WEBVTT Kind: captions Language: en-US 00:00:01.171 --> 00:00:02.664 [silence] 00:00:02.664 --> 00:00:07.749 I’m making my semi-annual appearance as seminar overlord, 00:00:07.750 --> 00:00:11.670 and I’m going to make this quick. We’re having a transition 00:00:11.670 --> 00:00:16.813 in seminar chairs, so I first wanted to thank Tim and Leah, who have hosted seminar 00:00:16.813 --> 00:00:20.119 for the last six or seven months, for all their hard work and actually 00:00:20.119 --> 00:00:24.313 starting some of the hybrid seminars back in person 00:00:24.313 --> 00:00:28.980 and hosting five different additional seminars in the last two weeks. 00:00:28.980 --> 00:00:33.020 So thank you very much to them for taking this on. 00:00:33.020 --> 00:00:38.500 And our two new seminars chairs are Shanna Chu, who is a Mendenhall here 00:00:38.500 --> 00:00:42.563 at Moffett Field, and Evan Hirakawa, who was a Mendenhall and 00:00:42.563 --> 00:00:48.640 is now a full-time research geophysicist. So I’m going to pass it off to them. 00:00:48.640 --> 00:00:51.750 Thanks to them for taking on this responsibility. 00:00:51.750 --> 00:00:53.164 Take it away. 00:00:53.164 --> 00:00:54.920 - All right. Thanks, Justin. 00:00:54.920 --> 00:01:00.400 So welcome to the Earthquake Science Center seminar for September 7th. 00:01:00.400 --> 00:01:06.000 As a reminder, turn off your cameras and mute your microphones today. 00:01:06.000 --> 00:01:11.119 And live captioning is available, which is available on the top tab 00:01:11.119 --> 00:01:15.688 under the More tab with the three dots there. 00:01:17.740 --> 00:01:21.929 Only announcement that I have is that there’s no seminar next week 00:01:21.929 --> 00:01:24.751 because of the SCEC meeting. 00:01:25.079 --> 00:01:29.610 And now today our speaker is Ethan Williams from Caltech. 00:01:29.610 --> 00:01:32.880 And I’ve been told that he will actually accept questions 00:01:32.880 --> 00:01:36.780 in the middle of his talk. Just try to raise your hand. 00:01:36.780 --> 00:01:41.938 And now I’m going to pass it off to Lisa Schleicher to introduce. 00:01:43.055 --> 00:01:46.979 - Hey, everybody. I’m excited to introduce to you Ethan Williams. 00:01:46.979 --> 00:01:50.314 He’s a graduate student at Caltech Seismological Laboratory 00:01:50.314 --> 00:01:55.200 working with Zhongwen Zhan. Previously, he received his B.S. in geophysics 00:01:55.200 --> 00:01:58.470 from Stanford University in 2017. 00:01:58.470 --> 00:02:03.689 And Ethan’s research primarily focuses on fiber optic sensing of the seafloor 00:02:03.689 --> 00:02:08.209 with diverse applications from seismic tomography to monitoring 00:02:08.209 --> 00:02:10.369 ocean waves and currents. 00:02:10.369 --> 00:02:15.689 And, under the guidance of Tom Heaton, he has recently forayed into engineering 00:02:15.689 --> 00:02:20.877 seismology, leveraging the so-called large-N and large-T data sets for 00:02:20.877 --> 00:02:24.470 structural – seismic structural health monitoring and site characterization, 00:02:24.470 --> 00:02:28.480 which is the subject of today’s talk. And I’m presenting Ethan just because 00:02:28.480 --> 00:02:33.659 I happen to run into him at SSA and was excited to see his working with 00:02:33.659 --> 00:02:36.290 the Millikan Building, and the Strong Motion Project 00:02:36.290 --> 00:02:39.340 has a seismic array there. And, although he’s the one – 00:02:39.340 --> 00:02:42.400 the one streaming station there, so you’ll see more about that. 00:02:42.400 --> 00:02:46.377 I was just super curious to talk to him about kind of some of the research 00:02:46.377 --> 00:02:50.069 interests out there in structures and, you know, the kind of things people 00:02:50.069 --> 00:02:53.030 want to do these days in engineering seismology. 00:02:53.030 --> 00:02:56.072 And we also had a lot of great discussions on DAS, 00:02:56.072 --> 00:02:58.431 and I knew people in our group here were interested in that. 00:02:58.431 --> 00:03:01.280 So, with that, Ethan, take it away. 00:03:01.280 --> 00:03:04.376 I’m looking forward to discussion. 00:03:04.376 --> 00:03:07.269 - Thanks so much, Lisa. Thank you for the invitation. 00:03:07.269 --> 00:03:08.909 I’m really happy to give this seminar today. 00:03:08.909 --> 00:03:11.909 Before I get started, I want to thank the collaborators 00:03:11.909 --> 00:03:16.970 that have put a lot of work into this. This is really the combination of two 00:03:16.970 --> 00:03:18.829 different talks on two different subjects. 00:03:18.829 --> 00:03:22.769 And so on the first – has been working with Tom Heaton who’s had 00:03:22.769 --> 00:03:26.170 a longstanding interest in the – it’s now called Caltech Hall, 00:03:26.170 --> 00:03:29.680 but formerly Millikan Library on the Caltech campus, 00:03:29.680 --> 00:03:32.379 which is a structural engineering test bed. 00:03:32.379 --> 00:03:35.065 And then secondarily, we’ve had a lot of support from people at 00:03:35.065 --> 00:03:39.579 the Universidad de Alcalá in Madrid and Marlinks, a cable company in Belgium, 00:03:39.579 --> 00:03:44.310 in acquiring interesting engineering- related fiber optic data in Europe. 00:03:44.310 --> 00:03:47.409 So thanks to them. 00:03:47.409 --> 00:03:50.315 The motivation for today’s talk is pretty simple, 00:03:50.315 --> 00:03:54.069 at least as I’m – kind of thematically organized it. 00:03:54.069 --> 00:03:56.459 Data is exploding in contemporary seismology. 00:03:56.459 --> 00:04:01.319 There was just a fantastic review out on big-data seismology. 00:04:01.319 --> 00:04:07.019 And, if you just look at the size of the IRIS archive, which is just one of the, 00:04:07.019 --> 00:04:10.129 you know, modern seismological data archives, you can see that there’s 00:04:10.129 --> 00:04:15.440 near exponential growth of seismic research data in the last decade or so. 00:04:15.440 --> 00:04:17.300 And then has enabled some really amazing things. 00:04:17.300 --> 00:04:21.410 The so-called large-N sensing, especially with nodal arrays, 00:04:21.410 --> 00:04:23.710 where you have many sensors, and therefore, you’re able to 00:04:23.710 --> 00:04:27.850 see things in unprecedented resolution, like this example recently from 00:04:27.850 --> 00:04:30.870 Daniel Trugman in Oklahoma looking at the radiation pattern 00:04:30.870 --> 00:04:35.160 from a small earthquake, completely unaliased. 00:04:35.160 --> 00:04:39.878 Or large-T applications, now that we have decades of continuous digital 00:04:39.878 --> 00:04:43.970 seismographic data, we can do things like reconstruct groundwater levels 00:04:43.970 --> 00:04:48.128 using seismic velocity perturbations over long terms in order to 00:04:48.128 --> 00:04:52.800 estimate long-term changes in groundwater storage. 00:04:52.800 --> 00:04:55.710 But unfortunately, the structural and geotechnical communities 00:04:55.710 --> 00:04:59.539 are really lagging behind. If you look at this IRIS plot here 00:04:59.539 --> 00:05:03.440 on the left, and you look at this little purple – whoops, let me switch 00:05:03.440 --> 00:05:08.120 to the laser pointer – this little purple, or lavender, strip, 00:05:08.120 --> 00:05:11.979 that’s the volume of engineering data. 00:05:11.979 --> 00:05:14.800 And admittedly, while a lot of the engineering data isn’t archived by IRIS, 00:05:14.800 --> 00:05:17.680 you can see that it’s only growing about linearly. 00:05:17.680 --> 00:05:20.380 Whereas, the rest of the data is growing about exponentially. 00:05:20.380 --> 00:05:23.128 And one of the reasons for that is that, even in the U.S., 00:05:23.128 --> 00:05:25.753 but especially internationally, it’s still very common for 00:05:25.753 --> 00:05:27.680 strong motion networks to be triggered. 00:05:27.680 --> 00:05:31.379 K-NET and KiK-NET, the two largest strong motion networks in the world, 00:05:31.379 --> 00:05:34.920 are still only providing triggered waveforms. 00:05:34.920 --> 00:05:39.816 Unfortunately, that means that you might spend thousands and thousands of dollars 00:05:39.816 --> 00:05:43.816 deploying hundreds and hundreds of stations, but over a span of decades, 00:05:43.816 --> 00:05:47.370 each of those stations might only record a few minutes of data. 00:05:47.370 --> 00:05:51.816 And so the engineering community, kind of state-of-the-art in terms of 00:05:51.816 --> 00:05:57.979 monitoring structures for damage in particular, it really falls behind 00:05:57.979 --> 00:06:00.120 where we are in seismology, and that’s unfortunate. 00:06:00.120 --> 00:06:03.629 And my case in today’s talk is really just to say there are frontiers 00:06:03.629 --> 00:06:08.004 in large-N and large-T seismology for engineering and structural dynamics that 00:06:08.004 --> 00:06:12.500 are at least as exciting and maybe a little bit easier to do because of the, you know, 00:06:12.500 --> 00:06:17.620 limited scale of urban infrastructure compared to the planet Earth. 00:06:17.620 --> 00:06:21.590 And so I’m hoping that there’ll be some interest from people here, 00:06:21.590 --> 00:06:25.240 like the National Strong Motion Project group, in working towards 00:06:25.240 --> 00:06:30.210 an implementation of more big-data-friendly approaches. 00:06:30.210 --> 00:06:33.490 And so today’s talk is in two parts. The first part focuses on this 00:06:33.490 --> 00:06:36.920 large-T idea – long time series. So I’m going to look at 20 years 00:06:36.920 --> 00:06:41.120 of continuous data from a single strong motion station in a building. 00:06:41.120 --> 00:06:44.520 And, among other things, this allows us to test the question, 00:06:44.520 --> 00:06:47.340 is elasticity in structures really linear- and time-invariant? 00:06:47.340 --> 00:06:51.610 And the answer is no, not even a little bit, to spoil the conclusion. 00:06:51.610 --> 00:06:53.509 And then, for the second half, we’re going to be talking about 00:06:53.509 --> 00:06:56.442 distributed acoustic sensing, which I know some people at the USGS 00:06:56.442 --> 00:07:00.363 have started exploring for earthquake early warning and other applications. 00:07:00.363 --> 00:07:04.419 And I’m going to talk about looking at structural vibrations in the far-field. 00:07:04.419 --> 00:07:08.020 Kind of a novel observation. The seismic waves that are actually 00:07:08.020 --> 00:07:11.319 radiated away from structures when structures vibrate. 00:07:12.535 --> 00:07:14.819 So, on to the first part. 00:07:15.824 --> 00:07:18.050 The subject of this study is Caltech Hall. 00:07:18.050 --> 00:07:22.220 It was called the Robert Millikan Memorable Library up until last year. 00:07:22.220 --> 00:07:27.020 And it was constructed in 1967, and it’s a nine-story reinforced 00:07:27.020 --> 00:07:31.509 concrete building. And, for its size, it’s really quite a stiff building. 00:07:31.509 --> 00:07:35.800 It has two 30-centimeter-thick reinforced concrete shear walls, 00:07:35.800 --> 00:07:38.692 which are on the east and the west faces of the building, 00:07:38.692 --> 00:07:42.850 represented here in the structural diagram. That’s the face here that you’re seeing. 00:07:44.317 --> 00:07:49.443 And it’s got a reinforced concrete moment frame, and it’s got a very thick 00:07:49.443 --> 00:07:53.970 reinforced concrete central elevator core and stairwell. 00:07:53.970 --> 00:07:58.150 And so the building – naturally, because of 00:07:58.150 --> 00:08:03.020 the asymmetric shear wall design, has asymmetric stiffness. 00:08:03.020 --> 00:08:05.460 We can think about the deformation of a building like this, 00:08:05.460 --> 00:08:08.860 in terms of three simple contributions. 00:08:08.860 --> 00:08:10.930 There is rigid body translation 00:08:10.930 --> 00:08:13.530 of the building, when it undergoes some shaking. 00:08:13.530 --> 00:08:18.278 There’s the solid-body rocking, meaning – or rigid-body rocking, 00:08:18.278 --> 00:08:23.319 meaning that you can imagine the foundation is basically tilting, 00:08:23.319 --> 00:08:25.039 or rotating about an axis. 00:08:25.039 --> 00:08:28.068 And then there’s the fixed-based shearing, where you can assume that 00:08:28.068 --> 00:08:31.289 there’s some drift of the upper floors relative to the base. 00:08:31.289 --> 00:08:34.255 And, because of the shear wall design, meaning that the north-south direction 00:08:34.255 --> 00:08:38.110 is much more resistant to shear than the east-west direction, 00:08:38.110 --> 00:08:40.979 the north-south motions more approximate rigid-body rocking, 00:08:40.979 --> 00:08:43.279 which means that they’re much more sensitive to the soil properties 00:08:43.279 --> 00:08:47.300 because large deformations of the structure involve large deformations 00:08:47.300 --> 00:08:50.680 of the soil at the foundation level. Whereas, the east-west direction 00:08:50.680 --> 00:08:55.810 is much softer, and the motions approximate fixed-base shearing. 00:08:55.810 --> 00:08:59.779 So just keep those two ideas in mind, that this building is really testing 00:08:59.779 --> 00:09:02.100 two different things in two different directions, 00:09:02.100 --> 00:09:04.790 has different sensitivity in each direction. 00:09:04.790 --> 00:09:07.670 But this building has been studied for a very long time. 00:09:07.670 --> 00:09:11.881 You know, Caltech had civil engineering going in 1967 when it was built, 00:09:11.881 --> 00:09:15.560 and they monitored the building’s stiffness and did actually a full-scale 00:09:15.560 --> 00:09:20.510 forced vibration test before the building was even completed in 1967. 00:09:20.510 --> 00:09:22.990 And the goal of that kind of study – of forced vibrations tests, 00:09:22.990 --> 00:09:26.410 or an ambient vibration test – in the structural health monitoring zone 00:09:26.410 --> 00:09:29.680 is to identify the natural frequency of a building. 00:09:29.680 --> 00:09:33.420 Engineers like to think of buildings as lumped systems of harmonic oscillators. 00:09:33.420 --> 00:09:36.940 And, in that context, the natural frequency, f, 00:09:36.940 --> 00:09:39.480 is proportional to the square root of the stiffness. 00:09:39.480 --> 00:09:43.260 And so it’s basically a proxy for the elastic state of the building. 00:09:43.260 --> 00:09:47.820 And, because it’s been monitored for so long, both through dedicated studies – 00:09:47.820 --> 00:09:51.740 there’s over 50 papers on this building – and also because it’s been a test bed 00:09:51.740 --> 00:09:53.980 for civil engineering classes at Caltech, 00:09:53.980 --> 00:09:56.180 there’s a long record of detailed knowledge. 00:09:56.180 --> 00:09:59.881 And so, when this building was first built in 1967, it was much stiffer 00:09:59.881 --> 00:10:04.310 than it is today. It was damaged in the San Fernando earthquake. 00:10:04.310 --> 00:10:06.640 People don’t really know how or what happened. 00:10:06.640 --> 00:10:09.757 There’s actually quite a large debate in literature over whether or not 00:10:09.757 --> 00:10:12.550 we should think that the foundation slab is cracked. 00:10:12.550 --> 00:10:14.334 Because you can’t actually see the foundation slab, 00:10:14.334 --> 00:10:15.730 so people have fought back and forth 00:10:15.730 --> 00:10:18.632 over various seismic structural health monitoring approaches, 00:10:18.632 --> 00:10:23.970 trying to determine whether there’s a change in foundation flexibility. 00:10:23.970 --> 00:10:28.070 But the general idea is that it experienced large motions 00:10:28.070 --> 00:10:30.069 during the San Fernando earthquake. There was a dramatic 00:10:30.069 --> 00:10:34.330 reduction in stiffness. And that reduction was permanent. 00:10:34.330 --> 00:10:37.130 Then we believe that the building was relatively stable across 00:10:37.130 --> 00:10:41.380 The ’70s and the ’80s, but there was less study during that period. 00:10:41.380 --> 00:10:45.819 And, in the subsequent moderate-sized earthquakes in southern California 00:10:45.819 --> 00:10:48.700 throughout the ’80s – Whittier Narrows, Sierra Madre, Northridge – 00:10:48.700 --> 00:10:53.459 the building’s natural frequencies dropped a little bit each time. 00:10:53.459 --> 00:10:57.640 Not enough to really indicate damage, but maybe something happens 00:10:57.640 --> 00:10:59.380 like basically a jackhammer effect. 00:10:59.380 --> 00:11:03.132 You know, you have 10%g in the building, and it basically compacts the soil, 00:11:03.132 --> 00:11:06.249 the foundation, a little bit and changes the bulk stiffness. 00:11:06.249 --> 00:11:08.829 Those are the kind of ideas that have been proposed to describe this. 00:11:08.829 --> 00:11:13.400 And then, up until the present, this is basically the complete record. 00:11:13.400 --> 00:11:17.642 Because, historically, the building only had triggered stations in it. 00:11:17.642 --> 00:11:21.870 And so what you’re basically looking at is a – is a synthesized history based on 00:11:21.870 --> 00:11:26.220 only about 20 or so strong motion records. 00:11:27.910 --> 00:11:31.000 What we can do – let’s go forward. 00:11:31.000 --> 00:11:35.220 What we can do, though, is improve that with continuous monitoring. 00:11:35.220 --> 00:11:39.757 So, in 2001, Southern California Seismic Network installed a station 00:11:39.757 --> 00:11:44.040 on the 9th floor of Caltech Hall called CI.MIK. 00:11:44.040 --> 00:11:46.616 And it’s an EpiSensor. And it has some 00:11:46.616 --> 00:11:49.991 other instrumentation associated with it, which I won’t discuss today. 00:11:49.991 --> 00:11:54.089 There’s some rotational sensors and other things in the building that are interesting. 00:11:54.089 --> 00:11:58.445 But just this one basic normal EpiSensor has been recording continuously 00:11:58.445 --> 00:12:01.210 since May 2001. And so, instead of looking at 00:12:01.210 --> 00:12:05.041 a scattered record of ambient vibration, forced vibration, 00:12:05.041 --> 00:12:08.320 and strong motion records, we can look at a continuous record 00:12:08.320 --> 00:12:12.920 of ambient vibrations over a 20-year period in order to track 00:12:12.920 --> 00:12:17.079 how the building is really changing in between these large earthquakes. 00:12:17.079 --> 00:12:20.470 And the answer is, if we compare here, May 2001 to May 2021, 00:12:20.470 --> 00:12:22.079 that there have been substantial changes in the building 00:12:22.079 --> 00:12:25.480 despite no damaging earthquakes occurring in the past 20 years. 00:12:25.480 --> 00:12:29.070 So, if we look on the top here, this is the response spectrum for 00:12:29.070 --> 00:12:32.579 the east-west component. You see a whole lot of things going on. 00:12:32.579 --> 00:12:37.790 All of these little spikes relate to motors in the building or residences related to, 00:12:37.790 --> 00:12:45.380 like, the elevator or A/C system and are not seismological in origin. 00:12:45.380 --> 00:12:49.540 But these main big peaks here – this is the East-West 1 fundamental mode. 00:12:49.540 --> 00:12:52.660 This is the torsional mode – basically rotation of the building 00:12:52.660 --> 00:12:55.889 about its central axis. 00:12:55.889 --> 00:12:59.329 This is the overtone, and then there’s a torsional overtone. 00:12:59.329 --> 00:13:04.560 And then the same thing – sort of thing we start to see on the other component, 00:13:04.560 --> 00:13:07.279 just with a different set of modes. And these have been extensively 00:13:07.279 --> 00:13:11.389 identified using forced vibration tests to really eke out what the mode shape 00:13:11.389 --> 00:13:14.700 is and be sure that these are really the true modal frequencies. 00:13:14.700 --> 00:13:17.470 But if we just kind of summarize here briefly, you can see that 00:13:17.470 --> 00:13:22.190 this second torsional overtone has gone down in frequency. 00:13:22.190 --> 00:13:25.870 You can see that the shape of this east-west overtone 00:13:25.870 --> 00:13:27.220 is really dramatically changed. 00:13:27.220 --> 00:13:30.060 It’s hard to really track that over the last 20 years. 00:13:30.060 --> 00:13:32.884 And you can see that, not only do we have more peaks down here 00:13:32.884 --> 00:13:37.440 by the fundamental frequency, but the fundamental frequency has gone up. 00:13:37.440 --> 00:13:41.721 And that’s true for both the north-south and east-west directions. 00:13:42.899 --> 00:13:46.550 If we look at that in detail, you can see that increase here from black to red. 00:13:46.550 --> 00:13:48.259 And you can see that there’s this additional frequency. 00:13:48.259 --> 00:13:52.071 There’s both crosstalk between the two components and beading between 00:13:52.071 --> 00:13:55.690 the two components. And that could be a result of 00:13:55.690 --> 00:13:58.630 a seismometer installation issue, to be quite frank, right? 00:13:58.630 --> 00:14:00.821 The EpiSensor could be rocking back and forth, and that’s why we see 00:14:00.821 --> 00:14:04.220 the harmonic mean of these two frequencies and increased crosstalk. 00:14:04.220 --> 00:14:08.009 But it could also mean that there’s a redistribution of mass inside the building 00:14:08.009 --> 00:14:11.240 or that there’s some sort of damage that’s causing basically 00:14:11.240 --> 00:14:15.480 coupling between the two modal systems and increased torsion. 00:14:15.480 --> 00:14:17.884 But this fundamental – regardless of how you want to interpret that, 00:14:17.884 --> 00:14:23.080 which we really can’t, with one station, say anything more definitive about, 00:14:23.080 --> 00:14:25.634 the fundamental observation here is that, over the last 20 years, the building 00:14:25.634 --> 00:14:29.920 has had a 5.1% increase in frequency in the east-west direction, 00:14:29.920 --> 00:14:33.322 and a 2.3 increase in the north-south direction, which, when recalling 00:14:33.322 --> 00:14:35.759 the frequency is proportional to the square root of stiffness, 00:14:35.759 --> 00:14:39.399 means that the stiffness has increased by up to about 10%. 00:14:39.399 --> 00:14:41.920 Which is surprising. Why is their long-term healing going on? 00:14:41.920 --> 00:14:43.884 This is the opposite of what we expect to be happening 00:14:43.884 --> 00:14:46.200 in reinforced concrete structures over their lifetime. 00:14:46.200 --> 00:14:49.880 You know, there’s many known mechanisms of loss of stiffness. 00:14:49.880 --> 00:14:52.290 There’s chemical carbonation 00:14:52.290 --> 00:14:55.810 and chlorination reactions that can cause a degradation of concrete. 00:14:55.810 --> 00:14:58.230 There’s thermal stressing, which can lead to surface spalling 00:14:58.230 --> 00:15:01.190 and fracture formation. Of course, small earthquakes 00:15:01.190 --> 00:15:03.839 can cause the growth of pre-existing fractures. 00:15:03.839 --> 00:15:07.940 And rebar inside reinforced concrete can corrode over time, too, 00:15:07.940 --> 00:15:10.100 especially if it’s exposed to water. 00:15:10.100 --> 00:15:13.385 So we suspect that reinforced concrete buildings should be getting 00:15:13.385 --> 00:15:17.260 softer and softer over their lifetimes, but yet here Caltech Hall 00:15:17.260 --> 00:15:21.535 appears to be getting stiffer and stiffer, at least over the last 20 years. 00:15:22.970 --> 00:15:27.885 Since we have a continuous record and not just a couple of triggered motions, 00:15:27.885 --> 00:15:32.019 we can actually map this out on a detailed way from one week to the next. 00:15:32.019 --> 00:15:35.760 So this is the weekly medians power spectrum, zoomed into 00:15:35.760 --> 00:15:39.240 these two fundamental frequencies – the east-west and north-south directions. 00:15:39.240 --> 00:15:43.323 And we can see that actually that just comparison – 2001 to 2021 – 00:15:43.323 --> 00:15:50.736 even still oversimplifies what’s going on. The building exhibits 9.7% variation 00:15:50.736 --> 00:15:54.350 in the fundamental frequency over the last 20 years, meaning that 00:15:54.350 --> 00:15:58.600 there’s about 20% variation in stiffness just passively going on. 00:15:58.600 --> 00:16:03.020 And we can see that some of this has clear associations. 00:16:03.020 --> 00:16:07.135 There were some minor renovations that perplexingly were supposed to be 00:16:07.135 --> 00:16:10.930 completely non-structural renovations in 2003 and 2004. 00:16:10.930 --> 00:16:14.885 These were the removal of library books from three floors of the building, 00:16:14.885 --> 00:16:19.010 and instead, they installed office partitions, which aren’t attached to 00:16:19.010 --> 00:16:22.170 the moment frame, and they only abut the false ceiling. 00:16:22.170 --> 00:16:25.011 They don’t actually connect to any of the structural components, 00:16:25.011 --> 00:16:27.990 and they’re just wood and plastic. 00:16:27.990 --> 00:16:31.880 So why there’d be as much as a 2 to 3% increase in the natural frequencies 00:16:31.880 --> 00:16:35.209 of the building during those construction events is a complete mystery 00:16:35.209 --> 00:16:38.198 since none of the major structural elements were involved, 00:16:38.198 --> 00:16:42.460 and the mass of the building didn’t change substantially as far as we can tell. 00:16:42.460 --> 00:16:46.136 Following each of those events, the building initiated some sort of 00:16:46.136 --> 00:16:50.770 softening trend, which is still yet perplexing because you expect that, 00:16:50.770 --> 00:16:54.011 if you add structural elements to increase the stiffness of the building, 00:16:54.011 --> 00:16:56.511 they’re bonded together with things like adhesive and plasters, 00:16:56.511 --> 00:17:01.430 and those generally stiffen with time over the months to years after application. 00:17:01.430 --> 00:17:04.800 So, for whatever reason – I don’t have a good explanation here – 00:17:04.800 --> 00:17:09.948 the building started softening after each of these construction events. 00:17:10.640 --> 00:17:12.460 Then we have the Chino Hills earthquake, and unfortunately 00:17:12.460 --> 00:17:18.472 there’s a data gap, but something fundamentally changed between 00:17:18.472 --> 00:17:23.501 2008 and 2015 or so, and both systems – the east-west direction and 00:17:23.501 --> 00:17:25.770 the north-south direction – started increasing at a rate 00:17:25.770 --> 00:17:34.580 of about 0.01 hertz per year, which is interesting because 00:17:34.580 --> 00:17:37.840 it’s pretty much a steady-state healing process, right, that there’s no significant 00:17:37.840 --> 00:17:42.191 change over this decade-long period in the rate of increasing stiffness. 00:17:42.191 --> 00:17:44.980 And, again, I don’t have a good explanation for that. 00:17:44.980 --> 00:17:48.750 With the single-station record, unfortunately we can’t decompose 00:17:48.750 --> 00:17:51.640 contributions from soil structure interaction of the foundation 00:17:51.640 --> 00:17:54.370 and from the actual superstructure itself. 00:17:54.370 --> 00:17:57.980 But clearly, there are very complicated long-term dynamics going on. 00:17:57.980 --> 00:17:59.910 What we can see, though, is that actually there was 00:17:59.910 --> 00:18:01.640 a significant response to the Ridgecrest earthquake. 00:18:01.640 --> 00:18:06.780 During the Ridgecrest earthquake, the peak accelerations in the building 00:18:06.780 --> 00:18:13.020 were about 70 centimeters per second for both the 6.4 and 7.1 events. 00:18:13.020 --> 00:18:18.570 And that actually led to about a 2% drop in the fundamental frequencies. 00:18:18.570 --> 00:18:22.449 So, looking at everything, you can also see, if you zoom in, 00:18:22.449 --> 00:18:25.880 in addition to those really large discrete events and long-term secular trends 00:18:25.880 --> 00:18:31.150 that are hard to explain because they’re quite large changes compared to 00:18:31.150 --> 00:18:33.730 the causative mechanisms that are considered, 00:18:33.730 --> 00:18:37.349 we also have these smaller variations, which are seasonal. 00:18:38.130 --> 00:18:41.887 We can see that the east-west mode and also the torsional mode 00:18:41.887 --> 00:18:45.480 increased abruptly during periods of rainfall, like the example you can see here 00:18:45.480 --> 00:18:49.790 in 2012, and in 2016, it’s very clear following rainfall. 00:18:49.790 --> 00:18:54.880 And this has been modeled previously by Maria Todorovska at USC, 00:18:54.880 --> 00:18:58.510 who looked basically at a poroelastic model of an embedded foundation. 00:18:58.510 --> 00:19:01.870 And the fundamental idea is that, if you, you know, add water, 00:19:01.870 --> 00:19:05.220 you saturate the soils on the sides of your foundation, and increase 00:19:05.220 --> 00:19:08.650 the horizontal stiffness of the foundation to shearing. 00:19:08.650 --> 00:19:11.810 And so this is a quick modulation because basically, 00:19:11.810 --> 00:19:16.630 as the shallow soil in the top 2 meters abutting the foundation drains out, 00:19:16.630 --> 00:19:20.912 you lose that temporary increase in stiffness. 00:19:22.170 --> 00:19:25.530 So that’s easy to explain, but this is a lot harder to explain. 00:19:25.530 --> 00:19:31.809 So, if you might remember from that first slide, the second torsional mode 00:19:31.809 --> 00:19:35.820 seemed like, between 2001 and 2021, it had decreased by a lot. 00:19:35.820 --> 00:19:40.420 And, if we look at it in detail, actually, there’s a gigantic variability 00:19:40.420 --> 00:19:44.930 in excess of 9% from one year to the next. 00:19:44.930 --> 00:19:51.180 And, just like the fundamental east-west mode, this change is correlated with 00:19:51.180 --> 00:19:55.160 the onset of winter rainfall. But, unlike the east-west mode, 00:19:55.160 --> 00:19:59.950 instead of decaying rapidly over a few weeks or a few months during the rainfall season 00:19:59.950 --> 00:20:02.513 as that shallow soil is draining, 00:20:02.513 --> 00:20:09.410 the perturbation to the apparent torsional stiffness in this mode is basically annual, 00:20:09.410 --> 00:20:12.180 that it decays over the scale of a year. 00:20:12.180 --> 00:20:14.080 And this is really quite hard to understand. 00:20:14.080 --> 00:20:18.325 You know, why is the drainage time scale that would be inferred from 00:20:18.325 --> 00:20:22.630 a simple poroelastic model different for different modes? 00:20:22.630 --> 00:20:25.261 One possible explanation is that there’s more 00:20:25.261 --> 00:20:27.770 vertical displacement in this frequency. 00:20:27.770 --> 00:20:31.940 If you remember, the east-west direction is the softest direction, and therefore 00:20:31.940 --> 00:20:35.840 it’s dominated by sharing of the superstructure with a fixed base. 00:20:35.840 --> 00:20:39.638 Whereas, the torsional mode, and particularly the torsional overtones, 00:20:39.638 --> 00:20:41.576 because Caltech Hall has this asymmetric design, 00:20:41.576 --> 00:20:44.890 it includes some non-negligible vertical displacement. 00:20:44.890 --> 00:20:46.610 And so, if there’s vertical displacement, 00:20:46.610 --> 00:20:51.295 then the stiffness of the embedded foundation is a function of basically 00:20:51.295 --> 00:20:56.220 the entire integrated stiffness of the soil underneath the foundation as well. 00:20:56.220 --> 00:20:58.240 And so you can imagine that we have seasonal rainfall. 00:20:58.240 --> 00:21:01.000 It brings up the water table. 00:21:01.000 --> 00:21:04.910 And then, over the course of a year, until the next rainfall, 00:21:04.910 --> 00:21:08.740 the water table is slowly drawing down. And it might be quite far underneath 00:21:08.740 --> 00:21:12.990 the foundation, but it’s still affecting the integrated vertical stiffness – 00:21:12.990 --> 00:21:19.318 basically the resistance of the soil to an applied vertical point load. 00:21:19.318 --> 00:21:22.660 So that’s one hypothesis, which could be tested with modeling. 00:21:22.660 --> 00:21:24.790 What’s convincing, I think, about this argument is that, 00:21:24.790 --> 00:21:29.370 if we look at a soil moisture model, which captures the combined effects 00:21:29.370 --> 00:21:34.224 of precipitation, drainage, and evapotranspiration, 00:21:34.224 --> 00:21:37.400 and is basically a proxy for the depth to fully saturated soil – 00:21:37.400 --> 00:21:41.430 the depth to the shallow water table, we can see that it basically matches 00:21:41.430 --> 00:21:43.900 the same time scale. So it seems like something about 00:21:43.900 --> 00:21:46.790 the depth to water is what’s affecting this part of the stiffness. 00:21:46.790 --> 00:21:48.930 But it’s – what’s amazing about this is how big 00:21:48.930 --> 00:21:53.890 of an effect that water has, right? It’s a 9% change in this frequency. 00:21:53.890 --> 00:21:57.790 And it’s, therefore, by application, you know, almost a 20% change 00:21:57.790 --> 00:21:59.970 in the stiffness of the components 00:21:59.970 --> 00:22:03.539 which are being represented by the second torsional mode. 00:22:04.740 --> 00:22:09.452 If we look instead at the north-south direction, we can see that 00:22:09.452 --> 00:22:11.880 there’s some actually effects – it’s a little hard to see here, 00:22:11.880 --> 00:22:14.220 but there’s some effects related to rain, but it’s dominated 00:22:14.220 --> 00:22:17.050 by temperature-related effects, that there’s a seasonal trend, 00:22:17.050 --> 00:22:23.084 which means that the building is stiffer in the summer and softer in the winter. 00:22:23.084 --> 00:22:27.110 And this is about a 1% change per 10 degrees C. 00:22:27.110 --> 00:22:31.389 And probably the easiest explanation for why the north-south direction is 00:22:31.389 --> 00:22:37.050 exhibiting this really strong temperature sensitivity is because of the shear walls. 00:22:37.050 --> 00:22:41.670 When concrete is warm, it expands and becomes stiffer. 00:22:41.670 --> 00:22:45.480 And the shear walls on the sides of the building the contribute to stiffness 00:22:45.480 --> 00:22:50.540 in the north-south direction are 30 centimeters thick. 00:22:50.540 --> 00:22:56.702 And that’s about 10 times the diurnal skin depth for forcing – 00:22:56.702 --> 00:23:01.015 for harmonic forcing at, you know, a 24-hour period, given the point of 00:23:01.015 --> 00:23:05.180 ordinary thermal diffusivity of concrete. And so what’s probably true is that, 00:23:05.180 --> 00:23:09.900 on a daily basis, the, you know, 12-hour or 24-hour variations 00:23:09.900 --> 00:23:15.270 in solar forcing on the sides of the building aren’t penetrating very far 00:23:15.270 --> 00:23:19.980 in and causing a relatively small effect, whereas, on a seasonal scale, 00:23:19.980 --> 00:23:27.110 there are large changes in the thermal expansion of the shear walls. 00:23:28.007 --> 00:23:32.390 So, to kind of integrate this all together, we’ve observed from this 00:23:32.390 --> 00:23:36.360 20-year continuous record that there are gigantic passive variations, 00:23:36.360 --> 00:23:39.060 both from environmental trends, minor construction work 00:23:39.060 --> 00:23:43.420 that we wouldn’t have expected to cause major structural changes, 00:23:43.420 --> 00:23:45.670 small earthquakes that we know didn’t cause damage, 00:23:45.670 --> 00:23:53.050 and also passive trends that we really don’t understand that are contributing 00:23:53.050 --> 00:23:57.270 to a really dynamic time-dependent elasticity in the structure. 00:23:57.270 --> 00:23:59.830 And so we can plot that here comparing with 00:23:59.830 --> 00:24:02.520 forced vibration tests represented by X’s, 00:24:02.520 --> 00:24:05.828 historical ambient vibration tests represented by triangles, 00:24:05.828 --> 00:24:09.367 earthquakes represented by dots. 00:24:09.367 --> 00:24:14.630 And the kind of inferred history of this building has been to assume, 00:24:14.630 --> 00:24:18.100 following the dashed lines, that the stiffness has been constant 00:24:18.100 --> 00:24:20.570 in between major earthquakes that could have caused damage. 00:24:20.570 --> 00:24:24.050 But what we see over the last 20 years is that that’s not true at all, 00:24:24.050 --> 00:24:25.930 that there’s this enormous variability. 00:24:25.930 --> 00:24:28.090 And that has some significant implications. 00:24:28.090 --> 00:24:30.640 One of them is that the building is about where it was 00:24:30.640 --> 00:24:34.420 before the Whittier Narrows earthquake, which is surprising. 00:24:34.420 --> 00:24:36.630 And we don’t have a good explanation for why. 00:24:36.630 --> 00:24:40.266 We know that there was cracking of some of the foundation components 00:24:40.266 --> 00:24:42.340 during some of these historical earthquakes. 00:24:42.340 --> 00:24:46.810 One possibility is that water flows through those components regularly, 00:24:46.810 --> 00:24:49.800 is depositing calcite all the time, and those components are 00:24:49.800 --> 00:24:54.150 healing back up again because of that mineral deposition. 00:24:54.150 --> 00:24:58.703 Another option is that the foundation soils are just getting stiffer and stiffer 00:24:58.703 --> 00:25:02.516 and stiffer because of, you know, the lifetime of pounding 00:25:02.516 --> 00:25:07.430 by the building’s vibrations during small earthquakes and forced vibration tests. 00:25:07.430 --> 00:25:12.140 And so the soil structure interaction is becoming dominant. 00:25:12.140 --> 00:25:16.700 Another potential explanation is that there are long-term trends related to groundwater 00:25:16.700 --> 00:25:19.590 that could be, you know, moving fine particulates around, 00:25:19.590 --> 00:25:23.829 redistributing, like, little silt particles, and therefore 00:25:23.829 --> 00:25:26.590 changing the bulk modulus of the sub-foundation soils. 00:25:26.590 --> 00:25:28.630 It’s very hard to tell with one record. 00:25:28.630 --> 00:25:32.039 This is something that I hope people investigate further. 00:25:32.039 --> 00:25:35.410 It’ll take some detailed work with monitoring – 00:25:35.410 --> 00:25:40.039 with array-based effort to, like, look at mode shape changes. 00:25:40.039 --> 00:25:43.235 But this certainly a novel observation. As far as I’m aware, 00:25:43.235 --> 00:25:48.700 I’ve never seen any other observations of buildings healing themselves. 00:25:48.700 --> 00:25:51.611 The other implication of this is that, if this is how this building 00:25:51.611 --> 00:25:55.600 just behaves passively, that it’s always increasing in stiffness, 00:25:55.600 --> 00:25:58.370 that, from the scattered record of forced vibration tests and 00:25:58.370 --> 00:26:02.200 triggered seismic records, it’s totally possible that the inferred reaction 00:26:02.200 --> 00:26:05.704 in stiffness during past earthquakes, which has basically been done by saying, 00:26:05.704 --> 00:26:08.770 oh, here was our last forced vibration test before the earthquake. 00:26:08.770 --> 00:26:11.380 Here’s our next forced vibration test after the earthquake. 00:26:11.380 --> 00:26:13.730 What’s the difference? That kind of simplistic analysis 00:26:13.730 --> 00:26:16.579 might significantly underestimate the reduction in stiffness 00:26:16.579 --> 00:26:19.810 from previous earthquakes. In particular, Whittier Narrows, 00:26:19.810 --> 00:26:23.704 which didn’t have a reference point before it for about a decade could have 00:26:23.704 --> 00:26:28.798 been significantly underestimated in how much it damaged the building. 00:26:30.798 --> 00:26:33.642 And, from the structural health monitoring perspective, this is, 00:26:33.642 --> 00:26:36.240 of course, concerning because we observe that there are these 00:26:36.240 --> 00:26:41.205 earthquakes in the history of Caltech Hall that caused some minor damage and 00:26:41.205 --> 00:26:47.306 that the passive changes over the last two decades exceed that change. 00:26:47.306 --> 00:26:50.540 And so it may actually be very hard, given a scattered record 00:26:50.540 --> 00:26:52.940 of triggered motion records and forced vibration tests 00:26:52.940 --> 00:26:55.910 just to even tell whether or not there is damage. 00:26:55.910 --> 00:26:59.610 As you can see, over – from these X’s, the forced vibration tests – 00:26:59.610 --> 00:27:02.980 over the last 30 years or so, the forced vibration tests themselves, 00:27:02.980 --> 00:27:05.392 which are theoretically, you know, supposed to be more stable and 00:27:05.392 --> 00:27:07.950 more detailed than these ambient vibration measurements, 00:27:07.950 --> 00:27:09.900 have had a comparable level of variability. 00:27:09.900 --> 00:27:13.517 So this is, of course, a concerning observation 00:27:13.517 --> 00:27:16.150 for the kind of current state of practice in structural health monitoring, 00:27:16.150 --> 00:27:19.540 which is to take before-and-after pictures. 00:27:20.494 --> 00:27:25.744 So, with this – with this 20-year record, we can also look at earthquakes. 00:27:25.744 --> 00:27:29.080 And this allows us to start to look at the response 00:27:29.080 --> 00:27:30.630 of the building under different levels of excitation. 00:27:30.630 --> 00:27:34.643 There’s over 600 events rated magnitude 4 in southern California over 00:27:34.643 --> 00:27:38.510 the last 20 years – well, in the greater southern California region 00:27:38.510 --> 00:27:42.910 in the last 20 years that have had a measurable acceleration 00:27:42.910 --> 00:27:46.697 at CI.MIK greater than the ambient noise floor. 00:27:46.697 --> 00:27:52.100 And you can see these plotted here on the left in terms of their frequency response. 00:27:52.100 --> 00:27:55.768 And you can see the color bar here is the log of the peak acceleration. 00:27:55.768 --> 00:28:03.103 And so, for small-amplitude earthquakes, the curve basically follows exactly along 00:28:03.103 --> 00:28:06.270 the trajectory we determined from ambient vibrations for the building’s 00:28:06.270 --> 00:28:10.314 self-healing and all these seasonal variations related to rainfall. 00:28:10.314 --> 00:28:16.439 But, for larger earthquakes, they drop down to a lower frequency. 00:28:16.439 --> 00:28:18.140 And this is a – you know, a relatively known effect 00:28:18.140 --> 00:28:24.780 that buildings soften dynamically under large excitation. 00:28:24.780 --> 00:28:28.981 But the form of this nonlinearity is interesting. 00:28:30.361 --> 00:28:34.542 Importantly, if we go all the way up here to 1 meter per second squared, 00:28:34.543 --> 00:28:42.143 which is around the level of the Ridgecrest earthquakes, we see 00:28:42.143 --> 00:28:48.190 that there’s, first of all, a gigantic – about 25% – reduction in stiffness. 00:28:48.190 --> 00:28:51.950 So the – we know those earthquakes didn’t damage the building. 00:28:51.950 --> 00:28:53.660 They recovered almost completely. 00:28:53.660 --> 00:28:59.550 And so the change in elasticity during strong motion is very large. 00:28:59.550 --> 00:29:04.769 Second is that the general curve here that you can see going from basically 00:29:04.769 --> 00:29:08.230 the intensity of ambient vibrations on the left to the Ridgecrest-level 00:29:08.230 --> 00:29:12.700 earthquake on the right shows that there’s really no linear elastic regime. 00:29:12.700 --> 00:29:18.144 That basically, the reduction in stiffness with increased amplitude is pervasive 00:29:18.144 --> 00:29:22.815 across the entire measurable band all the way down to the level of ambient noise. 00:29:22.815 --> 00:29:28.100 And this is also consistent with forced vibration tests. 00:29:28.100 --> 00:29:32.410 There’s a Kinemetrics orbital vibrator in the building, and we can run 00:29:32.410 --> 00:29:35.410 forced vibration tests at different levels of excitation. 00:29:35.410 --> 00:29:39.706 And these ones here, shown by the black X’s, which were conducted in 2019, 00:29:39.706 --> 00:29:43.290 exactly match the earthquake records. And that suggests that really 00:29:43.290 --> 00:29:47.900 this nonlinearity can be well-described using the peak acceleration. 00:29:47.900 --> 00:29:52.294 That things like the shaking don’t have as significant of an effect. 00:29:54.390 --> 00:29:57.090 So this was noted, to some extent, historically. 00:29:57.090 --> 00:30:00.644 From this very, very limited record of triggered records, 00:30:00.644 --> 00:30:06.850 people knew that there was this potentially power law nonlinearity. 00:30:06.850 --> 00:30:10.040 But what we can see is that it’s really not strictly power law, first of all. 00:30:10.040 --> 00:30:12.650 This is – this is a kind of schematic representation, 00:30:12.650 --> 00:30:15.660 but really there’s two regimes. There’s a weekly nonlinear regime 00:30:15.660 --> 00:30:21.780 here at low levels of excitation where a 10-times increase in acceleration 00:30:21.780 --> 00:30:26.270 is about a 3% drop in frequency. And then above some threshold that, 00:30:26.270 --> 00:30:29.360 you know, I’m kind of defining roughly here, 00:30:29.360 --> 00:30:34.040 we observe a much steeper slope such that a 10-time increase in acceleration 00:30:34.040 --> 00:30:38.560 is about a 9% drop in frequency. 00:30:38.560 --> 00:30:41.910 And this is consistent with what was observed for the strong motion records. 00:30:41.910 --> 00:30:46.160 One reason that we could have this really complex trend between 00:30:46.160 --> 00:30:49.780 two different kind of intensities of nonlinearity is that 00:30:49.780 --> 00:30:54.090 the soil structure interaction could take over at some critical point. 00:30:54.090 --> 00:30:57.820 We know that, during earthquakes, the reduction in stiffness 00:30:57.820 --> 00:31:04.060 of the soil foundation system becomes dominant for largest excitations. 00:31:04.060 --> 00:31:07.920 And so it seems likely that soil plasticity, or nonlinear elasticity, 00:31:07.920 --> 00:31:11.180 takes over and starts to contribute beyond a small strain threshold 00:31:11.180 --> 00:31:15.000 around here and that this probably represents the weak nonlinearity 00:31:15.000 --> 00:31:17.820 of the building materials themselves. 00:31:17.820 --> 00:31:20.833 We can also compare over this period when we know that 00:31:20.833 --> 00:31:24.970 the background level of stiffness increased. 00:31:24.970 --> 00:31:26.630 And we can see that basically, 00:31:26.630 --> 00:31:31.520 regardless, up until this threshold, everything just shifted up. 00:31:31.520 --> 00:31:36.170 That there’s no change in the nonlinear response over time. 00:31:36.170 --> 00:31:37.500 And that basically, above that threshold, 00:31:37.500 --> 00:31:41.215 we don’t have enough events to say anything statistically significant. 00:31:42.973 --> 00:31:48.645 But, of course, we know that, for most events, we get a recovery 00:31:48.645 --> 00:31:52.860 back to the pre-event frequency. And this is an interesting question. 00:31:52.860 --> 00:31:53.860 Well, how does this happen? 00:31:53.860 --> 00:31:56.145 We have this nonlinear response, a drop in the apparent stiffness 00:31:56.145 --> 00:31:57.980 during strong motion. What happens? 00:31:57.980 --> 00:32:01.460 Well, the answer is it kind of relaxes back slowly, 00:32:01.460 --> 00:32:03.010 and it really follows about a log-linear curve. 00:32:03.010 --> 00:32:08.240 So here’s the record for the magnitude 5.4 Chino Hills earthquake in 2008. 00:32:08.240 --> 00:32:11.790 At the onset on strong motion, we have this sudden drop in frequency. 00:32:11.790 --> 00:32:15.540 And the frequency starts to recover. And this recovery curve persists 00:32:15.540 --> 00:32:21.260 long after the strong motion has abated, right, that the frequency 00:32:21.260 --> 00:32:25.470 is no longer proportional to the excitation level beyond some 00:32:25.470 --> 00:32:28.973 very short point after the peak acceleration. 00:32:28.973 --> 00:32:33.000 And, if we look at another earthquake with a similar shaking duration, 00:32:33.000 --> 00:32:35.040 but a 10-times-lower level of acceleration, 00:32:35.040 --> 00:32:38.450 we see that it follows a very similar curve. 00:32:39.731 --> 00:32:43.575 And, if we formalize this for all events, we can see that – 00:32:43.575 --> 00:32:46.208 well, there’s quite a long scatter, probably due to the kind of 00:32:46.208 --> 00:32:49.600 order of magnitude differences in shaking duration for different earthquakes. 00:32:49.600 --> 00:32:53.911 The recovery is approximately log-linear for all events, 00:32:53.911 --> 00:32:56.140 suggesting some sort of universal time scale. 00:32:56.140 --> 00:32:59.521 About 20% of the stiffness is recovered in the first 30 seconds, 00:32:59.521 --> 00:33:03.210 and about 80% is recovered by five minutes, on average. 00:33:04.255 --> 00:33:06.781 And this is not the first time something like this has been observed. 00:33:06.781 --> 00:33:09.080 It’s been observed in about three other buildings and been 00:33:09.080 --> 00:33:12.610 quite widely observed in the rock physics literature. 00:33:12.610 --> 00:33:15.870 And there’s been an argument that I think has some issues, 00:33:15.870 --> 00:33:19.646 but there’s an argument out there that this sort of log-linear healing is 00:33:19.646 --> 00:33:26.410 really truly multi-scale, that it occurs over the scale of seconds to months, 00:33:26.410 --> 00:33:28.300 maybe even years. 00:33:28.300 --> 00:33:30.958 And there’s been a lot of interesting explanations, ranging from kind of 00:33:30.958 --> 00:33:35.708 Arrhenius’ law governing the thermodynamics of material bonds, which re-organize after 00:33:35.708 --> 00:33:40.310 significant motion, to the possible role of viscoelastic relaxation and fractures 00:33:40.310 --> 00:33:44.771 that might be opened during vibration and have to close again after 00:33:44.771 --> 00:33:47.200 the earthquake is over, to even rate-and-state friction. 00:33:47.200 --> 00:33:49.708 You know, if we think about all these little fractures in those 00:33:49.708 --> 00:33:52.659 concrete structure being governed by [inaudible] of friction, 00:33:52.660 --> 00:33:58.505 then the aging law sticks in this log time dependence automatically. 00:33:59.255 --> 00:34:02.981 But this is still very much an area of debate. 00:34:03.684 --> 00:34:06.646 But, so to kind of wrap up Part 1 here, which is really the meatier part 00:34:06.646 --> 00:34:12.510 of this talk, I want to make some broader generalizations. 00:34:12.510 --> 00:34:16.396 So, first of all, the key observations that we observe this gigantic variability 00:34:16.396 --> 00:34:18.840 in apparent stiffness in time scales greater than one week. 00:34:18.840 --> 00:34:21.333 And this should really be taken as a minimum estimate because 00:34:21.333 --> 00:34:25.583 that underestimates, one, the effect – it doesn’t include 00:34:25.584 --> 00:34:28.620 the effect of the nonlinear response during strong motion, 00:34:28.620 --> 00:34:33.970 which could add easily another 20% reduction in stiffness on top of that. 00:34:33.970 --> 00:34:37.458 Then it also neglects the events that happen on a sub-weekly time scale, 00:34:37.458 --> 00:34:40.889 which a study from 2006 by John Clinton, when he was a Ph.D. student 00:34:40.889 --> 00:34:44.310 at Caltech, that showed that, on top of this seasonal change 00:34:44.310 --> 00:34:48.330 in rainfall, there’s actually a change on the scale of one to three days 00:34:48.330 --> 00:34:52.911 right after a rainfall that’s another 3% increase. 00:34:52.911 --> 00:34:56.210 And, of course, there’s also changes during things like wind forcing 00:34:56.210 --> 00:34:58.460 and other meteorological conditions that are [inaudible]. 00:34:58.460 --> 00:35:03.380 So this 20% variability should be really taken as a minimum. 00:35:03.380 --> 00:35:07.700 The second key observation is that there is strong nonlinear elasticity, 00:35:07.700 --> 00:35:11.180 it’s time-dependent, and there is no linear elastic regime. 00:35:11.180 --> 00:35:15.810 The nonlinearity persists down to the level of ambient vibration. 00:35:15.810 --> 00:35:17.460 And I want to make two conclusions. 00:35:17.460 --> 00:35:21.850 One conclusion from this first part is that – is a practical conclusion – 00:35:21.850 --> 00:35:25.570 is that we should really have strong motion engineering networks 00:35:25.570 --> 00:35:27.830 be continuous all the time. 00:35:27.830 --> 00:35:31.240 Because, as I discuss before, the passive variability means that, if you just have a 00:35:31.240 --> 00:35:36.770 before-and-after picture, you can’t really necessarily identify damage reliably. 00:35:36.770 --> 00:35:40.220 I think the approach we should be going for is to treat these parameters of interest 00:35:40.220 --> 00:35:42.560 for structural health monitoring, like the natural frequencies 00:35:42.560 --> 00:35:45.896 of the damping of a structure, like a geodetic time series, where we can 00:35:45.896 --> 00:35:50.410 model and remove these seasonal variations and secular trends 00:35:50.410 --> 00:35:53.810 in order to estimate, just like we would for, like, a – 00:35:53.810 --> 00:35:57.570 try and estimate the coseismic displacement of a GPS station, 00:35:57.570 --> 00:36:01.860 we can estimate the coseismic change in stiffness of a building. 00:36:01.860 --> 00:36:04.646 The other thing that I think is really important to point out here is that 00:36:04.646 --> 00:36:06.990 this kind of order – five-minute time scale 00:36:06.990 --> 00:36:11.690 of post-event recovery is longer than most triggered records are. 00:36:11.690 --> 00:36:15.350 So it may be that you can’t even identify damage from a triggered record, 00:36:15.350 --> 00:36:18.640 no matter how fancy your analysis method, because it’s still recovering. 00:36:18.640 --> 00:36:21.880 And so you can’t really infer the asymptotic post-event frequency 00:36:21.880 --> 00:36:24.290 from a short record. 00:36:25.559 --> 00:36:29.700 And finally, just to – for this first part, just to kind of get a little bit 00:36:29.700 --> 00:36:32.500 speculative, but to, you know, think about how 00:36:32.500 --> 00:36:36.146 broad sweeping this is in terms of implications, if we, like, 00:36:36.146 --> 00:36:40.540 go into a PSHA-type framework – you know, there’s lots of things you can do 00:36:40.540 --> 00:36:42.760 in this zone, but you want to use an intensity measure, 00:36:42.760 --> 00:36:45.640 like the 5% damped spectral acceleration. You’re inherently 00:36:45.640 --> 00:36:52.890 baking in this single degree of freedom, linear time-invariant elasticity. 00:36:52.890 --> 00:36:56.817 And that means that, you know, for any given building, you may be looking at, 00:36:56.817 --> 00:37:00.040 assuming that the sorts of variations you observe in Caltech Hall are universal 00:37:00.040 --> 00:37:04.530 in reinforced concrete structures, a hidden 20% variability in stiffness 00:37:04.530 --> 00:37:11.044 that’s not accounted for in your current PSHA-type uncertainty propagation. 00:37:11.044 --> 00:37:15.833 And that, because the stiffness of the building is always decreasing during 00:37:15.833 --> 00:37:19.050 strong motion, and there’s this time-dependent recovery, you’re basically 00:37:19.050 --> 00:37:25.083 going to always overestimate the stiffness using the kind of, you know, 00:37:25.083 --> 00:37:29.720 background reference level for the frequency, and therefore the intensity 00:37:29.720 --> 00:37:32.030 measure becomes a little bit dangerous. 00:37:32.030 --> 00:37:35.896 Because, for a stiff building like Caltech Hall that’s – 00:37:35.896 --> 00:37:38.591 you know if it were to fail would probably fail because of, 00:37:38.591 --> 00:37:43.960 like, a stress concentration at some joint causing fracture of the concrete, 00:37:43.960 --> 00:37:46.970 right, it’s probably good to overestimate the stiffness. 00:37:46.970 --> 00:37:52.110 Because you’re then overestimating the stresses and planning for 00:37:52.110 --> 00:37:55.480 a scenario that’s much worse than what you’re actually probably 00:37:55.480 --> 00:37:57.442 going to get because of the dynamic softening. 00:37:57.442 --> 00:38:01.370 But, for soft structures, or flexible structures, like high-rise buildings, 00:38:01.370 --> 00:38:04.540 this is a really bad idea. Because what you really care about 00:38:04.540 --> 00:38:06.750 from the failure perspective is the inter-story drift. 00:38:06.750 --> 00:38:09.970 And the inter-story drift in a stiff building is much less than 00:38:09.970 --> 00:38:11.580 the inter-story drift in a soft building. 00:38:11.580 --> 00:38:16.833 And so, overestimating stiffness by neglecting nonlinear dynamic elasticity 00:38:16.833 --> 00:38:22.320 is a little bit perilous in that context. 00:38:22.320 --> 00:38:24.458 So that’s the main meat of the talk. 00:38:24.458 --> 00:38:28.271 I’m going to now go through a couple interesting novel observations 00:38:28.271 --> 00:38:32.010 on the other side of big data. 00:38:32.010 --> 00:38:35.100 Instead of looking at one station for a long time, 00:38:35.100 --> 00:38:39.740 we’re now going to look at a lot of stations for a short time. 00:38:39.740 --> 00:38:41.390 And I mentioned forced vibrations tests already. 00:38:41.390 --> 00:38:44.646 The forced vibration tests in Caltech Hall are carried out with this 00:38:44.646 --> 00:38:51.271 1972 orbital mass vibrator, which we rebuilt in 2019, the general idea being 00:38:51.271 --> 00:38:54.021 that there are two counter-rotating buckets – here’s our summer intern 00:38:54.021 --> 00:39:00.021 for scale – that apply a sinusoidal force along one direction of the building. 00:39:00.021 --> 00:39:03.458 And there’s substantial amplification, obviously, at the structural – 00:39:03.458 --> 00:39:06.660 at the structure’s resonant modes and the natural frequencies. 00:39:06.660 --> 00:39:10.510 And it’s been known for a long time that this generates seismic radiation. 00:39:10.510 --> 00:39:16.083 In 1970, Paul Jennings, who was a professor of civil engineering at Caltech, 00:39:16.083 --> 00:39:19.940 realized that you could see the forced vibrations of Millikan Library 00:39:19.940 --> 00:39:22.850 on the nearby Mount Wilson Observatory seismograph, 00:39:22.850 --> 00:39:25.830 which is 11 kilometers away and at the top of a mountain. 00:39:25.830 --> 00:39:28.860 And you can see them in the raw data in the time domain on this seismograph. 00:39:28.860 --> 00:39:31.350 So it’s pretty impressive. 00:39:31.350 --> 00:39:33.583 And what’s happening here is that basically, you know, the building is 00:39:33.583 --> 00:39:37.271 rocking and shearing, and that’s causing displacement at the surface 00:39:37.271 --> 00:39:40.646 that’s vertically polarized. And so it very efficiently excites Rayleigh waves in 00:39:40.646 --> 00:39:44.270 its normal modes. And these can actually be observed throughout southern California. 00:39:44.270 --> 00:39:48.860 There have been a few attempts – one most recently by Toshiro Tanimoto 00:39:48.860 --> 00:39:55.567 at UCSB – to use this data for imaging for subsurface investigation. 00:39:55.567 --> 00:39:58.780 Toshiro showed that, actually, if you stack four hours 00:39:58.780 --> 00:40:04.010 of continuous shaking, you can see this data as a – as a discrete spectral peak 00:40:04.010 --> 00:40:07.540 as far south as Mexico, which is just astonishing. 00:40:07.540 --> 00:40:11.050 I think that you have to probably kind of do some back-of-the-envelope calculations 00:40:11.050 --> 00:40:15.320 about the kinetic energy involved in a building like this size 00:40:15.320 --> 00:40:17.980 once it starts to get moving before you can really believe that, 00:40:17.980 --> 00:40:20.190 over four hours, you could put enough energy to see in Mexico 00:40:20.190 --> 00:40:22.600 into the ground by vibrating this building. 00:40:22.600 --> 00:40:25.833 But, nonetheless, it’s true that you can do this. 00:40:25.833 --> 00:40:31.740 And it turns out you can’t really use these for structural investigations very well, 00:40:31.740 --> 00:40:34.520 these building forced vibrations, because they’re extremely narrow-band. 00:40:34.520 --> 00:40:37.708 He tried to make some group velocity estimates based on the dispersion, 00:40:37.708 --> 00:40:41.370 but it’s very challenging to do. 00:40:41.370 --> 00:40:45.021 What probably you could learn more about if you were to develop a better 00:40:45.021 --> 00:40:47.960 physics-based model in the building is the soil structure interaction. 00:40:47.960 --> 00:40:54.708 From a perspective of a civil engineer, they often refer to something called 00:40:54.708 --> 00:40:57.620 radiation damping, that, when you have soil structure interaction, 00:40:57.620 --> 00:40:59.950 one of the ways that you lose energy in a building is through 00:40:59.950 --> 00:41:03.400 leakage of seismic waves out the base of the building. 00:41:03.400 --> 00:41:05.340 And they usually just refer to this as radiation damping 00:41:05.340 --> 00:41:08.403 and make a damping ratio for it and say end of the day. 00:41:08.403 --> 00:41:12.208 But, in reality, you know, the flexibility of the foundation, 00:41:12.208 --> 00:41:17.020 the shape of the foundation, the relative amount of rocking versus shearing, 00:41:17.020 --> 00:41:21.294 how flexible the foundation is to vertical displacement, 00:41:21.294 --> 00:41:25.160 are all going to affect what’s radiated away from the building. 00:41:25.160 --> 00:41:26.919 And so I think, with a physics-based model, 00:41:26.919 --> 00:41:30.796 we could probably learn a lot about soil structure interaction this way. 00:41:32.263 --> 00:41:36.000 But I just want to show some quick observations of this 00:41:36.000 --> 00:41:38.650 now in the time that remains. So distributed acoustic sensing – 00:41:38.650 --> 00:41:40.870 I won’t go through this, but I think actually you’ve had 00:41:40.870 --> 00:41:43.442 several seminars in distributed acoustic sensing. 00:41:43.442 --> 00:41:47.271 But it’s basically a way of converting a fiber optic cable into a dense array 00:41:47.271 --> 00:41:50.270 of sensors, and it allows us to see these structural vibrations 00:41:50.270 --> 00:41:52.780 in the urban environment in extremely high resolution. 00:41:52.780 --> 00:41:55.920 So, for example, here at Caltech we have this Pasadena DAS array. 00:41:55.920 --> 00:41:59.280 It’s a 37-kilometer loop in the city of Pasadena. 00:41:59.280 --> 00:42:03.333 We have one DAS instrument shooting clockwise, but there’s a really bad 00:42:03.333 --> 00:42:05.780 reflection here, so then we have a cheaper, older instrument shooting the 00:42:05.780 --> 00:42:10.080 other direction in a secondary fiber just to get good S and R on this segment. 00:42:10.080 --> 00:42:15.278 And here at Caltech is where Caltech Hall, or Millikan Library, is. 00:42:15.278 --> 00:42:17.790 And, when we shake the building, we can see these Rayleigh waves 00:42:17.790 --> 00:42:22.290 all throughout the city. They’re discernable in the raw time domain data. 00:42:22.290 --> 00:42:26.630 You can see this is a couple kilometers’ distance and time. 00:42:26.630 --> 00:42:28.820 You can see these are cars. 00:42:28.820 --> 00:42:31.583 And then there’s some long-period drift and other crap going on in the – 00:42:31.583 --> 00:42:33.130 in the data. 00:42:33.130 --> 00:42:35.146 But, when we filter this, you can see these beautiful Rayleigh waves 00:42:35.146 --> 00:42:37.640 traveling along the cable around 400 meters per second. 00:42:37.640 --> 00:42:41.833 This happens to be an example from one of the overtones at 5 hertz. 00:42:41.833 --> 00:42:48.740 And we can track the velocities on a sub-block scale throughout the city. 00:42:48.740 --> 00:42:53.583 And we see that there are variations in Vs30 – the Rayleigh wave velocity 00:42:53.583 --> 00:42:58.208 of 4.9 hertz is not that dissimilar from Vs30 – on the scale of 00:42:58.208 --> 00:43:01.540 20 to 40% from one block to the next. 00:43:01.540 --> 00:43:05.208 So that’s notable, and we’re in the process of doing a more 00:43:05.208 --> 00:43:08.260 comprehensive investigation throughout the city. 00:43:08.260 --> 00:43:15.070 But I want to talk about a much more, I don’t know, out-there example. 00:43:15.070 --> 00:43:20.021 We have also looked at this problem of building-generated variations 00:43:20.021 --> 00:43:21.708 in the context of offshore energy development. 00:43:21.708 --> 00:43:24.720 Because wind farms generate a lot of seismic energy. 00:43:24.720 --> 00:43:29.708 You can use arrays quite far away to locate high-frequency seismic radiation 00:43:29.708 --> 00:43:31.800 back to wind developments. 00:43:31.800 --> 00:43:37.646 And we wanted to know, you know, if we actually go in and look at a DAS on a cable 00:43:37.646 --> 00:43:41.840 in a wind farm, can we do things to, like, figure out, for example, 00:43:41.840 --> 00:43:46.333 the shear wave velocity at the bottom or the structural vibrations of 00:43:46.333 --> 00:43:50.920 the individual turbines using that pre-existing fiber. 00:43:50.920 --> 00:43:54.958 Because there’s fiber all throughout wind farms that’s used in the same cables 00:43:54.958 --> 00:43:58.500 that carry the power back to shore and also do internet 00:43:58.500 --> 00:43:59.990 telecommunications between the cables. 00:43:59.990 --> 00:44:02.771 So it’s basically this giant spider web of fiber optic cables that 00:44:02.771 --> 00:44:05.850 we can utilize for investigations. 00:44:05.850 --> 00:44:11.541 And so, in 2018, we collected this data from a pre-existing wind farm 00:44:11.541 --> 00:44:15.280 in the Belgian North Sea along with 42-kilometer cable here 00:44:15.280 --> 00:44:18.640 that’s buried that’s meant for power transmission. 00:44:18.640 --> 00:44:21.610 Each of these boxes is a wind farm development. 00:44:21.610 --> 00:44:26.208 And the raw data is really messy, but it contains a component of seismic waves, 00:44:26.208 --> 00:44:28.722 as you can see here in the [inaudible] domain around, you know, 00:44:28.723 --> 00:44:32.559 500 to a couple kilometers per second. 00:44:34.567 --> 00:44:38.160 I’m just going to skip through some of this in the interest of time. 00:44:38.160 --> 00:44:42.146 One thing we can do with this is, just using normal ambient noise, do tomography 00:44:42.146 --> 00:44:46.247 and extract parameters like Vs30 all along the cable. 00:44:46.247 --> 00:44:48.390 That’s not very new. 00:44:48.390 --> 00:44:51.146 But, in the process of doing this, you know, we’re doing 00:44:51.146 --> 00:44:53.146 ambient noise correlation. We expect to see this 00:44:53.146 --> 00:44:58.550 nice direct wave propagating away from our cross-correlation source. 00:44:58.550 --> 00:45:03.521 And, instead – okay, we do see that, but we also see all sorts of crazy, 00:45:03.521 --> 00:45:07.560 wacky hyperbole that make no sense. And what these really represent 00:45:07.560 --> 00:45:11.320 are secondary sources of scatter. If you imagine that you have 00:45:11.320 --> 00:45:15.520 a linear array and you’re doing cross- correlations and you have some additional 00:45:15.520 --> 00:45:19.620 source of scatter off here, in a common source gather where we have – 00:45:19.620 --> 00:45:23.646 like here, our source and then all the channels we’re cross-correlating it with, 00:45:23.646 --> 00:45:25.850 the travel time curves for these two different waves – 00:45:25.850 --> 00:45:29.030 the direct wave along our array and the secondary wave from 00:45:29.030 --> 00:45:33.850 this off-axis source – are quite similar in their move-out. 00:45:33.850 --> 00:45:38.750 Whereas, if we rearrange this in terms of pairs of consistently separated channels, 00:45:38.750 --> 00:45:41.830 common-offset gather, we flatten the direct wave. 00:45:41.830 --> 00:45:45.840 As long as the lateral variations and shear wave velocity are pretty similar. 00:45:45.840 --> 00:45:50.190 And we can see this secondary source very clearly. 00:45:50.190 --> 00:45:53.920 And, if we do this along this cable, which passes by all these wind turbines, 00:45:53.920 --> 00:46:00.540 we can see that there are all of these secondary sources at high frequencies, 00:46:00.540 --> 00:46:03.320 which are presumably related to wind turbines. 00:46:03.320 --> 00:46:06.950 We can apply – because of the really dense array-based nature of DAS, 00:46:06.950 --> 00:46:12.067 we can apply sort of imaging or migration to localize these 00:46:12.067 --> 00:46:15.400 high-frequency vibrations to their sources, and we can see that 00:46:15.400 --> 00:46:18.550 they’re actually associated with, as you might have guessed, 00:46:18.550 --> 00:46:22.110 individual wind turbines along the array. 00:46:23.645 --> 00:46:25.896 In particular, here, this Rentel wind farm, where we 00:46:25.896 --> 00:46:28.641 only see a few of them, was under construction at the time of this, 00:46:28.641 --> 00:46:31.740 so only a handful of the turbines were active. 00:46:31.740 --> 00:46:34.302 We went back for a short data set in 2019. 00:46:34.302 --> 00:46:36.720 Here’s the same wind farm – Rentel. 00:46:36.720 --> 00:46:41.000 There’s nine turbines installed there now, and you can see all nine turbines 00:46:41.000 --> 00:46:44.270 in this common-offset data. And there was an entire other 00:46:44.270 --> 00:46:46.790 wind farm here, Norther, that was built in between the two acquisitions, 00:46:46.790 --> 00:46:50.960 and you can see all six of the closest wind turbines. 00:46:50.960 --> 00:46:54.208 And so these are Rayleigh waves, or Scholte waves, the seafloor radiated by 00:46:54.208 --> 00:46:57.030 the vibrations of the turbine under the cyclic loading from 00:46:57.030 --> 00:47:03.255 the rotating prop that excites all of the resident modes of the building. 00:47:03.255 --> 00:47:09.180 And we can actually look even closer and verify this interpretation because, 00:47:09.180 --> 00:47:11.770 if you see – I know this image is a little bit saturated on the right, 00:47:11.770 --> 00:47:13.860 but you can see that there are all these little teeny cables. 00:47:13.860 --> 00:47:17.450 These are the umbilicals that cross between individual turbines 00:47:17.450 --> 00:47:20.410 and bring the power back to one big substation. 00:47:20.410 --> 00:47:24.440 And so what you see here is a DAS acquisition filtered 2 to 10 hertz 00:47:24.440 --> 00:47:28.960 on one of those umbilical cables. And this is – here it’s on the seafloor. 00:47:28.960 --> 00:47:33.833 It goes up in the J-tube, which is kind of like a – basically a walkway 00:47:33.833 --> 00:47:40.150 into the side of the turbine for the cable. And then it’s hanging freely 00:47:40.150 --> 00:47:44.510 in the turbine column, and so you’re getting all this random noise. 00:47:44.510 --> 00:47:48.240 And then it comes out the other side. Here it runs near another turbine, 00:47:48.240 --> 00:47:50.760 goes into a turbine, then out of a turbine, etc. 00:47:50.760 --> 00:47:54.700 And what you can see is that each turbine is radiating energy away from it. 00:47:54.700 --> 00:47:57.220 And you can see this continually as you go up into the J-tube 00:47:57.220 --> 00:48:00.583 and actually part of the structure. And then, when you’re out 00:48:00.583 --> 00:48:02.990 onto the seafloor in between, you can see the contributions 00:48:02.990 --> 00:48:06.320 interfering from two adjacent turbines. 00:48:06.320 --> 00:48:10.130 And so what are we going to – you know, what can we do with this? 00:48:12.075 --> 00:48:15.059 Obviously, you know, I breezed through it, 00:48:15.059 --> 00:48:19.146 but we can use these waves as an active source for subsurface investigations to 00:48:19.146 --> 00:48:22.470 measure things like shear wave velocity, both in an urban environment where 00:48:22.470 --> 00:48:26.958 we have the vibrations at Caltech Hall, and offshore where we have the vibrations 00:48:26.958 --> 00:48:31.505 of offshore platforms or drilling rigs or wind turbines. 00:48:31.505 --> 00:48:35.980 But, in terms of the structures themselves, we should be able to do quite a lot because 00:48:35.980 --> 00:48:38.670 we can clearly identify the resonant modes – 00:48:38.670 --> 00:48:42.890 at least a couple of the resonant modes in these far-field vibrations. 00:48:42.890 --> 00:48:45.760 We can localize these to individual turbines 00:48:45.760 --> 00:48:48.900 without having to have instruments in those turbines. 00:48:48.900 --> 00:48:53.390 And, because the vibrations are generated 00:48:53.390 --> 00:48:55.690 by the rocking and shearing of the foundation, 00:48:55.690 --> 00:48:58.800 we should have information about the health of the foundation. 00:48:58.800 --> 00:49:02.458 In particular, wind turbines – offshore wind turbines have a very, very short 00:49:02.458 --> 00:49:06.640 design life – something around 20 to 25 years for most of them. 00:49:06.640 --> 00:49:09.278 And this is because there’s damage to the blades and other things 00:49:09.278 --> 00:49:10.919 that can happen. But one of the big things 00:49:10.919 --> 00:49:14.720 that can happen is, you know, you stick turbines in the soft mud offshore 00:49:14.720 --> 00:49:17.700 Belgium in the North Sea, and they can start to tilt 00:49:17.700 --> 00:49:20.119 over a long operation life because they’d just be jackhammering 00:49:20.119 --> 00:49:23.630 in their foundation like a piledriver in this soft clay mud. 00:49:23.630 --> 00:49:29.910 And even a tilt of less than 1 degree can be enough to decommission a turbine. 00:49:29.910 --> 00:49:33.646 Second, they modify the flow – the hydrodynamic conditions around them 00:49:33.646 --> 00:49:36.110 when they’re in the water column. And so they have this problem called 00:49:36.110 --> 00:49:40.146 scour, where basically the locally perturbed currents create, like, 00:49:40.146 --> 00:49:46.028 vortex shredding off of the – off of the cylindrical turbine structure. 00:49:46.028 --> 00:49:52.396 And that’ll lead to increased sediment removal around the base of the turbine, 00:49:52.396 --> 00:49:56.260 and that can also cause, you know, issues with the foundation and tilting. 00:49:56.260 --> 00:50:00.083 And so the conclusion is that there’s a lot of value in having 00:50:00.083 --> 00:50:05.820 dense urban- and infrastructure-associated arrays for engineering applications. 00:50:05.820 --> 00:50:08.340 For this particular application, we really need 00:50:08.340 --> 00:50:10.520 more physics-based models in order to exploit it 00:50:10.520 --> 00:50:14.640 for soil structure interaction studies, but I think it’s very promising. 00:50:14.640 --> 00:50:18.010 So I’m happy to take questions on either of the two topics. 00:50:18.010 --> 00:50:20.510 I hope I’ve convinced you that big data should be 00:50:20.510 --> 00:50:25.130 a frontier in engineering seismology. Tell your civil engineer friends 00:50:25.130 --> 00:50:28.966 that they should get excited about databases and whatnot. 00:50:28.966 --> 00:50:32.080 And, if you want to talk more about this, 00:50:32.080 --> 00:50:35.560 I have a SCEC poster next week where I’ll be talking about the first part 00:50:35.560 --> 00:50:40.411 of this talk on the changes in the natural frequencies of Caltech Hall. 00:50:40.411 --> 00:50:43.958 And I’d be happy to chat more with you then. 00:50:46.263 --> 00:50:50.199 [silence] 00:50:50.200 --> 00:50:52.521 - Nice. Thank you, Ethan. 00:50:53.460 --> 00:50:55.372 So do we have any questions? 00:50:55.372 --> 00:50:58.921 People can raise their hand or type a question in the chat. 00:51:01.395 --> 00:51:07.638 [silence] 00:51:07.639 --> 00:51:08.888 Andy? 00:51:09.780 --> 00:51:11.242 Andy Michael? 00:51:11.243 --> 00:51:12.243 - Yeah. 00:51:13.481 --> 00:51:16.170 Hi, Ethan. Not sure if my camera will turn on or not. 00:51:16.170 --> 00:51:21.646 It strikes me that Caltech Hall is sort of unique for monitoring from 00:51:21.646 --> 00:51:25.560 outside the building – doing the monitoring of its impact on the far-field in that 00:51:25.560 --> 00:51:29.330 it’s really tall, and therefore has lower frequencies 00:51:29.330 --> 00:51:31.220 than pretty much anything else on campus. 00:51:31.220 --> 00:51:35.310 There’s nothing – I don’t think there’s anything else close to its height around. 00:51:35.310 --> 00:51:37.833 But I was starting to think about, like, applying some of this 00:51:37.833 --> 00:51:40.780 to other environments. If you went downtown San Francisco, 00:51:40.780 --> 00:51:43.645 there’s a ton of buildings of similar heights. 00:51:43.645 --> 00:51:46.980 Or I know we’ve talked about doing some – you know, running some 00:51:46.980 --> 00:51:51.083 fiber tests and seeing if we could manage to get fiber that we could 00:51:51.083 --> 00:51:54.230 use at Moffett Field. But, again, there, most of the buildings are fairly low. 00:51:54.230 --> 00:51:56.419 There’s some that are pretty massive. 00:51:56.419 --> 00:51:58.888 I don’t know if NASA wants us monitoring their wind tunnels. 00:51:58.888 --> 00:52:01.660 But what do you think it takes 00:52:01.660 --> 00:52:06.910 to be able to do structural monitoring from the far-field like that, 00:52:06.910 --> 00:52:09.670 you know, with fiber going past buildings? 00:52:09.670 --> 00:52:14.708 What sort of setup do you need for the – sort of the neighborhood? 00:52:15.833 --> 00:52:20.010 - Well, first of all, I think you need the building to actually emit enough energy. 00:52:20.010 --> 00:52:24.050 And this is probably going to be an issue for shorter buildings. 00:52:24.050 --> 00:52:27.050 It turns out, at Caltech, you’re right. It’s very uneven. 00:52:27.050 --> 00:52:31.021 We can actually – people have played around before with using the forced vibration 00:52:31.021 --> 00:52:34.794 of Caltech Hall to do forced vibration tests of adjacent buildings. 00:52:34.794 --> 00:52:38.940 Because it emits enough energy, because the building is just so massive, right, 00:52:38.940 --> 00:52:43.350 that once you get it going, even off resonance, it weighs so much 00:52:43.350 --> 00:52:48.030 that it’s just creating very large displacements in the soil. 00:52:48.030 --> 00:52:50.121 So I think that that’s going to be a problem for short buildings 00:52:50.121 --> 00:52:54.075 is you can’t get them to move that much. The overall displacements and 00:52:54.075 --> 00:52:58.410 the overall accelerations are just going to be smaller. 00:52:58.410 --> 00:53:02.860 I think that, in the context of multiple buildings, though, this is – this is the – 00:53:02.860 --> 00:53:07.958 so when I showed it very briefly here with the imaging, these turbines 00:53:07.958 --> 00:53:10.880 are only separated by a few hundred meters from each other. 00:53:10.880 --> 00:53:12.050 - Oh, yeah. 00:53:12.050 --> 00:53:14.220 - And they all have the exact same frequency because they 00:53:14.220 --> 00:53:15.890 have identical construction. 00:53:15.890 --> 00:53:19.150 And so the advantage with DAS is that you really have enough channels, 00:53:19.150 --> 00:53:20.880 and they’re in a complex enough geometry, 00:53:20.880 --> 00:53:24.730 like this bend here, where you can actually do imaging, right? 00:53:24.730 --> 00:53:26.630 When you have a line, it’s not very helpful 00:53:26.630 --> 00:53:30.440 from a knowledge-gain perspective. But, if you have a lot of channels 00:53:30.440 --> 00:53:35.100 in a diverse orientation, you can totally localize these vibrations 00:53:35.100 --> 00:53:37.450 back to structures that are only separated by a few hundred meters, 00:53:37.450 --> 00:53:40.930 even if they have the same frequency, based on, you know, the relative travel 00:53:40.930 --> 00:53:45.950 time of those vibrations across the array. So I think that just the large-N nature 00:53:45.950 --> 00:53:49.552 of DAS itself should make that possible in a place like San Francisco where it’s, 00:53:49.552 --> 00:53:53.231 you know, building after building right next to each other. 00:53:53.231 --> 00:53:55.970 - Yeah, of course, I mean – right, a few hundred meters would be 00:53:55.970 --> 00:53:58.980 a pretty big spread between buildings in a downtown area. 00:53:58.980 --> 00:54:00.980 - Right. - But, with more data, 00:54:00.980 --> 00:54:02.981 you may be able to knock that down. 00:54:02.981 --> 00:54:04.458 - Well … - I mean, if you knock it down 00:54:04.458 --> 00:54:07.000 by an order of magnitude, you’re there, yeah. 00:54:07.000 --> 00:54:09.458 - And if you actually go in and out of the buildings, right, then you’ve got 00:54:09.458 --> 00:54:12.497 some non-ambiguous … - Right. 00:54:12.497 --> 00:54:14.153 - … channels. 00:54:14.153 --> 00:54:15.786 - Cool. Thanks. It was a really interesting talk. 00:54:15.786 --> 00:54:17.396 Thanks, Ethan. 00:54:19.794 --> 00:54:22.145 - Jamie, do you want to ask a question? 00:54:23.317 --> 00:54:26.028 - Yeah, sure. Great talk, Ethan. 00:54:26.028 --> 00:54:33.208 Really happy to see somebody advocate for recording continuously in structures and 00:54:33.208 --> 00:54:38.820 geotech arrays. That’s really where my heart is, the geotech arrays. 00:54:38.820 --> 00:54:45.271 And it was interesting that – if I – if I got you correctly 00:54:45.271 --> 00:54:48.771 in the first half of this talk – I was watching on my phone, and so correct me 00:54:48.771 --> 00:54:54.614 if I’m wrong, but is – the building stiffens when you have precipitation. 00:54:54.614 --> 00:54:56.880 Is that correct? - Yes. 00:54:56.880 --> 00:55:02.550 - Yeah, which is – you know, what we see in – at Garner Valley, 00:55:02.550 --> 00:55:07.278 one of our geotech arrays, where we have a small slab structure 00:55:07.278 --> 00:55:11.120 and a cross-hole array that runs underneath that slab, 00:55:11.120 --> 00:55:14.114 and we can measure the velocity of the soil 00:55:14.114 --> 00:55:17.820 as the water table goes up and down with, you know, seasonal changes. 00:55:17.820 --> 00:55:24.208 We actually see the velocity of the soil decreases as the water table goes up. 00:55:24.208 --> 00:55:27.220 And we attribute that to excess pore pressure, 00:55:27.220 --> 00:55:31.190 reducing the stiffness of the soil. And maybe it’s – I don’t know 00:55:31.190 --> 00:55:35.983 if you have measurements of where the water table is at Millikan. 00:55:37.230 --> 00:55:40.492 Or maybe it would be interesting to put a cross-hole array in and measure 00:55:40.492 --> 00:55:42.330 the properties of the soil. 00:55:42.330 --> 00:55:49.271 But it’s possible, if you’re in the unsaturated, where you have, you know, 00:55:49.271 --> 00:55:51.400 some moisture, but it’s not saturated, 00:55:51.400 --> 00:55:53.119 you could see an increase in velocity. 00:55:53.119 --> 00:55:56.490 So maybe that’s what’s happening there. I don’t know. 00:55:56.490 --> 00:56:00.369 Do you have a feel for where the water table is below the building? 00:56:00.369 --> 00:56:02.296 - I don’t. 00:56:03.390 --> 00:56:05.750 When they built the building, they said it was 15 meters. 00:56:05.750 --> 00:56:09.020 And subsequently, people have never found it again. 00:56:09.020 --> 00:56:11.614 So it’s a open question. 00:56:11.614 --> 00:56:14.790 I don’t know what – maybe we’ve changed it somehow by building things around there. 00:56:14.790 --> 00:56:18.770 But theoretically, it’s quite shallow, but we’ve not been able to find it. 00:56:18.770 --> 00:56:21.434 - How many meters you said? - Fifteen. 00:56:21.434 --> 00:56:24.271 - Fifteen. Okay, yeah. So that … 00:56:26.544 --> 00:56:27.497 Yeah, okay. 00:56:27.497 --> 00:56:33.208 So that could be – it could be that your – when you add water in the soil, 00:56:33.208 --> 00:56:35.010 but it’s not actually saturated because 00:56:35.010 --> 00:56:38.609 the water table doesn’t come up that high, that you do actually stiffen the soil. 00:56:38.609 --> 00:56:41.609 And that could contribute to what you’re seeing in the data. 00:56:41.609 --> 00:56:44.583 So it could be a combination of – maybe it’s not – 00:56:44.583 --> 00:56:46.240 has nothing to do with the structure. 00:56:46.240 --> 00:56:50.000 Maybe it just has to do with the stiffening of the soil. 00:56:50.000 --> 00:56:52.692 But that’s something that you could measure separately, 00:56:52.692 --> 00:56:54.184 which would be interesting. 00:56:54.184 --> 00:56:55.900 - You’re right. It would be interesting to put, 00:56:55.900 --> 00:56:58.800 like, an array in just for that purpose here. 00:56:58.800 --> 00:57:02.451 I think it’s also a little hard to tell what’s important, whether it’s an increase 00:57:02.451 --> 00:57:04.720 in the shear modulus or the bulk modulus of the soil 00:57:04.720 --> 00:57:05.970 that really matters most, right? 00:57:05.970 --> 00:57:08.583 If you talk about shearing of the foundation, then you care about 00:57:08.583 --> 00:57:11.710 the shear velocity adjacent in the soil. 00:57:11.710 --> 00:57:15.146 But if you care about the rocking of the foundation, then you care about 00:57:15.146 --> 00:57:18.091 the bulk modulus of the soil under the foundation. 00:57:18.091 --> 00:57:19.890 Right, it’s going to provide the main resistance. 00:57:19.890 --> 00:57:24.060 And so it’s hard to know whether you really should be looking at Vp or Vs, which, obviously, 00:57:24.060 --> 00:57:28.067 for some saturation models, have an opposite trend. 00:57:29.895 --> 00:57:31.396 Yeah. Yeah, exactly. 00:57:31.397 --> 00:57:36.036 [echoing and feedback] 00:57:36.036 --> 00:57:40.850 [silence] 00:57:40.850 --> 00:57:43.570 Not sure what was going on there. [laughs] 00:57:45.138 --> 00:57:46.590 Well, thanks. That was really great. 00:57:46.590 --> 00:57:49.934 I’ll look – I’ll look for you at the SCEC meeting because I’m interested 00:57:49.934 --> 00:57:51.684 in talking to you some more about it. Great. 00:57:51.684 --> 00:57:53.067 - Yep. 00:57:56.388 --> 00:57:59.122 - Alan Yong, you have a question? 00:57:59.122 --> 00:58:00.286 - Yeah. 00:58:00.286 --> 00:58:04.833 Nice talk, Evan – or, Ethan. I’m sorry. [laughs] 00:58:04.833 --> 00:58:11.350 I understand, in your last slide, you indicated that, you know, 00:58:11.350 --> 00:58:17.080 for your arrays, you could use phase velocity to get to the time average, 00:58:17.080 --> 00:58:21.390 shear wave velocity at the upper 30 meter, or Vs30. 00:58:21.390 --> 00:58:27.771 And I wonder if you can expand a little bit on, like, when you say phase velocity 00:58:27.771 --> 00:58:36.247 to Vs30, are you talking about the customary inversion component in between? 00:58:36.247 --> 00:58:41.290 - Right. I just mean, like, MASW, or SASW, the classic surface wave 00:58:41.290 --> 00:58:43.610 methods, which is basically what we did here. 00:58:43.610 --> 00:58:46.780 We applied interferometry to ambient noise, picked, you know, 00:58:46.780 --> 00:58:50.810 some of these different modes out for their dispersion using 00:58:50.810 --> 00:58:53.780 array-based transformation. Because basically, 00:58:53.780 --> 00:58:57.570 what you can do here is treat it like a – you know, a small geophone array. 00:58:57.570 --> 00:59:04.021 And then inverted that for – in this case, we used a modified power law velocity 00:59:04.021 --> 00:59:07.460 because that’s really a pretty good approximation of shallow seafloor 00:59:07.460 --> 00:59:11.850 sediments, but, you know, you could do more arbitrary layered models, 00:59:11.850 --> 00:59:16.990 or whatever you would normally do for MASW. 00:59:18.067 --> 00:59:24.739 - Nice. I would – I’d encourage you to visit Jose Gomez’s poster at SCEC. 00:59:24.739 --> 00:59:28.890 - Mm-hmm? - He’s leading an effort with 00:59:28.890 --> 00:59:34.208 a bunch of us to circumvent this whole inversion component 00:59:34.208 --> 00:59:40.290 in between phase velocity to Vs30. And Jose is on this call right now. 00:59:40.290 --> 00:59:48.660 He may or may not be able to speak up on this, but I think that going from, 00:59:48.660 --> 00:59:55.070 say, just phase velocity to inversion to Vs30 is – that middle step 00:59:55.070 --> 00:59:59.325 is actually not – I feel it’s not necessarily needed … 00:59:59.325 --> 01:00:00.880 - I agree. - … if you’re in sedimentary 01:00:00.880 --> 01:00:03.390 environments. You can look at the phase velocity 01:00:03.390 --> 01:00:08.271 and almost have a one-to-one, you know, estimate of phase velocity to 01:00:08.271 --> 01:00:11.790 Vs30 itself, right? So that’s … 01:00:11.790 --> 01:00:13.030 - I think you could even go further with it. 01:00:13.030 --> 01:00:17.410 I had a conversation about this once with Victor Tsai, who has been 01:00:17.410 --> 01:00:20.500 trying to develop some sort of, you know, semi-empirical 01:00:20.500 --> 01:00:23.802 approximations that work well for surface wave properties. 01:00:23.802 --> 01:00:26.521 And I think that, you know, if you take this from – like Thomson-Haskell 01:00:26.521 --> 01:00:29.070 framework, right, you really care about two things. 01:00:29.070 --> 01:00:31.860 One is you’re measuring phase velocity dispersion, which is basically 01:00:31.860 --> 01:00:35.500 the determinant of your Haskell-Thomson matrix. 01:00:35.500 --> 01:00:39.630 And on the far other side of things, in geotechnical analysis, 01:00:39.630 --> 01:00:43.830 what you care about for, like, Seed and Idriss-type site response 01:00:43.830 --> 01:00:48.780 is just the product, you know, of that Thomson-Haskell matrix 01:00:48.780 --> 01:00:51.210 multiplied by some additional damping term. 01:00:51.210 --> 01:00:53.771 And so it seems like there should really be – like, you shouldn’t first 01:00:53.771 --> 01:00:58.650 invert for the modulus of that matrix and then multiply it by that matrix again. 01:00:58.650 --> 01:01:02.646 Right, that there should be basically one unified step that should be 01:01:02.646 --> 01:01:08.930 at least approximatable for most – you know, certain classes of layered soils. 01:01:08.930 --> 01:01:09.930 - Right. 01:01:09.930 --> 01:01:12.970 - And nobody’s really gone far enough to develop that yet so that we can go straight 01:01:12.970 --> 01:01:16.646 from phase velocity into ground motion without having to make any sort of 01:01:16.646 --> 01:01:22.190 layered medium assumptions in between or do any sort of damping in the inversion. 01:01:23.075 --> 01:01:26.920 - Right, right. I remember Victor’s paper with – I think – it’s somebody else 01:01:26.920 --> 01:01:33.350 at Golden at USGS, but this is only applicable for sedimentary 01:01:33.350 --> 01:01:39.869 environments which also – you know, one has to say that Vs30 only works, really, 01:01:39.869 --> 01:01:43.622 at sedimentary environments, right, as a proxy for site amplification. 01:01:43.622 --> 01:01:45.069 - Yeah. - So – right, right. 01:01:45.069 --> 01:01:46.458 Thanks. Thanks, Ethan. 01:01:46.458 --> 01:01:48.069 I really appreciate this. 01:01:48.069 --> 01:01:50.942 And I do encourage you to visit Jose Gomez’s poster at SCEC. 01:01:50.942 --> 01:01:52.590 - Will do. 01:01:52.590 --> 01:01:53.819 - Okay, thank you. 01:01:57.817 --> 01:02:02.040 - Ethan, there’s one question in the chat from Paul Bodin. 01:02:02.040 --> 01:02:03.641 Connecting the two parts of your great talk. 01:02:03.641 --> 01:02:07.458 Is there potential to use DAS in fibers running into all those buildings 01:02:07.458 --> 01:02:09.546 to monitor their responses? 01:02:10.990 --> 01:02:13.966 - I think the answer is probably yes. 01:02:13.966 --> 01:02:19.609 We have – we’ve thought about doing this with Caltech Hall. 01:02:19.609 --> 01:02:22.146 And part of the problem is – if I just, like, schematically go back 01:02:22.146 --> 01:02:24.358 all the way to the beginning … 01:02:25.848 --> 01:02:28.030 Where is the fiber actually in the building, right? 01:02:28.030 --> 01:02:31.458 You know, when you’re coming down here across campus, the fiber is – 01:02:31.458 --> 01:02:33.583 can you see my mouse? I’m going to switch back to the laser pointer. 01:02:33.583 --> 01:02:36.050 It’s buried in some conduit, right? 01:02:36.050 --> 01:02:40.750 And then it comes up into the building, and then there’s a giant spool in the wall of, 01:02:40.750 --> 01:02:43.430 like, you know, some closet where they do the telecom. 01:02:43.430 --> 01:02:44.619 And then it hangs freely 01:02:44.619 --> 01:02:48.000 in a wall cavity as it goes in between the floors going up. 01:02:48.000 --> 01:02:50.960 And so, when we looked at this building-related data, 01:02:50.960 --> 01:02:55.210 at least for most civil structures, you don’t really get many channels 01:02:55.210 --> 01:02:58.372 that are coupled to anything inside the actual building. 01:02:58.372 --> 01:03:02.208 And, if you did, at least theoretically, you’d mostly be measuring 01:03:02.208 --> 01:03:06.325 the vertical strain in the walls, which is a really strange metric. 01:03:06.325 --> 01:03:10.208 You know, as we saw – just looking at the east-west versus the north-south 01:03:10.208 --> 01:03:14.580 accelerations in the roof of the building, you would get dramatically 01:03:14.580 --> 01:03:19.521 different results because it really is two only mildly coupled 01:03:19.521 --> 01:03:24.050 separate vibrational systems in orthogonal directions that are mostly horizontal. 01:03:24.050 --> 01:03:28.271 And therefore, you know, measuring the vertical strain in a wall component or 01:03:28.271 --> 01:03:32.619 some – you know, it would be really hard to interpret that data. 01:03:32.619 --> 01:03:37.521 That being said, there’s been some great work by the Soga group at Berkeley 01:03:37.521 --> 01:03:42.550 civil engineering working on embedding DAS arrays into wind turbines, bridges, 01:03:42.550 --> 01:03:45.160 and other structures. And so, if you actually know 01:03:45.160 --> 01:03:47.940 where the channels are, and you’ve made effort to tape them down 01:03:47.940 --> 01:03:51.020 or lay them into some sort of epoxy so that the fiber is actually coupled 01:03:51.020 --> 01:03:54.950 to the building, and you know where it is, then you can learn a lot. 01:03:54.950 --> 01:03:57.000 They’ve done some really great analysis with that. 01:03:57.000 --> 01:04:00.522 But, as far as pre-existing fibers, I think you learn more next to the building 01:04:00.522 --> 01:04:04.359 than you do in the building, mostly because of the vertical hanging problem. 01:04:05.598 --> 01:04:07.983 - Cool, thanks. Makes sense. 01:04:10.277 --> 01:04:15.519 [silence] 01:04:15.520 --> 01:04:20.046 - Any other questions for Ethan before we sign off? 01:04:23.747 --> 01:04:28.160 Okay. If not, let’s thank Ethan. 01:04:28.160 --> 01:04:31.580 And we can stick around in this chat if anyone wants to 01:04:31.580 --> 01:04:37.170 kind of casually talk, but otherwise, that concludes today’s seminar, 01:04:37.170 --> 01:04:40.396 so thank you very much for coming.