Smart Bridges

Adding sensor networks to infrastructure will make them cyberphysical systems

Loading the podcast player...

Steven Cherry: Hi, this is Steven Cherry for IEEE Spectrum’s “Techwise Conversations.”

Back in May, a truck on Interstate 5’s Skagit River bridge hit some overhead beams. The bridge collapsed, and 50 meters of it fell into the river, along with two other vehicles containing three people. Fortunately, no one died.

For a month, more than 70 000 vehicles a day faced a half-hour detour. Many didn’t bother, and nearby businesses lost up to 80 percent of their sales. Could technology have helped?

It might seem that bridge sensors, for example, would be powerless to help us when a truck clips an overhead beam at high speed. But it turns out that last year, as a local newspaper reported, bridge inspectors had “identified eight different points on the bridge that had high-load damage, including some portions in which components were deformed by the impact.” Also, last fall a truck had ripped an 8-centimeter gash in the bridge and tore off the surrounding paint.

Bridges seem to be collapsing around us nearly every day. At least some of those failures could be predicted in advance and averted with sensors that could be monitoring their structural worthiness. My guest today is just the person to tell us how sensors can help.

Chenyang Lu bridges two worlds, pun intended. He teaches computer science and engineering at Washington University in St. Louis; he’s an active member of both the ACM and the IEEE; and his research areas include real-time systems, wireless sensor networks, and what he calls “cyberphysical systems.” He joins us by phone.

Chenyang, welcome to the podcast.

Chenyang Lu: Hi, Steven. Thanks for having me.

Steven Cherry: First, could sensors have helped understand the ongoing pounding that bridge in Washington state was taking? And second, is this a common scenario for bridges?

Chenyang Lu: Well, certainly there’s this area of structural-health monitoring that has exactly focused on this area, this problem, of trying to use sensors to detect damages on structures such as bridges. There’s been a long history of studying this problem, and actually ranging all the way from airplanes to bridges. So, certainly, I think, of course we would have to take a close look at the specific problems, and certainly there’s potential for structural-health-monitoring technology to solve these problems, at least in certain cases.

Steven Cherry: So how do—what do the sensors measure? I mean, they could detect, I guess, the impact of a truck banging a beam, and do they also detect those losses of paint, for example?

Chenyang Lu: That’s a good question. So there’s a variety of sensors that could tell you more information about the structural integrity and damages on a bridge. So it ranges from strength gauges—so that would measure the strength of connection between joints—then there’s fiber-optic sensors that would be able to detect deformation of certain damages in the structures as well. And also there’s these general technologies, such as accelerometers, that efficiently measure the vibration of the bridges. Then you can use modal analysis, signal processing, essentially you are doing pattern recognition, and try to detect if there are changes in the vibration patterns in the bridge that would be indicative of damages, potentially.

Steven Cherry: So you’ve been modeling physical structures in software in the lab [PDF] and then looking at what sensors can tell you about them. How does that work?

Chenyang Lu: You mean, how to do the experiments?

Steven Cherry: Yes, the modeling that you do.

Chenyang Lu: Okay, so, certainly what you can do, and what we have done, is to take a structural model, you know, these are in the structural-engineering, civil-engineering labs, where there are—for example, the bridge in Washington state was a truss bridge, which is quite common. So we actually have done experiments on trusses in the lab, where, for example, there’s a reconfigurable truss prototype at UIUC [University of Illinois at Urbana-Champaign] belonging to my collaborators, so it’s actually, you can easily damage any segment of the truss intentionally, or you can replace them, like replace a thick one with a thinner one, so that would represent some kind of structural degradation.

And you can remove one if you want, and then you can run—and then you have these shake tables that would generate a bit of a vibration on these trusses, and then you have sensors deployed on these trusses that can run the algorithm and the software to detect the damage that you intentionally created and try to mimic, of course, the real-world damages to a certain degree. Then it goes all the way up to larger and larger structural specimens. Like, for example, there’s this full-scale highway truss that we used at Purdue University—that’s another civil-engineering collaborator that we work with—so there, it’s actually a fairly large truss that you can run similar experiments on.

Steven Cherry: So these models are based on real-world data that have been collected. I guess one such instance is the Jindo Bridge in South Korea. What’s so special about this bridge, and what have you learned from it?

Chenyang Lu: Yes, so first of all, I should be clear that that whole project is really done by my collaborators at UIUC, so these are Professor Bill Spencer with civil engineering and Gul Agha in computer science. So they have been working with their Korean collaborators. Jindo Bridge is a relatively new bridge; it’s a very large bridge. Jindo is an island in South Korea, so they are basically connecting the island with the peninsula, you know, the mainland of South Korea. So they did deploy hundreds of sensors on that bridge. So the unique thing about their system, and the kind of research I do here at Washington University in general, is these are wireless sensor networks that are much easier to deploy onto the smart structures. And this is especially meaningful to America, actually, because you have all of these old bridges, and you have to essentially retrofit a monitoring system onto these bridges. So that’s what the UIUC team did, and they deployed this system over multiple years in Korea, where they continuously monitor the vibrations of the bridge, including cases where there are typhoons and so on blowing by.

Steven Cherry: So I gather the sensors themselves are fairly sophisticated now, and in effect the limitation of the system is the network itself, that is, getting the data from the sensors back into the lab or whatever. So tell us about the networks themselves.

Chenyang Lu: Sure. So, traditionally—so we build a new bridge, so, for example, the bridge incident before the Washington state one was, of course, the I-35 bridge in Minneapolis, right? So when that bridge collapsed, they built a new one, so they were able to make one out of wired sensors onto that bridge. So there are two downsides of the wired networks. First is, when you really have severe damages—think about an earthquake or some other—the bridge is really suffering very severe deformation, and the wires tend to break. And then the structural-health-monitoring system stops working when it is most needed. So that’s one downside.

The other downside, of course, is that it’s really expensive to install wires into these structures, into the old structures, right, because you have to drill holes and there’s a lot of engineering that needs to be done in order to put wires into these systems. So that’s why in more recent years we’ve been working more on wireless structural-health monitoring, which is a very natural fit. So basically you have these wireless sensors running on these low-power wireless networks, right? So there are a variety of technologies. For example, you can take this technology called IEEE 802.15.4 that’s for sensor networks running at very low power, then you form a mesh network.

Steven Cherry: So besides the sort of ongoing traffic across the bridge, there are specific events you mentioned: earthquakes and storms. There are also rampaging trucks hitting into beams. It seems like in creating a sensor network you would use the data that involves these events, but it seems like a chicken-and-egg problem, right? You need to know a lot about the events in order to even know how to deploy the sensors, and you need the sensor data to develop that knowledge, so how do you break out of that?

Chenyang Lu: Well, that’s exactly why you need experiments in a lab. You need simulation tools that try to, you know, study these type of problems. For example, one of the research projects we did, so we developed this tool that we call WCPS— Wireless Cyber-Physical Simulator—where one of the things we did was exactly to take these bridge models, so the particular thing we did was this Emerson Memorial Bridge over the Mississippi River at Cape Girardeau, Missouri.

So we take the fairly high-fidelity structural models of that bridge, and then we implement it in software so that we can simulate the impact of different events. So one of the things we did, for example, we took one of the traces from the—essentially the earth vibration—from the earthquake in California, then we replay that trace onto these simulation models. And then that would allow us to create the impact in software and then study it accordingly.

Steven Cherry: So it seems like a very manual, and time-consuming, and ultimately expensive process to first model a bridge and then to deploy a network. How expensive would it be to put sensors on all of the major bridges in, you know, in the country?

Chenyang Lu: Actually, you know, that’s one of the major motivations of using wireless networked structural-health-monitoring systems, is to reduce the cost. So these sensors, sensors have this whole different range, so in general, for example, the acceleration-based, model-analysis-based approach, actually the sensors are not expensive. You know, these are essentially relatively good accelerometers, which are generally, you know, quite cheap. And then they run on these low-power wireless radio interfaces, and those are pretty cheap as well. And then again, the big benefit is, it’s really easy to deploy. So you essentially attach them to the right parts of the structures and then turn them on.

Steven Cherry: And I guess there are some savings as well, right? I mean, first of all, it costs money to inspect bridges, and maybe that would be reduced, but more importantly we can repair a bridge early on, when it’s less expensive to.

Chenyang Lu: Right, so there are two major benefits of these kinds of structural-health-monitoring systems, based on wireless networks that essentially are always on. Think of these as real-time monitoring systems. So the one-time benefit is, it can now do these routine inspections, right? So right now the current regulations mandate these bridges to be inspected once every two years manually. So that’s a very low frequency, and bad things can obviously happen within the two years. And it’s very labor-intensive, because actually it’s a manual process. You send teams of engineers up into the bridge; they essentially look up and down the bridge and try to visually detect damages. Sometimes they have some newer tools, like X-ray and ultrasound that help out a little bit. So the frequency is low, because the labor cost is high.

And the wireless structural-health-monitoring system, on the other hand, so because it’s always there, you can look at the bridge anytime you want. You can periodically monitor it. You can do this once every day. You can do it even more often if you want to. So as you said, it would allow you to detect damages much earlier that would make it not only safer but also much cheaper to fix. So the second benefit is during impacts, such as an earthquake. So first, of course, during the earthquake, because these systems can be turned on on demand, so you would essentially monitor the integrity of these structures in real time. If bad things happen, you can automatically close down the highways and bridges on demand. And then what’s also important is what happens after the earthquake, for example. So in some cases I’ve seen reports that after an earthquake it took them two years to inspect a structure, a building, and so on, before they can allow people to move back in, right? So these are residences, but sometimes manufacturing facilities as well, so that is a very, very costly consequence.

Steven Cherry: Now, one issue sometimes in sensor networks is the amount of data that they throw off can be kind of overwhelming. Is that ever a problem here?

Chenyang Lu: That’s certainly a very important problem. And that’s why it’s never enough just to put sensors in, networks in, and collect a ton of data. Indeed, I have certainly heard incidents where they have these brand-new bridges with a lot of sensors, and they accumulate a ton of data but don’t know what to do with them. So that’s why there’s this—you have to have this intelligence that’s built into these systems. It’s not just collecting data, but also through all these sophisticated model analyses, and then you actually not only collect data but also detect damages, localized damages. Recently, here, there’s this cutting-edge research on prognosis. So not only that, you assess the current state of the bridge, you also try to predict what is the remaining lifetime of this bridge.

Steven Cherry: So once enough bridges are wired up, and there’s enough data from them, can that data be mined also to help in the design of new bridges? Is there any information for the civil engineers, for example?

Chenyang Lu: I would anticipate that to be very helpful, because I do think they don’t have enough data. So they design a bridge, they don’t know the future loading situation on that bridge, and they—I’ve seen in the reports when you design a bridge you claim that this is going to last for 50 years, this is going to last for 100 years. There’s really no—a very rough guess, because they simply don’t have any information once the bridge is deployed into the field. They don’t know the loading. They don’t know the wear and tear. They don’t know the most severe impacts over time. And they don’t know the weather conditions as well. So I think by having this new data, indeed it can be a sent back to the civil engineers for much better science and engineering for bridge design.

Steven Cherry: And I guess in the absence of a lot of confidence, engineers have just overbuilt these bridges, and that just makes them very expensive.

Chenyang Lu: Exactly. So one thing I’ve seen people claiming that—so there’s a lot of reports that say, “Many of our bridges have exceeded the 50-year design lifetime of these structures.” But in reality it really varies. You know, some bridges can still last much longer beyond that 50-year lifespan, while some others need to be fixed right now. So that’s why you need this much better science and measurements and analysis to know what is the real status and remaining lifetime of these bridges.

Steven Cherry: Well, Chenyang, I guess our cars are becoming—they’re up to about 30 percent software in terms of value. It had not occurred to me, but I guess bridges will soon be significantly software based. Thanks for being part of that movement, and thanks for joining us today.

Chenyang Lu: Thank you.

Steven Cherry: We’ve been speaking with Chenyang Lu about why it’s so hard to build sensor networks that will report on the health of today’s bridges and other infrastructure.

For IEEE Spectrum’s “Techwise Conversations,” I’m Steven Cherry.

Photo: Stephen Brashear/Getty Images

This interview was recorded Tuesday, 17 July 2013.
Segment producer: Barbara Finkelstein; audio engineer: Francesco Ferorelli

Read more “Techwise Conversations,” find us in iTunes, or follow us on Twitter.

NOTE: Transcripts are created for the convenience of our readers and listeners and may not perfectly match their associated interviews and narratives. The authoritative record of IEEE Spectrum’s audio programming is the audio version.