Welcome to the Pipe Liners Podcast, Episode 94, sponsored by Intersys Corporation, providers of Poins, the Pipeline Operations Excellence Management System, SCADA, Compliance and Operations Software for the Pipeline Control Center. You can find out more about Poins at intersyscorp.com. The Pipe Liners Podcast, where professionals, public eks, and industry insiders share their knowledge and experience about technology, projects, and pipeline operations. And now your host, Russell Treet.
Thanks for listening to the Pipe Liners Podcast. I appreciate you taking the time and to show that appreciation, we give away a customized Yeti Tumsler to one listener each episode. This week, our winner is John Bufled at the American Petroleum Institute. So John, congratulations, your Yeti is on its way.
And I'm very hopeful that you will share that and show it off so that we'll get more listeners from the API. To learn how you can win this signature prize back, stick around to the end of the episode. This week, we have Scott Williams, one of my fellow Bubba Geeks coming back on the podcast. And we're going to talk about the edge and data analytics and what's going on with technology around all of that.
Well, Scott, welcome back to the Pipe Liners Podcast. Glad to be here, Russell. You realize that you were the guest for episode 11. So that was like over a year and a half ago now.
I know it doesn't seem like it, right? That's a lot of episodes. Yes, it is. So I asked you to come home because we've been kind of having chats now and again about data analytics and internet of things and the edge and all of that.
And I thought it might be interesting for the podcast listeners to hear us get kind of geeky. So as always, I guess we'll do show notes so that we'll decompile or explain or define some of these terms we're going to use. But the idea here is for you guys to kind of hear some of the conversations we're hearing about all this cool stuff. Sounds good.
So the first thing I wanted to ask you about is Python. So I know, all I know about Python is that it's a coding language and it's what all the data analysts use. What do you know about Python? Well, I've started Newly, Python in my spare time because it looks interesting.
And as a developer, you see these various websites that you're purport to have a here's the list of the top languages of 2019 or lists of that nature. And I use C sharp and C sharp up there, but it's not at the top. It's at the top, which is really interesting. So if a lot of people are favoring this language, I ought to take a look.
So a lot of that popularity is driven by the areas it's being used for. So apparently the tools, it's not just the language itself, it's the tools that surround it. But the Python language and tools are apparently very good at data analytics and artificial intelligence and machine learning. And there's a lot of applications in some of the new academic and non academic environments that are focusing on Python.
And I've also seen that Python is not too hard for beginners to learn, which say, it sounds great. More people using doing programming better off will be. Well, yeah, that would be one perspective, I guess. So I actually, oh, it's probably been a few months ago, but I was noodling on the idea of getting a certificate in data analytics.
And there's some big name schools like Harvard and MIT and some others that offer, you know, a two year data analytics program. And at the end of this, you get a certificate and in some cases, depending on the program you select, you can get maybe even a master's degree in data analytics. But for every single one of those, the prerequisite is you got no Python. So I guess it's interesting for me to hear you say it looks like it would be easy for beginners to use because even though I've been working in software forever, you know, I'm not a coder and it's been a very long time since I've tried to write any codes.
I'm curious just how steep that learning curve would be. Well, it's a modern programming language. So I mean, learning programming is never easy, but it's all about barriers. At least with the C suite of languages, you'll see originally then C++ and L C Sharp.
And they're not, they're different languages. So don't take that the wrong way. But the syntax is rather rigorous. There's a lot of special, special punctuation that's required.
And if you're new to programming, that's a barrier that you don't need. I mean, I've been doing it for ages. So it's comes natural to me. But in the little bit, I've been to go figure out with Python, the structure of the language, the way you type it is much cleaner.
But a lot of that special punctuation is just gone. So from that perspective, I think that's probably why they're saying it is simpler because there's, you know, why does this program run? You forgot to send me a call in. A lot of that's gone with Python.
It doesn't have it. Interesting. That takes away some of the fear and trepidation about trying to get into it. Because, you know, I'm, I'm, I have a high degree of interest in statistics and have always, I'm just giving up just how big a geek I am.
I spent, I went through a phase for about three years where I was reading all kinds of books and magazines on technical analysis of stocks and commodities and basically all that is statistics. And I've built, you know, spreadsheets and algorithms and approaches to crunch data to look at it and, you know, all kinds of stuff around that. And, you know, when I first started hearing about analytics, my mind went to this place. Well, that'll be easy for me because I know that math.
And then I started investigating and it's like, well, you don't just need to know the math. You also need to know this programming language. I'm not sure at this point in my life I want to go and learn a programming language. So I guess, I guess you're taking a little of the concern off by saying that it's, you know, it's more, it's the language and the syntax and the approach to writing that kind of code is more straightforward.
I guess it'll be the way I would say it. Yeah. A lot of the, say the notions of modern object-oriented languages, no job of C sharp C++, all those notions are still in there. This is one of the barriers they took away.
The syntax is a lot simpler. It's streamlined. So have you actually, have you actually tried to do anything with Python or have you just been kind of new, you know, it? Yeah, just do it right now.
I'm at the stage of, except I've used enough different languages. So learning a new one is more about how is this different? How is it better? How is it worse?
You know, where does it fit? I'm still in that phase. I haven't written any big applications yet, but that's coming soon. I'm going to get overhead first.
Yeah, I think one of the other reasons it may be simpler. And I only know this for what I've read about the language, not anything I've actually done with the language is that it's, its function set is smaller because it's pretty much dealing with the math part of analytics. It's not dealing with all the other kinds of things that you might do with a programming language. I don't know if that's true, but that's kind of what I've, one of the conclusions I came to.
Yeah, I don't know about that one. It's possible, but I think Python's been around long enough that that may not have been the original, original impetus. I'm not sure how old it is, but it's not new. Right.
Right. Yeah, no, exactly. Well, we have now probably exhausted the full extent of our knowledge about Python. It should move on.
I think we're pretty close. Yeah. So the other thing that came up recently in conversation, so, and it is Docker. And so Docker, I know virtually nothing about Docker.
It's only recently come up and we were talking about this earlier before you on the mic. So maybe you could kind of tell people what is Docker and from a standpoint of what's going on in automation and analytics and edge, why would I even care about Docker? Stated kind of simpler. You know, Docker is an awful lot like virtual machine technology, except that it's a lot lighter weight.
So consider a typical virtual setup. You have VMware or Microsoft Hyper-V. There might be some other products, those are two biggies installed on a server. And then you create virtual machines, but your virtual machine is a whole complete separate install.
I install Windows, I install my apps, I install my tools, it's a full installation of Windows inside of every virtual machine that's running on the server. Whereas Docker is a little leaner than that, in that your Docker image is really just your application parts. All the Docker images running on a particular server, I'll share the operating resources of the server more directly. So your actual Docker image you're just dealing with, the technologies you need to run your program and not the operating system.
So they're a lot smaller, because you don't have to install Windows and do all that business. It's installed the tools you need to run your program that goes on the image. So now all of a sudden they're a whole lot smaller. And machine can run more of those.
I'm pretty sure various of the cloud providers have a mechanism to run Docker images. And that's going to be at a lower cost point than a virtual machine would be. Interesting. Well, so the conversation I was having was about with a manufacturer that has an edge device.
And they were talking about their chipset and enabling Docker to own their edge device. And I'm like, when you're a nerd and you think you're on top of everything right and then somebody throws a new buzzword at you and you're like, huh, I haven't heard that one. That's kind of what my reaction was. But so I wasn't really clear what Docker was.
So I guess if I'm talking about, you know, a chipset or an edge device that's where the processor is a, you know, I've got a right kind of embedded firmware, then Docker would be something where I could have several different embedded firmware programs running and they could be isolated from one another. Would that be a way to frame that? Yes. Well, it's, you know, going back to the virtual machine analogy, it's because Docker is much lighter weight.
I mean, they still the resource, they still share the resources, multiple Docker images running at once. They share the resources of the host computer. They're on, you know, just like a virtual machine does. But there's no, there's no replication.
The Docker image has what it needs. Nothing extra. The operating system is separate. So that's shared.
So with Docker, it's not the isolation isn't as complete as with a virtual machine. Right. And you know, even that's not as isolation is not as complete as with dedicated server. Right.
But it's, yeah, it's, yeah. All about what you're trying to get done. Right. Right.
That makes sense. And then like, like you're, like you're eluding to, I mean, Docker will run on arm chips, it'll run on Intel chips, it'll run on a Linux system, it'll run on a Windows system. I believe I, I recently that somebody got Docker running on a Raspberry Pi. Oh, really?
With the newer, with the newer higher horsepower Raspberry Pi. Oh, that's interesting. Yeah. Raspberry Pi cost to 35 bucks.
Yeah. Plus the hardware, but the main board is 35 bucks. Interesting. And why you're looking at that, you know, think Raspberry Pi, you're thinking, you know, Linux, because that's the native operating system.
There is a Windows 10 IoT that runs on Raspberry Pi. So that's, it's a Windows 10 subset running on a Raspberry Pi, which is an arm chip, not an Intel chip, like all your desktop machines are. So it's getting very cross platform. So I was just listening to all the things you're saying, and I'm listening to it, you know, just in the way I normally listen.
But I was always kind of, I was at the same time in that conversation we just had, I'm having this conversation with myself about, I wonder how the, how the podcast listeners are navigating through all those buzzwords. So there's that. Yeah. So probably, we probably ought to unpack that a little bit.
So Raspberry Pi is, is basically a controller. It's a board level controller. So it's an, an appranied integrated circuit that does all the things that a, you know, controller could do like a PLC or an RT or something like that. It's even beyond that.
It's, you know, a little bit bigger than it. So consider the size of like a five by seven index card. It's smaller than that. It's a single circuit board and it's computer.
HDMI monitor, USB keyboard mouse, it's a computer. Yeah. It's a full up computer on a very, very small profile, not only in terms of size. But also in terms of power and everything else, a lot of processing.
So it's a, it's the kind of thing that a lot of people start with when they're developing for the edge. So that's, that's interesting. So that's, that's what, that's what a Raspberry Pi is. The other thing we talked about was chip sets and ARM and Intel and, et cetera.
So the chip is a, you know, most people know what an Intel chip is because you go and, you know, you buy your new computer and it says powered by Intel and it's got some fancy marketing stuff about how that's the latest, greatest, fastest chip that ever existed. So Intel's not the only chip manufacturer, ARM and others, manufacturerships as well. And I guess the thing to be, you know, so why would a, like what chip set is a Raspberry Pi run? Raspberry Pi's run a flavor of ARM.
A flavor of ARM, okay. Our arms, our arms a whole, well, just like Intel, you go to Intel's website and you have the Pentiums and you have the Z-Nons and you got the different speeds and the i7s and the i yet, all these numbers, they're all different flavors of Intel chips. Okay. ARM has the same kind of thing.
So there's a lot of different flavors of ARM chips. Why would I pick? I mean, what would cause me to pick one chip set over another? If I'm, if I'm trying to figure out what I'm going to use as a platform for something.
Well, in the past, you think about your horsepower of the chip itself, you know, how much computation can chip do? But ARM chips are coming along. The modern arm shifts are getting pretty close in power to your, your desktop computer Intel processor. So the performance edge is kind of, kind of going away.
There's, there's architectural reasons for this. I won't go into it because we'll get a whole, it'll be worse than that last conversation. In general. Better.
It depends on your perspective. That's true. In general, ARM chips lose, use less power than Intel chips. You have to broadly state meters.
There's a thousand variance to that. But broadly stating that's the thing. Yeah. Yeah.
That's exactly right. You'll find ARM chips on a lot of lower horsepower machines, like the Raspberry Pi. I think the Raspberry Pi is not going to match up with your desktop computer. It's not that kind of horsepower.
It's maybe your desktop computer from a couple years ago, which is not bad, but it's not your modern desktop computer. Right. But now you take that argument and you look at all of Intel Apple's products. I think since 2010, they decided that the core CPU was going to be ARM with Apple products.
And they have, they have the names to be seen to advertise. I'm using the A10 something, the A11 something, something. That's as Apple's chip, which incorporates a lot of different stuff. One of the things it has on it is the ARM processor.
Yeah. It's, it's very clear if I'm running, if I've got an iPhone or an iPad or something like that or a, you know, even a MacBook, the advantage of running an ARM, if it's lower power, is I can get more out of the battery. Right. I'm not sure about MacBooks.
They might still be using Intel's, but all the mobile devices, you can. Yeah. But, but exactly. But my point is that, you know, battery life is becoming a bigger and bigger deal.
Now, it's interesting is when you start talking about the edge, so I've been using the buzzword edge. I should define edge. So edge is, if you think of what a current PLC or a flow computer does, that device is operating at the edge is the first computer that's attached to the instruments. That by definition is the edge.
So when you start talking about building a new kind of platform for the edge and taking and putting a computer out there, like a Raspberry Pi versus putting a controller out there, when you start talking about low power, particularly in oil and gas, particularly in pipelining is there's more to it than just batteries, because in order to get to division classified or explosion proof, I've got to get the power down, because it's by lowering the power, it's by lowering the voltage that I clear those, some of those division classification hurdles. So I think that in terms of what's going on out at the edge, that the arm chip is got some, there's some preference for the arm chip over the alternatives, because it's low power, and it also tends to have a higher environmental ability, meaning you can run at very much lower temperature and much higher temperature. Does that fit with your understanding as well? Yeah, I'm not sure about the temperature specs, but low power for sure.
Obviously that leads to if you're running a solar system, smaller batteries, smaller solar panels, and all those things matter if you're running solar for instance, or if you're running AC, you might not be constrained with your power if you're running AC, but anything short of that is something to think about, or if the AC goes out, battery backup, with arm units, you don't need as much battery backup. So and, again, processing capability isn't the issue, it used to be plenty of horsepower and arm chips to do whatever you need. Exactly. Well, so I think this is interesting because, I'm going to kind of do a recap, because Python, one of the things, there's a lot of people out there that are selling analytics platforms.
Most of those analytics platforms are going and they're getting things that are available to share where that is Python code, and they're putting some kind of application on top of it that makes it easier for a user to interact with the code. But what that means is for a lot of people, they could get to an appointment at a better price point just by using Python. So that to me, that's kind of interesting because there's a lot of there's a lot of products out there that when you start unpacking what's really available underneath, it's, you know, it's stuff that's widely available at very low cost. Right.
You know, I'm going to suggest a different word for you there, though, in share where is an old term for something else. The word you're really looking for today is called open source. Right. And even with open source, you still have to kind of keep an eye on the license models.
They have, they're open source projects that you can use freely for your own open source project. There are other open source projects that you can use freely on your own commercial project. So there's that limitation, you've got to be careful when you're looking for libraries you're going to use. But basically what the application providers are doing is they're getting you beyond that, plus they're making it easier to interact with the capability of that code.
Oh, yes. Yeah. Library's doing, you know, encapsulate a whole lot of functionality to make it really easy for you. Right.
So the other thing I want to talk about in this whole kind of what's going on at the edge, what's happening is new development is the idea of data management. So to me, one of the things that's going on that's really not being addressed yet is that, and I'm old enough to remember, you know, software and databases on mainframes and IBM DB2. So for those of, for anybody who understands what that is, you know exactly how old I am. So, so when we moved off of mainframes and we started moving to what at the time was called mini computers, what now we kind of call servers.
It's not the same thing, but it's a fair analogy. One of the big challenges was on those mainframes, you know, those, the mainframe vendor sold their software packages and the way they structured and organized data was extremely rigid and bound up in the application. But when we moved off the mainframe, we began to separate the data from the application. I think the same thing is going to have to happen at the edge if we're going to really get to the value proposition.
So let me unpack that a little bit. I want to tell you what I think I, what I mean by that, I want to hear your opinion about it. So right now, most of the data at the edge exists in a PLC or an RT or a flow computer and is not well structured. There's not common naming, there's not common units, there's, there's a lot of challenges around how the data, there's no, and the data is flat, meaning it's a list of data points with names.
There's no organization to it. Like it's not, well, here's all the data around a pump and this is what the, so, so to collaborate that a little bit more, a pressure is not always a pressure. Sometimes a pressure is a pump, sometimes it's a pump discharge, sometimes it's a compressor, sometimes it's a pipeline line pressure, sometimes it's a meter body pressure, sometimes it's a vessel pressure and all those pressures mean something different. So I think there's a big need to create tools like relational databases, but for use at the edge such that I can manage the entire enterprise of real time data.
And I'm not aware of anything on the market doing that, although I do know some people that are working on it, but I'm wondering what your thoughts are about that. What you mentioned is kind of the key bit, you know, every, well, actually every PLC program, every RTU vendor, you know, the naming of different pressures is all over the place, and which is fine, they did their thing for their device and it's great. But when you bring it all together, you want to normalize that somehow. So just like you said, it's just static pressure on for measurement versus a pipeline pressure versus suction pressure discharge pressure, for all different, that means something different.
So, you know, to each of these data points, you know, attach some kind of metadata, metadata meaning, you know, data about the data, you need to identify it. So you said it's a pressure, okay, which kind of pressure, suction discharge, what are the units? Is it PSIA, is it VAR, is it something else? Is there a time factor associated with it?
Is it a life pressure, is it a daily average, is it hourly average? If it's a daily average, hourly average, what hour is it for, what day is it for? And then you have to pull those together. You know, is this, is that a pressure off, the stage one, suction pressure of compressor A, or is it the stage two, suction pressure, high compressor A, or is it actually the suction pressure on pump six?
Yeah. If you can normalize that, now all of a sudden you have a mechanism for looking, you know, across your system, what does the amount of pumps, we look at my suction pressures, you know, you can do a lot of things once you've applied common identification. Right, and one of the key things that's required to do broad-based enterprise level analytics is you've got to have control over the data. The data has to be reliable, has to be well organized, and has to be accurate.
Exactly. And that's a fundamental requirement before you can do anything meaningful with analytics. And you know, if you go to, if you go to, you know, if you go to a single run flow computer from a major manufacturer, then it's pretty easy to get understanding of what that data represents. Although there's still complexity around, well, is that a Coriolis meter?
Is that a, you know, is that a Orpheus meter? Is it ultrasonic meter? Is it a turbine? You know, those kinds of differences aren't always clearly easy to ascertain, by the way, the data set up in the RTU.
So I actually think that's one of the big problems that needs to be addressed in our business. And we'll see how that shakes out. But once you start getting all this bigger power at the edge, and once the edge becomes a computer versus a proprietary device, now all of a sudden there's a lot I can do that I, that really, the there's not any technical obstacle to doing it now other than there's infrastructure obstacles. Right.
Well, and you go from, you know, PLC, which I would say it's not a general firm's computer. You know, it's, you could do a lot with it with your program, but it's not a general firm's computer. But where you put your edge device out there, you can write any program you can imagine to get out there next to your PLC. Yeah, and there's a lot that goes into actually making that viable.
And I think it's an industry where ways away from getting there. But it's a very interesting conversation. So here's how I want to kind of wrap this conversation up. A lot of times I'll do three key takeaways.
I want to do this a little different because we're talking about a whole bunch of technologies and a whole bunch of things that, you know, many people are playing with or experimenting with and may not yet have completely figured it out. And my question is this, why should I as a pipeline operator care about this conversation we just add? What do you think? Well, I think it's really interesting.
Yeah. Well, I do think I don't know that the five-pound operator should be thinking it's interesting. Yeah, I'll do that. With the cost of these things coming down, like we mentioned earlier, Raspberry Pi 35 bucks, general purpose computer, reasonable horsepower, 35 bucks, all of a sudden have the capability of substantial computational horsepower that uses very little power.
You can put anywhere you want. So the possibilities are endless. And there's a, let's say operational issues or maintenance issues that will come up that if you had that kind of capability, you might have been able to discover the situation before equipment failed or get better information about misconfigurations or the large, large class of problems you can solve by having, you know, a general purpose, you know, targeted, but general purpose computer next to your PLC, but looking for something else, the PLCs are running the system safely. The computer can be looking for other things.
Yeah. So I'll try to, I'll try to give a specific example because I think specific examples are often helpful. Well, I'm thinking, I'll give a couple of the first one I'll give is kind of a process example. But let's say I'm running a gas plant and I've got inlet gas coming into the gas plant and I've got a chromatograph there and then I don't have chromatographs upstream of the plant inlet.
That'd be pretty common. But I have measurement and the measurement has detailed analysis of the field. So, you know, once a month, I'm updating the flow computers in the field, but at the, at the plant, I'm looking at the real-time gas stream. I could actually create an algorithm that would tell me that the well-head composition is changing because it's deviating from what I think it ought to be.
So I can, and I could do that in real time. If I, if I use this kind of capability and by knowing that, I might be able to operate my plant more efficiently and consequently, you know, drive more profit. And further to that point, I might do that in a way that I, my algorithm becomes proprietary and then that becomes a way for me to outperform a competition. So that'd be one example, right?
Now, people that know what I'm talking about will go, would say, well, I can kind of do that now, but there's a difference when you're doing that on, you know, millisecond data versus every few minute kind of data. It matters. The other example I'd give would be something like leak detection, where by putting a computer in the field and by looking at the, you know, a pipeline pressure and how that pipeline pressure is moving in very small increments of time, I might be able to see things that I wouldn't be otherwise be able to see. I mean, another example.
So I think, and I'm with you, I think the examples are endless. There's lots of them out there. So ultimately, what this does is it creates an opportunity, one, to do further optimization and further asset integrity type things with real time data. And secondly, it creates an opportunity to create an algorithm that creates a competitive advantage and deploy it in a way that before now, you wouldn't have the ability to deploy it.
Right. Even the simpler model, so you have vibration sensors on your compressors, and you're monitoring those in your skates. So if they get high enough, you'll alarm them and you'll do something about it. But you know, now imagine you have a device out there in a field that can, you know, collect those vibration values once a second, once a second, once a second, something really fast.
And then it can pack them up and it can send them up to a back office system or, and they can be compressed. You can do a lot of pre-analysis in the field. You can also do more modern computational stuff, you know, compress it, get it small, you can get it to the back office because traditional scale systems don't deal with compressed data. Right.
Even even EFM collections, not really compressed. It's full size data. Right. That's exactly right.
I didn't even thought about that. So that was, and you could have a much richer data set in that office than you had before without burning up your communications bandwidth. Well, the other thing that this does too, kind of along the same vein, is if I'm running on a standard computer, well, then a whole bunch of things opened up to me in terms of data compression, encryption, firewalling, and security, all things I can do in software to harden up my processes. Right.
Because that's one of those other areas, not, if you're going to put this, if your control network is on open radio frequencies, unless you've got firewalls, that's a potential risk. Right. But with this kind of unit at the field, you got a firewall, you can't get to it from the host, it can transfer to the host only. There's a lot of things you do by having a proper firewall at the field.
Right. And you can do it in software. So I don't have to, you don't have to install, support and power, another piece of hardware. Right.
Yeah. Well, this is awesome. So, you know, at the top of the show, we always do the intro and we welcome all the Bubba Geeks. So what's clear in this conversation to anybody that's listening is Scott and I are definitely geeks.
What would also become clear if you hang out with us on the weekend sometime, we're likewise Bubba's. So, this is where Bubba Geeks comes from. We would be sitting out, if we were to go hunting together, we'd be sitting out in the blind, having coffee as the sun comes up and talking about edge computing. That's just, that's what, that's what a Bubba Geek does.
Right. Scott, thanks so much for coming back. It's been too long. This was fun.
I definitely learned some things and we need to do this more often. I hope you enjoyed this week's episode of the Pipelineers Podcast and our conversation with Scott Williams. Just a reminder before you go, you should register to win our customized Pipelineers Podcast Yeti Thumper. Simply visit pipelineerspodcast.com slash win to enter yourself in the drawing.
If you would like to support this podcast, the best way to do that is to leave us a review. You can do that on iTunes, Apple Podcast, Google Play, Stitcher or whatever, smart device, podcast app you happen to use. You can find instructions at pipelineerspodcast.com. If you have ideas, questions or topics you'd be interested in, please let me know.
Or either on the contact us page at pipelineerspodcast.com or reach out to me on LinkedIn. Thanks for listening. I'll talk to you next week.