CHAPTER 4 SETUP FOR THE EXPERIMENTAL STUDY
4.3 Sensing Instrumentations
Three types of sensors, including accelerometers, electrical resistance strain gages (RSG), and optic fiber Bragg grating (FBG) sensors, were installed on the specimen to measure the structural responses during the shaking table tests. The basic information, such as the specifications and arrangements of the sensors, about the sensing instrumentations is briefly introduced in the followings.
Five accelerometers were designed to monitor the acceleration responses of the test specimen when subjected to the simulated earthquakes. Figure 4.6 shows the deployment of the accelerometers. According to the figure, four accelerometers were placed along the central line of the frame at each floor to measure the structural responses, and one accelerometer was placed at the foot of the column to measure the input base excitation to the structure. Figure 4.7 and 4.8 show the actual installations of the accelerometers at the 2nd floor and base, respectively. A simple description of the employed accelerometers is listed in Table 4.4 in which the symbols A1 to A4 represent the accelerometers at the 1st to 4th floor, respectively, and the symbol Abase represents the accelerometer at the base.
Fiber Bragg grating sensors are one of the most exciting developments in the field of optical fiber sensors in recent years. Since the pioneering work done by Meltz et al. [21], subsequent interest in FBG sensors has increased considerably. One of the probable main reasons for this is that, FBG sensors have great potential for a wide range of sensing applications for important
physical quantities, such as strain, temperature, pressure, acceleration [22-24], etc. It this work, the FBG sensors are used for measuring the strain responses of the structure during earthquakes.
FBG sensors have a number of distinguishing advantages which make them a promising candidate for smart structures. When compared with RSG used for strain monitoring, FBG sensor have several distinguishing advantages, including [25]
(1) much less intrusive mass and size;
(2) much better immunity to electro-magnetic interference;
(3) greater capacity of multiplexing a large number of sensors along a single fiber link, unlike RSGs which need a huge amount of wiring;
(4) greater resistance to corrosion when used in open structures, such as bridges and dams;
(5) higher temperature capacity (typically about 300oC);
(6) longer lifetime for long term operation.
These features have made FBG sensor very attractive for health monitoring of smart structures.
Sensing principle of FBG sensor
An FBG is written into a segment of Ge-doped single mode fiber in which a periodic modulation of the core refractive index is formed by exposure to a spatial pattern of ultraviolet light in the region of 244-248 nm. The lengths of FBG sensors are normally within the region of 1-20mm and grating reflectivities can approach ~100%. Being a recent developed technique for civil engineering, the sensing principle of FBG sensor is briefly introduced herein. When the FBG is illuminated by a broadband light source, each FBG sensor in a fiber reflects a specific wavelength that shifts slightly depending on the strain applied to the sensor. The change in wavelength is directly proportional to the change in mechanical features such as strain or temperature.
In its simplest form a fiber Bragg grating consists of a periodic modulation of the refractive index in the core of a single-mode optical fiber (Figure 4.9). According to Bragg’s law, reflected Bragg wavelength, λB, is given by
Λ
=2neff
λB (4.1) where neff represents the effective refractive index of the fiber core and Λ is the period of the index modulation.
The Bragg wavelength is the free space center wavelength of the input light that will be back-reflected from the Bragg grating. The Bragg grating resonance, which is the center wavelength of reflected light from a Bragg grating, depends on the effective index of refraction of the core and the periodicity of the grating. The effective index of refraction, as well as the periodic spacing between the grating planes, will be affected by changes in strain and temperature.
While the strain effect, which corresponds to a change in the grating spacing and the strain-optic induced change in the refractive index, is considered (Figure 4.10), the Bragg wavelength change can be expressed as
∆Λ
=
∆λB 2neff* (4.2)
where n*eff is the changed effective refractive index. Hence the wavelength shift,
B
where κε is a strain-sensitive coefficient. Furthermore, the Bragg wavelength shift due to the temperature effect may be written as
T expansion coefficient; and κT represents the thermal-sensitive coefficient. Combining equations
(4.3) with (4.4), the Bragg wavelength shift due to strain and thermal effects can be expressed as
Further details about the FBG sensor technology and its applications can be found in related books or review articles, such as Othonos and Kalli [26], Kashyap [27], and Rao [25, 28].
According to the experiments the strain-sensitive and thermal-sensitive coefficients provided by the Prime Optical Fiber Corporation (POFC, Hsinchu, Taiwan) are about 0.80 and 5.88×10-6, respectively, which make equation (4.5) becomes
T
Then the strain or temperature variation can be converted from the wavelength change by the following relationship.
C pm 0.8 strain 0.1o
1 ≈ µ ≈ (4.7)
FBG Data Acquisition System
An FBG data acquisition system, including the MOI’s (Micron Optics, Inc.) Fiber Bragg Grating Swept Laser Interrogator (FBG-SLI) and a notebook (Figure 4.11), is adopted to monitor and restore the FBG wavelength data.
The FBG-SLI is a high-power, fast, multi-sensor measurement system that provides a major advancement for mechanical sensing applications. The FBG-SLI combines the speed of MOI’s unique Swept Laser technology and the accuracy of the patented picoWave reference technique to resolve changes in optical wavelengths of approximately 1pm (<1µ strain) and achieve high calibrated wavelength accuracy. It is a complete system that includes a swept source used to illuminate the FBG sensors and the four detectors, which simultaneously measure the reflected optical signals on each fiber. All sensors (maximum of 64 FBG sensors per fiber) on all channels are scanned simultaneously at a maximum rate of 108Hz. Table 4.5 lists the specification of the
A block diagram of the optical layout is shown in Figure 4.12. The swept laser illuminates the Bragg gratings and each FBG sensor reflects its corresponding wavelength. The Fiber Fabry-Perot Tunable Filter (FFP-TF) simultaneously scans the reflected wavelengths from the FBG sensors and the picoWave reference. Through the detector circuitry and software, the detected signals are converted to wavelengths. A PC or notebook provides the on-line calibration, data display/storage, and the FBG sensors under test.
FBG sensors arrangement
There are 12 FBG sensors along two fibers were employed in this work to monitor the strain responses of the test frame during excitations. Eight FBG sensors were arranged along one fiber link which was connected to the first channel (terms as Channel 1) of the FBG-SLI, and four FBG sensors were arranged along another fiber link which was connected to the second channel (terms as Channel 2) of the FBG-SLI. The descriptions of the specification and arrangement of the FBG sensors along Channel 1 and Channel 2 are shown in Tables 4.6 and 4.7 and Figure 4.13, respectively. Figures 4.14 and 4.15 display the transmission and reflection spectra of the FBG sensors on Channel 1 and Channel 2, respectively.
The FBG sensors were attached to the columns at each story to measure the strain responses of the test frame. The FBG sensors along Channel 1 were located near the top and the bottom of the story columns on east side; meanwhile, the FBG sensors along Channel 2 were located near the bottom of the story columns on west side. A sketch of the deployment of the FBG sensors is also demonstrated in Figure 4.5. The actual attachments of FBG1 and FBG2 on Channel 1 and FBG9 on Channel 2 are shown in Figure 4.16, Figure 4.17, and Figure 4.18, respectively. Note that, FBG1 and RSG1 in Figure 4.16 were installed near the bottom of the column of 1st story on east side, FBG2 in Figure 4.17 was installed near the top of the column of 1st story on east side, and FBG9 in Figure 4.18 was installed near the bottom of the column of 1st story on west side.
In addition to the FBG sensors, four RSGs were also adopted as a reference to the FBG sensors. Therefore, one RSG was attached right beside the FBG sensor to the bottom of the
column of each story, as shown in Figure 4.16. The RSGs configuration was also depicted in Figure 4.6.
4.4 Damage Simulation
There are several kinds of mechanisms for damage simulation according to the study objectives of interest. By reviewing the numerical or experimental studies in damage detection or assessment during these years, the simulations of structural damage are classified into the following categories.
(1) For beams-like or bridge structures:
─ decreasing the stiffness of the elements numerically [29];
─ reducing the thickness or cross-section of the selected elements [4, 14, 30-34];
─ support failure and/or crack degradation [2, 11, 35, 36].
(2) For truss structures:
─ reducing the cross-section or Young’s modulus of the bars to simulate the axial stiffness failures [31, 37, 38];
─ loss of stiffness and mass of members [39].
(3) For building or frame structures:
─ loosing the beam-column joints to simulate joints failures [40, 41];
─ weakening the story stiffness via the reduction in bracing areas [19];
─ reducing the flexural stiffness of the beams belonging to the corresponding floors [42].
4.4.1 Strengthening Column
The damage in a structure is assumed in this work to be the change in story stiffness. In most experimental studies of damage detection which is based on such an assumption commonly used
bracing elements as simulations [19]. This work, however, employs a different type of mechanism, called the ‘strengthening column (SC for short)’, as simulations to the structural damage.
Most of the past experimental studies focus on high-level damage extent, and most of the successful diagnoses were usually based on the high-level damage. Small extent damage is not as easy and evident as large extent damage to be identified and assessed because of many uncertainties and errors such as boundary conditions, measurement errors, ambient noise, and computation errors. Even though, this work attempts to investigate that if the small extent (or low-level) damage as well as the large extent (high-level) damage can be successfully identified and diagnosed.
Compare with the experimental specimens in the works conducted by Elkordy [19] and Koh [7], the specimen in this work is heavier and bigger. Furthermore, according to the analysis result from ETABS software, even small cross-section of bracing can provide significant lateral stiffness which results in considerable changes in the natural frequencies of the structure. In order to investigate whether small damage in a structure can be detected and assessed, simulation on small damage scenario is also considered in this work. Consequently, instead of using the bracing, the SC is designed to provide the specimen with additional lateral stiffness. The element selected to perform as SC is the lightweight ‘C’ shape steel. Table 4.8 and Figure 4.19 show the detailed dimension of the SC and how the SC is connected to the beams, respectively. Note that, the symbols ‘SC-A’ and ‘SC-B’ in Table 4.8 denote the SCs whose cross sections are C100×50×20×2.3 and C75×45×15×2.3, respectively. The SC was fixed with 4 bolts at each end to the top and bottom of the story beams. The actual installation of the SC at the 1st story is depicted in Figure 4.20.
The clear frame that combines with 6 SCs-A at the 1st to 3rd stories (2 SCs at each floor from 1st to 3rd stories) is defined as the ‘intact (or healthy) structure’. Figure 4.21 shows a photo of the intact structure. Alternatively, the frame that incorporated with smaller SCs (to simulate a slight damage scenario) or without SC (to simulate a considerable damage scenario) is treated as
‘damaged structure’.
4.4.2 Simulated Damage Cases
Herein, both single-site damage cases and multiple-site damage cases are simulated. Table 5.9 lists all the damage cases in studying. For convenience and simplicity, certain notations to the damage cases are assigned. Refer to Table 4.9, the notation AAA represents an undamaged case in which the test frame was installed with 6 SCs-A at the 1st to 3rd stories (2 SCs at each floor). The rest cases which are pre-noted with ‘Dcase’ represent the damage cases. Furthermore, the symbols ‘A’ , ‘B’, and ‘N’ represent SC-A, SC-B, and without SC, respectively. For example,
‘A-B-N’ of case Dcase_ABN means that the frame was installed with the SCs-A at the 1st story, the SCs-B at the 2nd story, and without SC at the 3rd story. It is seen from Table 4.9 that, Dcase_BAA and Dcase_NAA are damage cases in which the damage was induced by reduction in the story stiffness at the 1st story. Hence, the damage class for these cases is denoted by Dclass_k1. Similarly, if the SCs at the 1st and 2nd story were removed from the intact structure, this damage case and its corresponding damage class are denoted by Dcase_NNA and Dclass_k1&k2, respectively.
4.5 Experimental Scheme
Starting from the shaking table test of the intact structure, the simulated damage events, listed in Table 4.9, are then in turn implemented on the shaking table. The base excitation that inputs to the shaking table is the Kobe earthquake whose intensity is reduced to the level of PGA 0.08g. The sampling rate of the acceleration and RSG records is 200Hz. Table 4.10 shows the operation sequence of the shaking table tests and how the response records are denoted for simplicity.
4.6 Pre-Analysis Of The Measured Data
Some statistical properties about the acceleration measurements are listed in Table 4.11.
(1) Because the input excitations for each shaking table test are the same, the influence of the induced damage to the test frame can be directly discussed basing on the structural response.
(2) The effect of replacing the smaller SCs (SC-B in Table 4.8), according to Table 4.11, is much lesser than that of removing the SCs. For example, the measurements of the damage case Dcase_BAA are slightly higher than that of the intact case AAA (the
max(acc.)= 0.001, 0.010, 0.009, 0.023); in contrast, the measurements of Dcase_NAA increase significantly (the max(acc.)= 0.012, 0.025, 0.039, 0.014).
Similar situations also happen to other damage cases, such as Dcase_ABA and Dcase_ANA, Dcase_AAB and Dcase_AAN, etc. This phenomenon is welcome because it meets the requirement for studying both the low-level and high-level damage in the structure.
(3) Due to the insignificance of replacing the SC-A with SC-B to the structure, the measurement discrepancy between each other is small (i.e. the responses are similar to each other), especially when the SCs were replaced from only one story. For instance, compare the case AAA with Dcase_AAB whose SCs at the 3rd story were replaced with SCs-B, the relative increments of the maximum response for each floor are only -3.9%, 6.0%, 4.9%, and 0.8%. Similar situations also happen to Dcase_ANA and Dcase_BNA, Dcase_NAA and Dcase_NBA, and Dcase_AAN and Dcase_ABN, etc.
(4) Following with the finding 3, this phenomenon could possibly lead to similar identification results of modal parameters.
As mentioned previously, 12 FBG sensors were configured along two fiber links, Channel 1 and Channel 2. These 12 FBG sensors are further classified, according to their locations, into three groups: the sensors which located near the bottom of the story columns on east side (i.e.
FBG1, FBG3, FBG5, and FBG7) will be shortly called ‘sensors at BE’; the sensors which located near the top of the story columns on east side (i.e. FBG2, FBG4, FBG6, and FBG8) will be
shortly called ‘sensors at TE’; and the sensors which located near the bottom of the story columns on west side (i.e. FBG9 to FBG12) will be shortly called ‘sensors at BW’. Statistical results of the FBG sensors’ records, based on the three groups, are summarized in Tables 4.12 to 4.14, respectively. Meanwhile, statistical summaries of the RSGs’ records are shown in Table 4.15.
Based on the summarized tables of the acceleration and strain measurements (Tables 4.11 to 4.15), certain findings are presented below.
(1) Compare with the acceleration responses, the strain row data (from either FBG sensors or RSGs) shows more sensitivity to the system changes. The changes in maximum measurement of the acceleration responses (column IV, Table 4.11) are smaller than that of the strain responses (column IV, Tables 4.12 to 4.15). Take the case Dcase_NNN for example, the maximum relative increments in acceleration and strain responses are 68.5% (4F) and 122.3% (FBG5), respectively.
(2) One of the major advantages of the FBG sensors is that they have better immunity to EM interference. Therefore, the signals of the FBG sensors is lesser noise-corrupted.
This situation can be validated from two aspects. Firstly, according to the comparison between the FBG sensors’ and RSGs’ records, a number of disturbances containing in the RSGs records. Secondly, the ratio, (
) max(
) ( std
ε
ε ), between the standard deviation of
the response (std(ε )) and the maximum response (max(ε )) for the RSGs’ records (Table 5.15) is larger than that for the FBG sensors’ records (Table 4.12).
(3) As mentioned before, one of the functions of the RSGs is to be a reference to the FBG sensors.
(4) The FBG sensors at BE were placed at the opposite position of the story columns to the FBG sensors at TE. Therefore, the measurements they obtained should be with similar magnitude but negative phase. Take the case AAA once again for example, the correlation coefficients between the measurements of the FBG sensors at BE and the measurements of the FBG sensors at TE are {-0.9993,-0.9998,-0.9998,-0.9995}
which mean that these two sets of data are highly correlated in negative phase.
(5) Four FBG sensors (from FBG9 to FBG12) were placed at BW to have parallel location with the FBG sensors at BE so as to check torsional effect of the specimen. The correlation coefficients between the data of these two figures are
0.9985}
0.9995, 0.9989,
{0.9997, which mean that these two sets of data are highly correlated.
C HAPTER 5
H EALTH M ONITORING O N T HE T EST F RAME
5.1 Introduction
In Chapters 3 and 4, neural network-based system identification methods and a two-stage damage assessment approach were proposed and examined by either numerical examples or laboratory measurements. The examined results have preliminarily shown their capabilities of dealing with the associated problems. By conducting a series of shaking table tests for the health monitoring study, the proposed methods and approach are further investigated by the experimental measurements.
In implementing the health monitoring of the test structure, three strategies are carried out.
By using the first strategy, the acceleration measurements of each simulated damage case are first analyzed using the ANNSI model to generate the modal frequencies and displacement modal shapes of the test structure. The structural condition of the specimen can then be diagnosed based on the identified modal data change. In the second strategy, the health monitoring of the test structure is basing on the changes in strain mode shape information. The strain mode shapes are extracted from the FBG sensors and RSGs measurements by also using the ANNSI model.
Moreover, the global and decentralized monitoring networks are adopted for the purpose of health monitoring using the structural acceleration and strain measurements in the third strategy.
The three strategies are sequentially introduced in the subsequent sections. Notably, according to the nature of the damage detection procedure in the strategy, the first strategy is model-based;
while the second and third strategies are non-model-based. Moreover, though three different strategies for structural health monitoring are utilized, they should produce similar diagnostic results.
5.2 Modal Analysis Using The ANNSI Model
the Fourier spectra of the experimental measurements, the appropriate architecture of the modal analysis network (MAN) in ANNSI model is determined. The acceleration measurements are first analyzed to obtain the corresponding modal parameters. Subsequently, the strain measurements
the Fourier spectra of the experimental measurements, the appropriate architecture of the modal analysis network (MAN) in ANNSI model is determined. The acceleration measurements are first analyzed to obtain the corresponding modal parameters. Subsequently, the strain measurements