Sep 27, 2004 - The application of the monitoring system under discussion to a long span, flexible bridge in the metropolitan Los Angeles region is described.
INSTITUTE OF PHYSICS PUBLISHING
SMART MATERIALS AND STRUCTURES
Smart Mater. Struct. 13 (2004) 1269–1283
PII: S0964-1726(04)85525-3
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems S F Masri1 , L-H Sheng2 , J P Caffrey1 , R L Nigbor1 , M Wahbeh1 and A M Abdel-Ghaffar1 1
Department of Civil Engineering, University of Southern California, Los Angeles, CA 90089-2531, USA 2 California Department of Transportation, Sacramento, CA 91234, USA
Received 22 April 2004, in final form 18 June 2004 Published 27 September 2004 Online at stacks.iop.org/SMS/13/1269 doi:10.1088/0964-1726/13/6/001
Abstract The system architecture of a novel structural health monitoring system that is optimized for the continuous real-time monitoring of dispersed civil infrastructures is presented. The monitoring system is based on a highly efficient multithreaded software design that allows the system to acquire data from a large number of channels, monitor and condition this data, and distribute it, in real time, over the Internet to multiple remote locations. Bandwidth and latency issues that impact the operation of monitoring systems are discussed. The application of the monitoring system under discussion to a long span, flexible bridge in the metropolitan Los Angeles region is described. The bridge had previously been instrumented with 26 strong motion accelerometers. Sample ‘quick analysis’ results continuously provided by the monitoring system are presented and interpreted. System identification results, obtained through off-line batch processing, are presented for a data set from a recent earthquake that automatically triggered the recording capability of the system. It is shown that, using a time domain system identification approach, the bridge stiffness and damping matrices can be identified from the earthquake data set and subsequently used to determine the bridge modal properties, such as frequencies and damping ratios. In this approach the bridge is modeled as a multi-input/multi-output system with order compatible with the number of available sensors. Implementation issues requiring further investigation are presented and discussed. (Some figures in this article are in colour only in the electronic version)
1. Introduction 1.1. Background Interest in the field of structural health monitoring (SHM) has been growing at a fast pace in the recent past due developments in the fabrication of innovative sensors, the ease of deploying sensor networks and data acquisition systems, and the associated growth in the computational power that is becoming readily available with PCs. Furthermore, the development of 0964-1726/04/061269+15$30.00 © 2004 IOP Publishing Ltd
sophisticated digital signal processing tools for the analysis of vibration signatures of dispersed civil infrastructure, in conjunction with real-time monitoring approaches, allows continuous condition assessment (of limited scope) of an instrumented structure. Numerous national and international workshops and conferences have been convened in the recent past to deal with various aspects of this broad, interdisciplinary field of structural health monitoring. Information concerning some notable conferences, meetings and papers focusing on SHM
Printed in the UK
1269
S F Masri et al
Figure 1. The real-time monitoring system flow diagram.
are available in the publications of Agbabian and Masri (1988), Natke and Yao (1988), Housner and Masri (1990), Housner et al (1992), Housner and Masri (1993), Natke et al (1993), Szewezyk and Hajela (1994), Housner et al (1995), Chen (1996), Housner and Masri (1996), Housner and Kobori (1996), Masri et al (1996), Housner et al (1997), Chang (1997, 1999), Casciati and Magonette (2000), Smyth et al (2000), Smyth and Masri (2001), Chang (2001), Casciati (2002), Wolfe et al (2002), Chang et al (2002), FWSHM (2002), IABMAS (2002), ICANCEER (2002), SHMISIS (2002), IMAC (2004), Chang (2003), Wu (2003), and SPIE (2004). Notable applications of SHM to full-scale bridges include: the work of Nigbor and Diehl (1997) who developed and implemented an ‘on-line alerting of structural integrity and safety’ system to remotely monitor two large bridges in Thailand and Korea; the work of Aktan and colleagues who have applied SHM approaches to several bridges in the USA (Aktan et al 2003, Helmicki et al 1997); and bridge monitoring activities in Europe (see, e.g., Rohrmann et al 2003). 1.2. Scope This paper provides a comprehensive overview of the underlying system architecture, deployment, and utilization of a Web-enabled, real-time structural health monitoring system that is optimized for dealing with realistic situations encountered in the SHM of ‘intelligent’ civil infrastructure systems such as long span bridges provided with damping augmentation devices. Section 2 of the paper describes the system design and performance, section 3 discusses the deployment of the real1270
time monitoring system on an important long span bridge in the metropolitan Los Angeles region, and section 4 demonstrates how acceleration response measurements, obtained during a recent earthquake to which the test bridge was subjected (shortly after the monitoring system was installed), can be analyzed to extract important features of the structural characteristics of the bridge that are useful in performing condition assessment of the bridge. Field implementation issues that influence the data collection strategies as well as feature extraction approaches for practical SHM applications are presented in section 5.
2. Real-time monitoring system architecture Members of the research team have collaborated on the planning and deployment of an innovative real-time monitoring system that is optimized for use with realistic situations encountered in the SHM of dispersed civil infrastructure systems. This section gives an overview of the main system architecture features of the software, as well as discussing the tradeoffs involved in selecting appropriate values for some essential performance parameters that have a significant influence on the data acquisition and distribution capabilities of the system. 2.1. System design Figure 1 shows a flow diagram of the real-time monitoring system used in this study. The following explanations apply to this diagram.
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems
2.1.1. Overview of system architecture. The monitoring system is based on a multithreaded software design. This highly efficient software architecture allows the system to acquire data with multiple channels, monitor and condition this data, and distribute it, in real time, over the Internet to various remote locations. The TCP/IP data distribution interface of the system is designed around a publish/subscribe interface. This means that the system can send data of different types (publications) to different clients (subscriptions) requesting that specific type. For example, on a particular structure, one can group the sensors in different categories based on type, locations, orientation, and so on. The system can make those separate types available for distribution in the manner in which they were grouped. In order to optimize communication and productivity, remote clients can use the ‘publishing’ concept to receive only the channels that they are interested in. Now, the clients can subscribe to only the data that they are interested in. The system will distribute to multiple clients for each subscription. That means that it can maintain multiple subscriptions per data type. More than two clients can subscribe and receive data in real time for each data type that exists on the system. Per each data type, the system will distribute the data to each client (‘subscriber’) in sequential order. The switching time is on the order of microseconds. The number of clients that can be connected to the system and the number of data types that can be specified are directly proportional to the bandwidth available to the system. 2.1.2. Server software. The server software has three main threads. One is the local monitoring (recording) thread, another is the data distribution thread, and the third is the acquisition thread. They all revolve around a limited size queue. Data from the acquisition thread goes into the queue. From here, each packet is analyzed for triggering conditions, and the same packet is then made available for distribution to the publishing communication interface. Each packet, based on its content (which data channels it contains), is routed to the corresponding publishing interface. Another very important feature of the publish/subscribe interface is that in the case of a disconnect, the server retains the subscribers, and when the connection is re-established, the server will start sending data to the subscribers without the need to restart the clients. 2.1.3. Data monitoring and distribution. The local monitoring (recording) thread and the data distribution thread use exactly the same packet as input. The distribution thread takes the packet just as it was acquired, without any conditioning, and distributes it to the publishing interface. The only signal processing applied to the data is scaling (conversion to cm s−2 if the sensors are accelerometers). Now, the monitoring thread, besides scaling, performs signal conditioning for trigger monitoring, and local recording (eventually) with pre-event (user-specified). The local monitoring thread will record an event locally regardless of what happens to the data distribution thread. Once the event is recorded, the system will try to notify a list of users (via e-mail), and ftp the event to another site. If this process fails, the system will retry every 5 min until successful.
2.1.4. Trigger parameters. The triggering mechanism operates as follows: the raw data coming from the server is filtered using the settings of the trigger filter, and subsequently each specified channel/sensor is monitored for threshold exceedance. If one or more of the monitored channels exceeds their specified trigger level, then the system starts recording and storing the data in specific directories. 2.1.5. Filters. Three separate and independent filters can be set by the user: the (1) integration filter, (2) global filter, and (3) trigger filter: (1) The integration filter is used to filter out the DC component of the signal in order to perform the double integration of the acceleration signal to obtain the corresponding velocity and displacement. A typical choice for this filter is Butterworth type, high pass, with a lower cut-off frequency fc of 0.1 Hz. (2) The global filter is used to apply a filter to the signal plotted in the time domain window. This allows the user to eliminate some high frequency components that are not a significant part of the overall structural vibration or local structural element(s) vibration. It is worth noting that if an event triggers the monitoring system (on the basis of the criteria discussed above), this filter is ignored, and the recorded data is the exact raw data (i.e., no signal processing is applied to the recorded data). A typical setting for this filter, when used in conjunction with the Vincent Thomas Bridge (VTB), is: Butterworth type, low pass, with f c = 6 Hz. (3) The trigger filter is used to specify the frequency range within which a trigger level exceedance should be monitored. A typical setting for this filter (also known as the ‘Classic Earthquake Strong Motion Trigger Filter’) is: Butterworth type, band pass, spanning the range from 0.1 to 12.5 Hz. 2.2. Operation Using the ‘quick analysis’ capability of the RTM system, various measures of the monitored system’s response can be displayed in near real time. The only delay (on the order of a few seconds) is due to filling data storage buffers needed to perform digital signal processing on a selectable time segment. The screen image shown in figure 2 presents the FFT of any two arbitrary channel choices, and their transfer function. The screen in figure 3 shows the short-term rms and long-term rms of one selectable channel. The short-term window displays the updated channel rms level once every second, while the longterm window has an averaging time that can be set to show the moving average rms of the same channel. Another choice of the quick analysis that can be conveniently displayed is the cross-correlation between any two channels. The screen image in figure 4 shows the crosscorrelation between channels 15 and 17, both on the bridge deck, with orientation in the vertical direction. This plot contains a lot of useful information about the interaction between the dynamic loads on the bridge and its modal characteristics. It can be used for a rapid (rough) estimation of the dominant bridge mode being observed in the selected time window, as well as being a crude estimator of the corresponding bridge damping parameter. 1271
S F Masri et al
Table 1. Speed comparison of Internet connection types. Internet connection type Fast (>10 Mbps) Ethernet Analog modem ISDN DSL/cable Stationary satellite DS1/T1
Figure 2. FFT and transfer function of selected nodes.
Latency 500 µs, mostly due to high orbital elevation 2–5 µs
up under discussion, a 1.5 Mbit up-link is more than enough to satisfy five clients or more in real time. The commonest installation of such systems however, sits on digital subscriber lines (DSL). Speed (latency) and capacity (bandwidth) are two very separate things. The combination of latency and bandwidth gives users the perception of how quickly a Web page loads or a file is transferred. The most common analogy for comparing latency and bandwidth is to imagine water running through a pipe: the flow speed is latency; the area of the pipe is bandwidth. If one has a wide pipe but low speed, one can move more water through the pipe but at a slower rate. On the other hand, if one has a narrow pipe but high speed, one can move less water but at a faster rate. 2.3.2. Real-time system performance. Table 1 below provides typical latency values for different kinds of connections. For the VTB installation discussed in the next section we used a commercial DSL connection with complete success. Reliable real-time remote monitoring has been achieved at low cost, less than $100/month. Throughput rates of 500 samples per second per channel with 32 channels and 24 bit resolution are achievable. Indeed, several weeks of continuous data have been monitored and recorded by a remote client for the VTB system. Modern Internet connectivity allows us to realize the potential of this real-time health monitoring technology.
3. Application of the SHM system to Vincent Thomas Bridge 3.1. Background Figure 3. Sample RTM results showing short-term and long-term rms levels of selected channels.
2.3. Performance 2.3.1. Bandwidth and latency. Bandwidth is a measure of data throughput over different data lines. In the present case, the discussion concerns bandwidth over TC/IP channels. The important parameter for the system is the up-link data rate of the connection. The system is sending high amounts of data on the line, which makes it different from the regular client browsing the Web, generally receiving high amounts of data. One connection scenario is on a LAN layout. LAN speeds are usually 100 Mbps and are usually symmetric (uplink throughput is equal to the down-link rate). For the set1272
The VTB is located in San Pedro, California, and is a major transportation artery connecting Los Angeles with its harbor. It is a cable-suspension bridge, approximately 1850 m long, consisting of a main span of approximately 457 m, two suspended side spans of 154 m each, and a ten-span approach of approximately 545 m length on either end. The roadway accommodates four lanes of traffic. The bridge was completed in 1964, and in 1980 was instrumented with 26 accelerometers as part of a seismic upgrading project. The geographical location of the bridge is shown in figure 5, and a bridge photograph is shown in figure 6. The VTB has undergone a major seismic retrofit during the period 1996–2000, which involved a variety of strengthening measures in addition to the incorporation of about 30 large scale, nonlinear passive (viscous) dampers.
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems
Figure 4. Sample results showing the cross-correlation function between two data channels.
Figure 5. Geographical location of the VTB.
Figure 6. A photographic overview of the VTB.
3.2. Vincent Thomas Bridge sensor network Currently the sensor network is maintained by the State of California Department of Conservation (CDC) Office of Strong Motion Studies, through the California Strong Motion Instrumentation Program (CSMIP). Figure 7 shows the location layout of all 26 sensors mounted on the bridge. Notice that the eastern half of the bridge is more densely instrumented than the western half. This is because the analog recorder is
housed in the eastern cable anchorage. Sixteen accelerometers are distributed at various locations in the lateral, longitudinal, and vertical directions on the superstructure itself. 3.3. Operation of streamer software The monitoring system discussed in section 2 was installed on the VTB in early 2003. Sample measurements from the system are shown in figures 8 and 9. 1273
S F Masri et al
2
16
15
14
4
8
18
11
17
1
3 23
10 5 12 22
6
21 19 7
9
457.2 m 20
13 26
North
24
25
Figure 7. The VTB sensor location.
Figure 8. The time history screen. Figure 9. The trigger selection and filtering screen.
The screen image shown in figure 8 shows the time history of the ‘raw’ sensor data (acceleration sensor measurement in the present example). The index of the displayed channel is shown on the left-hand side (LHS) of the panel, and the corresponding sensor index (as denoted in figure 7) is shown on the RHS of the panel. The displayed time history segment corresponds to a 10 s window. Automatic amplitude scaling is 1274
performed for each displayed channel. Any arbitrary choice of channels can be displayed. To enhance visual monitoring, data channels are color coded so that sensors in the same direction of motion (bridge global x, y, and z directions) are displayed with the same color on the computer screen.
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems
Figure 10. The geographical location of the M5.4 22 February 2003 Big Bear City, California, earthquake; the epicenter is 3.1 miles north of Big Bear City, California (N34.31◦ , W116.85◦, depth 1.2 km).
4. Analysis of earthquake data from the RTM system 4.1. Triggered measurement of the 22 February 2003 Big Bear earthquake response In the early morning of 22 February 2003, a relatively small earthquake (magnitude M = 5.4) occurred in the vicinity of the city of Big Bear, California (figure 10). The epicenter is located about 180 km from the VTB. All acceleration channels were triggered on the VTB, and a complete data set with 26 channels was obtained by the RTM system. A collection of all 26 acceleration records obtained during this earthquake are shown in figures 11(a) and (b) using a different amplitude scale for each component, in order to enhance resolution. Figure 11(a) shows the acceleration time histories of the 16 data channels corresponding to the bridge deck, while figure 11(b) shows the corresponding accelerations of the ten sensors at the base of the VTB. A sample of three acceleration components at the base of the VTB (channels 1, 23, and 14, corresponding to the lateral, longitudinal, and vertical components, respectively), as well as three response components on the bridge deck (channels 5, 12, and 17, corresponding to a lateral, longitudinal, and vertical components, respectively) are shown in figure 12. To help with visualizing the dynamic amplification effects of the deck motions with respect to the base motions, identical amplitude scales and timescales are used for all displayed data channels in figure 12. The recorded accelerations were subsequently processed to obtain the corresponding velocities and displacements for all the channels. The resulting velocity and displacement time histories for the set of channels shown in figure 12 are displayed in figures 13 and 14. To obtain accurate velocity and displacement time histories from the measured acceleration data, special
attention is needed to use the most suitable digital signal processing approach. The potential problems and serious computational pitfalls encountered by using ‘naive’ data processing approaches to integrate acceleration measurements are discussed in detail by Worden and Tomlinson (2001) and by Smyth and Pei (2000). Comparison of the dynamic environment acting on the VTB during the subject earthquake shows that there is a drastic difference of the measured accelerations at the bridge base (on the order of milligrams) as compared to the response on the bridge deck. Note that, for convenience, identical amplitude scales are used in figures 12–14 for similar kinds of response measures. It is also worth noting the drastic change in the spectral content of the motion at the base and deck of the bridge. Figure 15 shows the FFT of the acceleration time history records shown in figure 12. Notice that the input records have a relatively small amplitude, but a broadband nature, while the bridge response records have a much higher amplitude level and exhibit behavior of a low pass filter. As will be shown in the following section, the response peaks are directly correlated with the dominant (identified) system natural frequencies. 4.2. Overview of the time domain analysis method The system identification procedure used in this study is based on a self-starting (non-iterative) time domain, leastsquares-based method, which develops a reduced-order multiinput/multi-output (MIMO) nonlinear (i.e., not necessarily linear) model whose order is compatible with the available number of sensors. Consider a MIMO system having n 1 degrees of freedom and subjected to n 0 excitations. In the context of the example bridge at hand, the number of response sensors mounted on the bridge is 15 (keeping in mind that one of the deck sensors was malfunctioning); hence n 1 = 15. Similarly, the number of sensors corresponding to the base excitation is 10; hence n 0 = 10. 1275
S F Masri et al
(a) 10
20 15
5
10
2
0 -5
3
5 0 -5
-10
-10 -15 850
900
950
1000
1050
850
1100
1.5
900
950
1000
1050
1100
15
1
10
0.5 0
4
5
-0.5
0
-1
-5
5
-1.5 -10 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
15
6
5
7
10 5
0
0 -5
-5
-10
-10 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
6
15
8
10
10
4 2
5
0
0
-2
-5
-4 -6
-10 850
900
950
1000
1050
1100
-8 850
900
950
1000
1050
1100
6 4
11
2 0
12
5 0
-2 -4
-5
-6 -10
-8 850
900
950
1000
1050
850
1100
30
900
950
1000
1050
20
15
20 10 0
1100
16
10 0
-10
-10
-20 -30
-20 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
40 40
18
17
20
20
0
0
-20
-20
-40
-40 850
900
950
1000
1050
850
1100
900
950
1000
1050
1100
40 20
21
10
30
22
20 10
0
0
-10
-10 -20
-20
-30 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
Figure 11. (a) Acceleration time histories of the deck channels only recorded during the 2003 Big Bear earthquake. (b) Base acceleration time histories of ten data channels recorded during the 2003 Big Bear earthquake.
Treating the problem as a MDOF system subjected to prescribed base excitations, the governing equation of motion can be expressed in the form −1 −1 −1 C11 x˙1 (t) + M11 K 11 x1 (t) + M11 M10 x¨0 (t) M11 −1 −1 (4.1) + M11 C10 x˙0 (t) + M11 K 10 x0 (t) = −I x¨1 (t1 ) where: • M11 , C11 , and K 11 are the typical mass, damping, and stiffness matrices, which characterize the forces 1276
associated with the unconstrained DOF of the system. Each of these is an n 1 × n 1 matrix. • M10 , C10 , and K 10 are constant matrices, which are associated with support or input motions. • x1 (t) is a vector of order n 1 of the active degrees of freedom (measured response) displacement, velocity, and acceleration, respectively:
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems 2
(b)
2
1
9 1
1 0
0
-1 -1 -2 -2
-3 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
3 2
0.75 0.5 0.25
13
1 0
14
0 -0.25
-1
-0.5 -0.75
-2 -3 850
900
950
1000
1050
850
1100
0.6
900
950
1000
1050
1100
0.6
19
0.4
20
0.4 0.2 0
0.2 0 -0.2
-0.2
-0.4
-0.4 -0.6
-0.6 850
900
950
1000
1050
1100
850
900
950
1000
1050
1100
2 3
23
2
24
1
1 0
0 -1
-1
-2 -2
-3 850
900
950
1000
1050
1100
2 1.5
850
900
950
1000
1050
1100
1
25
1
26
0.5
0.5 0
0 -0.5
-0.5
-1 -1
-1.5 850
900
950
1000
1050
850
1100
900
950
1000
1050
1100
Figure 11. (continued). 10
10
Ch.1
Ch.5
5
5
0
t
-5
0
t
-5
-10
-10 850
900
950
1000
1050
1100
850
10
900
950
1000
1050
1100
10
Ch.12 5
5
Ch.23
0
t
-5
0
t
-5
-10
-10 850
900
950
1000
1050
1100
850
10
900
950
1000
1050
1100
10
Ch.14
Ch.17
5
5
0
t
-5
0
t
-5
-10
-10 850
900
950
1000
Base Acceleration
1050
1100
850
900
950
1000
1050
1100
Deck Acceleration
Figure 12. Sample acceleration records measured at the VTB base (LHS column) and its deck (RHS column) during the Big Bear, 22 February 2003, earthquake. Plot units are cm s−2 .
1277
S F Masri et al 1
1
Ch.1 0.5
Ch.5
0.5
0
t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
850
1
900
950
1000
1050
1100
1
Ch.23
0.5
Ch.12
0.5
0
t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
850
1
900
950
1000
1050
1100
1
Ch.14
0.5
Ch.17
0.5
0
t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
850
900
Base Velocity
950
1000
1050
1100
Deck Velocity
Figure 13. Velocity time history records of selected data channels. 1
1
Ch.1
0.5 0
Ch.5
0.5 t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
850
1
900
950
1000
1050
1100
1
Ch.23
0.5 0
Ch.12
0.5 t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
850
1
900
950
1000
1050
1100
1
Ch.14
0.5 0
Ch.17
0.5 t
-0.5
0
t
-0.5
-1
-1 850
900
950
1000
1050
1100
Base Displacement
850
900
950
1000
1050
1100
Deck Displacement
Figure 14. Displacement time history records of selected data channels.
x1 (t) = [x11 (t), . . . , x1n1 (t)]T
x0 (t) = [x01 (t), . . . , x0n0 (t)]T
x˙1 (t) = [x˙11 (t), . . . , x˙1n1 (t)]T
x˙0 (t) = [x˙01 (t), . . . , x˙0n0 (t)]T
x¨1 (t) = [x¨11 (t), . . . , x¨1n1 (t)]T
x¨0 (t) = [x¨01 (t), . . . , x¨0n0 (t)]T .
and x0 (t) is a vector of n 0 order of the support displacement, velocity, and acceleration, respectively: 1278
The elements of the above-mentioned matrices can be obtained by expressing the time history measurements as
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems 10 2
8 1.5
Ch.1
6
1
Ch.5
4
0.5
2 Hz
Hz 0.5
1
1.5
0.5
2
Ch.23
1
1.5
2
5
2
Ch.12
4
1.5 3
1 2
0.5
1
Hz 0.5
1
1.5
0.7
Hz
2
Ch.14
0.5
1
1.5
Ch.17
25
0.6 0.5
20
0.4
15
0.3
2
10
0.2 5
0.1 Hz 0.5
1
1.5
2
Base FFT of acceleration
Hz 0.5
1
1.5
2
Deck FFT of acceleration
Figure 15. The FFT of the recorded VTB accelerations of selected data channels.
a series of overdetermined equations and then determining the unknown parameters by utilizing standard least-squares solution techniques. This is done by writing each row of these matrices of the above-mentioned equation of motion at every discrete time step, t = [t1 , . . . , t N ], where N is the number of time steps used, resulting in the following system of equations: −1 M11 C11 x˙1 (t1 )
+
−1 M11 K 11 x1 (t1 )
+
−1 M11 M10 x¨0 (t1 )
−1 −1 + M11 C10 x˙0 (t1 ) + M11 K 10 x0 (t1 ) = −I x¨1 (t1 )
.. .
(4.2)
−1 −1 −1 M11 C11 x˙1 (t N ) + M11 K 11 x1 (t N ) + M11 M10 x¨0 (t N ) −1 −1 + M11 C10 x˙0 (t N ) + M11 K 10 x0 (t N ) = −I x¨1 (t N ).
The above set of equation can be represented in the following compact form: Rˆ αˆ = bˆ (4.3) where Rˆ is a block diagonal matrix whose diagonal elements are equal to R, in which the elements of R consist of the measured accelerations, velocities, and displacements; vector αˆ = [α1T , α2T , α3T , . . . , ανT ]T , whose elements correspond to the unknown system matrices; and bˆ is the corresponding vector of excitation measurements. Note that Rˆ is of order m × n where m = N n 1 and n = 3n 1 (n 1 + n 0 ). Thus,
R 0 Rˆ ≡ : 0
0 R
···
0
. R
n 1 N ×n 1 (2n 1 +3n 0 )
α1 α 2 . αˆ ≡ . . αn1
n 1 (2n 1 +3n 0 )×1
b1 b 2 . . bˆ ≡ . . bn1
(4.4)
n 1 N ×1
One can solve for the unknown parameter vectors αˆ by determining the pseudo-inverse of Rˆ or R matrices. This can be written as follows: αi = Rˆ i bi
(4.5)
where Rˆ stands for pseudo-inverse of matrix Rˆ (Golub and Van Loan 1983). Once the equivalent system matrices have been identified as indicated above, the residual terms, found by taking the difference between the LHS and RHS of equation (4.1), can be used to determine the accuracy obtained by modeling the reference structure as an essentially linear system. However, if the norm of these residual acceleration terms is not sufficiently small, then the second phase of the identification approach 1279
S F Masri et al −1 Table 2. Sample output of the M11 K 11 matrix (15 by 15).
32.81 −0.62 0.22 −15.57 −0.67 −9.00 1.14 −0.07 −5.44 0.26 6.22 −0.14 0.30 −1.35 0.34 −7.55 0.13 −0.20 22.04 −0.04 1.64 −0.17 0.06 0.76 1.12 −4.31 −0.28 0.04 −8.58 −1.15 1.31 0.28 0.14 −1.50 −0.80 −0.55 −0.24 0.02 0.25 1.13 −1.36 0.03 0.19 −0.47 −0.13 7.14 −2.53 0.36 −1.00 0.10 −1.29 2.57 −0.30 −7.42 −0.20 4.67 0.05 −0.26 −3.33 0.36 −3.33 0.57 −0.16 4.78 −0.71 −0.69 0.42 −0.63 −3.88 −2.86 −0.90 −0.35 0.41 −0.89 3.15
2.17 1.57 −4.35 1.75 −1.25 33.41 0.08 −0.48 0.81 −2.18 2.61 0.03 −0.39 −0.14 −0.01
5.04 −3.80 −2.52 1.81 −1.86 0.25 0.14 0.00 −1.25 1.58 0.79 −4.53 4.26 0.19 −0.10 −0.08 −3.29 3.73 −0.28 0.30 −0.26 0.15 −0.08 0.74 1.43 −4.12 0.33 1.14 −1.15 −0.14 0.02 −1.56 −5.77 5.73 −1.46 0.07 0.02 −0.04 −0.01 0.30 6.07 −8.96 0.51 0.36 −0.73 0.19 −0.01 −0.72 10.75 −9.75 0.98 −0.30 0.36 −0.01 −0.02 −0.37 −7.39 8.76 1.71 −0.20 0.26 −0.13 0.08 −0.45 1.63 −1.62 5.23 −0.07 0.17 −0.19 0.22 −0.28 1.36 −1.14 −0.89 12.37 −10.14 −0.28 0.08 −0.50 −1.22 1.26 0.77 −13.32 13.78 0.83 −0.29 0.94 −8.98 7.05 0.63 −0.73 1.06 1.52 −0.33 1.14 7.51 −10.03 1.01 −0.73 1.04 0.25 0.65 −1.78 0.41 3.91 −1.26 −1.29 1.34 0.11 −0.37 10.63 11.60 −1.09 −2.47 1.03 −0.91 −0.24 −0.12 −10.62
0.46 0.32 −0.73 0.71 −0.52 −0.06 0.61 0.61 0.43 0.57 −0.64 −1.60 1.44 −8.90 16.10
−1 Table 3. Sample output of the M11 C11 matrix (15 by 15).
−0.07 0.10 −0.58 0.28 −0.17 0.43 −0.09 −0.14 0.40 0.24 −0.24 0.16 −0.16 0.33 −0.38 0.58 −0.17 0.30 −0.27 −0.08 −0.36 −0.70 0.55 0.47 −0.30 0.20 0.00 0.12 0.01 −0.02 0.20 −0.14 −0.01 −0.24 −0.72 0.68 0.76 −0.51 −0.35 0.14 −0.26 0.06 0.00 0.25 −0.32 −0.45 0.08 0.23 0.45 0.20 −0.31 1.14 −1.28 −0.83 0.08 0.04 −0.02 −0.08 −0.27 0.24 0.91 −0.31 0.75 −3.25 −0.14 0.33 −1.02 1.36 −1.32 −0.15 0.08 0.20 −0.19 0.29 −0.28 −0.87 0.17 −0.69 1.18 0.21 −0.26 1.50 −0.12 −0.45 −0.10 −0.03 0.13 −0.15 −0.06 0.26 0.09 0.11 −0.39 0.38 0.44 −0.48 −1.32 1.69 0.11 −0.09 0.12 −0.15 0.15 0.20 −0.09 0.01 −0.05 0.05 1.04 −0.12 −0.37 −1.68 1.08 0.62 0.02 0.05 −0.12 0.14 0.11 −0.29 −0.43 −0.20 0.11 0.66 0.45 0.01 0.37 −0.69 0.13 0.01 0.02 −0.05 0.04 −0.10 0.05 −0.78 −0.42 1.11 −0.58 0.00 0.70 1.08 −1.77 0.21 −0.20 0.13 0.41 −0.31 −0.21 0.10 0.18 −0.94 2.12 −0.57 0.12 0.22 −2.29 0.97 1.10 −0.21 0.26 −0.19 0.25 0.11 −0.26 −1.29 −1.40 0.88 0.72 −0.89 0.15 0.48 −0.16 0.30 −0.39 0.16 −0.08 0.21 −0.02 −0.04 −0.24 −0.88 0.74 0.20 0.46 −0.65 −0.96 −0.09 −0.30 −0.10 0.20 −0.14 0.15 −0.27 0.14 −0.44 −0.13 0.06 −0.19 −0.11 0.13 −1.31 0.38 0.28 0.00 0.03 −0.16 0.18 0.11 −0.20 −0.27 −0.82 0.86 −0.44 −0.11 0.30 0.33 0.63 −0.59 −0.07 0.12 −0.07 −0.02 0.01 0.07
under discussion can be used to develop a parsimonious nonparametric representation of the dominant nonlinear accelerations in terms of suitable basis functions corresponding to the generalized structure’s state variables. Further details regarding the use of this approach to model nonlinear MIMO systems are available in the work of Masri et al (1987a, 1987b). The application of this approach to the analysis of strong earthquake motion recorded in previous earthquakes at the VTB is reported in the work of Smyth et al (2003), where significant nonlinear contributions from the bridge response were detected and modeled with fairly good accuracy by using the identification procedure under discussion. 4.3. System identification of the VTB As previously mentioned, there are 16 sensors (strong motion accelerometers) mounted on the bridge deck, and 10 sensors at its base. However, measurements from sensor number 14 were not available due to instrumentation problems. Consequently, for the purposes of the present analysis, the model to be identified will be considered as having n 0 = 10 inputs and n 1 = 15 outputs. Hence, the order of the system matrices to −1 −1 C11 and M11 K 11 , while it is 15 be identified is 15 by 15 for M11 −1 −1 −1 by 10 for M11 M10 , M11 C10 , and M11 K 10 . For illustration, −1 the elements of matrix M11 K 11 are shown in table 2 and those −1 for M11 C11 are shown in table 3. −1 −1 K 11 and M11 C11 are obtained as a bySince both M11 product of this identification technique, one can obtain not 1280
Table 4. Comparison of frequency estimates from other research studies of the VTB earthquake response records. Smyth et al (2003) Dominant modes (all directions)
(This study) Dominant modes (all directions) ( f c = 0.15 Hz) Whittier Northridge Whittier Northridge Big Bear 2003 0.212 0.242 0.317 0.531 0.570
0.225 0.240 0.358 0.390 0.448
Lus et al (1999) Vertical component modes
0.234 0.388 0.464 0.576 0.617
0.225 0.304 0.459 0.533 0.600
0.194 0.234 0.283 0.361 0.515
only the classical normal modes from the solution of the −1 eigenvalue problem associated with M11 K 11 , but also the damped natural frequencies and corresponding ratios of critical −1 damping which are furnished by knowing M11 K 11 as well −1 as M11 C11 . Table 4 lists the first few (dominant) modal frequencies and damping ratios obtained through the solution of the underlying eigenvalue problem. Technical details on how to solve the complex eigenvalue problem associated with −1 −1 M11 K 11 and M11 C11 are available in the work of Smyth et al (2003). Notice that the method under discussion does not assume anything about (or restrict the nature of) the damping mechanism (whether it is of the ‘proportional’ or ‘non-proportional’ type). It is worth noting that the results obtained from this earthquake have close correlation with other frequency and
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems
Table 5. Comparison of modal frequencies from ambient vibration analyses. Frequency index
Abdel-Ghaffar and Housner (1977)
Pridham and Wilson (2001)
Lin and Betti (2003)
1 2 3 4 5 6 7 8 9 10
0.168 0.216 0.234 0.366 0.487 0.494 0.542 0.623 0.678 0.740
0.155 0.220 0.232 0.368 0.477 0.531 0.571 0.618 0.633 0.687
0.200 0.393 0.517 0.881 1.078 1.747 1.872 2.063 2.094 2.228
damping estimates made by several other investigators who studied the behavior of this bridge and whose results are compared in table 4. Other researchers have studied the ambient dynamic response of the VTB and obtained corresponding modal frequencies. Their results are summarized in table 5 in which the first (dominant) modes identified by the respective researchers are listed. Comparison of the identified frequencies in tables 4 and 5 shows that, while there is a reasonably close agreement between the range of frequencies identified from ambient and earthquake-induced motions, there is also a significant difference in the absolute values reported for the indexed modes. Among the numerous factors that can account for the observed discrepancies are: • The indexes of the modes identified by different investigators do not necessarily correspond to each other, since different modes may be excited during different episodes of motion used by the different investigators. • Significant retrofit measures were incorporated into the bridge in 2000. • Modeling errors in the mathematical model being used to analyze the data. • Errors in not accounting for missing excitations when using ambient response measurements. • Uncertainties and fluctuations in the environmental conditions corresponding to different time windows used by the investigators. • Physical changes made in the bridge when it was retrofitted so as to reduce its vulnerability to seismic events.
5. Discussion and recommendations On the basis of the valuable experience gained from this ongoing demonstration study, certain issues require further attention as detailed below. 5.1. Spatial resolution of the sensor network While the installation of the 26 sensors was a pioneering feat in the field of strong motion instrumentation back in 1970s when the sensors were installed, the sensor network is clearly very sparse for applications that need to have an accurate estimate of the complex, three-dimensional motion time histories that the complete bridge system is undergoing. A heterogeneous sensor network involving one to two orders of magnitude more sensors than are currently deployed at the bridge would be
needed for an adequate spatial resolution. Fortunately, the emergence of a relatively inexpensive class of modern sensors makes this quite achievable in the near future. 5.2. The need for heterogeneous set of sensors Considering the complex nature of the failure mechanisms that various components and subassemblies of the bridge can suffer from, there is a need for a broad spectrum of sensors to capture different measurands of monitored infrastructure systems such as bridges. For example, strain gages, corrosion sensors, direct displacement meters, and vision-based approaches can collaboratively provide comprehensive diagnostic snapshots of the target structure. Nigbor and Diehl (1997) provide an example of the use of heterogeneous sensors; they used a combination of acceleration, displacement, strain, wind, and temperature sensors in their installations. Another important factor to keep in mind is that extensive variations in the spectrum of identified frequencies can be obtained, depending on what value of f c is chosen in high pass filtering of the acceleration data. This problem is exacerbated by the sparseness of the sensor network and the presence of varying degrees of low frequency components in the response of flexible bridges. Consequently, the availability of direct displacement measures for selected locations on the bridge can be used to calibrate the digital signal processing algorithms. Among the promising, practical approaches to achieving such a direct displacement measurements is the use of optical techniques or GPS-based methods (Wahbeh et al 2003). 5.3. Data processing issues Due to the inherent limitations in using structural health monitoring techniques based on system identification approaches (modeling errors, unknown excitations, varying environmental conditions, etc), it is important to select a suitable data processing approach for interpreting the voluminous measurements. As an example, the VTB monitoring system can generate data files on the order of 125 MB h−1 of monitoring time. Not only is the field of damage detection and condition assessment based on vibration signature a wide open area for research and development, but there is also a significant issue that underlies all situations related to SHM: the probabilistic nature of the problem and the associated issues of the quantification and propagation of uncertainties in extended nonlinear systems.
6. Summary and conclusions This paper has presented the main features and the system architecture of a novel structural health monitoring system that is optimized for the continuous real-time monitoring of dispersed civil infrastructures. The monitoring system is based on a highly efficient multithreaded software design that allows the system to acquire data from a large number of channels, monitor and condition this data, and distribute it, in real time, over the Internet to multiple remote locations. Bandwidth and latency issues that impact the operation of monitoring systems are discussed. 1281
S F Masri et al
The application of the monitoring system under discussion to a long span, flexible bridge in the metropolitan Los Angeles region is described. The bridge had previously been instrumented with 26 strong motion accelerometers. Sample ‘quick analysis’ results continuously provided by the monitoring system are presented and interpreted. System identification results, obtained through off-line batch processing, are presented for a data set acquired through the monitoring system and corresponding to a recent earthquake that automatically triggered the recording capability of the system. It is shown that by using a time domain system identification approach in which the bridge is modeled as a multi-input/multi-output system, in conjunction with the data set acquired from the recent earthquake, the bridge stiffness and damping matrices, whose order is compatible with the number of available sensors, can be identified and subsequently used to determine the bridge modal properties, such as frequencies and damping ratios. It is shown that the analysis results from this study correlate well with similar results previously obtained by other investigators who studied the Vincent Thomas Bridge. Also presented and discussed in this paper are some implementation issues requiring further investigation: the spatial resolution of the sensor network, the need for a heterogeneous set of sensors, and data processing issues, particularly those related to filtering approaches.
Acknowledgments The authors appreciate the assistance of the many individuals at USC, the California Department of Transportation (Caltrans), the California Strong Motion Instrumentation Program (CSMIP), and Digitexx, Inc., who have contributed to the success of the real-time health monitoring of the Vincent Thomas Bridge by helping in the various facets of the project. Particular thanks are due A Shakal of CSMIP, and M Sereci, D Radulescu, and C Radulescu of Digitexx, Inc. The assistance of E Kallinikidou in the preparation of the manuscript is appreciated. This study was supported in part by grants from the National Science Foundation (NSF), the Air Force Office of Scientific Research (AFOSR), and the National Aeronautics and Space Administration (NASA).
References Abdel-Ghaffar A M and Housner G W 1977 An analysis of the dynamic characteristics of a suspension bridge by ambient vibration measurement California Institute of Technology, Earthquake Engineering Research Laboratory EERL 77-01 Agbabian M S and Masri S F (ed) 1988 Proc. Int. Workshop on Nondestructive Evaluation for Performance of Civil Structures (Civil Engineering Department, University of Southern California, Los Angeles, CA, July 1988) Publication No M8805 Aktan E, Ciloglu K, Grimmelsman K, Pervizpour M and Qin X 2003 Monitoring operations, security and structural health of major long-span bridge hyper-systems Proc. Structural Health Monitoring 2003 ed F-K Chang (Lancaster, PA: DEStech Publications) pp 188–96 Casciati F 2002 Proc. 3rd World Conf. on Structural Control (Como, Italy, April 2002)
1282
Casciati F and Magonette G (ed) 2000 Structural control for civil and infrastructure Proc. 3rd Int. Workshop on Structural Control (Paris, France, July 2000) Chang F K (ed) 1997 Structural health monitoring: current status and perspectives Proc. Int. Workshop on Structural Health Monitoring (Stanford University, Sept. 1997) Chang F K (ed) 1999 Structural health monitoring: current status and perspectives Proc. 2nd Int. Workshop on Structural Health Monitoring (Stanford University, Sept. 1999) Chang F K (ed) 2001 Structural health monitoring: current status and perspectives Proc. 3rd Int. Workshop on Structural Health Monitoring (Stanford University, Sept. 2001) Chang F K (ed) 2003 Structural health monitoring: current status and perspectives Proc. 4th Int. Workshop on Structural Health Monitoring (Stanford University, Sept. 2003) Chang F K et al 2002 Proc. 1st European Workshop on Structural Health Monitoring (Ecole Normale Superior, Cachan (Paris), July 2002) Chen J 1996 Proc. 2nd Int. Workshop on Structural Control (Hong Kong University of Science and Technology, Hong Kong, Dec. 1996) FWSHM 2002 Proc. 1st European Workshop on Structural Health Monitoring (Ecole Normale Superieure de Cachan, Paris, July 2002) Golub G H and Van Loan C F 1983 Matrix Computations (Baltimore, MD: Johns Hopkins University Press) Helmicki A J, Aktan A E, Comfort L K and Kam M 1997 Information technology in civil infrastructure system monitoring: opportunities and issues Proc. Structural Health Monitoring 1997 ed F-K Chang (Lancaster, PA: Technomic Publishing) pp 396–409 Housner G W, Bergman L A, Caughey T K, Chassiakos A G, Claus R O, Masri S F, Skelton R E, Soong T T, Spencer B F and Yao J T P 1997 Structural control: past, present and future ASCE J. Eng. Mech. 123 897–971 (Special issue) Housner G W and Kobori T 1996 Proc. Joint US–Japan Workshop on Mitigation of Urban Disasters: Cooperative Research on Structural Control (Kyoto, Japan, March 1996) Housner G W and Masri S F (ed) 1990 Proc. US National Workshop on Structural Control Research (University of Southern California, CA, Oct. 1990) USC Publication No. M9013, ISBN 0-9628908-0-4 Housner G W and Masri S F (ed) 1993 Proc. Int. Workshop on Structural Control (Honolulu, Hawaii, Aug. 1993) USC Publication No CE-9311, ISBN 0-9628908-2-0 Housner G W and Masri S F 1996 Structural control research issues arising out of the Northridge and Kobe earthquakes Proc. 11th World Conf. on Earthquake Engineering (Acapulco, Mexico, June 1996) Housner G W, Masri S F, Casciati F and Kameda H (ed) 1992 Proc. US–Italy–Japan Workshop/Symp. on Structural Control and Intelligent Systems (Sorrento, July 1992) Housner G W, Masri S F and Chassiakos A G (ed) 1995 Proc. 1st World Conf. on Structural Control (Civil Engineering Department, University of Southern California, Los Angeles, CA, Aug. 1994) ISBN 0-9628908-3-9 IABMAS 2002 Proc. 1st Int. Conf. on Bridge Maintenance, Safety and Management (Barcelona, Spain, July 2002) ICANCEER 2002 Proc. Int. Conf. on Advances and New Challenges in Earthquake Engineering Research (Harbin and Hong Kong, Aug. 2002) IMAC 2004 IMAC XXII Conf.: Proc. Int. Modal Analysis Conf. (Society for Experimental Mechanics, Bethel, CT, 2004) Lin J W and Betti R 2003 On-line identification and damage detection in non-linear structural system using available forgetting factor approach Earthq. Eng. Struct. Dyn. 33 419–44 Lus H, Betti R and Longman R W 1999 Identification of linear structural systems using earthquake-induced vibration data Earthq. Eng. Struct. Dyn. 28 1449–67 Masri S F, Miller R K, Saud A F and Caughey T K 1987a Identification of nonlinear vibrating structures: Part I—Formulation ASME J. Appl. Mech. 54 918–22
Application of a Web-enabled real-time structural health monitoring system for civil infrastructure systems
Masri S F, Miller R K, Saud A F and Caughey T K 1987b Identification of nonlinear vibrating structures: Part II—Applications ASME J. Appl. Mech. 54 923–9 Masri S F, Nakamura M, Chassiakos A G and Caughey T K 1996 A neural network approach to the detection of changes in structural parameters ASCE J. Eng. Mech. 122 350–60 Natke H G, Tomlinson G R and Yao J T P (ed) 1993 Proc. Int. Workshop on Safety Evaluation Based on Identification Approaches Related to Time Variant and Nonlinear Structures (Lambrecht, Sept. 1992) (Weisbaden: Vieweg) ISBN 3-528-06535-4 Natke H G and Yao J T P (ed) 1988 Structural Safety Evaluation Based on System Identification Approaches (Braunschweig: Frieder Vieweg and Sohn) Nigbor R L and Diehl J G 1997 Two year’s experience using OASIS real-time remote condition monitoring system on two large bridges Proc. Structural Health Monitoring ed F-K Chang (Lancaster, PA: Technomic Publishing) pp 410–7 Pridham B A and Wilson J C 2001 Subspace identification of Vincent Thomas suspension bridge ambient vibration data Proc. 20th Int. Modal Analysis Conf. (Los Angeles) pp 596–604 Rohrmann R G, Said S and Schmid W 2003 Load and condition monitoring of the Putlitz bridge in Berlin-Moabit Proc. Symposium Topics from Civil and Bridge Engineering (BAM, Berlin, 2003) (in German) SHMISIS 2002 Proc. 1st Int. Workshop on Structural Health Monitoring of Innovative Civil Engineering Structures (Structural Health Monitoring Intelligent Sensing for Innovation Structures) (Winning, MB, Sept. 2002)
Smyth A and Pei J 2000 Integration of measured response signals for nonlinear structural health monitoring Proc. Workshop on Nonlinear Structural Control (University of Southern California, CA, 2000) Smyth A W and Masri S F 2001 Nonlinear system identification and structural health monitoring of bridges through the use of reduced-order models Proc. IABSE Conf. on Cable-Supported Bridges (Seoul, Korea, June 2001) Smyth A W, Masri S F, Caughey T K and Hunter N F 2000 Surveillance of mechanical systems on the basis of vibration signature analysis ASME J. Appl. Mech. 67 540–51 Smyth A W, Pei J-S and Masri S F 2003 System identification of the Vincent Thomas suspension bridge using earthquake records Earthq. Eng. Struct. Dyn. 32 339–67 SPIE 2004 Proc. Int. Society for Optical Engineering (SPIE), Smart Structures and Materials 2004 (San Diego, CA, March 2004) Szewezyk P and Hajela P 1994 Damage detection in structures based on feature-sensitive neural networks J. Comput. Civil Eng. Trans. ASCE 8 (2) 163–78 Wahbeh A M, Caffrey J P and Masri S F 2003 A vision-based approach for the direct measurement of displacement in vibrating systems Smart Mater. Struct. 12 785–94 Wolfe R W, Masri S F and Caffrey J C 2002 Some structural health monitoring approaches for nonlinear hydraulic dampers J. Struct. Control 9 (1) 5–18 Worden K and Tomlinson G R 2001 Nonlinearity in Structural Dynamics: Detection, Identification and Modeling (London: Institute of Physics) Wu Z S 2003 Proc. 1st Int Conf. on Structural Health Monitoring and Intelligent Infrastructure (Tokyo, Nov. 2003)
1283
