal centerline of the specimen and spaced frequencies, and simple geometry of the experimental
used to impact the specimen at specific locations along its test configuration. The modal parameters were obtained from
length. The custom-fabricated impact head/slider assembly was the plot of frequency response function using the peak picking
designed to maintain a horizontal impact vector perpendicular method [19]. The mode shaped for each specimen were drawn
to the specimen as the assembly was moved to each impact by measuring the magnitude and phase of frequency response
location. functions corresponding to impact location 1 through impact
Instrumentation used to conduct the modal tests consisted location 13 at specific frequencies (e.g., resonant frequencies
of a PCB 298A10 piezoelectric load cell mounted in a of the first five modes). The measured resonant frequencies
spring-loaded impact head/slider assembly, a PCB 352A10 for each specimen are presented in Table 3. A typical set of
piezoelectric response accelerometer attached to the specimen, measured mode shapes (#14 BP1 specimen) are plotted in
a 4-channel DSP Technology 20-42 Digital Signal Processor Fig. 8. In the figure, the modes represent the first through the
(DSP), and a portable computer (PC). Data acquisition software fifth bending mode in the horizontal direction.
was provided by the DSP vendor, Siglab Version V3.10 (11-
Sep-98 or later version). The DSP unit and portable computer 4.4. Determination of fractional stiffness changes in the
used in these tests are shown in Fig. 6. Instrumentation and test weldments
settings used for the modal tests are summarized in Table 2.
All of the specimens under the test had 13 equally spaced The baseline model for each specimen was constructed using
impact locations and an accelerometer response location. A the resonant frequencies extracted from the modal testing. Note
typical impact/response location configuration is shown in that the baseline structure is assumed to be damage-free with
Fig. 7. The Frequency Response Functions (FRFs) measured resonant frequencies near those of the welded specimens. To
at each impact location were derived from an average of develop such a baseline model, a finite element (FE) model
10 response measurements. Time data from the response of the welded reinforcing bar specimen was developed [18]. A
accelerometer and impact head were converted in the DSP schematic of the FE model for #14 BP1 specimen is shown in
hardware to the frequency domain and the associated frequency Fig. 9. The specimen was modeled using 779 beam elements.
response functions (FRFs) were generated. It was possible to Also, linear spring elements were used to model the cable to
extract modal parameters (i.e., frequencies and mode shapes) suspend the specimen in the free–free boundary condition. The
458 S. Park et al. / Engineering Structures 28 (2006) 453–465

Fig. 7. A typical configuration of a bar specimen (#11 BP1).

Table 3
Measured frequencies of the reinforcing bars

Size Specimen Frequencies (Hz)

Mode 1 Mode 2 Mode 3 Mode 4 Mode 5
#14 No weld 124.38 339.38 660.63 1081.3 1595.6
BP1 124.38 340.00 659.38 1082.5 1594.4
BP2 118.75 327.50 633.13 1041.3 1528.8
BP3 121.88 335.63 649.38 1068.8 1569.4
PH3 121.88 334.38 647.50 1065.0 1565.6
S1 120.63 330.63 641.88 1052.5 1551.3
S2 121.88 334.38 649.38 1064.4 1568.8
S3 122.50 335.63 651.25 1066.9 1573.1
W1 121.25 332.50 645.00 1058.8 1558.8
W2 122.50 338.75 654.38 1078.1 1580.6
#11 No weld 103.13 283.13 551.88 906.25 1341.9
BP1 100.00 274.38 532.50 879.38 1295.6
BP2 100.00 276.25 533.75 883.13 1299.4
BP3 100.63 276.88 536.88 886.25 1305.0
PH2 100.63 276.25 535.63 883.75 1302.5
S1 100.00 276.88 536.25 885.63 1304.4
S2 100.63 276.88 537.50 886.88 1308.1
S3 100.63 276.25 536.88 883.75 1306.3
W1 100.63 277.50 536.25 888.13 1311.9
W2 103.75 285.63 555.00 912.50 1349.4
#8 No weld 74.375 205.63 401.88 662.50 985.00
BP1 73.125 200.63 389.38 645.00 953.13
BP2 73.125 200.00 388.75 643.13 952.50
BP3 78.125 210.00 405.00 674.38 991.25
PH1 73.750 206.88 398.75 665.00 976.25
S1 72.500 200.63 390.00 646.88 955.63
S2 72.500 199.38 388.13 642.50 951.25
S3 73.125 200.63 390.63 645.63 956.25
W1 71.875 198.75 384.38 639.38 941.88
W2 72.500 200.63 388.13 646.25 950.63

spring elements have negligible stiffness and mass properties. Table 4

The other specimens were modeled based on the measured Initial values of material properties for FE models
length, location of the weldments, location of the sensors, Properties Value
and location of the support cables. All beam members of the
E (GPa) 210
FE models were assigned to one group, and initial material ρ (kg/m3 ) 7850
properties are summarized in Table 4.
The procedure of baseline parameter identification for the
#11 BP1 specimen is illustrated here as a typical example. the appropriate group stiffness reduced by a known amount,
With the initial estimates of material properties and with the stiffness sensitivity matrix, F, which relates changes in
 S. Park et al. / Engineering Structures 28 (2006) 453–465 459

(a) Mode 1. (b) Mode 2.

(c) Mode 3. (d) Mode 4.

(e) Mode 5.

Fig. 8. Mode shapes of #14 BP1 specimen.

Fig. 9. Schematic of FE model for #11 BP1 specimen.

Table 5
System identification for specimen #11 BP1

Mode Frequency of initial Updated frequencies (Hz) Frequency of target Error (%)
FE model Iter. 1 Iter. 2 structure Initial Final
1 105.58 99.65 99.68 100.00 5.58 0.32
2 293.53 274.67 274.68 274.38 6.98 0.11
3 562.93 532.81 533.00 532.50 5.71 0.09
4 939.16 878.85 878.88 879.38 6.80 0.06
5 1369.1 1295.0 1295.5 1295.6 5.67 0.01

element stiffness to changes in resonant frequencies, was
developed. The 6-step algorithm was utilized until the system
converged. The convergence of the system identification
scheme is demonstrated in Table 5. After two iterations, the
percentage differences in the corresponding five frequencies
of the specimen and the FE model have been reduced from,
respectively, 5.58%, 6.98%, 5.71%, 6.80% and 5.67% to
0.32%, 0.11%, 0.09%, 0.06% and 0.01%. Therefore, the
updated model with frequencies of 99.68 Hz for the first mode,
274.68 Hz for the second mode, 533.00 Hz for the third mode,
878.88 Hz for the fourth mode and 1295.5 Hz for the fifth mode
is designated as the baseline model. The corresponding material
parameter for each specimen is provided in Table 6.
Using the baseline models and the extracted mode shapes,
the fractional changes for the weldments were identified using
the following steps: (1) the damage index for the weldment
was calculated; and (2) the effective stiffness change, α, was
estimated by Eq. (2), i.e., α = 1/β − 1. The identified change
of the effective stiffness of the each weldment is summarized in
Table 7. In the table, negative values indicate that the stiffness
of a welded splice is less than the effective stiffness of the
corresponding unwelded specimen. Note that the fractional
460 S. Park et al. / Engineering Structures 28 (2006) 453–465

Table 6
Identification average stiffness parameter (EI) for baseline models

#14 specimen Average stiffness EI (N m2 ) #11 specimen Average stiffness EI (N m2 ) #8 specimen Average stiffness EI (N m2 )
BP1 30 701 BP1 14 627 BP1 3762
BP2 30 667 BP2 14 487 BP2 3780
BP3 30 693 BP3 14 573 BP3 4150
PH3 30 851 PH2 14 429 PH1 4098
S1 30 377 S1 14 633 S1 3768
S2 31 218 S2 14 432 S2 3762
S3 31 427 S3 14 573 S3 3765
W1 30 543 W1 14 642 W1 3751
W2 31 003 W2 14 625 W2 3768

Table 7
Summary of change of the bending stiffness

#14 α #11 α #8 α
specimen specimen specimen
BP1 0.075 BP1 0.228 BP1 0.025
BP2 0.018 BP2 −0.045 BP2 0.080
BP3 0.015 BP3 −0.025 BP3 −0.140
PH3 0.105 PH2 0.053 PH1 −0.050
S1 0.028 S1 0.021 S1 0.050
S2 0.021 S2 0.055 S2 −0.070
S3 0.178 S3 0.079 S3 −0.070
W1 0.092 W1 −0.081 W1 −0.040
W2 −0.056 W2 −0.027 W2 −0.020

The welds for the reinforcing bar were apparently welded

Fig. 10. Radiographic images for a typical specimen (#11 BP2).
change in effective stiffness is based on the baseline stiffness
value of each specimen presented in Table 6.
5. NDT evaluation of welded bar specimens
The welded reinforcing bar specimens were subjected to an
independent NDT process. The objective of this evaluation was
to provide an independent review of specimen weld quality
using a traditional NDT technique—radiographic inspection.
Prior to shipment, CALTRANS collected radiographic images
of each specimen weld. Images of specimen welds were
provided from two perpendicular viewing directions (Fig. 10).
The quality of the weldments was determined using the criteria
provided in ANSI/AWS D12.1 [20]. Section 4 of the code
describes the workmanship requirements for the direct butt
splices. Reinforcement of the weld is limited to 0.3175 cm
(0.125 in.); no cracks in the weld metal of the heat affected
zone (HAZ) are permitted and complete fusion across the entire
cross-section is required. No overlap is permitted, undercutting
is very limited, and porosity or any discontinuity related to
fusion is limited both in single size and total dimensions.
Using the code as a reference, the welds were inspected and
discontinuities were noted. Welds were accepted or rejected
based on the requirements as summarized above.
with an excessive heat input. The welds were characterized by
underfill and overlaps visible on the side of the welds at the
ends of the weld beads. The backing bars and their associated
tack welds were still in place on most of the specimens. All
of these tack welds had excessive reinforcement and large-
diameter cluster porosity, as well as some lack of fusion to the
base material. All the welds had the same types of discontinuity,
with some samples being worse than the others. The results
of the interpretation of the radiographic images are provided
below for each specimen.
BP1 (#8): Some porosity is visible at the fusion line to base
metal. Shading of radiograph indicates variations in the weld
cross-section dimension. The size of porosity falls within the
acceptance criteria. The weld is accepted.
BP2 (#8): Lack of fusion or very connected linear porosity is
evident on one sidewall of the joint at the fusion line to the base
metal. The weld is rejected due to lack of fusion.
BP3 (#8): Significant underfill can be seen on the face of the
weld. There is lack of fusion at sidewalls of joint. Some porosity
can be seen in the radiographs. The weld is rejected due to the
insufficient throat.
PH1 (#8): Large-diameter porosity is scattered throughout the
weld. A portion of the porosity may be characterized as linear
porosity at the fusion lines between the bead layers. The weld
is rejected as a result of the summed size of the porosities (it
must be less than 4.76 mm (0.1875 in.) for this dimension of
reinforcing bar).
S1 (#8): Many large-diameter porosities are present in the weld.
Linear porosity can be seen at the fusion line of the tack weld
 S. Park et al. / Engineering Structures 28 (2006) 453–465
to the base material and the weld. The weld is rejected because
of the excessive porosity.
S2 (#8): Large-diameter porosity is observed in the body of the
weld and is located primarily at the fusion line. There is also
scattered porosity in the weld. The weld is rejected as a result
of the summed size of the porosities.
S3 (#8): Large-diameter and pinhole scatted porosities are
found throughout weld. The total amount of porosity requires
that the weld be rejected.
W1 (#8): Lack of fusion exists under the top fill pass. The weld
is rejected as a result of this lack of fusion.
W2 (#8): Cracks are at both toes of the weld and may extend to
the HAZ. The sides of the weld are very irregular. Since cracks
are not permitted, the weld is rejected.
BP1 (#11): The weld appears to be of good quality, with very
small internal porosity but heavy surface irregularity. Since the
size of the porosity is under the single size and the summed size
limits, the weld is accepted.
BP2 (#11): A small crack or lack of fusion, probably the latter,
extends from the toe of weld along the fusion line. Isolated
instances of large-diameter porosity appear near the top surface.
No other discontinuities are observed. The weld is rejected due
to the possible existence of the crack.
BP3 (#11): A small area underfill is located on one side of
the weld. The weld has fairly good internal quality, but has an
irregular surface area. Since the underfill is an insufficient cause
for rejection, the weld is accepted.
PH2 (#11): Scattered porosity is distributed evenly across the
weld cross-section. The last weld at the face has heavy large-
diameter porosity. Lack of fusion appears at the fusion line to
the base metal. The weld is rejected on the basis of the size of
the porosity and the lack of fusion.
S1 (#11): One large-diameter pore is located near the centerline
of the weld. There is also a lack of fusion between beads. The
weld is rejected on the basis of these observations.
S2 (#11): Porosity and lack of fusion are evident on the
bond line. Linear porosity is present across the weld. Some
large-diameter porosity and lack of fusion appears in the root
area. The weld is rejected as a result of the total size of
discontinuities.
S3 (#11): Large-diameter porosities are present throughout the
weld, especially in the top third of the weld. Lack of bond line
fusion occurs in several locations especially at the root and in
the center of the weld. Porosity is also concentrated at the bond
lines between passes. On the face of the weld, porosity and
lack of fusion can also be observed. Because of these numerous
discontinuities, the weld is rejected.
W1 (#11): A crack is located in the HAZ. Large-diameter
porosities are observed at the center near the face of the weld.
Lack of bond line fusion or probably underfill may be at the
root of the weld. In addition, the original edges of the joints are
distinct. Finally, large-diameter porosities are observed at the
bond line on same side as the crack. The weld is rejected.
W2 (#11): Wormhole porosity is located at the face of the weld.
Some lack of fusion appears between the weld beads. Large-
diameter porosities are observed at the bond line near the face
of the weld. The surface of the weld is extremely irregular with
461
underfill. One side appears almost to have a crack extending
down into the weld metal. The size of the porosity leads to the
rejection of the weld.
BP1 (#14): Lack of bond line fusion is observed on one side of
the weld joint. One large-diameter pore is evident in the weld.
The weld is rejected as a result of lack of fusion.
BP2 (#14): No discontinuities in the weld are visible except for
the lack of fusion at the backing bar tack welds which is not in
the weld joint. Although the weld is very irregular in density, it
is accepted.
BP3 (#14): Some lack of bond line fusion is visible near the
bottom third of the weld. Also, the surface of the weld is very
irregular. The weld is rejected due to lack of fusion.
PH3 (#14): Cluster porosity is seen near the center of the weld
cross-section. Small surface irregularities appear at the face.
The weld is acceptable.
S1 (#14): Many large-diameter cluster porosities appear in the
last welds in the joint face. There is lack of fusion near the root
that may be in the backing weld. The weld is rejected because
of the summed porosities and the size of the porosities.
S2 (#14): Large-diameter porosities are scattered throughout
the entire weld cross-section. The weld is rejected.
S3 (#14): Large-diameter porosities are scattered throughout
the weld. The weld is rejected due to the quantity and size of
the porosities of the weld.
W1 (#14): There is a lack of fusion to the base material near the
face of the weld. The lack of fusion (underbead) also appears
between passes near the top third of the weld. The weld is
rejected as a result of the lack of fusion.
W2 (#14): There is a lack of fusion to the base material
at the face of the weld. In an overlap, there exists a large-
diameter porosity. The bottom two-thirds of the weld cross-
section exhibits underfill. The weld is rejected as a result of
the lack of fusion.
6. Destructive test of field prepared reinforcing bar
Destructive tensile tests were performed on the thirty welded
reinforcing bar specimens. Yield strength results from these
specimens were used in combination with the predicted yield
strengths to evaluate the accuracy of the models proposed
in Section 3. The destructive tensile tests were conducted
in accordance with ASTM Procedures and Standards A370
and A615. The reported data included yield strength, tensile
strength, breaking strength, load–displacement plots, and
percentage elongation. A 1780 KN Baldwin–Satec testing
machine was used for all tests (Fig. 11). The distance between
the upper and lower grips was 40 cm, with the weld centered
between grips.
All welded specimens failed in or very near the welded
connection. Unwelded control specimens failed near one of the
grips. Several specimens exhibited significant bending at the
welded joint. Specimen failures generally occurred in the weld
perpendicular to the longitudinal axis of the specimen. Fig. 12
presents the end and side view of the #11 control specimen
failure surface. Necking of the specimen is apparent in the side
view of the specimen near the failure surface. Fig. 13 presents
462 S. Park et al. / Engineering Structures 28 (2006) 453–465
Fig. 11. A tensile test machine.
a similar end and side view of the #11 BP2 specimen. The
specimen exhibited a perpendicular failure through the weld
(i.e., the typical failure mode noted above).
7. Comparison between measured and predicted weld
capacity
7.1. Computing the predicted weld capacity
The equation to predict the yield strength, σ Dy , is given by
Eq. (8). The values of the stiffness changes, α, are provided
in Table 7. Note that the stiffness changes are based on the
estimate using the first, third, and fifth bending modes so that
the weld part avoids node points at which numerical errors often
occur. The value of σ y used in this exercise is 460 MPa, since
E8018-C3 electrodes were used in the welding process [1].
The system being evaluated in this study consists of the bar
comprising the base metal, the weld itself, and the interface
between the weld and the base metal. When subjected to an
axial load, the system will fail when any one of the elements
(i.e., the bar, weld, or interface) comprising the system fails.
Since the system described here is a weakest link (series)
system, the predicted strength of the system is the strength of
the weakest element. Note that, from the destructive test, the
yield strength for unwelded #8, #11, and #14 control bars were
453 MPa, 492 MPa and 506 Mpa, respectively. In summary, the
predicted yield strength of the bar–weld system is given by:
σ y(system) = Min{σ y(bar) , σ y(weld) , σ(interface) }.
Because we had no information on the strength properties of
the weld–base metal interface, it was assumed that the yield
strength of the interface was equal to the yield strength of the
filler material.
7.2. Results of the comparison between the predicted and
measured weld capacities
The measured weld capacities from destructive tests and the
predicted weld capacities from the fractional stiffness changes
are listed in Table 8. Note that, in order to predict the yield
strengths in the table, the nominal yield strength of the E8018-
C3 electrode was utilized. To provide an indication of the
Fig. 12. End and side view of failed #11 control specimen.
accuracy of the prediction of the yield strength, the deviation
between the measured yield strength and the predicted yield
strength is provided in the last two columns of the table. Note
also that the mean and standard deviation of the percentage
error between the predicted and the measured values is provided
for the entire test sample. Cases in which the failure of the
weld did not control the failure of the system are indicated in
the table. The results in Table 8 are presented graphically in
Fig. 14. Note that a line of unit slope has been included in the
figure, to aid in the interpretation of the accuracy of the results.
In Table 8, it can be seen that the mean deviation is 4.51
(MPa) with the standard deviation of 29.13 (MPa). It is also
observed that the mean deviation percentage error of the
predictions for all specimens is 5.05% with a standard deviation
of 4.40%. While unacceptable deviation errors are observed for
some specimens such as BP1(#11), S1(#11), and BP3(#8), most
deviation errors are within 10%. This observation can also be
seen in Fig. 14 which indicates that the predicted and measured
yield strength tend to cluster around the line of unit slope.
The predicted and measured performances of the weldments
are presented in Table 9. In the table, the performance of the
weldments were determined on the basis of the smaller value
between the yield strength of the unwelded control bars and
the nominal yield strength of the E8018-C3 electrode. Thus
the threshold value for the quality of weldments are 453 MPa,
460 MPa, and 460 Mpa for #8, #11, and #14 bars, respectively.
Assuming that no defect is present in the base metal, specimens
with the (predicted or measured) yield strength below the
threshold values were classified as bad quality. In the table,
compared to the results of the destructive tests, the vibration-
based method correctly predicted the performance of 21
(11) outcomes of the reinforcing bar specimens of 27 specimens
(78%) while the radiographic inspection technique predicted
the correct performance of 14 specimens (52%).
7.3. Areas for potential improvement for field implementation
of the methodology
Even though the methodology proposed here to predict the
capacity of a weld–bar system is quite good, there are still
many areas for potential improvement in the accuracy of the
predictions. Three obvious areas include the following: (1)
improving the strength–damage models, (2) accounting for
uncertainties in the parameters used to predict the strength, and
(3) reducing errors associated with experimental measurements
and stiffness change determination.
 S. Park et al. / Engineering Structures 28 (2006) 453–465 463

Fig. 13. End and side view of failed #11 BP2 specimen.

Table 8
Comparison of predicted and measured yield strength

Specimen Measured yield strength (MPa) Predicted yield strength (MPa) Deviation Error (%)
#14 BP1 508.1 494.5 −13.60 2.68
BP2 506.3 468.3 −38.00 7.51
BP3 476.6 466.9 −9.70 2.04
PH3 479.1 506.0a 26.90 5.61
S1 508.3 472.9 −35.40 6.96
S2 478.9 469.7 −9.20 1.92
S3 479.8 506.0a 26.20 5.46
W1 505.4 502.3 −3.10 0.61
W2 476.9 434.2 −42.70 8.95
#11 BP1 415.5 492.0a 76.50 18.41
BP2 424.2 439.3 15.10 3.56
BP3 410.9 448.5 37.60 9.15
PH2 484.0 484.4 0.40 0.08
S1 412.7 469.7 57.00 13.81
S2 486.7 485.3 −1.40 0.29
S3 491.3 492.0a 0.70 0.14
W1 396.0 422.7 26.70 6.74
W2 412.0 447.6 35.60 8.64
#8 BP1 440.8 453.0a 12.20 2.77
BP2 443.2 453.0a 9.80 2.21
BP3 441.9 395.6 −46.30 10.48
PH1 444.9 437.0 −7.90 1.78
S1 442.1 453.0a 10.90 2.47
S2 446.3 427.8 −18.50 4.15
S3 443.3 427.8 −15.50 3.50
W1 432.7 441.6 8.90 2.06
W2 432.1 450.8 18.70 4.33
Standard deviation 29.13 4.40
Mean 4.51 5.05
a Bar controls.

The strength–damage model for weld systems can be
improved or refined in several ways. First, the model may
be extended to include the strength properties of the interface
and/or changes in the properties of the base metal in the HAZ.
Second, the impact of the incremental damage incurred during
loading on the strength may be considered. Finally, the irregular
geometry in the weld zone and the impact of the mass due to the
weld and the backing can be included in the analysis.
The equations used in this study assumed that all of the
parameters, such as yield strengths, tensile strengths, and
nominal areas, are deterministic. In reality these variables
are random variables. Thus, the model can be significantly
improved by treating the parameters as random variables. To
accomplish this task, statistical descriptions of the parameters
of interest must be developed. For scientific reasons, the
dynamic testing portion of this study involved a weld
reinforcing bar system that had precisely defined boundary
conditions; namely, free–free boundary conditions. Such ideal
conditions are unlikely to be encountered in the field.
Measurement of modal parameters must be accomplished
for systems with more complicated boundary conditions and
system identification procedures must be extended to account
for the variable boundary conditions. The system identification
technique used in this work is capable of identifying such
464 S. Park et al. / Engineering Structures 28 (2006) 453–465

strength of a weld. The inputs to the method were the yield

strength of the weld filler material and a nondestructive measure
of the fractional stiffness change in the weld region relative to
an undamaged reinforcing bar. The output of the method is a
prediction of the resisting capacity, i.e., the strength expressed
in units of stress, of the weld system. If a weld system is
evaluated using the method presented here, the information
passed on to the designer is an estimate of the resistance of
the system in terms of its yield strength. The measures of
accuracy utilized in this study are the magnitude of deviation
and the percentage deviation of the predicted capacity from the
measured capacity. As noted above, here capacity is measured
in terms of yield strength of the weld system.
On the basis of the results presented in this paper, the
following conclusions ensue from this study:
Fig. 14. Correlation between predicted and measured yield strength.
1. the performance, measured by the yield strength, of a welded
Table 9 joint can be related to the fractional change in stiffness at the
Comparison of predicted and measured quality of weldments (O: Good, ×: joint location due to the weld;
Bad) 2. fractional stiffness changes at the location of a weld can be
Specimen Measured Predicted routinely measured using vibration-based methods;
X-radiography Vibration-based method 3. a good agreement is observed between the predicted
#14 BP1 O × O
strengths of welded reinforcing systems and the correspond-
BP2 O O O ing strengths measured in the laboratory; and
BP3 O × O 4. the vibration-based method yields more accurate predictions
PH3 O O O of the actual condition of the welds (78%) as compared to
S1 O × O the qualitative traditional radiographic inspection technique
S2 O × O
S3 O × O
(52%).
W1 O × O
W2 O × × Acknowledgement
#11 BP1 × O O
BP2 × × × Funding for this study was provided by State of California
BP3 × O × Department of Transportation (CALTRANS).
PH2 O × O
S1 × × O References
S2 O × O
S3 O × O [1] ANSI/AWS. Structural welding code — reinforcing steel. 4th ed.
W1 × × × American Welding Society; 1992.
W2 × × × [2] Chang C, Chen I, Shee H, Chen H. X-ray inspection reliability for welded
#8 BP1 × O O joints. In: Proceedings of the sixth international conference on structural
BP2 × × O safety and reliability, 1993. p. 991–6.
BP3 × × × [3] Askeland DR. The science and engineering of materials. Boston: PWS
PH1 × × × Publishing Company; 1994.
S1 × × O [4] Doebling SW, Farrar CR, Prime MB, Shevitz DW. Damage identification
S2 × × × and health monitoring of structural and mechanical systems from changes
S3 × × × in their vibrational characteristics: a literature review. Los Alamos
W1 × × × National Laboratory: LA-13070-MS; 1996.
W2 × × × [5] Rytter A. Vibrational based inspection of civil engineering structures.
Ph.D. thesis. Aalborg (Denmark): University of Aalborg; 1993.
Number of 14 21 [6] Stubbs N, Kim JT, Topole KG. An efficient and robust algorithm for
correct damage localization in offshore platforms. In: ASCE 10th structures
predictions congress ’92. 1992, p. 543–6.
[7] Kim JT, Stubbs N. Damage detection in offshore jacket structures from
limited modal information. International Journal of Offshore and Polar
variable boundary conditions. The capability to deal with more Engineering 1995;5(1):58–66.
complicated boundary conditions is the key to extending the [8] Stubbs N, Park S, Sikorsky C, Choi S. A global nondestructive damage
technique to the field. assessment methodology for civil engineering structures. International
Journal of Systems Science 2000;31(11):1361–73.
8. Conclusions [9] Choi S, Stubbs N. Damage identification in structures using time-domain
response. Journal of Sound and Vibration 2004;275(3–5):577–90.
[10] Choi S, Park S, Yoon S, Stubbs N. Nondestructive damage identification
The approach proposed here attempted to provide the in plate structures using changes in modal compliance. NDT&E
nondestructive evaluation method to directly evaluate the International 2005;38(7):529–40.
 S. Park et al. / Engineering Structures 28 (2006) 453–465
[11] Kim JT, Stubbs N. Model-uncertainty impact and damage-detection
accuracy in plate girder. Journal of Structural Engineering 1995;121(10):
1409–17.
[12] Park S, Stubbs N, Bolton R, Choi S, Sikorsky C. Field verification of
the damage index method in a concrete box-girder bridge via visual
inspection. Computer-Aided Civil and Infrastructure Engineering 2001;
16:58–70.
[13] Farrar CR, Jauregui D. Damage detection algorithms applied to
experimental and numerical modal data from the I-40 bridge. Los Alamos
National Laboratory, LA-13074-MS; 1996.
[14] Cawley P, Adams RD. The location of defects in structures from
measurements of natural frequencies. Journal of Strain Analysis 1979;
14(2):49–57.
465
[15] Choi S, Park S, Stubbs N. Nondestructive damage detection in structures
using changes in compliance. International Journal of Solids and
Structures 2005;42(15):4494–513.
[16] Lemaitre J, Chaboche J-L. Mechanics of solid materials. USA: Cambridge
University Press; 1990.
[17] Kachanov LM. Introduction to continuum damage mechanics. Dordrecht
(The Netherlands): Kluwer Academic Publishers; 1986.
[18] ABAQUS. Version 5.4 user’s manual. Providence: Hibbitt, Karlsson &
Sorensen Inc.; 1994.
[19] Ewins DJ. Modal testing: theory and practice. London (England):
Research Studies Press; 1984.
[20] ANSI/AWS D12. 1. Reinforcing steel welding code. American Welding
Society; 1992.