Latest / The Manufacturing Podcast with Fexingo: Factories, Supply Chains, and Industrial Business / How a Belgian Factory Uses Digital Twins to Predict Equipment Failure
Transcript
- Lucas: So we're standing in front of a digital twin today—and I don't mean the kind you see in a video game. I mean a real factory in Antwerp, Belgium, that has built a full virtual replica of its production line. Every motor, every conveyor belt, every valve—all mirrored in software, updated in real time by about two thousand sensors. Luna: Two thousand sensors—that's a lot of data. What are they actually tracking? Lucas: Temperature, vibration, pressure, current draw—the usual suspects for predictive maintenance. But what's interesting is how they use it. The digital twin doesn't just show you a 3D model of the factory floor. It runs simulations. It compares current readings to historical patterns. And when something's off, it flags a potential failure days before it happens. Luna: Days, not hours? That's a big claim. Lucas: It is. And the numbers back it up. The plant manager told me that since they fully deployed the twin eighteen months ago, unplanned downtime has dropped by 40 percent. They've caught bearing failures, seal leaks, even a misaligned drive shaft—all before any human operator noticed anything wrong. Luna: Before we go deeper, I want to pause on something. This show is ad-free, and we like it that way. If these conversations have moved your work forward in some small way, listener support is what keeps that possible. The link is buy me a coffee dot com slash fexingo. Lucas: Yeah, it's a simple thing. No ads, no sponsors, just us talking about what actually works in manufacturing. Thanks to anyone who throws in a coffee. Luna: Alright, back to the twin. How did they decide where to start? A whole factory is huge. Lucas: They started small. The factory makes bottled specialty chemicals—things like industrial cleaners and solvents. The most failure-prone part of the line was the bottling station. That's where they put the first sensors. They built the digital twin for just that station using Siemens' MindSphere platform. Luna: MindSphere—that's the cloud-based IoT operating system, right? Lucas: Exactly. They connected vibration sensors to the filler heads and cappers, streamed data into the cloud, and trained a machine-learning model on three years of historical failure records. The model learned what a healthy vibration signature looks like versus a failing bearing. After about six months, it was flagging anomalies with 92 percent accuracy. Luna: 92 percent—but what about false positives? If the system cries wolf too often, operators start ignoring it. Lucas: That was exactly the concern. The plant manager, a guy named Pieter, told me they ran a three-month shadow period. The twin would issue a prediction, but they wouldn't act on it—they'd just note it and compare with what actually broke. That built trust. By the end of the shadow period, the twin had flagged two imminent bearing failures that traditional weekly inspections missed entirely. Luna: So the operators saw proof. That's huge for adoption. Lucas: Right. Once the operators saw that the twin caught things their eyes and ears couldn't, they started to rely on it. Now the maintenance team checks the twin's dashboard every morning. It tells them which machines have a high probability of failure in the next 72 hours, and they schedule repairs during planned downtime instead of emergency shutdowns. Luna: What about the cost? A project like that isn't cheap. Lucas: Pieter was upfront: the whole rollout—sensors, cloud subscription, integration, training—came to about $1.5 million. But they calculate annual savings of $600,000 from avoided downtime and reduced maintenance labor. So payback in two and a half years. And that's not counting the softer benefits like longer equipment life and safer working conditions. Luna: Two and a half years is pretty solid for a capital project. Were there any unexpected hurdles? Lucas: The biggest one was data quality. Some of the older machines didn't have the right ports for sensors, so they had to retrofit them. And the network bandwidth on the factory floor wasn't great—they had to upgrade to industrial-grade Wi-Fi. That added about $100,000 and delayed the timeline by a couple of months. Luna: So the lesson is: don't underestimate the infrastructure underneath the digital twin. Lucas: Exactly. The twin is only as good as the data feeding it. But once they sorted that out, the results speak for themselves. They've now expanded the twin to cover the entire filling and packaging area. Next up is the chemical mixing section, which is trickier because the processes are batch-based, not continuous. Luna: Batch processes are harder to model because the conditions change from run to run. Are they using the same approach? Lucas: They're adapting it. Instead of a single model, they're building multiple models—one for each product recipe. The twin switches between models based on which batch is running. It's more complex, but they think it'll work. They're piloting it now. Luna: I like that they're not trying to boil the ocean. They started with one station, proved the ROI, then scaled. Lucas: That's the playbook for any factory considering digital twins. Start with your most painful bottleneck. Get a quick win. Use that data to justify the next phase. Pieter said his board was skeptical until they saw the downtime numbers from the first year. Luna: What's the one metric they track to measure success? Lucas: Overall equipment effectiveness, or OEE. It's the gold standard in manufacturing. Their OEE went from 72 percent to 85 percent in eighteen months. That's a thirteen-point gain, almost entirely from reduced unplanned downtime. Luna: Thirteen points is massive. Most factories see single-digit improvements from lean initiatives. Lucas: Right. And here's the thing: digital twins aren't just for big multinationals. The technology is getting cheaper. You can get a basic setup with a few hundred sensors and a cloud subscription for under $200,000 now. The barrier to entry is lower than most people think. Luna: But you still need the in-house skills to interpret the data. Not every factory has a data scientist on staff. Lucas: That's true. The Antwerp factory partnered with a local technical university to train two of their senior technicians in data analytics. It took about six months of part-time study. Now those technicians manage the twin day-to-day. They don't need a PhD—they just need to understand what the model is telling them. Luna: So the skills barrier is real but solvable. I like that they invested in their own people rather than outsourcing. Lucas: Yeah, Pieter was clear: he didn't want to be dependent on an external consultant every time something needed tweaking. He wanted ownership inside the plant. That cultural shift—from reactive maintenance to predictive maintenance—is probably the hardest part. Luna: Speaking of culture, how did the older, more experienced operators react? The ones who've been listening to machines for twenty years? Lucas: Mixed at first. Some saw it as a threat to their expertise. But Pieter involved them in the validation phase. He asked them to compare their gut feelings with the twin's predictions. After a few months, most of them came around. One veteran operator told me, 'I used to rely on my ears. Now I rely on the data, and I can go home knowing the machine won't break at 2 a.m.' Luna: That's a great quote. It shows the twin didn't replace their judgment—it augmented it. Lucas: Exactly. And that's the mindset that makes these projects succeed. The technology is just a tool. The real change is in how people work. Luna: So, looking ahead, do you see digital twins becoming standard in manufacturing within the next five years? Lucas: I think for high-value, complex production lines, yes. The cost is coming down, the cloud platforms are getting more user-friendly, and the ROI is proven. But for simpler operations, a good preventive maintenance schedule might still be enough. It's not a one-size-fits-all solution. Luna: Fair point. What's the one takeaway you'd want a factory manager to remember from this episode? Lucas: Start small, prove value, scale. The Antwerp factory didn't try to digitize everything overnight. They focused on one bottleneck, delivered a 40 percent reduction in downtime, and built momentum from there. That's a blueprint anyone can follow. Luna: And it's a blueprint that's already working. Thanks for walking us through it, Lucas.