Podcast: Oil analysis can boost equipment reliability, but only if you use the data

In this episode of Great Question: A Manufacturing Podcast, chief editor Thomas Wilk talkes with Mike Holloway, president of 5th Order Industry, about the intersection of artificial intelligence and oil analysis. Mike has more than 40 years experience in industry and holds 16 professional certifications, a patent, an MS in polymer engineering, a BS in Chemistry, and a BA in philosophy. Mike is a subject matter expert in tribology, oil and failure analysis, reliability engineering, and designed experiments for science and engineering. He is currently publishing a blog series with Plant Services on AI and oil analysis, and previously published an article on tribology certifications with this magazine.

Below is an excerpt from the podcast:

PS: The article that came out of our initial contact was one on certifications in lubrication, but you're writing a new blog series for us. It's a multi part series and in the first installment you talk about oil analysis customers who do and don't use those data to improve reliability. Maybe we can start there because I was struck by one of your statements, that when someone asks you how many customers use those data for the oil analysis companies to improve reliability, you put a thumbnail on that of about 15%. I was curious about that number, what are they doing right and what are the other 85% doing wrong?

MH: This really came about because for a short time I worked at SGS, and my boss who passed on recently (his name was Patrick Runchey), actually, two days before he died of an aneurysm, he called me up after I'd been working for my own company. And he said, hey, I’ve got a question for you: How many customers of ours do you really think use oil analysis to improve reliability and whatnot? I said that pretty much it's between 10-15%.

He's like, no, I don't believe that, it’s got to be more like 50%. I said no, Patrick, I know for a fact because of all the years I've done training, a lot of times I ask this question to all end users and it's about that. He's like, that's ridiculous, why do we do this then? I said, we do it for the 10-15%. He's like, is that good enough? I said, well, that's a question you have to ask yourself; to me .oil analysis is an exercise bicycle. He said, what do you mean by that?

I said: listen, exercise bicycles have been around as long as bicycles have been around. They really have been. But the interesting thing about it is that they're not a really good business, because the idea is that you buy the bike, you use it for about a month with all great intentions, and then you just kind of forget about it. We saw this with Peloton, and that was a high tech exercise bicycle. But it really failed, and the reason is that people don't make it a habit, and they don't see the immediate benefits from it. And that's always been the problem with so many things. To transition into the series about AI, I talk a lot about the innovation of AI, but we see a lot more immediate gratification and influence through AI than we do through other things. When we actually do take a look at oil analysis, it could take months if not quarters, if not a couple of years to really see the overall benefit that this thing brings to the table.

Many years ago I worked for ALS and we had a really large account – Wal-Mart – and they're pushing maybe 35,000 samples a month, some ungodly number. And I was watching the numbers and all of a sudden one month nothing was coming in. So I called the account manager and I said, can you look into this? Why haven’t we gotten any samples in the past week from Wal-Mart? And she did, she called back and said, I can't get through to the decision-maker. I said, well you’ve got to, something’s up. She finally did and said they just decided they didn't see any value in it anymore. I go, why is that? She said, everything's green, everything looks good. I said, ok, however, there's still value in that. She said, apparently they just said, “hey, we don't need this anymore, because everything's fine.”

I said, you know, that's analogous to when we go to the doctor for an annual physical and the doctor says, you look great, keep on doing what you're doing. Then you think to yourself, do I have to go next year? Everything has been fine for the past 20 years, why should I go next year? Well, that next year, if you don't go, maybe something's not going to be fine. You don't know, it’s really to keep an eye on things, but people don't realize that. They just think that if there's not a problem we’re just going to move on and attack only problems, because human beings are really good at solving problems and innovating, we're really great at it actually. It's probably one of our only things we're good at, and it has to do with our ability to communicate and work as a team.

What's interesting is that with oil analysis, is that it’s a cool idea, you can't find anybody that disagrees with it, but you can't find many people who embrace it to that degree. Now, that's not necessarily always the same with like thermography or vibration analysis. Because with vibration analysis it's a physical thing that we can touch a machine and feel it shake a certain way; or with thermography, we know if it’s hot, it's not right. Those are all physicalities but oil analysis, the best you can do with that is maybe smell the oil, and if it smells acrid or if it looks really dark then you could have physicality to it, that maybe there's something wrong with it.

Normally oil is like blood – you don't touch it, you stay away from it. You can analyze it remotely or through a lab, but you don't have that visceral feel for if it's any good or not. Therefore when we don't have that, I think we lose our ability to truly understand the value of it. People don't understand an atom or a molecule. They just conceptualize it. So how can we expect them to really understand the nuances of used oil in order to bring to the table?

PS: You're putting your finger on something – the human element of the kind of folks who go into this field. I hadn't thought about it this way, that the people who like getting their hands on the machines and their hands on the equipment, they like tactile, sensory data. And what you're saying here is when it comes to oil analysis, some of these people, if it's not tactile enough or if the results come back positive all the time, there's a tendency just to put those data in the background in favor of the more immediate: “I can hear it, I can feel it. I can smell it.”

MH: Sure. And there's a distinct reason why, and it has everything to do with how we think and how we learn. Years ago when I started my company, I realized that by presenting and teaching and instructing in person, a lot of times the reason why it's so effective is because I could tailor my approach to my crowd. If I was in a place that had mostly technicians and millwrights, I would get a lot more kinetic, I would put things in their hands. If I was with a bunch of engineers, there'd be graphs and tables and visual things. If I was teaching sales folks, it was all auditory stories, they're very auditory.

I got to thinking, everybody's different but really not that different. If you think about it, there's probably about 27 different types of characteristics of a personality. But when I start to look into it, there's several different ways in which people learn, and there's a dozen and a half types of intelligence. Now, you've probably heard about this, “oh this guy's got street smarts,” or “he's really logical, he's not street smarts.” But then there's also those that have kinetic smarts that are incredible athletes, but are horrible human beings. There are also those that are great musicians, great mathematicians, there's also people that have an intelligence for dealing with other people or being introspective and understanding themselves.

So I started doing some research into that as well, and I found out that yes, there's about 8 different ways in which people learn depending upon your neurology, and there's probably about 18 different forms of intelligentsia. (I actually started finalizing a book on different forms of intelligence on how to improve what you don't have and take advantage of what you do.) But when I built my company, it was really predicated around customized training, and I even did that online, where you could go online and take a customized assessment to find what kind of learner you are. And then if you want to, I can actually build a course according to your learning modality.

Well, it comes into play if we think about this oil analysis too, because the people that we're dealing with are mechanics, technicians, and millwrights. They have a propensity to work with their hands. If they weren't, if they're more visual and auditory, they would have gone into a different profession. The people that I deal with (mechanics, millwrights, and technicians) don't like sitting in a classroom. They like getting their hands dirty doing something, you know? And so it makes sense if they can't understand an oil molecule or a test result because it's not physical for them, they're not going to gravitate towards it. They will go towards vibration analysis, because that's physical for them. Now, if you take a look at vibration analysis versus tribology, it's really completely different, and quite frankly vibration analysis to me is like dark magic or something. It’s really complicated, but it makes perfect sense to many of these guys.

To me, I like the more esoteric abstract. I can completely understand chemistry, it’s like second nature, so it's not so abstract to me. But with oil analysis, if it just doesn't fit, then it doesn't gravitate towards them, unfortunately, because the way their brains are built.

PS: You mentioned in the first blog post that portable oil analysis units that you can use within factories and mobile equipment, that's getting closer within reach. Do you think the closer we get to those kind of technologies being available to everyone would make a difference, if these technicians get their hands on them?

MH: Yeah, absolutely. Back in the day, maybe 100 years ago, if you were driving a vehicle, it had a few different gauges and people who really started getting into automotives really had a certain sense of what the engine sounded like, and had a feel for it. It wasn't until many, many years, decades later, that we started getting warning lights and indicators and gauges that respond to what's going on, that we get a true appreciation for the reliability and performance of the vehicle.

It's the same way with machinery. Gone are the days (or will be soon) of the old mechanics that are just touching machine and telling you what's wrong with it. Now, you have your PLC and it’s telling you exactly what's up and what you have to do about it. I think in the first blog, I talked about a project that I worked on many years ago and it was for the Joint Task Force Fighter, the F35, to build a very small oil analysis unit that could test on the fly if the oil was good, cautionary, or bring the bird home as fast as possible. Part of the specification was, and they left it up to me to say what do you think we should look at as far as parameters and sensor technology, was that the footprint is 2” by 2” by 6”. It's not very big. We had to miniaturize a lot, which was interesting.

I said, how do you want the indicator to be? A gauge or something? “No, -- three lights: green, yellow, and red. That's it, and we'll put it on the IP and it will just say Oil Condition Indicator” or something.” When you're a fighter pilot, you’ve got too many things to look at. They just want to keep it simple. I said, OK, that's great.

Now, if we have that with machinery, like with an air compressor or hydraulic press, or for that matter a haul truck engine that says, hey, we’re approaching some problems, we're going to yellow, so bring it back into the state and we can do a quick diag by plugging a USB port into the undercarriage and then boom we're good. This is exactly the way we're going now. In the past, we never really saw on-board oil analysis really become mainstream because of cost.

Years ago, GM did have something along those lines where they said “you have so much oil life left”. Today, you and I, when we drive our vehicles, it'll tell you if it needs an oil change, but it has nothing to do with oil condition. It has to do with engine revolutions. It just counts many millions of revolutions and they assume that at some revolutions the oil is going to be burned. Not necessarily, but they're going to take the safe way out and say I should change the oil, better safe than sorry. But GM did have a sensor technology because I remember meeting a woman who came up with it, and it was really interesting. It actually tested the effectiveness of an antioxidant package that's found in all oils, and as soon as that went down to a certain level – it was a form of sweet volumetry actually – it would indicate that, hey, listen, our antioxidant package in our oil is no longer any good, or it's going to really start to become challenging, so it’s time to change it all.

That was really the first thing, but it didn't really catch on and it didn't provide great tech in terms of being repeatable, reproducible, accurate, and precise, but it was a great step in the right direction. I wish I still had her business card because I'd probably give her a call, find out what she's been up to on it. But they kind of just forgot about that, nobody really went back to it because they didn't have to. They felt, you know what? We can come up with a better idea by revolutions and just say, “through our development work on the very front end, we know that it can only last in a box for so long. We're going to assume that that's what it’s going to be like with engines going forward, and just say, change the oil at 10,000 km or 200 million revolutions or something like that.”

Is that true? No, I mean seen oil last in fuselages 100,000 miles without being changed. I know the Navy, they don't even change in the boat, sometimes they don’t even change the oil. They just add to it and filter it, and that’s it, the never change the oil. I think going forward that's probably the direction it’s going to go with most engines, if we stay with combustion engines. You won't have to do it, just change the filter and add oil, and that’s it. Will you need oil analysis for an engine? No. Should you have this for something like a power turbine or hydraulics? Yes, absolutely. But for transportation stuff, it might not be something that someone's going to gravitate towards because of expense.

However, now we're starting to see the cost of these sensors go down dramatically. And what's also interesting is let's say you have a sensor device and it could tell you if you have a certain amount of water in there, a certain amount particulate, a certain amount of wear debris, maybe even give you indication of your oil condition, acidity or neutralization capability, or even anti-oxidant concentration. That's all really good, but where does the data go? That goes down to your instrument panel and tells you, well, you're allowing a customer try to figure that out? You’re not going to do that. Well, then you just do the red, yellow, green, and then bring it in for a diag at your local mechanic.

Then we have the concept of, what you do with the data? I worked on some projects where we could actually take these sensor boxes and send that information to the cloud and then it goes into a bank in there for diagnosticians to assess what's going on and then make a recommendation, and that took time. We're now looking at saying, well, what if we just really threw that over to Chat GPT? I've read some of these reports and they looked fantastic. A couple years ago I said, who wrote this report, and one of the lab managers said we got a ChatGPT version of what we have internally. I said, that's amazing, it probably did it instantaneously, you could convert everything to that and probably put 250 diagnosticians out of a job within a week. Then he kind of looked at me.

So now that comes into a whole other ethical dilemma. I always loved looking at the industrial revolution back in the day because it was really transition, the first big transition of our country. Next came the Interstate Highway system. But really, the industrial revolution changed the complexion of our nation, in fact changed the complexion of Britain. There was a movement in the early 1800s, called the Luddite movement, and weavers, people who used to weave with looms or whatnot, they went absolutely berserk because they're being replaced by automation, by mechanized weaving machines. And weavers are primarily artisans, so they can't really put up too much of a fight, but what was really interesting is that they condemned automation and innovation because they knew that their life was changing.

It was interesting too, because in that time these weavers actually have certain social status. They were considered artisans and highly valued in their society, and all of a sudden they're being replaced by essentially a very simplistic robot. They lost complete value and worth, and of course people are going to go nuts over that, right? The British Government, not only did they crush the revolution, they executed many of these revolutionaries. I was like, gosh, we think we got it bad these days.

Now, what did we see last year in the year before with the Screenwriters Guild and SAG, then the longshoremen? The Screenwriters Guild and SAG feel that they’re going to be replaced by AI, because you could ask AI to generate a story and it will! It’s not a great story, but with a little tweaking it could be great. You can ask AI now to generate a particular image and we see this now, you can't tell the difference! There’s a program now, I messed with it two years ago, where all I had to do is talk into the screen and not move my face a lot, and it could take my image and translate whatever I said into any language I want, and made it look like I was actually saying it. Incredible. I was speaking Portuguese, I was speaking Hindi, I was speaking Mandarin. And I showed to some of my friends who actually speak these languages, and they said, wow, I can almost tell that's not real because I know you don't speak that, but gosh, that's pretty good.

What are we going to do when these things take over, like people unloading ships who make a great wage. People who do acting, and let's face it, screenwriters. These people are going to be taken out.

PS: It's a great question. At the at the trade shows I attend, especially at the Automate show, of course the answer is that we're taking the hard physical labor out of the equation, we’re keeping people safer and we’re driving greater speed of productivity. And the jobs will be different, they'll be a little bit higher skilled. Once you start innovating and increasing productivity, that horse is out of the gate, to use an old metaphor. The funny thing is, as an industry, we're experiencing this jobs crisis where you know what, maybe automation can help paper over some of these functions. I've been struck by how quickly AI modules have been built into CMMS systems, particularly what IFS is working on to streamline the FMECA process. Why run one per quarter when you can run 10 per week, and have them be pretty good and then refine those 10? That's the kind of physical grind-it-out labor, which at least I see AI doing in our industry, and you focus on some of that in your blogs too.

MH: Yeah, it's built exactly for that. It's like an Excel spreadsheet. Gone are the days of the notebook / ledger book type of thing when you can just use an Excel spreadsheet. In fact, if you think about this and nobody does this, my little friend Patrick was also my boss, the late Patrick Runchey, he was an Excel spreadsheet genius. This guy could do things on Excel that I've never seen people do. Excel is an incredible program, but nobody really uses it to that degree, right?

AI is exactly like that. There's certain artificial intelligence tools out there that are incredibly powerful, but the power remains on one very simple thing. It's all about organization and the speed of analysis. It takes a whole bunch of data points and completely crunches them through an algorithm, and then spits out something instantaneous to you. Whatever it should normally take you and me and whatever to balance the spreadsheet, these guys are doing it instantly, that's the power of computing, what it’s always been, ever from the abacus all the way through to quantum computing now – the ability to solve a problem for a whole bunch of different data points incredibly quick. That's the magic of it, and that's also going to be what's going to innovate because as soon as we start doing that, we'll never going to look back.

PS: To bring it back to the example you started with, which is I think it was Wal-Mart, the customer that had thousands of samples per month and decided to stop that program. That's their right of course but the nuance there was that someone on the human side of the side decided that this program was not returning because everything was baseline / green light. That's the kind of nuance that a reliability program manager is going to have to have with AI to understand, OK, what can it do for us, what are the limits? And are there prognostics that a might help us say, OK, if someone does want to cut this program off, what can AI tell us about the future? Instead of reporting just steady state month after month, can AI help us?

I was curious to know what are your thoughts on, is that the kind of nuance you want to see new reliability engineers train themselves up to? Is this the kind of thing where lubrication engineers would have these insights and would bring them to the table?

MH: That’s a great question, and here's what I think is going to happen, because history is a great indicator of the future. As we make things easier, we become more reliant on them, so if we rely on something like artificial intelligence to solve our problems for us, we're going to lose our problem solving skills. And if we lose our problem solving skills all of a sudden, we no longer are able to become innovative because innovation is nothing more than solving the next problem really. Without innovation, we become really stagnant.

Now, maybe it's good enough, but quite frankly, I think human beings have a long way to go before we become really decent creatures and masters of our earth. I feel what's going to happen is there's going to be much more reliance on this and they're going to lose sight of the analytical skill set, and they're not going to see the nuances.

Something else that I've noticed too: a lot of oil analysis labs, they have the diagnosticians, yet they don't really take full advantage of these folks. Some of these people are experts in what they do. However they tuck them away in a little office and they just have them look at spreadsheets, 300 or 400 spreadsheets a day – you can't really analyze more than 300 or 400 tests a day, I've never met anyone can do more than 500 really. These people do a lot more than they’re being utilized for; now, they're going to be completely eliminated, and all of a sudden there goes your skill set.

And then what you've done is now you rely on a databank that was built at about a 75 to 80% efficiency of understanding the problem, but there's a 20% gap of, “uh-oh, this is really not right.” Every so often I’ll play with ChatGPT, and I put a question in and the answer it gives me, I think, okay not bad, but it's about 75% right, and I don't know if that's good enough. It's great, and I talk to my kiddos about this too, because my daughter is still in school. I said, do you ever use this? She's like, oh yeah, we use it all the time, all students do. I said, wow, I can see people rely on it, and she said, but you’ve got to be careful with it. I said, yeah, I know.

It reduces a lot of time, however you really have to take it to the next step and still put in some work to make sure it's not fallacious. Then the problem is, what do you know that's right versus what do you know that's wrong. But then that's where the work comes in. It cuts your time down, but it doesn't eliminate it. And I'm thinking about that, “well, if that's the case, we're starting to rely on this to solve big problems, and it's only right 80% of the time?” Whoa, I don't like those odds. I mean, I might take a vaccine that's good 85% of the time, but if I'm relying on engine reliability of an aircraft, I need to have it 6 Sigma. I want it 99.999% right. I'll take a little bit of fluff, but come on, that's the big challenge.

PS: This is years ago. I met somebody at an ultrasound conference who was there because his employer, at the request of the finance and insurance team, wanted the technicians and the maintenance side to develop a secondary condition monitoring technology, to ensure system reliability. This happened to be a major sports broadcaster on the East Coast, with lots of electrical cables, and their primary condition monitoring tech was infrared, like you said before, to see if something’s hot. But they decided to add ultrasound and they did it for financial reasons to make sure the premiums wouldn't go up. That's what you're talking about, really, is in this age of AI, we've got to have ourselves conditioned to have humans who know what kind of context to add around the data they're seeing – whether it's a secondary condition monitoring technology for verification, whether it's a human insight to double check some of the numbers. That's the key, and it's going to be plant-specific, I guess, huh?

MH: Yeah, and it'll be asset-specific too. I mean, what you look at in the hydraulic system for a press in a plant is different than what you look like in a hydraulic system on say an excavator. But aren’t they both hydraulics? Not necessarily. It's like an automotive engine, a diesel engine for automotive, a haul truck engine, a marine engine, and a jet engine. They all run off of forms of diesel fuel really, jet fuel is nothing more than a type of diesel fuel. However, these things are very different in terms of what they experience in terms of contamination, stress levels, even the whole design of the asset, let alone the lubricant chemistry. So yeah, they might have some nuances but really it's truly asset specific.

And then it comes into case-specific, which once again if we can take the data of all the past and crunch it all down, we'll get closer to where we can actually start using artificial intelligence to make some pretty good assumptions. This is where the folks that have been spending years and years in the business understand when they take a look at a bunch of data. They can see trends implicitly and think, OK, here’s what's going on?

But there's something else that happens and I've seen this in labs with all oil analysis companies I worked for and with. I’ve worked with a lot of them, pretty much all of the big ones, and I worked for a couple of them. A diagnostician, when they see the data and it doesn't seem like it makes any sense to them, they don't assume that there's some crazy thing happening with the asset. The first assumption is there's something wrong with the test. Literally many of the good ones, they’ll go back to the lab and talk to the technicians say, “when we ran this, this is what we saw, what's going on here? I want to do a retest.”

And so we'll retest it. And if they get back the same value, then they realize that OK, you validated my question with a response that says, yeah, there's something going on with that asset; or, if they get the test result back and it's not the same as the initial one, it's OK you had a bad burn or a bad test, I'm glad you found it, now we've got it right, or we had a false negative, and now we’ve got it right.

AI won't be able to necessarily go back into the lab and ask the tech what's going on. We'd have to rely on our robustness of our test methods to say, “no, we’re spot on, we're accurate, we’re precise, repeatable, reproducible. We have this. Whatever data we're pushing out, you can rely on.” You can't do that in a lab. That doesn't happen! I can challenge anybody across the land because all I have to do is pull up your quality control stats and say, look, you've made errors and it's not your fault, it's just that you're asking instrumentation something that's not designed to do what it does. (Most instrumentation oil analysis labs were built out of the bioindustry to maybe test 10 samples a month of a blood- or aqueous-based solution. Really. I mean, ICP was never designed to test dirty oil at 1,000-2,000 samples a day. It just wasn't! And this thing in most ICPs they test part of elemental analysis down to a part per million? No they don’t, it’s a part per billion! We were just allowed to go to a part per million. These things go to part per billion, but we don't need that in oil analysis.

The thing is, we're asking so much of these sensitive pieces of machinery to do things in a very dirty, nasty environment, like dirty oil. Yeah, there's going be variables that are going to influence our results. We could count on these things much of the time, but sometimes we can't, and it requires us to go back and ask, hey, what happened here? Hey, I won't do that necessarily.

So, if we're going to do this AI thing, we're going to have to do it across the board and really hit back down to the very beginnings of it, which is, “I won't understand the instrumentation. I won’t understand robustness in my data and then when I completely am satisfied with that, I can then build my AI tables around the data I can count on, because otherwise it's not going to have a nuance to understand. It's going to have a built-in bias and we don't want to have that. A diagnostician with a bias is not a diagnostician. They're an idiot. You can't have a bias. It's wrong. And some of them go through confirmation bias, which is even worse. So once again we'll bring that philosophy stuff into it. But we have to remain and keep it with empirical evidence alone.

PS: We'll close with that prescription on how to use the human element to get our arms around AI. Mike, thank you so much for being with us on the podcast today. And for everyone listening, I've got links to Mike's blog series in the podcast notes and transcript, and also to his previous article for Plant Services on certifications. Thanks again, Mike, talk soon.

MH: I appreciate it. Thank you so much.

About the Podcast
Great Question: A Manufacturing Podcast offers news and information for the people who make, store and move things and those who manage and maintain the facilities where that work gets done. Manufacturers from chemical producers to automakers to machine shops can listen for critical insights into the technologies, economic conditions and best practices that can influence how to best run facilities to reach operational excellence.