For the first time on the Pipeliners Podcast, host Russel Treat dives into the topic of measurement analytics. To facilitate the discussion of this important topic affecting the future of the pipeline industry, Bruce Wallace of Peak AI Solutions joins Russel to discuss analytics, Big Data, learning algorithms, and more relevant issues.

Listeners will learn about the different schools of thought related to learning algorithms, the importance of a measurement data management system, and how artificial intelligence will affect the future of pipelining as new technology is adopted and implemented.

Measurement Analytics – Show Notes, Links, and Insider Terms

• Bruce Wallace is the President of Peak AI Solutions. Find and connect with Bruce on LinkedIn. Alternatively, email Bruce at Bruce.Wallace@Peak-AI.com.
• Measurement Analytics ensures that measurement data is correct and complete for consumption by downstream accounting services and for end customers’ reports.
• A flow computer averages and records all digital signals received from various types of flow meters, as well as analog or digital signals from temperature, pressure, and density transmitters.
• A Measurement Data Management System aggregates data, normalizes the data, and stores the data on a server database. This prepares the data for downstream processing.
• Big Data is two-fold: structured data contained in spreadsheets and database tables and non-structured data associated with communication.
• Predictive Analytics uses computer vision to analyze data on a chart using a learning algorithm, calculate the data, and analyze the data. Additionally, computers are programmed to look for scenarios to determine whether action needs to be taken, such as an edit.
• There are five main schools of thought about learning algorithms. Each approach has a defined methodology and their own principles.
  • Bayesian Statistics focuses on how to handle uncertainty. This approach starts with a hypothesis, gathers data, and updates the hypothesis as more data is gathered.
  • Connectivity focuses on “deep learning” based on connecting artificial neurons to a neural network. This is helpful for image recognition.
  • Symbolists start with premises or a conclusion and work backward to find the missing data using logic.
  • Evolutionaries focus on applying genomes and DNA to data processing. The idea is algorithms will evolve and adapt.
  • Analogizers focus on psychological techniques or models such as the “nearest neighbor” to provide results.
• Custody Transfer Measurement involves a metering point (location) where fluid is being measured for sale from one party to another. During custody transfer, accuracy is important to both the company delivering the material and the eventual recipient.
• FERC (Federal Energy Regulatory Commission) regulates, monitors, and investigates electricity, natural gas, hydropower, oil matters, natural gas pipelines, LNG terminals, hydroelectric dams, electric transmission, energy markets, and pricing.
• FERC Order 636 was issued in 1992 to relax service requirements on pipeline firms and gave customers greater purchasing flexibility by separating gas sales from transportation. The order also extended transportation to include storage and allowed end-users with firm transport contracts to sell unused capacity.

Measurement Analytics – Full Episode Transcript

Russel Treat:  Welcome to the “Pipeliners Podcast,” episode 12.

[background music]

Announcer:  The Pipeliners Podcast, where professionals, bubba geeks, and industry insiders share their knowledge and experience about technology, projects, and pipeline operations. Now your host, Russel Treat.

Russel:  Thanks for listening to the Pipeliners Podcast. We appreciate you taking the time to listen to the episode. To show our appreciation, we want to let you know about our signature prize pack. We are offering a free customized YETI tumbler to one listener each episode.

How do you register to win? Simply visit pipelinepodcastnetwork.com/win to enter yourself in the drawing. Very excited to announce our latest winner of the customized YETI tumbler. Congratulations to Dave Dodds with Dominion Energy Ohio. You should be receiving that shortly, Dave. Thanks for listening.

This week, we have a new subject matter. We’re going to talk about measurement and, in particular, measurement analytics, how big data and predictive analytics are impacting the measurement part of the pipeline business.

For that purpose, I’ve asked Bruce Wallace with Peak AI to join us today. I’ve always thought of Bruce as somebody who was on the cutting edge working on interesting technology and very forward thinking in our business in general. With that, let me welcome to the Pipeliners Podcast Bruce Wallace.

Bruce, welcome to the Pipeliners Podcast. So glad to have you on board.

Bruce Wallace:  Very happy to be here, Russel. Thank you.

Russel:  Bruce, when we first started talking about you coming on and I was asking you what you might want to talk about, you said, “Well, let’s talk about measurement analytics.” I said, “Okay. Let’s.” My first question for you is what is measurement analytics?

Bruce:  Measurement analytics is usually considered a very boring topic except for the people that are having to deal with it. The reason that that exists, it’s a pretty tight knit community of people that make sure that measurement data is correct and complete for consumption by downstream accounting services and for end customers’ reports.

When we talk about measurement analytics, we wouldn’t really need that if measurement data never was made to be bad. When I say bad, I’m talking about say at the meter source, somebody forgot to change an orifice plate and they entered into a flow computer that they did.

Anywhere that a human can touch a point in a metering system, whether it’s physical or during programming and configuration, there’s a potential for failure. It happens more often than people really understand.

Russel:  I would agree with that. I think some of the things that people don’t generally understand about measurement is that it’s a lot of detail. There’s a lot of things that you need to set in the flow computer to get the calculations to be performed correctly. It’s easy to get one little thing wrong.

That one little thing wrong, the risk of that is you get numbers that are in error, causes you to have imbalances or loss unaccounted for. The other thing about measurement, we like to refer to it as the cash register. It’s where the ownership of the products being transferred between one party and another. Consequently, there’s audit exposure.

When you get these numbers wrong, you create for yourself a lot of grief and aggravation down the road. I think people understand that. But I don’t think people really understand how many things you have to get correct in order for the measurement to be correct. Is that a fair statement?

Bruce:  Very fair statement. You can get down to the point of saying, “Okay, which compressibility method did I enter into my flow computer? Does that compressibility method match what is stated in the contract between my company and the other company?” If it’s not and you’ve had that wrong for a period of years, you may have to go back and make adjustments for a period of years. It can be costly.

Russel:  Exactly. In my experience, the field technician who’s putting in the flow computer and configuring everything is not generally carrying around a copy of the contract. [laughs]

Bruce:  That’s generally true. They’re not carrying a copy of the standards that the contract refers to for calculating volumes.

Russel:  That’s also true. Again, I guess we come back to briefly what is measurement analytics?

Bruce:  When we talk about measurement analytics, we basically have to have something that’s aggregating the data coming from all the meters that a company might pull data from and process for revenue accounting purposes, financial purposes, etc. and for operations.

To store that data and aggregate it and to normalize it, we have to have something that could be called a measurement data management system. In the old days, probably pre ’90s, that data management system was a mainframe computer or something very similar with a very strict and regimented program that was somewhat difficult to change with changing times.

Over the years, we’ve come to the point where we are using Microsoft SQL Server, Oracle, some kind of a SQL database in combination with a server or a hosted platform to store all that data. The data comes into the system. It runs through some traps, basically to throw out exceptions for data that’s being validated.

Those exceptions have to be reviewed or analyzed, and if they’re mixed material, they have to be edited or acted upon. Once the data’s edited, it’s complete and it’s ready for downstream processing.

Russel:  It’s interesting when I do this, I’m always mindful of somebody who’s not be listening to this that doesn’t know anything about measurement. That’s true in all these subjects when it talks about various things that pipeliners do. There’s always a certain level of jargon and so forth. I try to get some of these concepts down to something simple.

I think what I heard you say is I have to have a method to organize and go through all the data and make sure that it all makes sense.

Bruce:  Absolutely.

Russel:  Basically, that’s analytics where there’s some computer system that’s doing all of that and it’s kicking out things that somebody needs to look at.

Bruce:  That’s absolutely right.

Russel:  Cool. Analytics has been around for a long time.

Bruce:  Yes.

Russel:  We’ve been doing that since before you and I were in the measurement business which neither one of us…

Bruce:  Right. Even people looking at charts, saying, “This is right or this is not right.”

Russel:  Yeah, exactly. Looking at paper charts and saying that’s right or wrong just because they know enough to recognize a pattern that’s right or wrong. Now, things have moved from that kind of paper technology to electronic technology. Now, we have computers that are looking at that stuff. Where are we headed with analytics?

I think one of the things you mentioned is you want to talk a little bit about predictive analytics, what that is and how that might get applied to measurement?

Bruce:  It’s interesting that we talked a little bit about charts. Believe or not, there still are a significant number of charts, not necessarily on pipeline or high-end midstream systems, but certainly in production systems to the point of maybe having in North America and Mexico close to 800,000, 900,000 charts change per month.

That actually is one of the things that new technology with predictive analytics and with something called computer vision is attacking right now, such that instead of having a machine with a turntable retrace those chart lines and turn those analog signals into digital signals and eventually into numbers, companies are now going out with smartphones, taking a photo of the chart.

Within a minute or two, the data is being analyzed by a learning algorithm and is going through all the standard calculations and becoming volumes. It’s just that fast, and with the technologies that we’ve got, it’s everywhere.

That’s just in that one aspect, but when we talk about predictive analytics in the measurement data management system, we’re talking about prescriptive things that can be programmed into the computer to look for scenarios, and automatically determine whether an edit needs to be done or not, and actually make the edit as part of the machine’s responsibility.

Not only that, but leave an audit trail.

Russel:  Can you give me maybe a specific example of something that a machine could find and edit? To me it’s simpler to understand how a machine might find something. I tend to think about something like a meter freeze. With a meter freeze, you’re going to see your temperature’s going to be below a certain number and your differential is going to be rising.

Bruce:  Sure.

Russel:  That kind of thing. That’s a pattern that I can kind of visualize in my mind’s eye and see. But what might be an example of how a machine would see a pattern and then actually make a correction? That I have a hard time visualizing.

Bruce:  That’s a very good example that you just gave. Basically, when you can modify or correct the input variables going into the volume calculation, then everything else falls into place. To do this, you have to have, what everybody’s heard of, big data.

Big data’s not only the structured data that you see in spreadsheets and database tables, it’s also the non-structured data associated with communication such as if an analyst working in one of these systems is required to put an audit trail reason for the edit that was made.

The machine can take that contextual information and combine it with its pattern recognition, make the edit, and put in a reasonable audit trail comment, but it takes a lot of data.

Russel:  Interesting. How would you define a lot of data? Can you put a quantity on that?

Bruce:  Right now, I can’t. The technology being applied in this area is very new. Basically, you just start where you start. You keep metrics on how well your algorithm is performing looking at the results. In some cases, you create a learning model, feed the machine the data, have a separate set of data that the machine can’t see as a test set, and then score how the machine is learning.

Once you get to a certain acceptable percent of success, then you’re probably where you can be as far as what that algorithm can learn. At that point, you’ve reduced the amount of work that an analyst has to do tremendously.

Russel:  I was talking to one of our customers just recently. They’ve got operations up in the Northeast. We’ve had some really cold weather. In late December after Christmas, all of January, most of February, they were just getting hammered with all kinds of meter freezing problems, just absolutely hammered.

Finding them is relatively straightforward. I say relatively, if you were watching me on a TV screen, you’d see me doing air quotes right now when I say relatively.

Bruce:  The key thing is to keep feeding the machine data. More and more data will constantly refine the algorithm. The algorithms themselves are referred to as learning algorithms. If they do produce a bad result, you can make that change. The machine will learn from your change to its incorrect result.

Russel:  This is actually something I’ve been reading a bunch about. I’m a bit of a math nerd. Weird, I enjoy statistics and all that stuff. One of the things I know about these learning algorithms or these predictive analytics is they’re really good at looking at a normalized data set and using history, when they’ve seen something before, seeing it again. Seeing something they’ve never seen before.

Bruce:  That gets into different kinds of learning algorithms. I could go into it but understand there’s something like five different camps out there that have defined themselves and are somehow working to converge their methodologies and principles. One is a statistical tribe if, you will, using Bayesian statistics, I suppose.

Another one is a connectivity tribe which looks at how the brain functions and develops algorithms based upon that. Another one is the analogists such that if you were to consider an example where A and B are different entities, and A and B like two different things, then you find out A likes a third thing, you might be able to make the analogy that B will like that third thing also.

I can’t remember all of the different tribes, but they’re all working together to overcome some of what you’re talking about as far as predicting things that have not been seen.

Russel:  There’s certainly a whole lot of value in finding these things sooner and finding them in a way that you can say, “Well, this looks like a orifice plate mid-size,” or, “This looks like a meter freeze,” or, “This looks like a bad chromatography report.” That in and of itself is super valuable.

Bruce:  Not only these things, not only the learning from the algorithms, but there are other things that can be done. What I’m really interested in personally is to develop systems that basically streamline this part of the measurement process.

Some of the systems have become so complex in and of themselves that an analyst not only needs to know the measurement process but it’s come to the point where they almost need to have a training course in the system that they’re operating.

Russel:  I think that’s very well said. We talked earlier about just the level of complexity in getting a meter right. The complexity is only amplified when you get it into the back office and you start trying to put it together with all the other data that needs to be joined up, like lab reports and things of that nature.

Bruce:  Some systems allow you to create physical impossibilities that are totally outside of your contractual realm. You can rock along with those for a long time and not catch it.

What I envision is a system that becomes more minimalist such that just as you’re asking your iPad or your iPhone, “What’s the closest restaurant?” I envision being able to log into your data management system. The system prompts you upon log in and says, “What do you want to do?” From there, you build a meter.

Russel:  I can put my mind’s eye on that. There’s so many variations. If you just think about natural gas measurement, don’t even talk about liquids, just natural gas. Is it produced gas? Is it gathered gas? Is it rich? Is it lean? Is it wet? Is it dry?

All of those things impact not only how you do the measurement but what math you apply and what algorithms and standards you apply in order to get your measurement number.

Bruce:  Right, but all those things have got rules. The rules are your contractual requirements and the standards that you use along with your company’s policies. If you can put those rules into the system, you can have the system prompt you and guide you down the correct path instead of making mistakes because you don’t have the knowledge and experience.

Russel:  That’s a fascinating idea. One of the companies that I’m involved with, Gas Certification Institute, one of the things they do is, they provide standard operating procedures. We have literally thousands of pages that represent standard operating procedures.

When you say if you get the rules, to me that’s a bit mind boggling because I would assume — and maybe you can shed some light on this — it’s not just capturing the rules, but also capturing the ways the rules interrelate.

Bruce:  What we’re doing, though, is we’re putting those responsibilities into minds that are associated with bodies that have to carry out different processes and procedures. They may truly not have the years of experience to even understand why they’re doing what they’re doing.

Most of those rules, if you can put them into the system once, they’re in the system. Whereas you may have to revisit, revisit, revisit, you will make mistakes as a human making your own interpretations of things that you don’t quite understand.

Russel:  No doubt. Like yourself, I’ve been involved in measurement a long time. One of the things I’m way more clear about now than I used to be is what I don’t know.

Bruce:  Me, too. [laughs]

Russel:  Measurement is an interesting thing. You might get into a particular organization and be doing a particular kind of measurement, and really be all over those details. You can’t necessarily just pick that expertise up and apply it to another organization, because the nature of how they operate, the nature of how they contract and do business can materially impact how you do your measurement.

There’s just so many variables. I get what you say conceptually. To me, that sounds like a huge undertake. When might you be getting others to look at what you all are working on?

Bruce:  We already are. For instance, I mentioned the chart, the computer vision application. That’s already happening, both in the U.S. and in Canada. We’re exploring Mexico right now. Beyond that, on the predictive analytics, we have already created some training models. We’ve done some preliminary work. It’s a matter of getting the data in, which just means new customers. That’s picking up all the time.

Russel:  That’s awesome. To me, this is exciting stuff. I find the measurement back office to be challenging, in terms of what it is and what it does. I should probably put a little context around that.

What I mean by that is, as a guy who has never actually worked in a measurement group but has been a consultant to many, there’s always a fairly steep learning curve to get your mind around what is their particular operating model. Once you’re there, you can start adding value.

It’s getting that understanding of, “Where’s the swim lane that they’re working in?” That is really critical to gain that understanding. If you have tools, getting to that understanding becomes expedited. It becomes quicker.

Bruce:  That’s true. Also, you probably, along with everybody else in the industry, realize that the cycles of boom and bust have not always brought back the talent, knowledge and experience that was lost during each bust.

Because of that, we got a lack of experience. We got a lack of personnel, actually, that are coming in to try to fill these roles. That’s one of the things that AI and predictive analytics can help fill gaps with.

Russel:  I hear you. One of my concerns is, if you start making the “job” — again, I’m doing the air quotes thing — easier, then where do you find the people that really understand things from a bigger picture? I guess that has to do with, somebody has got to understand the rules, the algorithms, the processes, and how that all goes together.

Bruce:  That’s true. Most of those folks basically will depend upon people that actually are involved with standards organizations, people that do have a rich and varied experience in measurement, whether it’s measurement in oil and gas or otherwise, or parallels in all the measurement worlds.

Russel:  Without a doubt. It’s not just what you and I know as oil and gas. It’s not just oil and gas that has to get measured where this thing matters.

Bruce:  That’s right.

Russel:  There’s a vast area in petrochemicals, mining and a whole lot of other things. These issues, while different, there’s a lot of similarity, although there are important differences.

Bruce:  Here is an example. I’m afraid to ask you this, but I will, anyway. When was the last time you balanced your checkbook?

Russel:  [laughs] Oh, my gosh. I balance my checkbook at least once a week.

Bruce:  Do you really?

Russel:  Yeah.

Bruce:  There are so many people now that don’t, including myself, because it’s just so easy to go online constantly or if you want to, and look at what your balance is, and look at all your inputs and your outputs.

Russel:  Yes. If you do all your banking electronically, that gets easier. I don’t.

Bruce:  I do.

Russel:  It’s interesting. I’m a technology person, but I’m a late adopter of many things, because I want to have absolute certainty it works.

I grew up as an engineer first, and a measurement and an IT software developer person second. I want very high reliability. Because of that, it influences my willingness to adopt and experiment with new technology. Bruce, it’s really a good question, because your expertise is probably more in production and gathering than in pipelining. Would that be true?

Bruce:  Most of my experience outside of the vendor world was actually in midstream.

Russel:  If you talk about pipelining, midstream is certainly part of that, you talk about the big guys, the big transmission companies and all of that. These guys are highly risk averse, and for a good reason.

Bruce:  That’s right.

Russel:  For the most part, they’re running processes that they need to run very efficiently and very safely. The best pipeline operators are companies that generally — there’s exceptions — are not in the news very often. My point being that this risk aversion makes getting some of these new kinds of ideas and these new kinds of concepts challenging to adopt in some of these bigger organizations.

Bruce:  Their environment in pipelines has been changing probably since the late-’90s, too, whenever there was the big push for ethane rejection because of everything going topsy turvy in the liquids world. You had to really monitor and understand what was going in as pipeline quality gas from a processing plan perspective.

A lot of people were pushing the boundaries. You were ending up having situations like hydrate formation in your instrumentation and other places where the input was supposed to be “pipeline quality gas.” This technology can help, using those inputs for composition determine whether or not you do run the risk of having freeze-ups.

Russel:  That, I can see very clearly. This gets into a conversation about what is measurement, what is SCADA, how measurement and SCADA are different, and how analytics would work with measurement data versus how analytics might work with SCADA data. It’s the same but different, I would assert.

Also, the other thing that we’re probably…There’s probably going to be change coming out as a result of all these that we cannot anticipate. I remember when the EFM were early into the market, as I was getting started.

I was with a company that had a chart integration. That put a DOS computer with a chart integrator so that we could back print a volume right on the back of a chart. Let me tell you, in 1992, that was really cool, state of the art technology.

Bruce:  That was cutting edge.

Russel:  Yes, it was. [laughs] Younger people listening to this are going to get a giggle out of that. I just gave out on us both how long we’ve been doing this. Sorry, my apologies.

Bruce:  [laughs] No problem.

Russel:  My point being that at that time, FERC, which is the Federal Energy Regulatory Commission, did its Order 636. At that time, all the big pipelines primarily had charts. Most of their businesses were monthly in terms of their business process.

When FERC Order 636 came out, basically, before that, the big pipelines would buy the products, move it through their system, and sell it on the other side. Generally, they were price-controlled at the sales point.

In the ’70s, there was a huge gas shortage in the Northeast. They changed the rules and said, “Well, we’re not going to have the transportation companies buy and sell the commodity. We’re going to have them charge for transportation,” which meant, all of a sudden, everybody needed to know every day, “How much of my product is in their pipe? Where am I putting it in? Where am I taking it out?”

It drove us to electronic measurement. I don’t know where we’re headed, because I don’t know that I can predict that, but I think that this whole analytics thing is going to have just a big an impact. Do you have any ideas on how this might change the measurement part of our business going forward, as this technology begins to come out and become more dominant or widely used?

Bruce:  It’s going to impact the trading side before it hits the measurement so much. When we’re talking about energy trading and risk management, that’s where you’re going to see a lot of the benefits first.

Russel:  That makes sense. That’s not unlike what they’ve been doing on Wall Street for years. The guy with the best algorithm and the fastest computer ends up carving out an advantage in the market, until all the other algorithms figure it out and catch up and normalize it.

Bruce:  That’s right. Then it becomes just a very tight margin, regardless.

Russel:  One last question I want to ask you. I don’t know how this applies to measurement. I’ve been asked many times, probably over 30 times, in the last year, “Russel, what’s all the big deal about Internet of Things, big data, and predictive analytics? How is that any different than what we’ve been doing in SCADA for 30 plus years?” If I were to put that question to you, how would you answer it?

Bruce:  Over the years, you and I have both seen the cyclic tendency to go to centralized versus decentralized monitoring and control.

What we would possibly see with AI is a more centralized situation with the “cloud” being the big aggregator, such that each of the end devices are able to talk to the cloud    and to whatever device, system, or platform — as the algorithms — to aggregate and process and act upon those inputs. That’s probably one of the things that you’ll see.

Russel:  It’s an interesting answer, Bruce, because if you were to ask me the same question, I would tell you, it’s going to force us to decentralize.

Bruce:  At some point, we have to define what those terms mean. What is decentralize?

Russel:  [laughs] Exactly.

Bruce:  I have been on too many standards bodies. It’s terrible.

Russel:  [laughs] It will impact the way you think, I grant you that. Look, I really appreciate your coming on board and joining us with all these. Is there anything else that you want to talk about, as it relates to measurement analytics and just whole subject? Anything you think the listeners might be interested to know?

Bruce:  No, I’m just interested to hear what the community thinks about it all. Either going through you and your podcasting or through other resources, I would enjoy hearing feedback.

Russel:  There you go podcast listeners if you want to get in touch with Bruce. Bruce, what’s the best way to get in touch with you if somebody had questions and they wanted to reach out to you?

Bruce:  I’ve got access through LinkedIn. I also have an email that somebody can send me their questions, concerns, suggestions. It’s Bruce.Wallace@Peak-AI.com.

Russel:  Great. We’ll also make sure that all that’s on the show notes page. There’ll also be a profile for Bruce on the “Pipeliner Podcast” website. You can go there and find his contact information.

I would encourage you if you do have questions about all this, please reach out to him. I think very highly of Bruce. It’s one of the reasons I asked him on. I’m always curious what he’s working on because it’s always futuristic and very interesting, at least to me.

Bruce:  [laughs] Thank you, Russel. That’s a high compliment coming from you.

Russel:  Thanks for being on the show. We really appreciate it.

Bruce:  Thank you, sir.

Russel:  I hope you enjoyed this week’s episode of the Pipeliners Podcast. I certainly enjoyed the opportunity to talk measurement with Bruce Wallace and certainly learned a lot from the episode. I think Bruce and I have some different thoughts about where the future is headed, but hopefully the listeners will value the differences and learn from that.

Just a reminder before you go, you should register to win our customized Pipeliners Podcast YETI tumbler. Simply visit pipelinepodcastnetwork.com/win to enter yourself in the drawing.

If you have ideas, questions, or topics that you’d be interested or would like to see us talk about or if you have in mind guests that you think would be interesting and engaging and have information that pipeliners would enjoy, then please go to the contact us page at pipelinepodcastnetwork.com or reach out to me directly on LinkedIn.

Thanks again for listening. I’ll talk to you next week.

[background music]

Transcription by CastingWords