From Planned Shutdowns to Predictive Maintenance: Highland Pellets' AI Journey - with Andrew Rehm
Show notes
Welcome to the Trend Detection podcast, brought to you by Senseye Predictive Maintenance – which gives you visibility and insights into all your assets, from single machines to full plants to help you reduce downtime, increase knowledge sharing and accelerate digital transformation across your organization.
In this episode we're joined by Andrew Rehm, Director of Reliability and Planning at Highland Pellets, to explore how predictive maintenance is transforming industrial operations.
Andrew shares Highland Pellets' journey from traditional inspection-based maintenance to a more proactive, data-driven approach powered by AI. We discuss how the company increased plant uptime from 60% to 89%, uncovered critical issues before they became failures, and built trust in predictive maintenance across operations, maintenance, and reliability teams.
The conversation goes beyond technology, covering change management, workforce adoption, maintenance planning, and why predictive maintenance is becoming a core part of Highland Pellets' long-term strategy.
In this episode you will learn:
Why predictive maintenance looks very different today than it did ten years ago
How Highland Pellets identified the right assets to monitor first
The story behind a critical failure that was detected before it shut down production
Moving from scheduled inspections to condition-based maintenance
Building trust in AI-driven insights among maintenance teams
Connecting predictive maintenance with CMMS workflows and planning processes
Lessons learned from the first deployment and plans to scale further
Why the future of maintenance is data-driven, proactive, and predictive
You can find out more about how Senseye Predictive Maintenance can reduce unplanned downtime and contribute towards improved sustainability within your manufacturing plants, by visiting: www.siemens.com/senseye-predictive-maintenance
Transcript
We started doing downtime maintenance, scheduling the outages, stuff like that, and where we're at now. is really of how do we manage our uptime with predictive maintenance instead of planned inspection. Welcome to the trend detection podcast powered by Siemens. Hello and welcome to the latest episode of the Trend detection podcast. Uh, I'm really pleased to be joined once again by one of our fantastic Siemens customers, so Highland Peellets and Andrew from Highland Pellets today, and we're gonna talk about. What's really a fantastic success case, we've, we've had a pellet mill success case, because it's saved over 60 hours of unplanned downtime thanks to the monitoring of Sensai and obviously the work of, of the team at Highland Pellets as well, so we really want to dig into that, what happened and how they're using sort of Sensai day to day as well, but I guess before we get started, Andrew, it'd be great just to introduce yourself and a little bit about Highland pellets as well would be great. Thank you. Yeah. Thank you. Thank you for having me on here. We're really excited about it. My name's Andrew Ream. Um, I'm the director of reliability and planning for Highland Pellets. I've been with the company right at 2 years now, and I come from the oil and gas industry, which we started working on predictive maintenance about 10 years ago, so. We make woody biomass pellets for the energy market here in Arkansas and ship them overseas for coal replacement actually. So it's quite an interesting ecosystem that we work in. No, it does, and I guess it's a little, little bit off script straight away, but because you mentioned you're doing predictive maintenance 10 years ago, just to get your thoughts on what's changed in that time, and I guess there's one. Particular technology that might influence your answer, but yeah, just to get your thoughts on that from your experience. Yeah, I, I will say, so the first predictive maintenance program that I worked on was uh, on an offshore platform. Obviously, having personnel on site on an offshore platform in the middle of the Gulf, it's very expensive and has its own hazards associated with it. So part of my group. I was part of a continuous improvement group there. We started working on a project that used just the raw data signals from all of the temperature, pressure, everything that we can read there in telemetry, and set up a more progressive alarming system that would generate a work order. The biggest advancement From then till now, obviously, is AI. We did everything manually, so we went to every control point and wrote in code to say, if this pressure gets above this, and the vibration is at this level, then we want to know about it. And we did that with every single thing we could out on the plant. Which meant going through hundreds of work orders and looking at telemetry data at the point of failure and then backing up from there. That project took around 10 people, about a year and a half before we got our first alert. And it did eliminate some maintenance time out on the, on the platform. I believe they dropped 5 technicians that had to be on site, but obviously, AI has made that a huge. Difference along with the Sensi technology where the implementation for this project took around 3 months and it was pretty much myself and another technician that worked on it and we got results pretty much right away and a lot more results than we did by manually going through there because it picks up on things that you could never imagine when you're looking through. Work order history, you can imagine, for years, so. No, fantastic. Yeah, and, and of course we're gonna dive into, let's say, sen sizes impact at Island Pets, but maybe just to open it up to the audience a little bit, so, maybe you could start by explaining the role equipment reliability plays in your particular environment, so in the pellet mill and helping it operate most effectively. Yeah, it's, it's, it's a unique industry because. We have really dry fiber all throughout and conveyance is our biggest issue, right? Getting it from point A to point B, breaking it down, resizing all the way throughout. And it's very dependent upon each point in the process. So we don't have a lot of redundancy built into the system. It's huge volumes. We're, we're on track to do 600,000 tons this year. So, Reliability was one thing when I first started with the company that we focused on a lot. We started doing downtime maintenance, scheduling outages, stuff like that. And where we're at now is really of how do we manage our uptime with predictive maintenance instead of planned inspection maintenance. So that's really where our focus is right now. An hour of downtime for us is around $67,000 that we lose. So, it's, it's huge difference. So one outage is roughly $2 million that we lose in, in production, but we gain that back obviously in up time. But it's something that, that's that last 10% that's really hard to get. We started out at 60%. Up time, when I started, we're at 89 right now. And it's really using this new sensors that we just added on, hopefully, to get to that 95% of time at the plant. Yeah, I mean, it's amazing, yeah, you actually took the, the question right out of my mouth, cos I was gonna ask. No, no, no, no, no need to apologize at all. It's just what I was thinking about the impact of production stoppages and, and that, and you've illustrated that perfectly, that we're talking about 60 hours of downtime, so I think people can do the maths on this particular success case, right? Yeah, it's that particular case, we have 20 presses, so. We dry the fiber, obviously chip it up from round logs and break it down, run it through a dryer, and then run it through another hammer mill and into these pellet presses. We have 20 of them, so it's 60 hours on that unit. It wasn't plant-wide in that particular case. But we have had a couple of those events as well that we've been able to detect. But the main thing on that is just these parts are, are hard to get sometimes on an emergency basis. So we would have lost it for probably longer than that if we didn't have the parts available to do the replacement. Yeah, it's good fun. No, no, exactly, no, definitely, and I guess, just, just stepping back again, because you, you've illustrated perfectly that the significant savings have been made by spotting these issues early, but it all makes me think, what were you doing before you had Sensa deployed? How were maintenance decisions made? You mentioned, obviously some planned maintenance practices, but. Yeah, clearly that was in, and you mentioned about obviously trying to figure a. You're, you're already making effective decisions, but it's about those gains you can make on top of that, I think you were saying earlier. So, yeah, so, yeah, what were you doing before Sensai? I guess simply. Yeah, so, and, and, and we're changing that schedule this year, but basically we would bring the plant down and visually inspect every piece of equipment 3 times a year. That took around 3 days each time we had to go down to do that, to start up and everything as well. But yeah, our, our maintenance inspections were our only early detector of failures. And we have around, oh, I wish I had that number in front of me. I, I think it was 8% return. So while they were out doing inspections, 8% of the time, they found something that needed to be repaired, which is not huge. So really, it's Trying to get that up to a percentage where it makes more sense to bring the equipment down. So we adjusted our time slots, how we did inspections, all of the things that you try to do to drive that up. We got up to around 20% of accuracy, which is still, that means 80% of the equipment that we brought down did not need to be brought down, but it's without having the instrumentation in the live. Feedback from it, you just, that's the only option you have to prevent a catastrophic failure that causes more downtime. So like I said, that's really my focus now is how do we utilize this data and start pulling away from some of those planned outages and start reducing the number of equipment that we look at during them. So, This year, I plan on bringing the whole plant down 3 times, but 1 day apiece. So that in itself is, is quite a large change for the type of equipment that we're looking at. Some things obviously, you just have to look inside and do structural inspections and stuff like that. But everything that we can, we're putting sensors on and we're, we're loading those into Sensi and using that to really dictate when we need to bring that equipment down. Yeah, that makes it. And, and I was wondering as well, because you mentioned certain equipment requires monitoring, maybe versus others. So, it leads me to wonder, during the initial deployment of Sens I, like, how did you decide which areas of production to focus on first? Of course, it was in collaboration with the Siemens team as well, but just to, just to understand that decision process as well. Yeah, so, what we looked at first was our highest criticality. We do an annual criticality assessment on each asset. And we looked at what impacts it has safety, financial, and production. And then also the frequency of failures. So we take those into consideration and basically rank every piece of equipment by criticality. We took the highest criticality equipment, and then I went through it with our maintenance manager that's on site and looked at What sensors do we currently have, and what sensors could we put in place where we actually felt that we can detect a failure on there. So bearings were our first major implementation. A lot of high speed fans, low speed bearings on our feeder screws, stuff like that. And then we went into Bass plate vibration on our hammer mills, so high speed hammer mills, green and dry, and that, that was our first round. And then we looked at the decline in failure rates after that, and decided to go with the next group, which was the pellet presses themselves. We had some instrumentation on there, but everything was set up basically as an alarm basis. So trying to get more of the early identification of a failure before it fails. Historically, it just shut the equipment down, so it didn't damage anything else, and we would go fix it then. One good example of that that we found during the first round, and it was within a couple of weeks of starting the project, was a large bearing on our RTO fan that would have caused the whole plant to go down. And it's very expensive bearing, as you can imagine. It's a, I think it's a sixty-inch fan. It's quite large, and it moves the air through the dryers to allow for the drying to occur, obviously. But that picked up a, we had put sensors on it and picked up a small vibration that changed from the inboard to the outboard. So typically, your inboard bearing is gonna have a little bit more vibration. Then you're outboard just because it's the driven bearing. And when that vibration switched from the outboard being the highest to the inboard being the highest, we got an alarm from that, from SensIi, and we sent a technician out there. And it was one of those things that they had done maintenance on the oiler system for that bearing and trash had gotten in there and just stopped it up. So it just wasn't lubricating the bearing. So we were able to fix that, but after looking at it, we determined that that bearing would have failed within the next like 48 hours if we hadn't have found that greasing issue. So it's, that was a huge win right off the bat. And then we've had several more like that throughout the plant. One thing, uh, sorry, I keep just rambling into the next, but no, no, it's good, it's good. One thing that we found early on was that it wasn't just maintenance issues that we found. We have feeder screws that feed from a hopper into our hammer mills, and then they feed out from there. And we kept getting alerts. For RPM on those feeder screws, and they are adjusted by the driven equipment. So if the mill itself isn't getting enough feed, then it speeds up the screws and then slows them down, vice versa, right? And what we found was a mismatch. So the mill was calling for more feed, the screws were speeding up. But they weren't stopping. They were maxing out, and the mill was still calling for more feed. And at first we didn't know exactly what was going on, so it took us a, I think we went out and looked at it 3 times, thinking that it was a screw issue. So we took the bin apart, looked at the screw, nothing was there. And what we ended up figuring out out of that was, it was actually the screens on the bottom of the hammer mill. We had a hole in, a piece of concrete that went through there and blown a hole out of that screen. And it was, I, I guess it's maintenance fixed it, but it was really something that should have been picked up by our PSD, our particle size coming out of that hammer mill. So we set an alarm for that and we've actually been able to find about 12 of those. So, We find them before our quality does, just by monitoring that alarm and setting up that condition to say, hey, if this is still calling for feed and it's at max RP create a notification in our system for us to go out and look at it. So that was one interesting thing that just happened that we didn't expect as well. And we've taken that now into our pellet presses where we're Our next phase of this will be adding sensors to the rolls itself. So you have two rolls in there and a die that presses the wood through where we're looking at roll adjustments. We've done some small scale testing on that, and you can pick up the vibration of that roll losing contact with the die. And that's our next phase is right now we have people that go out there and shut down the presses, open them up, look at them every day, and we're trying to use this instrumentation to, to do away with that inspection and just use the vibration sensors on it. So sorry, I went on a bit of a tangent. No, no, no, no, no, if you, I always say, if you're talking more than me, that's always, that's always a good sign. So, no, it's pretty, but I think what, what's really interesting though, and, and to me, just hearing you talk, Andrew, is how, cos a lot of the time, the adoption of tools like Senites, how maintenance teams trust the, the. The insights it's providing, right, and actually taking action on it, but it seems like on your side, that, that seemed to happen quite quickly in terms of trusting the system, right, and what it's, what it's advising. So, interested to get your thoughts on that. Yeah, yeah, obviously we had some pushback at, at the beginning, is it gonna take my job away or is it gonna, gonna be accurate, are we just gonna be chasing our tails, all of those kind of comments. And, and understandably so until it's proven. But once we had a couple of major finds throughout this, the guys became a lot more trusting in it. And I honestly, I, I became better at reading the information from SensI. So knowing what was a false alarm before we sent somebody out there helped quite a bit as well. And then now we have our on-site planners are going through that data. And we just did an integration with our CMMS system where since I will create a case and it creates a work request in our CMMS so the planners can look at it directly in there and issue it out from the CMMS system into a work order. So it gets them more involved in it. I think we're right now we're at 63.5% accuracy on cases. We were a little higher than that. We were running around 80%, and then we added an additional 100 assets. So we're back in that training period for them right now. But overall we're at that 63% accuracy, which is fair to do on this scale. We have a right at 200 assets that we're measuring right now are monitoring out of our 700 total. So, and yeah, I guess another, another, another point that, that picked up. Up On is how between all the various teams, reliability, operations, maintenance, there seems to be a lot really close collaboration, but there, there must have been before, of course, but this, how has it changed the way you, you all work together? Because it's not, as we always say, it's not just a technology deployment, it's a, a change within the organization as well, right? Yeah, for sure. We've, we've been able to use the data collected to Encourage and give access to everyone that wants it or that needs it in the, in the plant to really understand that we need to be more of a data-driven mindset on maintenance, it's especially and on operations. We had, we still have a lot of really good employees that have been there for a long time, so they know they have the knowledge inside of them. So we encourage them to be part of this implementation and really help. Identify issues. So if we got an alert from SEI, we encourage those well-respected long-term employees to go out and do the analysis in the field and correlate it to those measurements that we got back from SensI. And that really brought on a lot more buy-in from those technicians. Because they feel like they're not just being replaced by AI, it's a tool for them to be better at their job as well. And then the benefits of it, right? So any downtime avoidance is, is better for the company. It's better for the employee, and you started a bonus program to help with that as well. So they're just as invested as, as I am in making sure that this works. And you mentioned that they beyond, we've talked about avoiding downtime and of course, this success case that brought us together is a significant downtime saving, but in terms of what other, what big difference Sensai has made in terms of planning, how, how you allocate resources, and I think you've touched upon some of it already, Andrew, in terms of when to go and investigate something versus not and these kind of things. But yeah, so beyond downtime, is there. Other real standout benefits of utilizing Sensai. Yeah, so our, which obviously downtime, but breaking work, as you're aware, obviously is just supremely inefficient. It's, we had a huge amount of break-in work when we first started with this, and what it's allowed us to do overall on the, on the 200 assets that we Monitor is, it typically gives us about 2.5 to 3 weeks to plan, get materials in, and, and move that into planned predictive work. And I'm looking at some numbers here, and obviously, this is just on a couple of pieces of equipment, but our break-in rate was, for one year was around $135,000 on one piece of equipment. And we still have those failures. Obviously, that doesn't eliminate a, a failure, but what it did was moved all of those failures, but I think 32,000 of break-in is all we had after we implemented the system. So it just gives us that efficiency on the repair side. And a knowledge going into the problem where it's not just the equipment shut down. Let's go spend an hour figuring out what it is. When we get an alert early on, we know what the issue is. Typically, it'll identify the very bearing that's going out, and it really just makes my planner's job easier. They can get all everything together before they send the first technician out to go work on that equipment. So everybody's a little happy with it. No, well, that's, that's what, that's what we'd like to hear. I was, I was also gonna comment on the, the fact, that was an interesting point you made about investigating something and, and fixing it and stuff, but what happens to the knowledge afterwards? When you, you fix something, you want to retain that knowledge for next time, right? When there's similar cases and things like that. Has it, has it helped from that perspective as well? Yeah, capturing the knowledge of that particular case and closing it, and then you can use that again for future similar cases to obviously speed up. Maintaining in the future. …