How AI Is Redefining Sensing and Data Science

Mike: Welcome to Science at Work, the podcast from Savannah River National Laboratory in Aiken, South Carolina. I’m your host, Mike Ettlemyer. Science at Work is a production of the Savannah River National Laboratory, SRNL Office of Communications. With this podcast, we’re trying to pull back the curtain a bit on what a national lab is and what it does so that the non-scientists among us understand how SRNL puts science to work—that’s our tagline—to advance our key mission areas of national security, environmental stewardship and energy resilience.  

With each episode, we talk with SRNL scientists, engineers and other professionals who are at the heart of who we are, what we do and why it matters to the nation. Recently, SRNL partnered with other national labs on a major initiative involving AI and scientific discovery, to put it in the simplest terms. This effort was led by SRNL’s Tom Danielson, who is our guest today on the podcast. 

He is the Applied Sensing and Data Science team lead in the Environmental and Legacy Management Directorate here at the lab at SRNL. Tom has played a key role in contaminant transport modeling for activities at the Savannah River site, and for the past several years, he has been applying artificial intelligence and machine learning to these activities. 

Welcome to the Science of Work podcast. Great to have you here.  

Tom: Thank you. Glad to be here. 

Mike: So, to start off, Tom, we can’t watch a TV show, read a media article or even look at social media without hearing the term AI. This term seems to mean many things to different people. In general terms, what is artificial intelligence? 

Tom: Yeah. Great question. I think it means a lot of different things to scientists as well, still at this stage. So, it’s a rapidly evolving field, but kind of at the most basic level, artificial intelligence is using computers to mimic, some human intelligence or human action. So, we can think of this as making decisions about certain things. Right?  

We can sort of map how processes work in the real world right there. So, yeah, It’s sort of a decision flow. Related to that. So, we can think of artificial intelligence as being able to sort of map that process and make the right decisions at the right times. AI is good with large data sets. 

So, extracting patterns to be able to make the right decisions at the right time, whatever that decision might be.  

Mike: Ok. And one of the things that I hear sometimes, or I read is, in simple terms, is that AI is only as good as the data or the information you put into it. Is that overly simplistic? Does that sort of encapsulate what it’s about?  

Tom: Yeah. So, at the start or it comes in lots of different flavors, some more data intensive than others, but certainly, data is a very valuable and important thing for, yeah, many, many different AI approaches.  

Mike: So why is AI so crucial to the environment, to national defense and energy resilience? These are the missions of SRNL. And how does this dovetail with DOE’s new Genesis undertaking? I mentioned at the start of this, that that’s a major AI undertaking with Department of Energy.  

Tom: Yeah, absolutely. So, artificial intelligence plays a big role or can play a big role in any one of those different fields. Here at SRNL, we often think of ROI metrics, right. We do applied science. And so, when we look at the environmental area, DOE ‘s environmental mission space, we start thinking about things like cost, like waste footprints, facility operations running safely, ensuring that contamination is not moving through the environment too freely and getting into things like drinking water or into our air. 

So, artificial intelligence can play a great role in, in all those aspects, reducing costs, ensuring robust safety measures are in place, running facilities more efficiently. We’re very rich with data, right? DOE was born sort of out of the Manhattan Project. Yeah, it operated through the Cold War. And so, we have lots and lots of not just mountains of data, but mountains of very unique data. 

We have very interesting facilities that are operating here on the Savannah River site, but across the complex really, that have lots of different process instrumentation, dictating sort of how almost any chemical facility would run across the nation. But, yeah, we get a really good inside view with a lot of that instrumentation that then we can develop algorithms to sort of streamline and optimize those processes. 

So, I think the reach extends beyond just the government operations that we have. In the national security space, the challenge is often that you have a lot of data, that’s creating noise, and your signal is maybe a little bit lower. Right. So, another area where artificial intelligence can really help us to draw out some of that signal or the predictive signal from the noise. 

Additionally, we might think of ROI metrics there as sort of lead time. Right? Can we detect something that is occurring and this really spans across all the mission areas as well. So, yeah, looking at a groundwater plume, something’s changing. There’s a shift. We need to take action. We want to catch those changes as early as possible. 

Because of energy resilience. It kind of goes back to how I was talking about the chemical plants. When we think about the grid and the complexity of the grid as it stands across the nation, artificial intelligence can help us with the operations that are occurring to supply power to the country. 

You know, we’re talking about it at sort of a systems level, but, yeah, there’s lots of fundamental research that goes on as well. Right? So, when we start thinking about what some of these facilities are doing, we monitor the process operations. There’s in some cases it’s glass that’s coming out of the building. And we need to tailor the material formulations. 

Help us at that level as well as sort of the more fundamental, scientific, materials development and discovery, type levels. Lots of different applications.  

Mike: So, in terms of energy resilience, so you’re talking about grid resilience is one aspect of that. Right? A major aspect of that. What are SRNL’s AI capabilities and initiatives, including those that you’re working on? Because I know beyond this, DOE mission, Genesis, there’s a lot of other things going on. So, if you could tell us a little bit about that. 

Tom: Absolutely. Yeah. There are many AI initiatives across all our different mission areas here at the lab, some that I’ve been involved in, over the course of the past number of years. 

One of them is called Advanced Long-Term Environmental Monitoring Systems. We call them ALTEMIS. And so, in this program, essentially what we’re doing is we’re using many different sensor types, to monitor different zones of vulnerability, around a facility where there’s an associated groundwater plume from the historical operations. So, we’re bringing together multiple different technologies, like I said, things like electrical resistivity tomography. 

Things like sensors that are placed in the ground, water wells that are measuring, things like pH and specific conductance. Where we’re measuring water levels to see sort of how the water is moving. And ultimately, what this is aiming to do is to sort of transform the monitoring paradigm. This is done across DOE, anywhere there’s a plume or a facility that’s operating, there are groundwater wells that are associated with that operation, to ensure that we’re not contaminating the environment, that it’s at manageable levels. 

So, the classical paradigm is to send a human out into the field, pull the sample, send that sample back to a lab, return sort of the analysis of that laboratory work, which takes time.  

Mike: It takes time and it costs money as well.  

Tom: Yeah. So, by using these integrated sensing technologies as well as AI and machine learning, what we can do is we can sort of forecast the plume migration. We can detect early if there’s a change in how the environment and geochemistry are behaving, so we can capture those that at earlier stages. 

Mike: And when you say plume, what are you referring to? 

Tom: Contamination that would be in the groundwater, right. That plume is sort of its migration between a few points. 

Mike: OK, just for members of our audience, that might not be as familiar with that as we are. You mentioned in terms of historical operations. So, you’re referring to the activities during the Cold War and other activities that happened on the Savannah River Site that Savannah River National Laboratory has been an intimate part of all of that, right? 

Tom: Yeah. In many cases, that’s what we’re looking at. Or those legacy operations. Still processing waste to this day. So, it’s also to ensure that, you know, the effluents coming from those facilities and the environmental impacts remain low as well. And many of these performance objectives are extending decades out into the future. So that’s where a big part of that cost ultimately ends up coming in even once, the handoff occurs between the Office of Environmental Management and goes over to legacy management, for example. There’s continued monitoring of groundwater wells and environmental systems. Once the facility is deemed closed. 

Mike: And AI makes all of this, I suppose, easier, more cost efficient, quicker.  

Tom: Right, right.  

Mike: So, it saves taxpayer dollars. Time and resources.  

Tom: Yeah, absolutely.  

Mike: That’s great.  

Tom: Call it kind of a proactive monitoring paradigm as opposed to a reactive monitoring paradigm. So, in this case we’re using AI. We’re actually sort of forecasting what, what may occur as opposed to, we find out that we may have an issue or a change in the environment.  

Mike: After the fact.  

Tom: Yeah. 

Mike: Can you think of other examples, besides ALTEMIS, of where we’re using AI here at the lab? 

Tom: Yeah, absolutely. So, AI is being applied in the facilities. So, there’s a handful of different test beds. We’re very fortunate to sit on the Savannah River site, and have these, again, one of a kind, facilities that are operating and processing waste from the tank farms. And so, some of my colleagues have worked in the past taking some of that process instrumentation and ultimately looking at the waste streams that are flowing through, for example, at the Salt Waste Processing Facility and basically being able to forecast different maintenance activities right there. Parameters. Sometimes there’s degradation of equipment within. Right. Maybe a filter or something of that nature. And maybe that filtration rate is one of your key parameters to making sure that you have an optimally functioning system. Both in terms of time, cost and performance of what the filter is doing.  

So, being able to forecast when you need to change that filter is kind of an important thing. I also sort of just tie it all back, right. Whenever we start looking at any of these facilities, I mean, there are a lot of different challenges associated with bringing all this data together. So you have all this process instrumentation. You have a very complex process. Or if we’re out in the field and talking about an environmental or groundwater plume, we have to be able to actually construct these models and understand what the data is telling us. So, we’ve got things like temporal alignment that need to occur. Sometimes just the timestamps on when measurements occur are not quite exactly in sync. 

So how do you make sure you’re comparing the right things or bringing the data together in the right ways? Another place, kind of we refer to it as knowledge management in a lot of cases. But there’s a lot more to it than just the knowledge. It’s synthesizing this data into useful information that can then be used as knowledge to sort of understand our system better. So, a lot in terms of that knowledge management aspect as well.  

Mike: OK, wow. So, it sounds like AI can really be helpful in many of our operations and technologies that we employ. Are there drawbacks and limitations to AI that you would say, both in terms of, let’s say, what, SRNL is developing, and maybe AI in the home, that we’re becoming somewhat familiar with. If so, how can they be avoided or mitigated if there are those drawbacks?  

Tom; Yeah, absolutely. So, I think we talked a little bit about the dependency on the data that’s coming in, right? The data is very important. But it’s relatively easy in this day and age to train an AI or machine learning model. And if you train it on not-so-great data, you may get not so great responses.  

So, we need to be very cautious in our design of these systems. There are certain challenges that AI is not actually going to overcome, right. So, in some of our processes we’re throughput limited. In some cases, there are DOE sites where material has to be transported between two locations. And think of robots and we can think of lots of different solutions that might allow us to do that in an optimal manner. But at the end of the day, there is a theoretical limit to how quickly we could do that. So, AI is not necessarily going to overcome some of those challenges. 

So, those are maybe some of the limitations, I think, understanding and where there’s a good bit of research still going on is understanding, uncertainty in AI, in machine learning predictions, understanding how uncertainty propagates through. 

So again, kind of going back to the data, but also simply understanding our physical system and the things that maybe we don’t know or that the data doesn’t actually capture about our system. Some of the challenges that we’re continuing to work through, with respect to that. And then I think on a much grander scale, there’s a lot of conversation about data centers and, and resource usage. Absolutely. Yeah. When it comes to electricity, when it comes to the water resources that are used. So, there’s a lot to think about there. And again, it’s an evolving landscape. So, I think there are, technological changes that we’ll continue to see over the course of the next, many years that, cause data centers to operate in a much more energy efficient manner and resource efficient manner. The computations can potentially become a bit cheaper. Yeah. So, a lot of research to happen yet and yeah. Exciting time.  

Mike: Yeah, it sounds like a really exciting field to work in. Because, not only, it seems like you can perhaps discover new ways of using AI to do the work. The important work that you’re doing, in science, but also, it’s changing and evolving every day as far as those AI models and how those are going to evolve over time. And we don’t even know, what’s ahead. Maybe.  

Tom: Absolutely. Right.  

Mike: We can make predictions, but who knows?  

Tom: Yeah, exactly. I don’t think we know yet what the theoretical limits are. 

Mike: So, if we shift gears a little bit here, could you tell us, how you got interested in science and, you know, what was your journey like, in getting to SRNL? 

Tom: Absolutely. So, I did my undergraduate degrees in physics and math. I kind of had always had a natural inclination toward science and engineering. When I was a kid, I wanted to be a civil engineer, so.  

Mike: Oh. So naturally, this is a little different.  

Tom: Yeah. Went the route of physics and math in pursuit of ultimately then moving on to go into civil engineering. As I got into quantum mechanics and, you know, higher math, became more and more interested in that and continued it and decided to pursue graduate school, where I studied material science and engineering. I was doing quantum mechanical calculations and modeling chemical reactions, on surfaces and at interfaces at the molecular level, during that time, I joined the American Nuclear Society group at Virginia Tech. 

Mike: Cool.  

Tom: I had the opportunity to take a tour of the High Flux Isotope Reactor over at Oak Ridge. That was pretty much that moment that I decided I wanted to work at a DOE national lab. I was just blown away by the lab in and of itself. And the facility. So, that was great. 

As part of that same group, we also then came to the Savannah River Site, and I toured the Defense Waste Processing Facility as a graduate student.  

Mike: Oh, wow.  

Tom: I love those robotic manipulator arms and seeing the process going on within the building. And it was kind of at that moment and I guess in the last couple of years of my graduate school, I had the opportunity to work at Oak Ridge, great place. 

But just sort of in deciding I was kind of between, you know, SRNL and ORNL, maybe the DWPF tour was the most recent, and that’s what got me here. It was fresh in my mind, but, yeah, both great places. DOE. It’s great opportunity to work with all the labs now, on various collaborating levels. 

And they’re all really impressive and really cool. I mean, that’s one of the most fun parts of the job is being able to go around and see the coolest stuff in history of all the different DOE labs. So, just an exciting place to be. 

Mike: Yeah, and the new Genesis mission, so it’s all 17 national labs that are involved, right? So, you’re able to work with your colleagues across the complex. That’s cool. Pretty cool.  

All right, well, and thanks to our viewers, for tuning in today. Thanks, Tom, for being here. We appreciate the time. Science at Work is a production of the Savannah River National Laboratory in Aiken, South Carolina.