The Monte Carlo method conjures images of a suave gambler beating the house in Monaco. In reality, Monte Carlo methods are computational algorithms that use randomness to solve problems.
In this episode we hear from Todd Palmer, professor of nuclear engineering, about his use of Monte Carlo simulations in nuclear fuel research. In addition, find out how Haizhong Wang , assistant professor in transportation engineering, is incorporating Monte Carlo to model tsunami evacuation routes in the Cascadia Subduction Zone.
NARRATOR: From the College of Engineering at Oregon State University, this is Engineering Out Loud.
STEVE FRANDZEL: One night in 1946, the brilliant young mathematician Stanislaw Ulam was rushed to the hospital with violent headaches caused by encephalitis. He lapsed into a coma, but survived and recovered. During his many of weeks of recovery, he played game after game of solitaire,
and like a true mathematician, wondered just what the odds were of winning the game. He quickly realized that solving the problem mathematically would be nearly impossible. Much more elegant would be to simulate 100 games, then count the number of wins.
[Music: Wes Hutchinson, Plastic or Paper, copyright free via YouTube Audio Library]
That was the humble birth of a technique that changed world and today is used in any field that applies statistics. That’s pretty much any one you can think of. Ulam called it Monte Carlo, supposedly for the casino where his uncle gambled
TODD PALMER: The Monte Carlo Method is a way of modeling physical systems using randomness and probability.
FRANDZEL: That’s Todd Palmer, professor of nuclear engineering, who frequently uses Monte Carlo simulations in his work. I’m Steve Frandzel from the College of Engineering, and in this segment of Engineering Out Loud, we’ll explore the Monte Carlo method, how it’s used to work out extremely complicated problems, and how it almost saved a nuclear reactor more than 6,000 miles from the Oregon State campus.
PALMER: Monte Carlo is the most mathematically simple, physically intuitive approach for addressing very complicated problems. You just have to envision what’s happening and assign distribution to the various possible outcomes and then you can model. We use these probability distributions to describe what happens to radiation that’s born in the reactor core when it moves through the background medium. If you know where the radiation is produced, its direction and its energy, now you can ask what’s the likelihood that it will travel a certain distance before it collides a nucleus or the chance that it escapes the fuel and interacts with another medium like the reactor coolant. Essentially you’re just simulating the life cycle of the radiation in the core.
FRANDZEL: As powerful as it is, Monte Carlo still has a glaring drawback.
PALMER: Well, the major one is that Monte Carlo is very slow.
Of all the algorithms, it’s easily one of the slowest ways of solving these types of problems. The time it takes to run a simulation varies depending on the speed of the computer and the number of samples that you use in the problem. The more samples, the smaller the error. Suppose we’re trying to estimate if a certain configuration of reactor fuel creates a critical state, meaning the reactor is operating safely at a steady state power level. A simulation like that can take on the order of 10 hours on computers that we have here at Oregon State, and that’s a relatively simple simulation. If you have a problem where time is a variable, simulations take much longer. For instance, if a reactor runs, it’s using up its fuel, this is a time-dependent problem, and that calculation can take on the order of two weeks of computer time. Essentially we’re trading speed for accuracy. Monte Carlo offers the benefit of modeling all the physics as accurately as you really can, but it can also be very slow.
FRANDZEL: And big – really big, when it comes to the amount of data involved.
PALMER: To get the fidelity we need tomake decisions about whether a reactor is efficient and safe, we need on the order of 2 billion individual pieces of information from a simulation – lots and lots of data, huge amounts. Other problems are simpler, but the scope of the reactor problem is particularly large.
FRANDZEL: Palmer also points out that nuclear engineering led directly to the high-speed computers that are capable of processing huge amounts of data and which makes Monte Carlo simulation possible at all.
PALMER: Digital computing came out of the same people who were doing nuclear engineering in the Manhattan Project. There’s a closer coupling between high-performance computers and nuclear engineering than there is with any other field. It sprang from the same well and I feel a really strong familial relationship with that. Back in the days of the Manhattan Project, the word “computer” referred to women who were handed calculations to do. They did their calculations, got a result, handed it to the next person, who took that result and did something else. They were doing computing on a very large scale with humans.
[mechanical printer and sheet of paper tearing]
FRANDZEL: In one of his research projects, Palmer used Monte Carlo to simulate what would happen if one type of nuclear fuel in a small but important research reactor in Uzbekistan was replaced by another type of fuel. But first, it helps to know the function and importance of research reactors, which are tiny versions of the giant nuclear power plants we’re used to hearing about.
PALMER: Primarily, the purpose of research reactors is to produce neutrons, and we use those neutrons to irradiate a known sample of a particular material that’s been sliced into a thin section, and we’ll measure what comes out. So if you know the thickness of that material and how many neutrons you’re using to bombard it, then you can estimate the number of the interactions, like absorption or scattering—all the different types of interactions that radiation can have. One of the applications that we’re interested in is nuclear forensics—trying to understand the origin of either pre-detonation or post-detonation material after a nuclear weapon is exploded. So that’s material that has been affected, for instance, by a nuclear blast. Essentially, you need neutrons to do this, and a reactor is a good source of that.
FRANDZEL: For many years, the Uzbekistan reactor has been a reliable source of those neutrons, and researchers from prominent laboratories, and the United States government itself, have come to rely on it.
PALMER: Researchers at places like Pacific Northwest National Lab and the National Nuclear Security Administration, which is part of the DOE, the U.S. Department of Energy, use the reactor in Uzbekistan to irradiate materials with neutrons. This reactor is very economical and the work can be done on a very short turnaround time. There are reactors in the U.S. that can do this work as well, but they’re availability is limited. Typically they’re booked three to six months, maybe to a year, in advance. So this reactor is available to us, but Russia supplies the fuel.
[MUSIC: military marching song]
It’s a particular type of fuel called plate fuel. The Russians have a monopoly on this particular plate fuel, and they’re very good at making it, but they keep raising the price and squeezing Uzbekistan for more money.
[cash register, cha ching]
Essentially they can’t afford it. Russia is saying if you don’t give us the money we need for the fuel, you won’t get any fuel, and without fuel the reactor will cease to operate. Those operators will lose their jobs, it’s a threat to those particular technical folks and their standard of living and what they’re doing with the science. It’s also not good for the United States because it disrupts a reliable source of neutrons.
FRANDZEL: There is one possibility: Alter the reactor so that it can use another type of fuel. But is that really practical?
PALMER: The replacement of the core will be expensive—millions of dollars. And we can’t afford to do that without knowing if it will work beforehand. So we use Monte Carlo to simulate it. We already know how the reactor works with plate fuel, we have models for that as well. We kicked around the idea of showing that you could operate this reactor using pin-type fuel, which is housed in long, cylindrical rods. It’s abundant and it’s inexpensive. We’re talking about redesigning the core using a totally different kind of fuel, but keeping the facilities the same. It’s a lot like yanking the combustion engine out of your car and putting in an electric engine. Hopefully, you get the same performance and all the safety systems and the brakes will still work. We needed to predict the behavior of the reactor using the pin fuel, things like how long will it run before it needs refueling, how much power will each fuel pin produce, what are the radiation fluxes inside the reactor, is it safe? So we ran these simulations and we said, hey, they look pretty close, so much so that you could probably do this.
FRANDZEL: For a while, Palmer was optimistic about the possibilities of saving the reactor.
PALMER: If we’d talked about this about six months ago, I would have said things are looking really good. The core is going to work out and there’s support for this in the Department of Energy. They certainly thought this when they funded the work in the first place when they funded the work in the first place to look at this. But now we’re hearing that the DOE isn’t all that interested anymore. It seems much more likely that when the current fuel supply is exhausted, this reactor will shut down and it may never start up again.
FRANDZEL: Unfortunately, Palmer was right about that. Uzbekistan decided to close the reactor permanently.
[Music: Wes Hutchinson, Plastic or Paper, copyright free via YouTube Audio Library]
Even though events overtook him, Palmer’s work on the reactor problem illustrates the power of Monte Carlo. But it also raises questions about the role, and even the definition, of data science in engineering. What exactly is it?
PALMER: This is a tough question. The way I think of it is it’s about understanding and using information in the best way possible to solve problems. It involves things like quantifying the uncertainty of the data, knowing how good the actual information is, visualizing it, interpreting it, and then making decision based on the results. I think this last piece is the hardest part. As scientists and engineers we tend to stop after we complete the calculation. So here’s the calculation and the associated uncertainty, and basically I’m out of here. But then the question is, really, what does that mean and how does a policy maker or decision maker use that information. The information we get from our calculation is it’s not just numbers, it has a relevance in the real world.
[cards shuffling, transitions to tsunami wave crashing into boat and glass breaking]
JENS ODEGAARD: Whoa, that turned scary quick. One minute, we're hanging out on the deck of a boat, playing cards, and the next, the big one hits.
RACHEL ROBERTSON: It's unpredictable.
ODEGAARD: Welcome to Part Two of the episode. I'm Jens Odegaard.
ROBERTSON: And I'm Rachel Robertson, and we are from the College of Engineering at Oregon State University.
Both of us actually grew up in Oregon, living in this zone, this earthquake zone. But, I don't remember anyone really talking about it – do you?
ODEGAARD: No, I don't. You know, I grew up just the Cascade Mountains on the east side, and certainly no one was ever really talking about the Cascadia Subduction Zone over there. It was more talk about, you know, there's a few above-ground faults in Eastern Oregon that historically happened, and big earthquakes over there, but that's totally a thing of the past. And then the other big talking point was always the San Andreas Fault, in Southern California – you know, the "big one" that's going to knock LA off into the Pacific Ocean, and cause major havoc.
ROBERTSON: Right. But, we've all become more aware of the danger. So, fortunately, we have scientists here at Oregon State University who are doing a lot of research on earthquakes and helping us build buildings that are safer. And also Haizhong Wang, who we're going to be talking to later in the podcast – he is doing research on looking at how people evacuate from an earthquake.
ODEGAARD: Yeah, actually Johanna Carson, who's one of the members of our podcast team, interviewed Dr. Wang a while back about his work, and we stepped in to take over her podcast, because she actually had to run off to Nepal – um, she got the opportunity to go interview folks doing fieldwork in Nepal.
ROBERTSON: Yeah, it's pretty cool that she got that chance to do that, and, you know, we're not bitter at all that she left us behind to finish the podcast, but, I got an opportunity to talk to Johanna on the phone about what she's doing there.
JOHANNA CARSON: Hello? Hi, Rachel.
ROBERTSON: Hey, Johanna! How are you?
CARSON: I'm very well, how are you?
ROBERTSON: Good. It's awesome to hear from you, all the way from Nepal, that's crazy.
CARSON: It is crazy.
ROBERTSON: So, why don't you just tell me what you're doing in Nepal.
CARSON: Yeah. Well, I'm here with Ben Mason, an Assistant Professor in Geotechnical Engineering here at Oregon State, as well as Domniki Asimaki, who's a Professor of Civil and Environmental Engineering at Caltech. And, they are here conducting a workshop for local engineers, here in Nepal, on ground failure risk and mitigation.
ROBERTSON: Okay, cool. So, what has been your experience so far with interacting with the local people there in Nepal?
CARSON: Well, it has been overwhelming in a wonderful sense. I've met a lot of warm, just wonderful, lovely people, and, particularly at the workshop, I was able to meet an engineer who took me out on a tour of a few of the schools that were impacted by the April 2015 Gorkha earthquake here in Nepal. One of the schools was completely pancaked - it just collapsed. Here in Nepal, they use the term sandwiched. But, subsequently, that school has reconstructed, so I was able to see what had occurred since the earthquake, and then we also went to another school that survived the earthquake quite well, and it was because it had been retrofitted prior to the earthquake, through the program called "NSET Schools," and NSET stands for The National Society for Earthquake Technology, Nepal. And, they're an organization that has been around for twenty years, and they've just done a tremendous amount of work in both, you know, the technical aspect of going out into the field and doing these retrofits to training engineers, training masons, training the general public in what to do during an earthquake, as well.
ROBERTSON: Okay, cool. So, there is hope that, you know, you can actually build the building so that it will withstand, then, a pretty big earthquake.
CARSON: Yes. Yes, there is. And I was able to see that.
ROBERTSON: Mhm. And so, what is your take-home message?
CARSON: Um, my take-home message is that, you know, not only are Dr. Mason and Dr. Asimaki here, providing their expertise to local engineers here in Nepal, to make Nepal a safer place, but we're also learning a lot, here, some things that can be applied to us back home in the Willamette Valley, to make that safer place for us as well.
ROBERTSON: Okay, great. Well, thanks so much for taking the time talk to me. I really appreciate it. I guess
CARSON: Yeah, well, you know what? I appreciate you – you aren't going to believe the fruit baskets that you and Jens are going to get.
ROBERTSON: [Laughs] Yeah, well we're having fun...
[End of recording]
ROBERTSON: So, there are actually several researchers in the College of Engineering that study earthquakes and tsunamis. For example, our dean, Scott Ashford, is an international expert on the impact of subduction zone-earthquakes in much of the Pacific Rim, including Japan's major disaster in 2011. He's also the director Cascadia Lifeline Program, which is a collaborative research initiative of eight partners from government and private industry, and what they're doing is they're trying to improve the infrastructure, like roads and bridges, so that they perform better in an earthquake.
ODEGAARD: Yeah, so his research is really related around the stuff that Johanna is over in Nepal with those guys about – it's more about the structure and about resiliency in the earthquake, or while the tsunami's happening. But, it's also, okay, so we have to have the structure and the infrastructure, but what about the humans involved? What do we do after the earthquake and the tsunami hits? And, you know, we mentioned that Johanna talked to Haizhong Wang, and his research is all about that – he's an assistant professor of transportation engineering here in the College of Engineering, and he actually models earthquake evacuation scenarios.
HAIZHONG WANG: The scenario we're trying to capture is the worst scenario, is the really big one we're talking about is a magnitude 9.0 earthquake and tsunami. So that's the worst event that could happen to the region.
ROBERTSON: So, this has been a long introduction about the earthquake research happening at Oregon State, but we want to get back to the topic of data science and engineering. So, Jens, bring us back to the point… how is this related to the Monte Carlo method?
ODEGAARD: Dr. Wang uses the Monte Carlo method in his simulations of how evacuee decision-making behavior impacts traffic flow and then he validates these models with real data. He takes various transportation modes, such as driving, or riding your bike, or walking into account. It’s computationally intensive. And as he explains, it can be tricky business.
WANG: the earthquake and the following multi-hazard tsunami if it's a near-field earthquake in the ocean that will trigger a tsunami and other type of secondary hazards, landslides and soil liquefactions and all those are, because that in nature is not just a one-type hazard it's a multi-hazard in nature. All these complications are actually creating the additional complexities in that modeling effort.
ROBERTSON: To simulate how individuals would flee hazardous area Dr. Wang uses a method called agent-based modeling.
WANG: there are other approaches, but agent-based is a microscopic simulation tool is more looking at an individual, and from the individual to study the collective, emotional collective behavior. It's not a predictive approach. We can understand it as more as a scenario-evaluation approach. We try to mimic a real-world setting. The agent-based approach provide a lot of flexibility in terms of, for example, when we try to accommodate a heterogeneous population group. For example, they are elderly and a different population types. And also, the decision making, the different kind of decision making that behavior that they have and how does that affect life-safety, the mobility and life safety in general.
ODEGAARD: So, now we come to how it involves the Monte Carlo method. To account for random events, Dr. Wang uses this method in his models. Developed at the Los Alamos National Laboratory, the Monte Carlo method is a way to insert randomness into models.
WANG: In this modeling work, or in the analysis, there's a lot of uncertainties. Even when taking into account a lot of the different decision making: people's departure time and there's a lot of uncertainties associated with the simulation itself. So if we for example using a one-shot simulation, only run the simulation once and the results are, if we draw conclusions from that one-shot simulation, the results is not as convincing. But the Monte Carlo simulation allows us to run the scenarios, for example, a thousand times, or even more than that, and to look at in that thousand or ten thousand times, how often does this scenario emerge.
ROBERTSON: Once Dr. Wang creates a simulation he validates it with real data, which can also be tricky.
WANG: In our case, in natural disaster research, you know one challenge is the lack of empirical data. We don't know when it will happen. So it's not something we can set up to collect the data. A lot of the data in this setting is either post-disaster, or post-event survey, and you can only survey people who survived.
ROBERTSON: So, as with many of the stories of research in our podcast season on data science and engineering, advances in the speed and storage capabilities of computers has completely changed what science is capable of. Wang said that when Monte Carlo simulations were first proposed computers did not have the capacity to even carry them out.
WANG: With the increasing, rapidly increasing computational development of high power computers, Monte Carlo simulation is very popular now, almost in every discipline.
ROBERTSON: And the smartphones we all carry around are another boon to science.
WANG: . For example, in 2011, the Japan, the Japan event the Tōhoku event, one company they actually unintentionally collected people's location data and the travel. For example, the trajectory. If you are using smartphones, if you are walking, leaving from one area, to a safety shelter and your data is recorded. If you're driving using GPS, your data is also recorded. You know with the people's access to the smart devices, and especially over the past few years, you know the smartphones and other smart devices could be connected and could be using as a tool to collect the data. And we're experiencing really from a data desert to a data ocean.
ODEGAARD: So, these scenarios are not just thought experiments. The work of Dr. Wang and his collaborators could have direct impact on those of us living in Cascadia Subduction Zone.
WANG: The model actually can be used as a more of a scenario simulation or management to inform decisions makings. For example, like city of Seaside, they wanted to design certain number of vertical tsunami evacuation structures and where those structures should be located. Or which are the optimum locations to put those shelters, the vertical evacuation shelters that can generate the maximum safety outcome. And this model can be used as a tool to evaluate that. For example, we put the shelters in the different locations in the model and ran that simulation using Monte Carlo simulation and to evaluate why this location is better than the others, and can provide the evidence or support to that decision-making process.
ROBERTSON: So, we've learned a little bit about what researchers are doing at Oregon State, you know, related to earthquakes and tsunamis. But, people might want to know what they can do, practically -- if they're on the coast, at Seaside or somewhere else -- what should they do if "The Big One" hits?
ODEGAARD: Yeah, that's a question I ask myself quite a bit. I surf quite often at the coast. Newport is just west of us, about fifty minutes from Corvallis here, and so that's the region I'm usually in, but it has similarities to Seaside, or any other coastal town up and down Washington and Oregon, and Northern California, and so, the question is, if you're out surfing, or if you're out on the bayfront, or you're just hanging out with family, how do you know "The Big One" has hit? And, how do you know the tsunami is coming? In the past, there's been physical air-raid sirens that'll blast, you know, that classic air-raid noise, and that will alert you, and some of those are still in effect. But, really, modern technology has shifted some of the alerts to wireless "push" notifications, like to your cellphone. It might not help you if you're surfing, but if you're actually on the coast, and you have your cellphone, it'll push a notification to alert you to get out of the area, and that the earthquake's hit, the tsunami's coming. So, that'll let you know that that's happening, and then I guess from there, it's, what's your response?
ROBERTSON: So, once it happens, what do you do?
ODEGAARD: Yeah, well, I mean, the research shows that if the really, the really big one hits, we have like maybe like, ten to fifteen minutes, and if it's just the "regular" big one, you know, it's a little bit longer, but not much longer, in response time, to get out of the inundation zone. So, the earthquake hits, and it sends this wave, the tsunami, toward the coast, and you have maybe ten to twenty minutes to get out of the low-lying areas and to a safer place, a higher vantage point. So, you know, really, you have to know what areas will be effected, if you're at the coast, and how to get out of them. I'll give you a practical example: in Newport, the bayfront is, really low-lying, and if you're on the south side of the bridge, there's really only one area to go. And, I often surf off of South Beach there. And, the only way out of the inundation zone would be to book it back up the sea wall, and run up this hill that's right there, and it's just tall enough to get you out of the inundation zone. But, if I was unprepared before I went surfing, I would have no way of knowing where I was supposed to go or how to get up that; that's the only safe hill. Because there's other sand dunes that look tall enough, but if the tsunami actually hits, you'd be underwater.
ROBERTSON: Okay, so be aware of where you are and where the safe areas are.
ODEGAARD. Yeah. And you can actually find that information out by some websites that we'll have linked on our webpage, that really show you, you know, the inundation zone, and also the evacuation routes for, like, every city on the coast.
ROBERTSON:That's great information, Jens. Thank you for all of that. As he mentioned all this information will be linked on our website, engineeringoutloud.oregonstate.edu. So this has been a podcast takeover by Jens Odegaard and Rachel Robertson for Johanna Carson who is off on assignment.
ODEGAARD: Steve Frandzel produced the first part of the episode, and we had help, as always, by Mitch Lea for our final audio editing.
ROBERTSON: Our intro music is The Ether Bunny by Eyes Closed Audio on SoundCloud and it's used with permission via Creative Commons 3.0 license.
ODEGAARD: For more episodes, visit engineeringoutloud.oregonstate.edu or search "Engineering Out Loud" on iTunes or your favorite podcast app. Take care.
ROBERTSON: That's good.
ODEGAARD: That's it, huh? Ba da bum ba ba.