Latest / The Operator's Manual with Fexingo: Daily Decisions, Tactical Wins, and Business Execution / How One Steel Mill Cut Scrap Waste 60 Percent With One Camera
Transcript
- Lucas: So there's a moment in a steel mill where a slab of steel is screaming hot, moving at about four meters a minute, and you have about three seconds to decide if it's good or if it's scrap. One minimill figured out how to get that decision right more often with a single camera. Luna: A camera. Not a whole new line, not a new furnace. Lucas: Exactly. And that's what we're digging into today—how one camera, placed just right, cut scrap output by sixty percent. The mill in question is a mid-sized minimill in the US Midwest. They were running around a six percent scrap rate, which may not sound terrible, but in steel, scrap is basically pure loss. You've already paid for the energy, the labor, the raw materials. Luna: Six percent sounds small, but I'm guessing the dollar figure is huge. Lucas: Huge. A single slab can be worth tens of thousands of dollars. So six percent scrap on their volume meant millions a year in lost margin. The traditional inspection method was a guy with a flashlight and a lot of experience, looking at the slab's surface as it passed by. But the slab is glowing, it's moving fast, and the defects are often subtle—hairline cracks, surface tears, inclusions that are barely visible. Luna: So the human eye has limits. But why a camera, and why one? Lucas: The key insight was that they didn't need to catch everything. They needed to catch the most common defect pattern early. So they studied their scrap data for a quarter and found that over half their rejects came from one specific type of defect—a transverse corner crack that happened during the cast. It was predictable, tied to a specific temperature gradient in the mold. Luna: So they knew the what and the when, but not the where. Lucas: Right. And that's where the camera came in. They mounted a single high-resolution thermal camera aimed at the corner of the slab right after it exits the mold. The camera runs at thirty frames per second, and it's pointed at a very specific spot where the crack typically first appears. They connected it to a small industrial PC running a simple image analysis script—not AI, just a threshold algorithm that looks for a sudden drop in temperature along a line of pixels. Luna: A temperature drop equals a crack. That's elegant. Lucas: It took their engineers about two weeks to tune the threshold. Once they had a false-positive rate under five percent, they hooked it up to a color-coded alert on a screen in the control room. Green means fine, yellow means a possible crack forming, red means definitively cracked. The operators had been trained to trust the system, and crucially, they were empowered to act on it—they could slow the line or even stop it if they saw red. Luna: But that's a big cultural change. Operators are usually told to keep the line moving at all costs. Lucas: That was the hardest part. The plant manager had to give them explicit permission to stop the line without a supervisor's approval. And they had to run a pilot where they compared the camera's calls to the old visual inspection. After two months, the camera was catching about eighty percent of the cracks that had been going to scrap, and the operators were making the right call ninety percent of the time. That built trust. Luna: So what did that do to the scrap rate? Lucas: It dropped from six percent to two point four percent in about five months. That's a sixty percent reduction. And here's the kicker—the whole system cost about forty thousand dollars. The camera, the PC, the cabling, the screen, and the integration. They got payback in under two months. Luna: Two months. That's a no-brainer. But I wonder, why didn't they just buy a commercial off-the-shelf inspection system? Those exist. Lucas: They looked at them. The big vendors wanted two hundred fifty thousand dollars or more for a full surface inspection system. And they were overkill—they'd catch all sorts of cosmetic defects that didn't matter for the customer. This mill makes rebar, not automotive sheet metal. Their customers care about mechanical properties, not a little surface roughness. Luna: So they bought the minimum viable sensor. That's a lesson in itself. Lucas: It is. And it's a pattern we've seen in other industries—start with the biggest pain point, find the simplest sensor that gives you a clear signal, and give the operators the authority to act on it. The camera didn't replace the human inspector; it gave the inspector a second pair of eyes that never blinks. Luna: That's a great framework. But let's talk about the numbers a bit more. They cut scrap by sixty percent, but what does that mean in terms of energy savings? Because steel is so energy intensive. Lucas: Every ton of scrap they avoided saved about one point one tons of carbon dioxide emissions, because they didn't have to re-melt and re-roll that material. Over a year, that's roughly eight thousand tons of avoided emissions. And in terms of energy, they saved about two thousand megawatt-hours. That's equivalent to the annual electricity use of around two hundred homes. Luna: And that's not just good for the planet—that's money saved on the energy bill. Lucas: Absolutely. The energy savings alone probably covered the cost of the system in a few months, not even counting the labor and logistics savings from not handling scrap. And there's a downstream effect—less scrap means less waste to transport, less re-processing, less everything. Luna: So where does this pattern apply beyond steel? I can think of a few industries. Lucas: The most obvious is food processing. Think about a potato chip line—a single camera can spot discoloration or foreign objects. Or automotive parts, where a crack in a cast component can be caught with a thermal camera just like this. The key is to find the one defect that causes the most scrap, and then put a cheap sensor at the point where that defect is most visible. Luna: You mentioned the operators were empowered to stop the line. How do you get that kind of buy-in? Because I feel like a lot of companies fail not because the tech doesn't work, but because the people don't use it. Lucas: That's the real challenge. The plant manager here did something smart—he didn't just send a memo. He held a series of small-group meetings with the operators, showed them the pilot data, and asked for their feedback. They tweaked the threshold together, and he made it clear that stopping the line for a suspected crack was never going to be penalized. He also made one operator the 'champion' of the system, someone who had been skeptical at first but became the go-to expert. Luna: So it’s about respect and ownership, not just a new tool. Lucas: Exactly. And the results speak for themselves. Now, other plants in the same company are rolling out similar systems, but they're adapting the camera placement to their own line layouts and their own defect patterns. It's not a one-size-fits-all tech—it's a one-size-fits-all principle. Luna: I love that. It reminds me of the old saying, 'Find the bottleneck, fix it with the smallest possible change.' Lucas: Which is essentially the philosophy of this show. And it's why we keep coming back to these stories—because they're replicable, and they're often cheaper than people think. Luna: Before we wrap, I want to get a little meta. We've been doing this show for a while now, and I think it's worth mentioning that we're able to keep this ad-free and independent thanks to a small group of listeners who chip in monthly through buy me a coffee dot com slash fexingo. Lucas: Yeah, it's a modest thing, but it genuinely covers the cost of production and the time we put into researching these cases. If you've gotten even one idea that saved you money or a headache, that's the kind of value we're trying to deliver. Luna: And we don't have a paywall or ads, which is the whole point. So if you're in a position to support, that link is there. If not, no pressure—just keep listening. Lucas: Right. So back to the steel mill—one last detail that I think is telling. After they installed the camera, they actually found that the defects were more likely to happen when the mold powder was slightly off-spec. So they ended up adjusting their purchasing specs for that powder, which further reduced the crack rate. The camera didn't just catch scrap—it taught them something about their process they didn't know. Luna: That's the hidden return on investment. The data becomes a diagnostic tool. Lucas: Exactly. The initial problem was scrap, but the camera opened a window into their own operation. That's the kind of win that compounds. Luna: So if you're running a factory, or any process with a visible defect, the takeaway is: find the one defect that costs you the most, put a cheap sensor where you can see it, and give your people the power to act. Lucas: Couldn't have said it better. That's our show for today. Thanks for listening.