Latest / The Robotics Podcast with Fexingo: Autonomous Systems, Industrial Robots, and Hardware / What Happens When a Robot Breaks the Factory
Transcript
- Lucas: You ever think about what happens when a robot breaks? Luna: I mean, I imagine it just stops. Some flashing lights, a service call. Lucas: Right. And that's basically correct in a small shop. But when you're talking about a line of fifty welding robots at a car plant, one breakdown can ripple through the whole system in minutes. Luna: How fast does it get expensive? Lucas: Fast. There's a well-known number in the industry — automotive assembly lines typically value downtime at somewhere between 20,000 and 100,000 dollars per minute. That's not hyperbole. That's lost production, idled workers, missed shipping deadlines. A single failed sensor can cost more than the robot itself. Luna: Wow. So what actually breaks? Is it the big mechanical arm, or the electronics? Lucas: More often than you'd think, it's the small stuff. Wires chafing through their insulation, connectors corroding, encoders drifting out of calibration. The robot itself is usually pretty robust — it's the peripherals that fail. And the really frustrating part is, many failures are predictable. You can see them coming if you're looking. Luna: So why aren't people looking? Lucas: That's the million-dollar question. Or in this case, the hundred thousand dollar per minute question. The answer is partly cultural. Manufacturing has this long tradition of run to failure maintenance — you don't fix something until it breaks. That works fine for a drill press, but not for a fully integrated robot cell. Luna: And partly economic, I'm guessing? Like, the maintenance team is understaffed? Lucas: Exactly. In many plants, the maintenance crew is lean, and they're busy reacting to breakdowns all day. They don't have time for preventive work. There's a concept called the maintenance maturity model, and most plants are stuck at the reactive stage. Luna: Let's talk about a concrete example. Last year, Ford had a big issue at their Kansas City plant — I remember reading about it. Lucas: Yeah, that was a doozy. In early 2023, Ford's Kansas City Assembly Plant had a major line stoppage because a robot controller failed. It wasn't the arm — it was the control cabinet. A power supply board died. The replacement part was on backorder. The line was down for something like three days. Luna: Three days. At a plant that builds the F-150. That's tens of thousands of trucks. Lucas: Exactly. Ford never released the exact cost, but analysts estimated it in the tens of millions. And the root cause? It wasn't a design flaw. It was a known issue — the cooling fan in that cabinet had been failing intermittently for months. They had work orders, but they kept getting deferred. Luna: So that's a textbook case of reactive maintenance catching up with you. Lucas: Right. And the irony is, the solution isn't even that expensive. You can instrument a robot cell for a few thousand dollars — vibration sensors on the motors, current sensors on the drives, temperature sensors in the cabinets. That data can feed into a simple dashboard that flags anomalies. Luna: But that requires someone to actually look at the dashboard, right? Lucas: That's the rub. The technology exists. It's called condition-based maintenance, or predictive maintenance. Big companies like Toyota have been doing it for years. Toyota's system is legendary — they monitor the health of every robot on the line in real time, and they schedule maintenance based on actual wear, not just calendar intervals. Luna: How do they pull that off? Isn't that a huge data management challenge? Lucas: It is. But Toyota standardised early. They use a common control platform across all their plants, and they developed their own monitoring software in-house. So a technician in Texas can see the same dashboard as one in Japan. When a sensor reading starts drifting, the system automatically generates a work order — no human has to notice it. Luna: That's the dream. But most plants aren't Toyota. They have a mix of robots from different vendors, different ages, different software. Lucas: That's the reality in most of manufacturing. And it's a real barrier. You can't easily retrofit a 10-year-old robot from one brand with the monitoring system from another. There's no universal plug and play standard. Luna: So what's a mid-size manufacturer supposed to do? Lucas: There are third-party solutions now. Companies like Augury and Uptake offer bolt-on sensors and cloud analytics. You attach a wireless vibration sensor to the robot's motor, and it sends data to their platform. The platform learns the normal vibration pattern and alerts you when it changes. It's not as deep as Toyota's system, but it's a huge step up from waiting for a crash. Luna: And the cost? Lucas: For a small plant with ten robots, you're looking at maybe 20 to 30 thousand dollars upfront for the sensors, plus a monthly subscription. Compare that to one hour of downtime. The payback period can be measured in weeks. Luna: So why isn't every plant doing this? Lucas: Partly awareness. Many plant managers still think of maintenance as an expense, not an investment. And partly it's the skills gap — you need someone who understands both the robots and the data. That's a rare combination. Luna: Let me push back on that a little. Isn't this also a cultural thing? I've visited plants where the maintenance team takes pride in being the heroes who fix things fast. Predictive maintenance almost threatens that identity. Lucas: That's a really good point. There's a tribal knowledge aspect — the veteran technician who can diagnose a problem by the sound of the motor. Predictive maintenance systems can feel like a replacement for that expertise, when really they should be a tool that amplifies it. Luna: Exactly. The best implementations I've seen treat the system as a decision-support tool for the technician, not a black box that tells them what to do. Lucas: And that's where the industry is heading. The next frontier is AI that can not only detect an anomaly but suggest a root cause. For example, a motor vibration pattern that correlates with a specific bearing wear pattern — the system could recommend replacing the bearing during the next scheduled downtime, before it fails. Luna: That would be a game-changer for scheduling. Instead of emergency shutdowns, you plan maintenance around production runs. Lucas: Right. And some companies are already doing that. Fanuc, the big Japanese robot maker, offers a service called Zero Downtime — they monitor your robots remotely and guarantee a certain uptime. If a failure is predicted, they send a technician with the right part before it breaks. Luna: That's like a subscription model for robot maintenance. Lucas: Exactly. It flips the economics from paying for repairs to paying for uptime. And for many manufacturers, that's a much easier budget line to justify. Luna: So what's the one thing a small manufacturer should do tomorrow to start moving away from reactive maintenance? Lucas: Start simple. Pick the robot that breaks most often, or the one that causes the most downstream disruption if it goes down. Put a cheap vibration sensor on it. Log the data manually for a month. You don't need a fancy platform — just trend the data in Excel. You'll start to see patterns. That's enough to make the case for a bigger investment. Luna: Low barrier to entry. I like it. And I think that kind of practical advice is exactly why people listen to this show. If today's conversation gave you something useful, and you want to help us keep making episodes ad-free, you can support the show at buy me a coffee dot com slash fexingo. Lucas: It really does make a difference. We put a lot of research into each episode, and listener support lets us keep the focus on substance over sponsors. So if you're able, we appreciate it. Luna: Alright, back to the factory floor. Lucas, you mentioned that Toyota's system is the gold standard. What's one specific practice they use that others could adopt? Lucas: One thing they do that's surprisingly low-tech: they log every unscheduled stop, no matter how brief. If a robot pauses for three seconds because a sensor flickered, that gets recorded. Over time, they build a heat map of transient failures. And often, those brief glitches are early symptoms of a bigger problem. Luna: So they're not just monitoring the robots — they're monitoring the monitoring system itself. Lucas: Exactly. And that's the mindset shift. Maintenance isn't a cost centre; it's a data source. The more you know about how your robots actually behave, the more reliably you can run them. Luna: That's a good place to leave it. Next week, we'll talk about something completely different — robot ethics in the age of autonomous decision-making. Lucas: Looking forward to it. Until then, keep your sensors clean and your bearings greased.