1. Introduction
In the past few decades, many damping devices have been developed and applied in civil engineering structures, especially in earthquake-prone regions [
1,
2,
3]. Viscous dampers generally represent a broad class of passive energy dissipation devices and have emerged as one of the most popular ones. By providing additional damping and energy dissipation, they can efficiently suppress earthquake-induced vibrations and, therefore, limit the damage to structural and non-structural components [
4,
5,
6]. To better understand the damping properties of such dampers, many experimental studies have been conducted, including damper element tests [
7,
8] and shaking table tests of small-scale and full-scale structures with viscous dampers [
4,
5]. However, studies focusing on in-service structures with such dampers when subjected to large earthquakes are relatively rare [
9,
10]. This practical issue requires accurate state quantification for dampers and structure itself, which holds a key position in structural health, safety, and risk assessments [
11,
12]. This paper presents a study on the damping estimation of an eight-story building structure equipped with oil dampers through a proposed framework, which consists of an enhanced strain-energy method and an improved direct method for model updating.
The strain-energy method is usually adopted at the design stage of passively controlled structures to assess the damping effect of added dampers quantitatively. Specifically, it incorporates the supplemental damping provided by dampers into modal damping ratios, which are frequently referred to as equivalent damping ratios and are defined as the ratios of energy dissipated by dampers and stored in structure system. Zhang and Soong [
13] provided a procedure to quantify the damping effect of viscoelastic dampers and showed their effectiveness in attenuating structural seismic response. Chang et al. [
14] carried out a comprehensive experimental program on a 2/5-scale five-story steel frame with viscoelastic dampers and applied this procedure to assess the equivalent damping ratios for measured modes. Ji et al. [
15] conducted a series of shaking table tests on a full-scale five-story steel building with three types of passive dampers, including viscous dampers and oil dampers. Using the strain-energy method and system identification, they investigated the damping characteristics of those dampers and presented simplified estimations for their supplemental damping ratios.
In this study, the strain-energy method is adopted to estimate the supplemental damping ratios provided by oil dampers for the investigated building under earthquake excitations. Additionally, it is notable that, except for the supplement damping, dampers can contribute to the increased lateral stiffness of the entire structure [
14,
15,
16,
17]. This stiffness contribution, usually not large, may affect structural dynamic behavior, such as increasing natural frequencies. For evaluating such modal changes, an enhanced strain-energy method is proposed herein, quantifying the damper stiffness in the form of equivalent natural frequencies, i.e., the natural frequency changes of structures with and without dampers. Besides, note that the strain-energy method adopts assumptions of harmonic motion and mode-proportional deformation for dampers and structure itself, which violate their realistic situation during an earthquake event [
15]. Consequently, the estimated equivalent damping ratios and natural frequencies may be questionable and, therefore, need to be verified under earthquake situations. In this study, it is achieved through model updating and prediction validations.
Model updating using dynamic measurements has been a topic of extensive focus for many engineering principles, including civil engineering [
18]. In structural dynamics community, it is well-understood that the predictions from an initial structural model are frequently inconsistent with those observed from its real-world counterpart [
19]. These predictions could be modal properties, time histories, or frequency response functions. The objective of model updating is to reduce such discrepancy, improving the correlation between numerical models and their real-world counterparts. Generally, model updating belongs to the family of inverse problems, and, in most practical applications, it can be approached from an optimization perspective. In recent decades, a significant number of model updating methods have been developed. Literature review for this topic is not presented here since is beyond the scope of this study. The interested reader is referred to references [
19,
20,
21]. This paper is concerned only with the so-called direct updating methods because of their particular features, the resulting updated model reproduces exactly the target/measured modal data, including natural frequencies and mode shapes. By taking advantage of such features, the damper stiffness effect on structural dynamic behavior can be easily captured in model updating.
The direct updating methods are essentially analytical methods which do not require iterations. In those methods, modal orthogonality, symmetry, and eigenvalue equation are usually imposed in correcting mass matrix or stiffness matrix or both, and minimum possible changes in updated matrices are achieved by using Lagrange multipliers or generalized inverse. Baruch and Bar-Itzack [
22] first formulated a procedure using Lagrange multipliers to correct stiffness matrix to satisfy modal orthogonality and the eigenvalue equation with nominal mass matrix. Berman [
23] questioned the assumption that nominal mass matrix is correct and developed a new procedure to modify mass matrix under modal orthogonality constraint. Later, Berman and Nagy [
24] integrated these two procedures into the so-called analytical model improvement technique to update a long-exposure facility structure. Based on this research, Caesar and Peter [
25] further described two direct methods for model updating using modal data. Wei [
26] considered the interaction between mass and stiffness matrices in their updating processes and proposed a united procedure to modify them simultaneously using the vibration test data. Yue [
27] developed an algorithm turning model updating to a forward problem, in which a transformation matrix is only required to update mass and stiffness matrices. Lee and Eun [
28] employed the Moore–Penrose inverse, rather than Lagrange multipliers, to reformulate the model updating problem. But the resulting corrected stiffness matrices lose symmetricity. Yang and Chen [
29] established a simple procedure to maintain the physical meanings in mass and stiffness matrices corrections, which allows the updated model to be compatible with the measured natural frequencies. All of these methods discussed above are computationally efficient and can be easily applied to large structural models [
24,
27,
29]. However, for applications in earthquake engineering, a problem arises that the predicted mode participation factors by updated models may not be consistent with those identified from earthquake measurements. To address this problem, an improved Berman–Nagy method is developed, using a new procedure to correct nominal mass matrix under the constraints of both modal orthogonality and mode participation factors.
A large earthquake event is chosen to demonstrate the effectiveness of the proposed framework for the damping estimation of the investigated building. The enhanced strain-energy method is first employed to calculate the equivalent damping ratios and natural frequencies provided by oil dampers, which are modeled by a linear Maxwell model consisting of a spring element connected in series with a dashpot element [
7,
15]. Then, the modal characteristics extracted from earthquake measurements are modified accordingly. The modified parameters are ultimately the data utilized to update an initial finite element model of the building by using the improved Berman–Nagy method. Finally, through prediction validations, the precision of the modified parameters as well as the estimated equivalent damping ratios and natural frequencies are verified. The remaining of this paper is organized as follows: in
Section 2, the investigated building and its structural monitoring system are introduced.
Section 3 presents the details of the enhanced strain-energy method and the improved Berman–Nagy method. The results of damping estimation for the building during the selected earthquake are illustrated in
Section 4.
Section 5 provides the conclusions of this study.
4. Application to the Investigated Building
The proposed framework is now implemented for the damping estimation of the investigated building using its earthquake measurements from the selected earthquake, as shown earlier in
Figure 3. The finite element (FE) modeling of the building is first introduced to provide preliminary information for the following analysis. Subsequently, using the enhanced strain-energy method, the equivalent damping ratios and natural frequencies provided by oil dampers are estimated, and the modal parameters extracted from the earthquake measurements are modified accordingly. Finally, the improved Berman–Nagy method is utilized to update the initial FE model using the modified modal parameters. Moreover, the accuracy of these estimates is demonstrated through prediction validations.
Figure 4 shows a high-fidelity FE model of the building, initially established in the SAP2000 modeling environment based on the available architectural and structural drawings. The structure was assumed to be fixed on the ground floor without considering any soil-structure interaction effect. All steel columns, beams, and damper braces were modeled by using frame elements, while concrete floor slabs were modeled using flexible shell elements [
32]. Following design guidance, the connections between structural components were assumed to be fully constrained. The oil dampers were modeled by linear Maxwell models with stiffness and damping parameters identified from their earthquake measurements, as detailed later. To reduce the computational complexity of this FE model, simplifications are adopted by invoking common assumptions used for building structures in earthquake engineering applications [
33,
34]. Note that the translational and torsional modes of this FE model are uncoupled due to the uniform distribution of mass and stiffness on floor plans, as illustrated in
Section 2, which allows the analysis through planar models for each horizontal direction. This facilitates damping estimation and model updating of each direction separately, using the earthquake measurements for the corresponding motions of the building. Additionally, the building mass is assigned as lumped mass applied to beam-column joints at floor levels, while the column deformation in the vertical direction is not considered [
34]. Therefore, rotational DoFs have zero mass, while vertical DoFs are neglected.
The resulting planar FE models have 7 translational DoFs (with mass) and 279 (zero-mass) rotational DoFs for the beam-column joints. Static condensation [
33] is further performed to remove the zero-mass rotational DoFs. The resulting nominal mass matrix is
Ma =
diag{446 348 337 334 328 324 311}/ton for the two horizontal directions, where
diag stands for a diagonal matrix with elements included in the bracket{·}, while the initial stiffness matrices(unit: MN/mm) are
The corresponding condensed models are referred to as the initial models, which are utilized in the following analysis. The equation of motion of the building in each of the two translational directions may be expressed by
where
M,
K, and
C are, respectively, the real mass, stiffness, and inherent damping matrices,
x represents the vector of floor displacement relative to the structure base,
is the acceleration input, corresponding to the acceleration of the ground floor of the building,
I is the vector of earthquake influence coefficients,
fD(·) denotes the vector of damper forces,
TD is the transformation matrix, relating movement across the ends of each damper to
x, and
cd, and
kd are the vectors of damping coefficients and stiffness parameters of linear Maxwell models, respectively, composed of the respective characteristics of the oil dampers for each floor. The damper force, for example, for the
ith floor damper is expressed as
where
fd,i and
xd,i are, respectively, the damper force and the damper deformation between its two ends. For the investigated building, the supporting brace stiffness is estimated based on their design information and is much larger than the identified damper stiffness. Since these two components are connected in series, the damper deformation is taken as the corresponding inter-story drift. Besides, note that a representation of the form of Equation (32) holds separately for each of the horizontal directions, though for notational simplicity, this distinction is not explicitly noted herein.
4.1. Damping Estimation of the Building
Implementations of the enhanced strain-energy method require the building modal data and the oil damper parameters. These characteristics are identified by adopting a standard time-domain identification technique [
7,
35] and using the selected earthquake measurements. The identification is separately performed in each horizontal direction, which facilitates a better consideration of the impact of the excitation/response amplitude for oil dampers. Precisely, for oil damper identification, the input-output earthquake data corresponds to the input displacement and output force data, with the governing equation given by Equation (33), while, for modal identification, the data is the acceleration of the ground floor (input) and two monitored floors (output), with modal-superposition models as governing equation [
35].
Table 1 lists the identification results for the four monitored oil dampers, in which I and II represent the damper types. Recall that Type I oil dampers were installed on the ground floor, while Type II oil dampers were placed on other upper floors. It is evident from the results that a discrepancy exists between the identified parameters of same-type dampers, verifying their anticipated amplitude-dependent characteristics.
Table 2 shows the identified (translational) modal parameters of the first two modes of the building. The remaining higher modes are of small importance, and identification of their properties suffers to a much larger degree by measurement noise [
35]. Notably, the identified mode shapes and mode participation factors in
Table 2 are determined by using the so-called mode scale factor, which is commonly adopted to (mass) normalize measured mode shapes. Furthermore, to obtain complete mode shapes, an expansion technique suggested in Reference [
27] is adopted, though note that other alternatives exist for resolving this issue [
19]. The implementations of this expansion approach are omitted here since it is a common practice. For clarity of explanation, the expanded mode shape is denoted hereinafter as
ϕi, for the
ith mode of the building.
Table 3 shows the estimated equivalent damping ratios (
ξeq,i) and natural frequencies (
ωeq,i) by the enhanced strain-energy method. Results of the inherent damping ratios (
ξs,i) and natural frequencies (
ωs,i) for the building itself are also reported in this table. They are obtained by subtracting those equivalent estimates from the corresponding identified modal parameters in
Table 2. It can be seen from
Table 3 that the estimated equivalent damping ratios have a sharp decrease when mode order increases and the second equivalent damping ratios are negligible. It indicates that the energy dissipated by oil dampers concentrates in the seismic motion of the fundamental modes. On the other side, for equivalent natural frequencies, an increasing trend is observed. Overall, they are relatively small compared with the identified natural frequencies of the building. However, it does not indicate that the estimated equivalent natural frequencies are unimportant. Instead, they can significantly affect the prediction accuracy of updated models, as will be shown in the next section.
Besides, the enhanced strain-energy method also allows quantifying the contribution from a specific damper to the equivalent damping ratios and natural frequencies, as evident from Equations (3) and (8).
Figure 5 plots the damping and stiffness contribution in percentages from the oil dampers of different floors along with the two horizontal directions. The dampers on the first floor dominate the contribution, up to 70% for damping and up to 50% for stiffness, which is attributed to its larger story drift. Because the damping and stiffness contribution from oil dampers is proportional to the square values of
ϕi,dj, as indicated in Equations (3) and (8). Higher floor dampers contribute more to the damping of the second mode, but the total damping contribution in that mode is negligible, as shown in
Table 3.
It is important to stress once more that the enhanced strain-energy method is built based on the assumptions of harmonic motion and mode-proportional deformation of structure and added dampers, which violate their real situations during earthquakes. Therefore, it is necessary to verify the method accuracy, which is achieved through model updating and prediction validations.
4.2. Model Updating of the Building
The model updating of the building itself is separately performed for its two horizontal directions. Specifically, for each updating case, the natural frequencies (
ωs,i), expanded mode shapes (
ϕi), and mode participation factors (
pi) of the first two modes are regrouped and utilized to correct mass and stiffness matrices by the improved Berman–Nagy method, i.e., Equation (27) for mass matrix updating and Equation (29) for stiffness matrix updating. Besides, for inherent structural damping, the damping matrix is modeled through a common modal damping assumption and is adjusted to match the estimated damping ratios (
ξs,i), as listed in
Table 3. Damping ratios for the remaining higher modes are generally large and are constrained to a predefined value (taken in the case study later as 10%). Note that using a large damping ratio for these modes, though not critical, can efficiently suppress higher-frequency components in the building response predictions.
First, mass and stiffness matrices are updated to demonstrate the effectiveness of the improved Berman–Nagy method. Two updating cases of using different modal data are considered to illustrate the impact of additional matrix Δ
M on updated mass matrix. Specifically, the regrouped modal parameters of the first mode and the first two modes are separately utilized to update the nominal mass matrix
Ma.
Figure 6 presents the new, updated mass matrix
M by Equation (27). For comparisons, the updated mass matrix
MB by the traditional Berman–Nagy method is also reported. The results in this figure correspond to the transversal direction FE model.
It is evident from the results in
Figure 6 that Δ
M, the difference between
M and
MB, has an increasing impact on the updated mass matrix when more modal data is utilized. In the first updating case, the predicted mode participation factor by
M exactly matches the measured one (48.952), while the prediction by
MB is 46.884. This prediction error is large but causes small changes in the two updated mass matrices, as shown in
Figure 6a. When the regrouped modal data of the first two modes is utilized,
M and
MB differ significantly in both diagonal and off-diagonal elements. The predicted mode participation factors are listed in
Table 4, where the improved method results are denoted as
pI,i. Again,
M reproduces exactly the measured mode participation factors, while
MB fails in the first two modes cases.
Figure 7 shows the new, updated stiffness matrices by the improved method and using the regrouped modal data of the first two modes. Unlike updated mass matrices, changes in updated stiffness matrices are small. Notably, the updated mass and stiffness matrices by the improved method exactly possess the modal data, including natural frequencies, mode shapes, and mode participation factors.
Then, the precision of the estimated inherent damping ratios (
ξs,i) and natural frequencies (
ωs,i) is considered. The updated models with new, updated mass and stiffness matrices are utilized for the building response prediction, according to Equation (34). The linear Maxwell models with identified parameters shown in
Table 1 are assigned to the same-type oil dampers, i.e., the parameter configures of Type I for the ground-floor oil dampers and those of Type II for the other floors oil dampers. Two updated models are utilized for this response prediction: In the first updated model, the damper stiffness is not considered, the modal data utilized for correcting mass and stiffness matrices is {
ωi,
ϕi,
pi,
ξs,i,
i = 1,2}, while, in the second updated model, the data is {
ωs,i,
ϕi,
pi,
ξs,i,
i = 1,2}. To distinguish them, they are labeled by updated model I and II, respectively.
Figure 8 and
Figure 9 present the measured and predicted floor acceleration and damper force, respectively. The results in these figures correspond to the longitudinal direction updated models. For clarity of illustration, the response predictions during the strong shaking phase (30 s~60 s) are shown, and the results of the two updated models are plotted separately. The normalized root mean squared error (NRMSE) between the measured and predicted response, defined as
is utilized to quantify the goodness of fit between them, in which
and y are the measured and predicted response, respectively. The results for the two updated models are listed in
Table 5.
It can be seen from
Figure 8 and
Figure 9 that the predicted responses by updated model II match very well the measured data with NRMSE values close to 0.2, which indicates a high precision in response prediction. On the other side, the updated model I has non-negligible phase-lags errors in all response predictions with larger NRMSE values up to 0.7. A similar trend of calculated NRMSEs is also observed for the transversal direction updated models, as shown in
Table 5. It verifies that, at least in this instance, the damper stiffness or the estimated equivalent natural frequencies, though small, can not be omitted in model updating for the building itself. Moreover, the high precision predictions of updated model II also show that the estimated equivalent damping ratios and natural frequencies are accurate, which addresses the concern at the end of the last section.