Data Science in the Energy Industry

It. Sam foreign.

And thank you for joining today's webinar. My name is Rhys and I'm going to be a moderator for today's session. We're going to kick off today's session in a couple of minutes. We're just waiting so everyone has a chance to join. In the meanwhile, though, we'd love to hear from you. So let us know where you're joining from using the chat or the comments, depending on what platform you're watching on. And yeah, tell us something that you'd like to learn in today's webinar. We are going to be taking questions throughout, but we will also be having a Q A section towards the end. So if you have any questions throughout, please let us know using the chat in the comments and I will save them for the Q and A at the end. Also, to note this, this session is being recorded and the recording will be emailed to everyone that's registered for the event. I'll be posting a link so you can register for the event and get the recording and resources sent to you afterwards as well. I'll be posting that on LinkedIn as well as in the chat shortly. But yeah, until we hit our start time, I will leave you with background music. So enjoy.

Foreign.

And thank you for joining today's webinar. My name is Rhys and I'm going to be your moderator for today's session. We're going to kick off in about 30 seconds or so. We've just been waiting so everyone has had a chance to join. Let us know where you're joining from using the chat or the comments, depending on where you're watching from. And yeah, tell us something that you'd like to learn from today's webinar. A few bits of housekeeping before we kick off. The session is being recorded and the recording will be emailed to everyone that's registered for the event in the next few days. I've put a link so that you can register and get the recording and resources emailed to you in the chat, but I'll also be reposting it shortly so that you can sign up for that. We are going to have a Q and A towards the end of the session, so if you have any questions, please type them in as soon as they spring to mind and I will save them for the Q and A.

Brilliant.

I think that's everything from me. So I'll hand you over to your host for today. Today's session.

Richie. Richie.

Please take it away.

Hello and welcome everyone. This is Richie. I Gotta say, just looking at where everyone's joining from around the world, I'm always very impressed with people who are joining from a very early time zone or a very late time zone. Well done for making it here. All right, so if you try and list the most important skills for a data scientist, most people will agree you need some sort of statistical skills and the ability to use data tools, whether it's Python and SQL or a business intelligence platform. And probably you'd throw in some soft skills like communication and self motivation. But something that often gets overlooked is that you really need domain knowledge of the business you're in. You need to understand what your data means and what the goals of any analysis are. So this is something I've definitely fallen over in my career. I've transitioned from one industry to a completely different one, said, oh, I know I have to do data science, it'll be easy, and then spent months being unproductive because I didn't understand the data. So today you're going to acquire some domain knowledge. You're going to learn all about what happens with data science in the energy industry. And I hope that for any of you considering working in this industry, it's going to save you having to ask a few silly questions once you start your job. And we have two guests from Pinnacle Technologies. So Ryan Myers is a product manager and his work focuses on digital transformation, the application of machine learning technology, product development and strategic consulting. And Drew Waters is director of data science. So he's an electrical engineer turned data scientist and his technical work focuses on mechanical reliability, natural language processing and Bayesian statistical methods. So I'm really excited to hear about what goes on at Pinnacle and in the broader energy industry. So I think over to you, Ryan.

Thank you, Richie. Thank you for everybody joining around the world. Very exciting to see that. I'm going to go ahead and we'll start the presentation brief agenda. We'll briefly cover who is Pinnacle and really where do we fit in in the industry as a whole throughout energy and specifically reliability and why that matters. Talk a little bit about the energy industry in general and then we'll start getting into more specific use cases. Right. So we'll be a blend of domain expertise as well as some technical. And then at the end we'll talk a little bit about various data roles and plenty of Q and A at the end for whatever y' all are interested in talking about. So who's Pinnacle really? We're pretty much the largest data analytics company that is solely dedicated to reliability and maintenance. We Serve customers around the world and their large industrial operations. For the most part. That's everything from refining to, to chemical plants, to mining facilities, water, wastewater, even wood mills and sawmills and some other industries as well. And really the unique perspective that we take is taking data and using that to help our customers make better decisions. Those could be micro decisions on when do I do maintenance on an asset or which assets should I put sensors on to then get more data, to do more data science to large scale decisions on when they should be shutting their plant down, what should they be doing when they shut down, where should they be investing capital, et cetera. We bring a lot of data into it, a lot of data science, various techniques, some interesting challenges in this industry which we'll cover in just a little bit. But it's a little about who we.

Are.

Why reliability matters. It's something quite frankly that I myself take for granted every day as well as most people in the US at the very least and throughout the world. Reliability is basically you get something when you want it at as low a cost as possible, right? That's everything from electricity to fuel to food to stuff that you order on Amazon. Having things, you know, when you need them and at reasonable cost is just, it's important to society and allows us to, you know, keep improving. And when any particular subset of the, of the energy industry or any particular facility doesn't operate reliably right, society, even in micro areas can break down. An example we have listed here is a couple years ago we had a historic freeze, got extremely cold, the right measures weren't in place. We had a variety of different facilities and assets go down, led to a lot of Texas not having access to electricity, which then had cascading effects. A lot of different things happened and then more in the industries that we serve. Some other examples, BP Texas City, back in 2005, there was an explosion, actually a series of explosions that occurred due to poor reliability and really injured a lot of people. Some people died. And unfortunately a benefit of that is it led to more rigorous programs being put in place and as the years have gone on, getting more and more data driven and using that to help make plants safer. Another example is a Flint water crisis in Flint, Michigan. There was a where they changed where they were getting their source water. There are some different contaminants and constituents in the water, led to things corroding, ended up leaking and then lead got into the water and other contaminants really bad for, you know, everybody who was drinking that water. And so reliability Is again, something we take for granted, incredibly important. And we'll be getting into some of those use cases as we go through before we dive deep into that. And some of the work we do at Pinnacle, just taking a step back, high level energy, right? It makes the world go round. It powers everything from our vehicles to our homes. A lot of byproducts that come from the various industries also are, you know, things we use every day, things we don't even really think about, but like plastics, for example. Almost everything that we have around the world is some way somehow impacted by this industry. There's a lot of different subsectors in it, and we're not going to necessarily cover everything there. But some of the main ones to call out are. And oil and gas, you know, upstream, where we're getting oil out of the ground midstream, where we're transporting that and moving that around and then downstream, which is things like refineries or chemical plants, where that gets processed and you can actually do something useful and valuable with that. There's other aspects like with power generation, so generating electricity, some of which may come from renewables, some of that may come from oil and gas products, and even things like mining, where a little bit of a stretch, but also a lot of raw materials that are produced, right. That, that cascades throughout the industry. So a lot of different verticals. That's why, as you see here in the columns, the things you'll kind of see throughout all of these is there's, there's different kind of functional groups. There's like the process itself or the technology that's being used to, to produce or distribute energy or energy sources. There's the supply chain. So the logistics between all these different facilities, every raw material coming in, what happens outside of the facility, and even within a given facility, there's a lot of logistics and things that happen as well, operations. So, you know, people who are actually running the facility and making decisions and just doing the day to day that needs to happen, if we need to turn valves on, move something and isolate it, et cetera. And then lastly reliability and maintenance, which is, that's the space that we play in is right. How do we help a facility produce as much as it can, be as safe as it can, and do that at the lowest cost possible. So that's, that's the space that we plan and so we touch these verticals and other verticals as well. But I wanted to give you all that context. It's a good way to kind of think about the industry. Go Ahead.

Yeah, I can talk about this. So first of all, I really appreciate Richie's introduction of the topic of really this is about talking about domain knowledge and the intersection of domain knowledge with data science. My background is in data science. My, my PhD is in electrical engineering. I've been working for as a data scientist for several years before joining Pinnacle. And I had no domain knowledge coming into my role at Pinnacle and Ryan was incredibly helpful to me and understanding our data better and understanding how those things work. But yeah, having that domain knowledge and being able to marry that with good technical data science skills is absolutely crucial in any role that you're going to have. Certainly crucial for the stuff that we're going to talk about today. So what we're going to do is we're going to go through a couple of examples of big data science problems that we've solved within this industry and some of those things are going to talk about asset degradation. So if I've got a piece of equipment at a refinery or some other kind of manufacturing plant trying to figure out when those things are likely to fail and cause problems, when they're going to break down and not work well, being able to make recommendations about when I should go out and do maintenance operations on certain pieces of equipment, on one hand I don't want to over maintain and overspend, but at the same time I don't necessarily necessarily want my equipment to fail because of, you know, loss of production or potential accidents that can happen. We've worked in areas like natural language processing. We're not going to talk a lot about NLP today, but this is a very near and dear topic to my heart. I work a lot in this, but being able to mine textual data in our industry is a very, very big deal. And being able to take words and derive insights from those classification and clustering, looking for patterns within data, where do we find similarities in safety, incidences that have occurred in the past or different types of failures that occur and what we learn and leverage from that forecasting is a big deal for us. Looking at what a plant's long term production capabilities are going to look like and being able to actionize those down to logistics. So things like trying to figure out when we should order spare parts, when we should do certain things so that when, when problems arise, we can respond quickly to them. So we'll be talking about a variety of these things.

Yeah, and some of the common data challenges. So it depends on which customer we're working with and kind of maybe even which subgroups and different use cases. For a lot of use cases, there's not always a lot of data. When we say a lot of data, having a few sensors worth of data, right. We're not talking about like Google level data or Amazon level data. Generally speaking, unless you're looking across the enterprise for one of our large customers might have 30 plus facilities. Now, that being said, throughout the industry, we're seeing a lot of our customers, they're investing time in standing up data lakes and various data pipelines. Some customers don't have that established fully or it's not something that anyone can really make use of. So they're going through a lot of IT projects to do those types of things. Or maybe they've got it partially implemented. Some data is available, but maybe not the data that we'd be looking for to solve a very specific reliability problem. Again, not the case always. That's just something that we're seeing a lot of investment go into is sustaining this type of stuff. People know data is valuable. The other thing we've also seen is some of our customers, our customers know data is valuable. Some of the challenges come into is, okay, what can I do with my data? Right? That's another thing. And this is where the domain expertise really comes into play with knowing what is possible with data science. But then knowing, hey, here are the problems that we need to go solve, here's the value of solving those problems, and, and here's all the little nuances that'll bite you as you deploy neural networks or other fun things.

Yeah, very good.

You want to cover the other two?

Yeah, can do so with our data too. So a lot of times we're dealing in situations where there's not a lot of volume of data like Ryan mentioned. But even when we have a lot of data, that data is noisy and sometimes it's not reliable. We're going to give a lot of concrete examples about that going forward. But those are two things that we do have to worry about with our industry too. We have to work around concerns that our customers have with the data that they, they already have access to. A lot of times there's a lot of distrust regarding the data. When we talk about inspection data, that's very much a real thing where there might be certain periods of time where they say, hey, we don't trust anything that's in here. Should we throw it away? Should we not throw it away? Or where we just have gaps. We have like data that's missing. We don't know what's going on in this One particular portion of the facility.

With high accuracy or the data is not in structured or unstructured data sources. There's a lot of data that still lives in people's heads. Yeah. Or paper documents. Or paper documents. Or whatever the case is.

Yeah, all that kind of stuff. Finally, there's a lot of tension between subject matter expertise in our industry and data science principles. We've staked out a very good space at Pinnacle on trying to marry those two things together. And we'll give some very concrete examples about that. But you will have a certain amount of tribal knowledge that exists among the veteran engineers who have been working in this industry for 40, 50, 60 years. And then, you know, up and coming data science techniques come in and try to revolutionize what they've been doing for a long time and doing very successfully. But trying to marry those two things together is not always straightforward. That takes a lot of, you know, both the industry knowledge and then also the statistical knowledge to be able to pull that off.

Yeah, I'll say the, what we found is most successful is blending the two together. Yeah, SMEs help point and point out, hey, you're missing this really critical piece of information. Worst case we get it from them, but best case we go poke around and find that and then vice versa. Right. Data science is pointing out interesting things they maybe didn't think about. Then they can, they can then reason, here's what is actually happening that's causing that, and then we can go help the customer make a better decision on how to prevent that issue. And so bringing both those worlds together is where the magic happens.

Great. So one of the first examples we're going to talk about is asset degradation. So when you're operating, say an oil and gas refinery, you're taking crude oil and you're turning this into, into products like gasoline or jet fuel or whatever else. The chemicals that you're, you're operating on, the crude oil, the things that you're doing chemically to, to transform that crude oil into useful things is a fairly corrosive process. It's some pretty nasty stuff that you're working with there. And so what I'm showing over on the right hand side of this particular slide are places where degradation and corrosion are happening on, on assets. So I've got a pipe in the upper right. Over time, after that pipe carries a lot of this chemical content, that pipe is going to thin out over time, it's going to corrode, it's going to rot out. Eventually this gets to the point where we're now springing leaks and best case scenario, that leak results in, hey, we have to shut down our facility for a little bit. We have to do some repairs. We maybe didn't get to produce as much product as we wanted during that day. Worst case scenario, that can start a fire, it can cause an explosion, it can be something very, very dangerous. Down on the right, I've got this picture in here just to highlight that there are different types of degradation that we deal with in our industry. So in the upper right, this pipe is just carrying corrosive chemicals that, that rots out the pipe over time. Down below is a special case called corrosion under insulation. So this is where we've wrapped a pipe in insulating material because we want to maintain temperature for this. If a little bit of moisture gets underneath that insulation, it can cause very, very rapid corrosion. Not so much because of the like corrosive chemical properties of that, just more, more mechanical type corrosion that happens there. So we deal with different types of corrosion, but the end result is the same. After time, when the asset gets to the point where it's no longer thick enough to hold the product, we have problems and we'd like to avoid that. And there's a lot of things that will determine what kind of damage we're dealing with, how quickly that damage is going to occur. Some of those I've listed on here. So the, the chemical content of the stream that the asset is working with, the metallurgy of our assets. So if we're dealing with carbon steel, things will manifest a certain way. If our pipe is stainless steel, it's a different thing thing. We have other alloy materials that are more high end for dealing with certain things, and they will all behave differently based on that interaction between the chemical content of the stream and the metallurgy of the asset. Operating conditions matter. So knowing what the temperature, knowing what the pressure is, knowing what the ambient temperature is, what's the humidity in the area that we're dealing with, all of that stuff is very important to us. Another thing that is a problem for us is that corrosion can be very localized. And by localized I mean very non uniform. So, so in some cases for certain types of damage, you might expect that damage to be fairly uniform over the content of say a pipe or a pressure vessel, whatever asset you're dealing with. Other times it could be very pinhole level corrosion that you're dealing with. So even if you're looking at one portion of the asset, you may not know anything about the real problem area of the Asset. And that's a problem. That's something that we have to deal with. So let's say I want to get down into modeling now. So I'm a data scientist. I want to do data science modeling on this. And what will happen is that over time, facilities will go out and they'll conduct inspection work on a particular asset. Say I'm dealing with a piece of pipe. There's a piece of machinery called an ultrasonic thickness monitor that an inspector can bring out into the field. They'll connect this up to the pipe, they'll take a reading on it, and they'll get some measurement of thickness for that particular pipe at that location at that particular point in time. And this is a small area that they're looking at. This is maybe at most like a foot of width that you're looking at on a pipe that could be very, very, very long. And what might happen over time is I get something that looks like that graph over on the right. Every red dot that I've got on there is an inspection data point that was taken. My gray cone that I'm drawing here is something of our expectations going into this problem. This is how we expected corrosion to happen on this. The red data ideally would be living inside of that cone. But as we're showing here, it's not doing that because it almost never does. What I want to point out on this, too, is that the data that we're dealing with, those red dots, they're quantitative. This is a machine that is taking a measurement and returning a value. There is confidence with that value. But there's also some noise that happens with this data right here. So depending on the skill of the inspector that goes out and does that inspection, that data point may be more or less reliable. If you have a very seasoned inspector who's very, very good at their job and very precise with things, that might be a very, very small error bar. If this is a newer inspector or somebody who's not very familiar with the equipment or the facility, that may be a very wide error bar that we're dealing with here. So there's. There's noise in this data. Other things that can happen, the equipment can be malfunctioning, so it can be returning faulty measurements. You will have times, too, where even if the inspector is doing their job correctly, they may have taken the measurement at the wrong location. So they may be a couple of feet down. So it's not necessarily the same data that you're getting over time. There's just. There's a lot of things that go into this problem that are difficult for us. But the real problem that we're interested in solving here is, hey, I've got some data that tells me about thickness over time. And because I'm expecting this asset to thin out over time, I'm kind of expecting a general downward trend. I should be getting thinner over time as things go on. And what I'm really ultimately after is I want to know when this thing is going to get so thin that I can't use it reliably anymore. That is not zero thickness. That's going to be some kind of a safety margin that's built into that. And so what we're after is, can I like have a data model that tells me, hey, based on your data, based on what you expect to have happen, when is this thing going to fail? Now one of the things that I actually throw this out as a job interview question quite often. So I know we've got some data science students on the call, so this is one for you guys. I will throw this problem out to prospective data scientists who apply for jobs at Pinnacle. And as you can see, I've got five data points on this plot right here. And I'll ask them, hey, like, go solve this problem for me. Go tell me like how you would go about figuring out when the thickness of this asset is going to get to the point that it's not usable anymore. And I'll give them some value of like, hey, when is it going to get down to 300? Now when I give that problem, I'm not looking for something as complicated as what we do at Pinnacle. Like, we've invested a lot of R and D in the modeling that we're about to talk about with you guys. But I'm basically looking for an answer like, hey, I'm going to use a regression model or maybe if they're really fancy, I'm going to use a Gaussian process or something like that. If the candidate says, I'm going to use a neural network on this five data point example right here, you're probably not getting a second interview. And so just, just something for the students out there. I know neural networks are definitely like the really cool thing. And if you're working at a Google or a Facebook, they're absolutely state of the art, best thing that you can do. But it's good to know other things that you can throw at data problems like this too. So just, just throwing that out there. But the two things that I'm going to balance out when I solve this problem. Here are 1. I've got my data. It's noisy data. It's infrequent data. I don't necessarily have a lot of it. But I also need to balance in the role of the subject matter expert. So, again, tremendous domain knowledge in this industry. We have people who have been working on these corrosion problems for 70, 80 years. I mean, we've just got this wealth of stuff in the industry that we have to draw on. And so what the subject matter expert brings to the table is they know, based on the parameters of this asset, what it's doing, what kind of stuff it's operating on, how much temperature it has. They know what that theoretical expectation for corrosion is, plus or minus. They maybe have worked on things that are similar. But one of the downsides of that subject matter expert is that that general theoretical laboratory knowledge may not translate to the exact asset that I'm working with. So I want to balance these two things out when I solve this problem. So what am I going to do here? Well, as when we were introduced, my background is in Bayesian statistics. So of course we're going to solve this with a Bayesian approach. So Bayesian statistics, real quick review. I have some data that I observe. So my data is Y in the diagram that I'm showing here. And what I really want to know is, hey, if I've got thickness data, Y, what is the true thickness of my asset or what is the true rate of corrosion that I'm seeing on this asset, given that observed data that I've collected? Now, my subject matter expert comes here in the role of being a Bayesian prior for this. So I, I have some kind of a prior belief on what thickness should look like over time for my asset, what corrosion should look like on this asset. I've got quantitative inspection data that is my observed data. What Bayes rule and what Bayesian statistics allow me to do is go backwards and say, hey, I've got data that tells me this and a prior belief that says this. What do I think now? Now that I've observed my data, so how do I get back to my X? How do I get back to that underlying ground, truth, reality based on potentially noisy data and some potentially noisy prior beliefs about what should happen here. So I'm going to marry my prior and my data together. I'm going to end up with a posterior distribution that tells me something about what's really going on based on all things that I have access to. And what I can do with that is I can Come up with a cool model like this. You want to talk to this one.

Or this is one of the models that we use which we call our. It's called a lifetime variability curve. But you can think about it as just a very fancy regression model that's got some various aspects to it that work in this domain very well. But basically a lot of times we like early in the NASA's life we may not have a lot of data. We're going to lean heavily on prior knowledge which could come from an SME, it could come from another machine learning model. A variety of different ways we get that and as we get information we're using that to update what our beliefs are. And what you see in the middle chart is we collected some data which is a little bit less severe than what the original expectation was from our subject matter expert. And as a result of that we don't have a lot of data points. So we're not going to overfit to the data itself, knowing that there's error in the data. And also the processes are very variable. These things are not operating exactly the same way all the time. They're highly dynamic. We see a wider band of uncertainty increase in our projection, meaning we don't have a lot of confidence exactly of what's going on. But that's going to help us understand when is it valuable to go collect more data. Big challenge we face. Data is not free. So it might be like I could totally predict failure if you go scanned every single asset and spent hundreds of millions of dollars. But that is just not simply practical to do on a regular basis. And so that helped this. Another big benefit of this knowing when do I go collect more data? And then lastly as we see here, and we've collected some more data in an appropriate point in time, we've seen that there's a shift, right? There's something that changed in the process or maybe we got, we're using a little bit better techniques, we're finding more of the damage now, which is, which is a challenge. And in this case we see that there's where we have a little bit tighter trend, it matches a little bit more what we expect. But we got a little bit of a life extension to an extent just based on the lower degradation operation we saw earlier in life. And so this is always going to be updating. There's also variations of this model we can take where we bring in other information like process data and other time series. So there's varying approaches we can take here, but that covers kind of one core approach we take there. And then I'll turn it over to Drew, another use case for this. Like I mentioned, collecting this data costs money. And it's not like it costs a dollar, it costs a significant chunk of change. But once we've got these types of analytics run across a variety of assets and across all the different locations where you might collect this type of data, we can start optimizing where we go collect it.

Yeah. So as Ryan was saying, data collection is costly. This isn't something that comes for free. This isn't a sensor that lives out in the. This is a human who has to go out there, spend a significant amount of time to do this. This costs money. Storing it costs money, like maintaining it costs money. So there's a cost associated with all of this. Now, what we found going over, I mean, facility after facility of client data, is that there's a lot of inspection that's being done that is useful, and there's a lot of inspection being done that is not useful. So as a good example, if I go back to my plot over here and then let's say I talk about that, that third plot over on the right, if my data points are in a perfect line, and every time that I am, like collecting a data point, I'm seeing exactly what I expect to see, collecting more data isn't super useful for me. That posterior distribution, that shaded confidence area, is going to get down to laser beam pretty quick. And what that's essentially telling me is that I don't need to go collect more data there. It's like, I know what's going on. I've got a good handle of things. Now. In a case where my uncertainty is quite wide, say in the second plot that we've got here, there, I do want to collect more data. It's like, I don't know what's going on here. Data can help resolve some of those issues, but what can we do above and beyond things? So once we run our model, the model gives us an idea of when things are likely to fail. Well, there's a couple of things that we can do with that that are interesting. One is we can start telling them, hey, for future inspections, these locations, the subset of locations, are like something with high uncertainty. We need to know more about it. There's risk that's associated with these locations. You should go collect more data here. But in other cases, we can say, you don't need to collect anything here right now. Like maybe in 10 years, 20 years, you might actually need to worry about this again, but right now you're good. Focus your resources where they matter and go look at the things that are important. So with the pipe that I've drawn over on the right red pipe, every blue location is one location that I'm inspecting. What we might find is that for every 10 locations that they could inspect, only one of them really matters. We can actually get quite a bit of reduction in the amount of inspection that they do, and that's a 10x cost savings. Now, if they can do that and do it well. Now, on the other hand, what will happen is because the inspection locations only represent a fraction of the total surface area of the asset. Say I'm talking about a pipe, I may have only sampled maybe 1 to 2% of the length of that large pipe out in the field, which means there's a whole lot of pipe left that could be experiencing other problems. We can also do some neat optimization work and some more Bayesian optimization work where we can say, hey, based on what you've seen already and based on concerns that your subject matter expert had, how bad could things get in those red regions of the pipe? The stuff that we haven't inspect, haven't inspected yet. A lot of times what will happen is, well, based on the data that we've got and based on the concerns of the SME, not really that much can happen there. You're okay, you don't need to go do more inspection work. You're fine, you're fine here. On this particular asset. For other assets, though, what we'll find is that there is quite a bit of risk that that hasn't been dealt with. So if I take some measurement data and I'm starting to see some really, like, wide variance on corrosion rates, I'm starting to see some really scary stuff. When I go and project over that unobserved region, we're going to say, hey, alarm bells. Like, there's a good chance that something really bad is happening out here. Given how little work you've done so far and given what you've seen already, you should go do more and you should bring maybe more better inspection techniques rather than just looking at point locations. Maybe do a bigger scan of this area. If there's a real threat of, you know, particular types of degradation and corrosion that you're worried about there. Now, when we've done this on real studies, this is the, the fun one here, we typically find that about 70% of the inspections that, that customers are doing are not value adding, and that's been as high as 90% with some customers. There's quite a bit of, like, room for growth here and for cost savings that you can have here. But in about 10% of the assets that we've looked at, not enough work was being done. So on one end, they're spending money where they shouldn't be spending it. They're failing to spend money where they should be spending it.

Might be experiencing failures as a result, which are costly.

Yeah. And so ultimately, we can come in there and try to, like, optimize things for them, tell them where they should do things and where they shouldn't.

Yeah. On the whole, they end up spending less money generally, and then they're also. Right. Mitigating risk as well. So it's big, big value add of data analytics with corrosion.

Good stuff.

Actually. There's more corrosion.

Yeah, more corrosion. We're going to do more corrosion now. So one thing that we mentioned in the previous portion of the talk was we're combining data and we're combining a subject matter expert's opinion of what should be happening with a particular type of asset. And so the SME may come around and say, hey, for this asset that's carrying this material, that's made of this metallurgy, operating at this temperature, operating this pressure, we expect to see 0.005 inches of degradation per year. That may be something that they say, but they wouldn't be surprised if it was maybe.03 or, you know, up to 0.07 or something like that. So they give some, like, error bars on that. That's the cone that we drew previously over here. That's. That's kind of the uncertainty that the SME might have. And the SME estimate, what the subject matter expert comes up with is based on a variety of things. It's based on tribal knowledge. It's based on laboratory condition, type of work.

There's a variety of industry studies that have been done, you know, over the years, and some models, quote, unquote, that exist.

So here's a fun anecdotal story right here. So one particular type of corrosion that happens out in the field is called sulfur dion. This is when you're carrying sulfur content in your material. And the way that we understand sulfuration today is that based on the temperature of the stream that you're dealing with, and based on the amount of sulfur that you have in. In your stream, that tells you something about what your corrosion would be. And this depends, again, on the metallurgy of the asset that you're dealing with. So in the plot that we're showing here. I've got temperature by corrosion rate, and we have a line for every different metallurgy that we've got. So carbon steel, this is the cheapest material that they can throw out here, and that's going to have the highest corrosion rate. It's not super resistant to this. If they start using more expensive metallurgies, they can bring that corrosion rate down. But the major trends are like, as I increase my temperature, I expect to see worse corrosion just because things are going to eat through things faster when temperature is active. And also as sulfur content increases, you're going to expect to see an increase in corrosion. Here. We're just showing one particular value of sulfur content of 0.6 here. But as you increase that sulfur content, things are going to get much worse. Now, the fun story on this one. So curves like this were first introduced in the industry back in 1961. So these very, very old amount of tribal knowledge that we're dealing with here, it's based on lab data, was based on SME experience. It was based on, like, best practice knowledge at the time. Now, by 1986, what they found out is that these curves were 2.5 times too excessive. So if it was predicting, say, 10 millimeters of corrosion per year, you could be pretty sure it was really about 2.5 millimeters of corrosion. They were too extreme in what they were doing. Now, conservativism is a very good thing in our industry. We want to err on the side of caution. We want to be scared about the things that could be happening in the field. But the downside of that is that if I'm overly concerned about corrosion, it means I'm going to go do more inspection work, I'm going to spend more.

Resources, I'm going to spend more money. And then two, at some point, you realize that, hey, stuff's way over conservative. And I kind of start, if I'm a business person, I start not trusting the technical as much when it is too inflated. Yeah, right. And then, then you might end up missing something that really matters.

Yeah. So, and so what we were driving at with the next part of the talk is, can we do something better than just rely on these standard industry curves where we leverage big data in some meaningful way? Now, the analysis that we're going to talk about is for a very specific type of asset in the industry. It's called a reformer unit. We have a fairly large comprehensive data set of reformers throughout the world right now, about 20 facilities, 36 reformer UN consisting of a very, very large number of assets back of the envelope. The calculation that we did said this is about 10% of the US reformer market, about 2% of worldwide reformers. But again, our overall goals here are can we leverage data and can we come up with a better predictor for corrosion based on all the things that the industry normally cares about? And then can we do a comparison with real industry standard methods and see what we come up with data that we're going to use in this model right here. So we've got our inspection data that we've talked about before. So this is thickness. And when the thickness measurement was recorded, we know which facility we're dealing with and we know where they're located. That's going to be a big deal. We'll talk about that more in a minute. We've got conditions of the assets. So what temperature are they running at? What's their pressure? What metallurgy are they constructed from? Are they insulated? Do they have. PWHT is a post weld heat treatment. It's something that gets applied selectively to certain assets. We have some knowledge of chemical information, so we know what our sulfide content is, we know what the ammonia content is, we know what the water content is. Things like that are typically available for a lot of this, at least the theoretical value is what we think we're going to run into. Now if I threw this content at you and I said, go solve this problem, a lot of you would probably be coming up with exactly what we did. And that is treated like a supervised machine learning model. So what we're going to do in supervised machine learning is we're going to simply take an algorithm and we're going to start spoon feeding it example data. So we're going to say, hey, here's an asset. It experienced this corrosion rate, it had this temperature, it had this pressure, it had this metallurgy, it had this sulfide content, it had this water mole percentage. And like, as we feed more and more examples to the machine, the machine is going to start to figure out, well, okay, what things matter for predicting corrosion, which things don't matter, and how do they all relate together? That's, that's how we tackle this problem. So very, very simple regression supervised machine learning problem. This is very data science bread and butter. And we simply just keep feeding all of these examples to the machine as it starts to figure these things out. Really important thing to know with our model here is we do not balance any type of SME information in this. We validated our model with the subject matter experts, but we don't actually, like, tell anything to the machine about that. The machine has no idea what corrosion is. The machine doesn't understand what metallurgy is. It doesn't understand what temperature is. It's just a data table that it's going to process on. Anything you want to say there?

No. If we wanted to, we could also bring in other SME knowledge or industry tool data as another attribute in the model that could help predict.

Yeah. And we've talked about doing this on our TBD list, what we're going to output from this. Ultimately, we want our model to understand, okay, how do I map all this input data to a corrosion rate? And then we're going to get some fun stuff with that. So corrosion rate estimates. Think of something like a random forest regression model or like a gradient boosting regression model. When everything is done, I'm going to get an estimate of what corrosion should be for any new thing that I throw into this model. But I'm also going to get uncertainty bounds on this. And this is important because even with the data that we've got, we may have assets that have the exact same conditions but experience slightly different corrosion. So we want to know, okay, well, what are the bounds on what corrosion could look like for this? Or.

Yeah. Or the conditions of the plant are changing every day.

Yes. Right.

So corrosion is never this perfect line. It might be flat and then it's severe or like everything in between at different. Different points.

Yeah.

Yeah. So there's some moving.

Yeah. Under. Quantifying the uncertainty of what's going on is hugely valuable for bottom line impact ultimately.

Yeah. And so when we query the model, we're going to get an estimated corrosion rate. That's the blue line say, that we've got on this graph over here. But we also get a confidence interval around that. So we have an idea of, okay, well, what are the. What are the bounds on this? Like how. How in best case scenario, worst case scenario, what are we looking at? Model actually fits really well. So, like, most of our predictions kind of fall within the 95% confidence interval of what our model produces. That's a good sanity check for us. Variable importance comes out of our model. So people who are scikit learn users on here, you know exactly what I'm talking about. We can query the model and say, hey, when you were doing your learning, what mattered and what didn't? And what matters depends a little bit on the type of asset that we're looking at and the particular Type of damage that we're looking at. But for one particular example, my feature important might look like the graph that I've got over here where for this particular example, temperature mattered quite a bit, pressure mattered quite a bit, location matters a lot. Now what do we mean by this? So we mentioned before, this is 20 different oil and gas facilities that we're looking at. Some of those are operated by big major oil companies and the names would be things that you would recognize there. Now it turns out the location is hugely important, like who is operating the facility matters a lot. A facility that is operated by one major refiner may look very different than a facility operated by another major refiner, even though they're producing the same product with the same process and the same equipment.

And really we see this as like, it's a proxy variable that blends a lot of things together. Like, you know, facilities that have a very consistent feed, like what they're bringing in and processing versus some facilities may get opportunity stuff that's hey man, this stuff's real cheap, let's run it. But then it starts tearing your stuff apart and they see all sorts of different things. So there's a lot of different variables that go, go into that. But that was a really interesting finding that that came out.

Yeah, good stuff. And I know we're getting close on time, so I'll get through model validation. We do all the normal data science things. We do five fold cross validation to verify that this model is doing what we expect it to do and we're not grotesquely overfitting to our data results when we do this actually look pretty good. So on average, we're off by about 2.2 thousandths of an inch per year in terms of the corrosion rates that we predict. That's actually really good for our industry right there. And then the other fun experiment that we did is we compared this against a real human corrosion SME. So we gave them a variety of data regarding the asset. It was the same stuff that our algorithm had. But then they also could dig deeper into the data. They could go pull up diagrams of how this facility operates and they could leverage a lot of that tribal knowledge that they had that our methodology didn't have access to. We had a subset of about 20 assets that we gave them. And generally our methodology outperformed the human SME by a reasonable margin. So on average the, the SME was off by about 5,000ths of an inch per year. We were off by 3.1. It was a nice validation of our methodology that this works well. And again the intent of this is not to try to replace that human corrosion SME. They bring a tremendous amount of value to the table. But we see this methodology is really a tool to kind of augment them.

And make their life easier or also. Right. You can use this also as kind of a screening approach. Right. You can mass run this model on an entire enterprise worth of data and then start following up with the highest risk areas or cross validate with SMEs. And the idea would be you continually improve this as you keep getting more and more data.

Cool.

Okay, then our last use case. So then we're talking about, we talk a lot about like corrosion and like fixed equipment. We'll talk a little bit about stuff that moves. So we're talking about an example project we did for a low pressure compressor. And so this is an asset that is compressing fluid. And in this case we have a lot more data than like the low data case with corrosion. There are about 250 sensors or time series available to us. And there was a specific failure event that happened. And so the challenge was, hey, can you tell me before this thing was going to fail and like how far in advance can you tell me that this thing had an issue that happened to it? So there's a few different things we did starting with anomaly detection and then there's a little bit of other work we did there as well. Well, as I mentioned, there was a failure that happened at the end of January a year or two ago. And like I mentioned, we had 250 different time series, extremely high volume data to the point where like my computer couldn't handle it. So we started getting into the realm of big data, ish, at least in this industry for a particular asset. So we had a lot of different information, again, a lot of cleansing and things that need to happen here. Right. Sensors may go bad or be, or the asset might be off just because, hey, the facility was down. That's not necessarily a failure. Right. So we've got to kind of bring some context into things and that affects how we cleanse the data and format and format and structure it and things along those lines. So like I mentioned, we had a failure on the 31st and we went through and you know, prior to leading up to that failure we were able to grab some data and we trained on it and which is basically where things were kind of acting much more normally. And then as we went into some more analysis, which I'm jumping ahead to. Well first start with is how do people Handle this today. There are some machine learning tools that are used in the industry, but a lot of what's happening is there's some form of alarms. It might be called a reliability operating window. There's some bounds on different parameters like vibration or temperature or flow rate or pressure or whatever the case is. And hey, if I'm in this window, I'm good. If I'm out of the window, I'm bad. Which typically results in a whole lot of alarms. And it will prevent some issues, don't get me wrong, it will prevent some issues. But if, depending on how you define those, you either are missing really important things, you're not getting enough time to actually react, to fix the issue, which is what happened here, which I'll show in a minute, or you're just getting a ton of alarms. And what typically happens is they tend to get ignored unless they're extremely severe or something someone was already worried about the first place. So, again, it's better than doing nothing and waiting sufferer failure. But this is very commonplace throughout the industry for. For a lot of assets. Drew, you want to cover this other bad idea?

Yeah. Yeah. So this is actually not a terrible idea. It's just one that's a little harder to pull off. So if we're thinking about failures on all the sensor data, so again, 200 plus sensor channels worth of stuff, a thought is maybe we could do something with supervised class classification. So we are looking for clusters of signals that seem to indicate a particular type of failure. And what would be cool about this is that if the asset can fail in multiple different ways, we could come up with a model that could predict, okay, you're about to. Not only are you about to fail, but you're about to fail in this particular way. Now, the downside of setting something like this up, even though we have voluminous sensor data, is we don't have voluminous failure data. So for a compressor or a pump, there may be 10 to 15 ways that that piece of equipment can fail for everyone. We would probably want to see sensor signatures, at least like 5 or 6 per failure to get a nice training data set. That's already 60 failures that we're talking about here. We typically don't get that much failure data to train on.

Yeah. And also it doesn't always transfer super well across assets. Right. You've got different makes and models, different services, and also the failure modes. The data signature might look a little different each time as well. So that's kind of the big challenge. Once you have Some failure. So, like, some of these tools are deployed in the industry, and we've done some of this as well. Once you've seen something, right, we can train a classifier just to point out, hey, we think this issue is happening. But again, to Drew's point, we don't have a lot of failure data, and failures are very expensive. So you're not incentivized to go create that data set. Yeah, put it that way.

Very good.

One of the things that we ended up deploying here that really boosted our performance was dimensionality reduction through principal component analysis. Made our analysis faster. Then the other thing too, that's a challenge with a lot of tools is having to select the specific features or the specific sensors to use, which ones to not use. By deploying this, we actually naturally got the useful features and reduced that down so the algorithm performs better. And then the ASME doesn't have to spend as much time trying to figure out, all right, what are all the things I need to do to have, what should I exclude, etc. In doing that, out of the 250 sensors, we reduced that down to about. There are about 60 that covered about 95% of the information. The bulk of it was in about you know, 10 to 10 to 20 components. And then from an anomaly detection standpoint, there's some different methods, you know, we've used. Like we're showing a Q statistic here recently. I threw an isolation force at it also performed somewhat well. There's a lot of different techniques you could use. This is a technique we ended up deploying. And what was interesting is, so we saw some anomalies happening when they started this thing back up on the 25th, particularly a little bit of something maybe going on the 27th is my hypothesis personally, that something acted on the asset. And then for whatever reason, on the 29th, things start going haywire across all of the data, right? And you can see that there is a very big shift. And then towards the end, it starts spiking a little bit as well. And so we detect that a couple of days before the failure actually occurs, which is a big benefit. This is just another view of the data just showing like, hey, the bulk of the anomalies are happening two days up to the failure point. From a sensor contribution standpoint, it was largely the two pressure sensors, which makes sense for a compressor, because you really can. The differential pressure is going to help inform, you know, how well is this asset performing, how well is it functioning, and when the function of the asset starts getting out of whack. That's, that's a sign. And then everything else from a vibration standpoint, that's kind of a lagging indicator is once you've got damage, it's going to start showing up that hey, mechanically, mechanically this thing's damaged, it's going to start experiencing some vibration until it, you know, reaches a failure point. And here's an example of like one of the vibration sensor data. And so you don't see the vibration going crazy whenever the anomalies were detected. You see the vibration start taking off the next day and then on the 31st when it failed, it rapidly went to failure. And the way the alarms were set that I talked about earlier, they provided about a 30 minute notice, ish, 30 minute to maybe an hour tops notice. That's just where the alarms were set. And so it didn't provide really quite enough time to react that by doing this we're able to get at least two days in advance in this case. And again, a lot of this is dependent on when something impacted the asset. We can't necessarily always predict that hey, this thing's going to happen. But once that does happen, we can pick that up in the data and give it as early of a notice as we can and then start figuring out when does it make sense to do the maintenance, depending on the failure mode we might experience, for example, I'll let you cover the last slide.

Sure. A lot of roles available in the industry, so we'll just talk to that very briefly. But, but lots of need for data scientists, data engineers, machine learning engineers, and people who are like, even if you're not specifically the data professional who's actually doing the coding and the algorithms, there's a lot of data adjacent fields that are very, very actively hiring in this industry. I mean, this is everything from a product manager, solution architect, people who understand the problems that this industry has and understands enough of the data science and the machine learning to be able to go like, solve those problems. And at least that's me. Yeah, get them on the radar. Yeah, that's him.

Right.

All right, so that's, I think everything that we were planning on presenting.

Yeah, that's what we planned on. Thank you all very much. And I'll turn it over to you, Richie, to facilitate the Q and A.

All right, thank you. That was absolutely fantastic. I must confess, like at the start of the session, most of my knowledge about the energy industry and was, it came from the Bruce Willis movie Armageddon. So it's like, you know, ask me to drill a hole in an asteroid, it's fine, but anything else, I don't know. So that was really nice overview of, like, yeah, what's possible and what you need to do. All right, so for everyone in the audience, if you have questions for Drew and Ryan, please do add them to the chat. We'll go through those. In the meantime, I've got a couple of questions for you. So first of all, I'm curious as to where you hire people from. Like, what sort of profiles are you looking for?

I can talk to that on the data science side of things. So typically what we're running into is finding a unicorn who understands both the data science and the industry is nigh unto impossible. Like that doesn't happen. When I got hired on, I knew absolutely nothing about this industry. In fact, when I was he, Ryan interviewed me and at one point I had mentioned, like, some of my friends were getting laid off in the oil and gas industry. And I was asking about, like, you know, long term stability in the industry. And he asked, well, does your friend work in upstream or downstream? And my response was, I have no idea what those terms mean, but I still got hired. So that was good. But typically we're finding a lot of, like, people who are very, very good at the data science. And if you're good at the data science, like, you can learn this industry. It's not a huge leap to do it. It takes some time and it takes some work. But I mean, ultimately, at the end of the day, the problems that we're solving in this industry are the same tools or use the same tools that we use to solve other data science problems elsewhere. We talked about pca, we talked about Bayesian statistics, we talked about supervised classification. That's like the bread and butter stuff that we've got. At the same time, we do hire a lot of people who are domain experts in this industry. They understand this industry backwards and forwards and they're interested in the data science. They want to learn to code, they want to learn how to do the math on this. And we can train those kinds of people up too. So, yeah, I'd say those.

Yeah, that's good.

The two extrema.

Yeah, yeah, excellent. That's good to know. And so one thing that sort of kept cropping up is that there seems to be a lot of things like chemistry and metallurgy in there. So as a data scientist in the industry, do you need to know anything about these things or. And if so, what sort of things do you need to know?

I would Say it's. It's useful in being able to communicate with people who are going to use the results of what you're producing and also kind of help you reason with, hey, what things should I include in my model? Which things should I not include? You do not need to be a corrosion engineer by any means. And a lot of times it's really. So I'll say where we tend to find the best successful projects are like, when, like Drew and I work together, for example, and kind of bring that together. So I think it's very useful for communicating and then also. Right. Just how you go about approaching the problem. So understanding some basics which you can. We can pick up relatively easily either with things on. On YouTube or online or, like, can work at Pinnacle. We'll train you on the basics. You do not need to be a domain expert to add value. It's going to help you in the long run. Like a test of. Drew knew nothing about the industry or corrosion or any half the terminology we threw out. He picked it up and now he's very able to solve problems and know, like, hey, I shouldn't use a neural network for this. I need to do this whole other approach or variational inference or whatever the right technique is. So I think that's how I see it. Adding value.

And when we do model validation too, I mean, we have a wealth of subject matter experts that work on premises here. I mean, we're in the building where they all sit. So we'll go sit down with one of them and say, hey, this is like, what my model is producing. This is what it thinks is important. Like, does this pass the smell test?

And you can do some of your own smell test. If it's like, hey, the cost of the task is the best thing, that's predicting risk, I'm gonna say that's probably not right. That's probably not right.

Yeah. Temperature, That's. That's a good temperature.

Okay, that makes sense for these reasons. And so it helps you reason about the data and the analysis you're going to do. Yeah, yeah.

Okay. So it really is all about communication. That's very useful to know. One last question from me before we go into audience questions. So you mentioned that the cost of data collection is often pretty high. And I'm sure there are many people in the audience who quite like to get their hands on a data set and play around, see what it looks like. Are there any sort of data sharing initiatives or public data sets that people might want to get started with?

I know we have Thought about creating data sets like this to be able to release out into the wild. We thought about like gamifying some of the algorithm work that we do.

I would say a lot of times, like when we work with our customers, right, it's their data, they own that and we have access to do to do the work that we're doing. So on the whole, like in this industry, it's kind of hard to find it now. For example, like the example I gave on the compressor, you could find like if you go Google, stuff like there's probably stuff on Kaggle or other places there might be some similar fault detection. Like on pumps, for example. I found, you know, a data set, but not troves of this, like for a whole facility, for example, I've not found that I am aware of. There are some public initiatives that have been started, but I don't know how well populated they are with data. I believe Google or Microsoft has created one. Again, I haven't used it personally. A lot of times we're using our customer specific data and so that's not really been an issue. But so I would say look around, you might find something, or at least if you can't find the exact data set, you could find data sets similar to the problem that you're trying to solve. Like there's a variety of fault detection and classification type data sets out there, not necessarily all from oil and gas or, you know, energy. So it's hit or miss, it's a little hard to find.

It seems like it's sort of early days in terms of data sharing within the industry. So look, I look forward to progress with that. All right, let's move on to audience questions. So this one comes from Dimaji saying, fantastic presentation. How do you deal with an asset that doesn't have any prior data or documentation?

Great question there. We have a variety of things. We thought about talking about this. Okay, so typically there's going to be some documentation, like almost always, but what does happen quite often is there are big holes in that documentation. So we might be missing say the temperature on a particular asset. Like that's just we've gone through, we've mined all the data that we've got and we're just missing it now. One thing that can be fun with this is you can start doing, say, imputation methods on that. So when we evaluate risk on an asset, temperature may be a very, very important variable for predicting risk. But based on the other parameters that we do have for that particular asset, and knowing what we know about all the other assets that we have access to, we may be able to predict that temperature very accurately. I mean, within five degrees or so, with very, very high confidence. If we can do something like that, we can use that imputation to kind of finish our analysis that we need to do on that, get good risk estimates. Other times we will be missing data that even if we don't know anything about it, and we may have, like, really, really wide uncertainty of what it could be when we start using it in risk calculations, it doesn't matter all that much either. It might be something that's very, very unimportant. So it doesn't move the needle very much, but we can at least, like, statistically make those kinds of decisions of, I don't have this data, what do I know about it? Can I say something about it? Or in the case where it doesn't really matter all that much anyway, we can kind of reliably not deal with it.

And around it, we have an ML model that we've trained on. I'd say a refinery data set. And then we can apply that to, hey, I know base information about. I know the type of the asset it's in. I know maybe where it is in the process in the facility, maybe one or two other parameters. We can estimate the other things. And to Drew's point, the way we prioritize, that is, we can understand generally what that impact has on risk. That helps us then go prioritize which things to go collect data on or not do. And so, yeah, we can. We can do something with little data. And then as we collect data, right, as that's operated, then we'll refine it through the other modeling that we've talked about.

So there's.

You can do something. You're gonna have to make some assumptions or have some uncertainty baked in. That's why we. Almost everything we do is Bayesian in that sense.

But anyways, yeah, borrowing strength across observations. Like, we may not have information on the one specific asset that we're looking at, but we have a lot of information on things that are very, very similar to it. We can leverage.

And this is also where context comes in. You got to be careful of how you apply it, because like we talked about, we're not concerned about what generally happens always. We're concerned about the rare extreme events. And so it applies in some cases really well and some. This is where understanding the industry and the interacting with other people who understand it, hey, we'll apply it here. Let's not apply it over Here until we get this information, stuff like that.

Okay.

And presumably if you're not quite sure what's going on, you can have a bigger margin of safety before the whatever disastrous thing is going to happen.

Okay.

Just since you mentioned Bayesian statistics again, I have to say I was very interested with the idea that, well, we have a lot of companies that worry about, well, you know, we've got all these domain experts and they've been in the company for 40 years and they really hate the data scientists kind of encroaching on the turf. And the idea that you can use Bayesian statistics to sort of reconcile like all this expert knowledge with data and sort of stop a few arguments between teams, I would think that seems pretty, pretty helpful.

Yes.

All right, so next question comes from an anonymous LinkedIn user saying, does your data come from spreadsheets or more robust sources? Actually, maybe you can talk a bit more generally about the technology stack using, like, what tools do you use?

So on the data science side, primarily everything that we're doing is custom Python scripts. Our data is generally stored in a variety.

The answer is yes, we get things in spreadsheets or maybe access databases. Where's also times where we're deploying ETL tools to grab things from SQL or other data lakes or a partner of ours, they have a data platform that has a variety of integrations. We can bring that into that platform and then extract or deploy scripts there. We do it all. It kind of just depends on the maturity of the customer typically. Or a lot of times if we're just doing a really small problem or a pilot, we'll just, hey, give us a dump of everything. Let's just do a quick analysis and then we go to a larger scale implementation. Let's deploy stuff live. Then that's where we start working with the IT groups and integrating systems and all that kind of stuff.

But then, yeah, the tech stack for the data science down. We're primarily a Python house. I will occasionally have to reach for R if I need some really niche statistical analysis that came out in the last four or five years where Python doesn't have it yet. But yeah, that's our stack for the most part.

All right, super. So we have about eight questions on a similar theme. This one's from Joel, but a lot of people asking sort of similar things, basically like where do I sign up, how do I get a job? This maybe. Do you want to talk through your hiring thing again?

Yeah, I'll talk to the data science roles. So the best thing right now I think like the industry is ripe right now and whether you're at Pinnacle or you're at one of the other major refinery companies, there's a big thirst for data science right now. People are getting the value that this brings. They will typically post jobs on their website. It's very helpful if you know someone at the company just to kind of set yourself apart from everybody else. Definitely brushing up on your interview skills, getting ready to be able to talk about if you're able to learn some of the domain before going into the interview. I think that it does help you stand apart from a lot of the other data science specific candidates that you have as far as getting jobs here. We will be releasing the slides pretty soon and there is a link to the job postings that we have here at Pinnacle and so you're welcome to apply for those kinds of things. We are based in the Houston area, so for out of country people that may be prohibitive. We do have a couple of other sites in Europe that.

Yeah, yeah, I know we at Pinnacle we have a lot of data analyst roles that are available. So a little bit less technical, less data scientist level roles. But then again, five years ago we had zero data scientists. So that's. Yeah, I echo what Drew said. I agree.

All right.

Yeah, it sounds like it's a sort of really up and coming area then a lot of enthusiasm. All right, so next question comes from Felix saying, I've noticed the need for better data quality and quantity. How do you suggest convincing management to invest more? How do you convince your boss of anything? I'm not sure, but I'll let you have a crack at it.

Yeah, I would say it comes down to the business value. And so that's a big thing that, that we do with all of our customer engagements as we put together. Here's the business value of whatever project we're proposing and if it is, hey, should we buy sensors for our assets to go do this thing? Should we go invest more in like the data collection for the corrosion aspect or whatever the case? Should we stand up a data lake, et cetera. It all comes down to business value. Right. So understanding, hey, if I have this data, I can do this analysis that's going to give me these results, figuring out how that affects, affects the business.

Right.

So almost everything is going to impact production, safety or cost. And so try to quantify those areas. Here's where you are today. If I can do this, I could cut out 50% of the failure you're having or I can reduce your spend by 30%, et cetera. That's ultimately what's going to get management's attention at the end of the day. And there's other soft benefits that you want to put in as well, like, oh, hey, I have access to data and that helps these people do these things faster or they don't or we automate this process they're doing. Etc, quantify time as well. But that's going to be my, that's my answer to almost anything like, why should you do it? Comes down to business value. So yeah, quantify it.

Yep.

I think that's absolutely great advice in general. Yeah, focus on the value. Now, we are running a little bit over, but there's a lot of really great questions from the audience. So I'd like to keep going for five minutes. For anyone who has to jump, please do catch up with the questions on the recording. Before we get to the more questions, I just want to say that we've got a few more webinars coming up. So tomorrow there is a session on getting return on investment from AI. Next week we've got a session on how to create a data portfolio using Data Camp Workspace. And we've got a session on how to build data and AI skills at scale. Next Wednesday and finally next Thursday there's a session on organizational data competency. Please do go to datacamp.comwebinars to sign up for those. I hope to see you in future webinars. All right, back to some audience questions. So Simeon is asking, are there any initiatives for getting investors and stakeholders to start incorporating tools for data collection with when these assets are built or installed? I'm not sure whether this is referring to like sort of Internet of Things type sensors or something else, but I'll let you answer it how you want.

Yes. And I would say more and more people are investing those technologies. Generally speaking, they're investing on their highly critical assets. So like the compressor example I gave, a compressor is highly critical, meaning if it goes down like that, that production, either unit or facility could go down, depending on the case. Those assets we're seeing, we're seeing there's definitely investment from a sensor standpoint. There's some other initiatives, you know, we talked about, like, hey, corrosion, there's not sensors. There are some people who are, you know, playing around with some thickness monitors and things like that. But I would say it's, it's more and more. Yes. And it's, it again, it comes down to business value. Right. I can't. It would be cost prohibitive to Censor everything I cared about. It's just not going to make economic sense. And so that's, that's a lot of the challenge that we have and it's part of the things that we can advise on. And like, here's where you should be investing your dollars and sensors because here's what we're going to do with the data and here's how much that's going to save you. So short answer is yes, and it still is somewhat. Now the cost of everything is coming down. So that we've been, you know, in talks with various companies that provide, you know, hardware for monitoring and things like that. The cost is coming down, but to still the scale we're talking about, it's still, you know, hundreds of millions of dollars across an enterprise.

Yeah, all right, that's certainly not cheap. And I have to say like working in sort of software company is like, you think, oh well, data's cheap. It's like, it's, it's, you know, cost nothing to capture a click, but hundreds of millions of dollars, that's expensive data collection. All right, so another sort of tooling question so Ambrose wants to know about, like, what are the go to libraries your team lets you. So.

Yeah, what? Yeah, so in the Python universe, if we are doing anything neural network related, it's, we're TensorFlow people for sure. SciKit Learn is a huge workhorse for us. But honestly most of the things that I do are, I do a lot of custom machine learning algorithms. So I mean, I'm really just writing everything with numpy scipy type functions, doing a lot of that pandas for any type of tabular data. That's definitely bread and butter for the things that we work on and then for a lot of the statistical modeling that we do. Sci Fi has really great functions for it. They're limited and so we usually have to do a lot of custom software on top of that to get anything really working. If there is some really niche statistical analysis that I need to do, again, like something that's been in the last couple of years, generally the stats community works in R more than they do in Python. So I may go crawling back to R if I, if I need something like that, but typically I'll just write my own. And yeah, those, that's, that's my tech stack.

All right, super.

Yeah.

Okay, so next question comes from Jacob saying, are there any cases where you've prevented an injury or worse using your, using your data skills?

Yes, I'll give an example. This was a few Years back, at least. But in one of the corrosion analysis projects we were doing, through the statistical analysis that we were doing at the time, we were able to flag a certain area that we. We went through. Actually, it popped up when we went through a data cleansing exercise more than the prescriptive predictive analytics. So we went through a data cleansing exercise. There's a lot of data that was like, basically at a high level. You just took like an average corrosion rate. It's like, oh, this thing's fine, don't worry about it. Went through a cleansing exercise, we went through our analysis, and that one popped up as like a highly critical one. And when they actually went out there and like inspected it, it was below a critical thickness value. So they took action and went and repaired it. So we were able to avoid that. And. Right, there's hydrocarbon in there. And so it's potentially. It was a extreme event that could have happened. And then there's a variety of other. I'll give another example. In a totally different industry, in a sawmill, we have our technology deployed live and we've got people at site helping the customer every day using the analytics. And every day they're seeing things pop up in priority and they're able to go figure out what the issue is and fix it. And so their production. So a little bit less on a safety incident side of things. Like the first example we've seen actually look at, like, their production targets were here. We get involved and then we start implementing and our technology comes on and then you see the production start going up and so they're fixing things before they break, etc. And so just two examples off top of the cup. I don't know if you've got any of those.

Perfect.

So, yeah, short answer is yes, that's brilliant.

Because, I mean, often you think, well, you know, we're using data science to save someone some money or, you know, make a bit more cash somewhere. But, like, actually saving lives, that's a pretty rewarding aspect of the job. That's. That's amazing. All right, I think we've got time for one more question. So this one comes from Mirella saying, can you talk a little bit more about the challenges of communicating insights within your industry?

I. I can talk to this one a little bit. You can talk to this, too. So I know from the data science side of things, being able to communicate highly statistical, highly technical results to engineers. I mean, very like intelligent, smart, educated people, but helping them understand what it is that we've done. So demystifying it as I think a really, really big deal. And this is a, this is an issue the data science industry faces in general as we start to work, you know, in other industries outside of, of the main tech sector of convincing people that the black box machine learning algorithm that you've created is reliable, that it's trustworthy and that it's doing something that actually makes sense is difficult. So I mean, demystifying that to end user customers that we have, or even just people within our own organization who maybe have some reservations about this is a really big deal. And it's a, it's a labor of love communicating this kind of information and going back and forth with them and making sure that everyone's on board with what we're doing. I think one of the other big communication hurdles that we've had is again dealing with the subject matter experts of convincing them that we are legitimately not trying to like get rid of them or make them unimportant. We know that data science cannot solve all of these problems. And any company that comes around saying, oh yeah, if you just dump a bunch of data on us, we'll like solve all your problems and we can like do it all ourselves. Like we know they're wrong, like they can't do it. But just making sure, sure to communicate that message very clearly in the industry of what is it that the data science can do? What can it not do? Where can it save you time? Where can it save you money? Where can it reduce your risk? Where can it not do certain things? It's definitely something that we have to do quite a bit.

And I would say just agree with what you said. Know your audience. When you're talking to an executive, they do not care about the technical details unless they ask. Talk about the business value. Here's how it changes what your people are doing. Right. So they don't have to do or new things that they know. So focus on the results and the high level. Why? And then when you talk like as you work your way, like if you're talking to more technical people, then you can get more into some of those weeks. So you know your audience.

Yeah.

Think about what do they care about? Why do they care about it? How does it affect their job? That's going to help you at the right level to communicate. I'll say most people don't care about the intricacies of how the algorithms work. Except for like we might talk about it. Right?

Yeah, we were so happy today, incidentally. We could say things like PCA and have equations in there.

That.

That was fun for us. We. We appreciate that opportunity. We don't get it a lot.

That's brilliant.

Yeah.

I do like that idea of knowing your audience and just making sure that you're speaking to things that people actually care about.

All right.

With that, I think we're going to wrap up. So thank you once again, Drew. Thank you once again, Ryan. That was really fantastic stuff. I certainly feel like I learned a lot today. Thank you to Rees for moderating. Thank you to everyone who asked a question. Thank you to everyone who showed up today. I hope to see you all again in future webinars. All right, goodbye, everyone.