Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How a Steel Mill Cut Unplanned Downtime by 80 Percent
Transcript
- Lucas: Luna, when you hear the phrase 'predictive maintenance,' what comes to mind? Luna: Honestly? A lot of sensors, a lot of data dashboards, and usually a pilot project that never scales. Lucas: Right. That's the stereotype, and it's not unfair. But there's a steel mill in Indiana that actually made it work — they cut unplanned downtime by 80 percent over 18 months. Luna: Eighty percent? That's not incremental. That's a transformation. What's the name of the mill? Lucas: It's a midsized operation — about 400 employees — owned by a private group. They produce specialty alloy bars for automotive and aerospace. Before the project, they were losing about 12 million dollars a year in lost production from unexpected breakdowns. Luna: And after? Lucas: Under 2.5 million. The interesting part is how they got there. They didn't just buy a fancy software platform and hope for the best. They built a very specific framework around three sensing technologies: vibration analysis, thermal imaging, and oil debris analysis. Luna: Those are standard tools in the reliability world. What made this implementation different? Lucas: The key was the decision protocol they layered on top. Most plants collect data, but the data sits in a specialist's report that comes out weekly. By then, the bearing has already failed. This mill set up a traffic-light dashboard that every operator saw in real time. Green means everything's normal. Yellow means there's a trend but no immediate action — just monitor. Red means stop the machine and call the maintenance team within the hour. Luna: So they pushed the interpretation down to the floor. That's a cultural shift. Lucas: Huge cultural shift. The veteran operators — some with 25 years on the job — initially hated it. They'd say, 'I can tell a bad bearing by the sound. I don't need a computer.' And to be fair, sometimes they were right. But the data caught things the human ear couldn't. For example, a gradual rise in vibration on a cooling fan that was spinning at 1,800 RPM — barely audible change, but the accelerometer picked it up 10 days before failure. Luna: Ten days of lead time is enormous. What did they do with that window? Lucas: They scheduled the replacement during a planned shift change, not during a rush order. The maintenance team had time to order the exact replacement part, pull the right tools, and do the swap in under two hours. Compare that to the old way: the fan seizes at 2 AM on a Saturday, the whole line stops for 18 hours while they scramble for a replacement. Luna: And the cost difference between a planned two-hour fix and an 18-hour emergency repair is enormous. Lucas: Exactly. The mill calculated that every hour of unplanned downtime cost them roughly $14,000 in lost output. So even preventing one major breakdown per quarter paid for the entire sensor installation within the first year. Luna: Let's talk about the failure modes they caught. Vibration analysis is great for rotating equipment. Thermal imaging catches electrical hot spots. Oil debris tells you about internal wear. Did they focus on one type of asset? Lucas: They started with the critical path assets — the rolling mill stands and the cooling systems. About 30 machines. Within six months, they expanded to 120 assets across the plant. The biggest wins came from the rolling mill stands themselves. Those are massive machines that shape red-hot steel billets. If one stand goes down, the whole line stops. Luna: What was the most surprising failure they caught? Lucas: An electrical cabinet that powers a main drive motor. Thermal imaging showed a hot spot on a contactor that was 15 degrees above normal. The operator flagged it red, maintenance opened the cabinet, and found that a bolt had loosened, causing resistance and heat buildup. If that contactor had failed, the motor would have stopped under load, potentially damaging the gearbox. They fixed it in 20 minutes during a lunch break. Luna: That's the kind of catch that never makes it into a ROI spreadsheet but saves you from a cascading failure. Lucas: Right. And that's the thing about predictive maintenance done well — it prevents the secondary damage. A bearing seizes, it can take out the shaft, the coupling, even the motor. By catching it early, you save not just the bearing but everything downstream. Luna: What about the human side? You mentioned the operators were skeptical. How did the mill get buy-in? Lucas: They ran a pilot on two machines for three months. They let the operators see the sensor data alongside their own observations. When the system flagged a yellow on a pump that the operator thought was fine, they'd check together. After a few times where the data was right and the human was wrong, trust started to build. They also made the traffic-light dashboard very simple — big green, yellow, red circles on a screen mounted at each workstation. No complex charts. Luna: Simplicity at the point of use. That's a lesson that applies way beyond steel mills. Lucas: Absolutely. And they tied it to a reward system. Each shift had a downtime target. If they hit it, the whole crew got a bonus. That aligned everyone's incentives toward catching problems early rather than just running through the pain. Luna: So the operators had ownership. They weren't just data entry clerks. Lucas: Exactly. And the maintenance team's role shifted too. Instead of being firefighters, they became planners. They spent more time prepping for scheduled replacements and less time scrambling. The mill reported a 40 percent reduction in overtime for maintenance staff. Luna: That's a quality of life improvement. I've heard maintenance crews burn out fast in reactive environments. Lucas: Yeah, the turnover in that department dropped significantly. People stayed because the work became more predictable and less chaotic. Luna: You know, speaking of keeping things running smoothly — and this is a bit of a sidestep — but shows like this one rely on a similar kind of support to keep going. If these conversations have moved your work forward in some small way, a couple of dollars a month is genuinely what keeps them ad-free and independent. Lucas: It's true. We don't run sponsor spots or track you for ads. That's only possible because listeners who find value in the show chip in at buy me a coffee dot com slash fexingo. It's a small thing that makes a real difference. Luna: Yeah. And we really appreciate it. Now, back to the mill — after the first year, did they hit any plateaus? Lucas: They did. The low-hanging fruit was easy. Once you've caught the failing bearings and loose connections, the remaining failures are harder to predict. They had to add oil analysis for gearboxes and start tracking mean time between failure more granularly to set better thresholds. Luna: So the first 50 percent reduction came quickly, and the next 30 percent required more sophistication. Lucas: Exactly. The last 20 percent is the hardest. They're now looking at using machine learning to combine vibration and temperature data to detect patterns that no single sensor would catch. But the foundation is already solid. They went from reactive to mostly predictive in 18 months. Luna: What would you say is the one takeaway for an operations manager listening who wants to try something similar? Lucas: Start small, but tie the data to a decision rule that the operator owns. The technology is the easy part. The hard part is changing the workflow so that the data actually triggers action. If you just install sensors and send reports to a reliability engineer, you'll get a shelf full of reports. But if you put a red light on the machine and say 'stop when this lights up,' you'll get results. Luna: And the traffic-light dashboard is a great example of making that decision rule visible to everyone. Lucas: Right. It's not about the fanciest AI. It's about closing the loop between data and action. This mill proved that with basic sensors and a clear protocol, you can cut downtime by 80 percent. And they're not a tech company — they're a 400-person steel mill in Indiana. Luna: That's the kind of story that makes you rethink what's possible in a traditional industry. Lucas: Yeah. It's a reminder that operations improvements often come from process design, not just technology. And when you get both right, the numbers speak for themselves.