Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Are Preventing Boiler Efficiency Failures in Real Time
Transcript
- Lucas: If you've got a building with a boiler — and most commercial and industrial buildings do — there's a good chance that boiler is running somewhere between 5 and 15 percent less efficiently than it could be. Not because it's broken, but because combustion is finicky and nobody's watching it in real time. Luna: So the boiler's still making heat, but it's burning more fuel than necessary — and maybe spewing more emissions — and nobody knows until the annual service visit? Lucas: Exactly. And in a lot of facilities, the annual tune-up is the only time anyone measures oxygen and carbon monoxide in the flue gas. That's a snapshot. In between, burner drift, air damper misalignment, fouled heat exchanger surfaces — all of that degrades efficiency. A plant I looked at recently, a food processing facility in the Midwest, put IoT sensors on its boiler stack and found a burner misalignment that was wasting 47,000 dollars a year in natural gas. That's one boiler. And the fix was a two-hour adjustment. Luna: That's a huge payback. But before we dig into the sensors — and I want to hear about the oxygen probes and the acoustic thermometers — a quick honest thing. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making shows like this one. No ads, just listener support. So if you get something out of these episodes, that's the channel that keeps it going. Lucas: Yeah, and we don't mention it often, but it's worth pausing on because it's the reason we can spend a whole episode on boiler combustion without shoehorning in a commercial. So — back to the sensors. Lucas: The standard approach for continuous boiler monitoring has been around for decades in large utility boilers — they use zirconium-dioxide oxygen sensors and infrared carbon monoxide analyzers. But those are expensive, high-maintenance, and they need sampling systems that pull and condition the flue gas. For a 200-horsepower industrial boiler — the kind you'd find in a food plant, a hospital, a university — that level of instrumentation doesn't make economic sense. Luna: Right, so the IoT play here is about making those measurements cheap and rugged enough that you can deploy them on smaller boilers. What's the sensor package look like? Lucas: The system I looked at uses an oxygen sensor based on, I think, a lambda probe — similar to what's in a car's exhaust — paired with a carbon monoxide sensor that's essentially a metal-oxide semiconductor. They also mount an acoustic temperature sensor that measures the speed of sound across the flue gas path. That gives you combustion efficiency in real time, because efficiency is basically a function of flue gas temperature and oxygen concentration. Luna: And the data goes to a cloud dashboard? Or do they have a local edge gateway? Lucas: In the food plant case, each of the fourteen sensor nodes had a cellular modem. They transmitted readings every ten seconds to a cloud platform that applied an algorithm to calculate combustion efficiency and recommend adjustments. The plant's maintenance team got alerts on their phones when the efficiency dropped below a threshold. No onsite server, no wiring beyond the sensor harness. Luna: Okay, fourteen nodes — that covers more than just the stack, right? Are they also monitoring burner temperature, pressure? Lucas: Some of them are on the boiler casing for surface temperature, and a couple are on the gas supply line to detect pressure fluctuations that could indicate a regulator issue. But the critical ones are on the stack. And the algorithm doesn't just say 'efficiency is low.' It correlates the oxygen and CO readings to identify the cause. If oxygen is high and CO is low, that's excess air — too much air in the mix. If oxygen is low and CO is high, that's incomplete combustion — not enough air. Luna: And each of those has a different fix. Excess air means you can close the damper a bit. Incomplete combustion means you need more air or a burner adjustment. So the sensor data is giving you a diagnostic, not just an alarm. Lucas: Exactly. And that's the difference between IoT monitoring and a simple temperature gauge. The food plant's 47,000 dollar waste came from a situation where the burner was running slightly rich — incomplete combustion — for months. The CO level was elevated but not high enough to trigger a safety shutdown. The cumulative gas waste was about 6 percent of the boiler's annual fuel consumption. That's typical. Studies from the Department of Energy suggest that industrial boilers average 5 to 10 percent efficiency degradation between tune-ups. Luna: So if you have a boiler burning 1 million dollars in natural gas annually, you could be losing 50 to 100 thousand a year. That more than justifies the sensor investment, even if each node costs a few hundred dollars. Lucas: The hardware cost for the fourteen-node system was about 8,500 dollars, plus a 200-dollar per month cloud subscription. The plant figured a payback period of under three months. And they've also reduced their NOx emissions, because running with excess air tends to produce more nitrogen oxides. So there's a regulatory compliance angle too. Luna: But I want to push back a little. How many false positives do these sensors generate? If the maintenance team gets alerts all the time for transient spikes — say, during startup or load changes — they'll start ignoring them. Lucas: That's a real concern. The system I saw uses a moving average filter — it doesn't alert on a single reading. It looks for a sustained deviation over a one-hour window. And it also learns the normal operating envelope for that specific boiler. So a spike during a load change is suppressed unless it persists. In the food plant's first six months, they got four actionable alerts. Three were real issues, one was a sensor that needed recalibration. The false positive rate was low enough that the maintenance team took every alert seriously. Luna: What about the sensor drift itself? Oxygen sensors, especially lambda probes, have a finite lifespan. Do the nodes self-diagnose? Lucas: Good question. The system does a periodic 'air-cal check' — it compares the oxygen reading in ambient air to the known 20.9% value. If it's off by more than a percentage point, it flags the sensor for replacement. The manufacturer claims a two-year lifespan for the oxygen sensor, but in dirty flue gas it's probably closer to 18 months. The CO sensor lasts longer because it's not directly exposed to the hot gas — there's a sintered filter. Luna: So there's some ongoing maintenance cost. But still, for a plant with multiple boilers — hospitals often have three or four — the economics scale. And I imagine the same platform could handle tens of thousands of boilers across a portfolio. Lucas: That's exactly what the vendors are aiming for. There are an estimated 100,000 industrial boilers in the US alone, plus hundreds of thousands of commercial boilers. Most of them are unmonitored or only manually checked. If you could improve the efficiency of even 20 percent of those by 5 percent, the energy savings would be in the billions of dollars. And the carbon reduction is significant too — natural gas is cleaner than coal, but it's still a fossil fuel. Luna: And I'm guessing the next step is closed-loop control, right? Instead of alerting a human to adjust the damper, the system just adjusts it automatically. Lucas: A few vendors are starting to offer that, but it's cautious. Boilers are safety-critical; you don't want a software glitch closing the damper too much and creating a dangerous oxygen-deficient condition. Most plants are taking a human-in-the-loop approach for now. But the sensor data is making those human decisions much faster and more precise. Luna: So we're in this middle phase — the sensors are proven, the analytics are smart, but the full automation is still coming. It feels like a classic IoT adoption curve. Lucas: It really is. And the interesting thing is that the technology is already cheap enough that the ROI is clear for any boiler over 100 horsepower that runs more than a few thousand hours a year. The barrier is more about awareness and change management — getting facility managers to trust a new system and integrate it into their workflow. Luna: Any idea what the next big sensor frontier is for boilers? Beyond oxygen and CO, what else matters? Lucas: I've seen some pilot projects using optical sensors to measure flame shape and color. The idea is that a healthy flame has a specific spectral signature. If the flame starts to lift off the burner or become asymmetric, that's an early warning of trouble. Also, acoustic sensors on the tube walls can detect soot buildup — soot insulates the heat exchanger and reduces efficiency. Those are further out, but the direction is clear: more data, more dimensions, all in real time. Luna: And with the data comes the potential for machine learning models that predict failure before it happens. That's the holy grail. Lucas: Exactly. The food plant I mentioned is already feeding their 6 months of data into a model that predicts efficiency loss trends. They want to be able to say 'in three weeks, this boiler will be running at 78 percent efficiency unless we adjust the burner.' That's the kind of prescriptive maintenance that IoT enables. And it's not science fiction — it's running in a food plant in Ohio right now. Luna: I love that we can point to a specific plant, a specific dollar amount, a specific sensor type. That's the kind of concreteness that makes IoT real for people. Lucas: And it's only going to accelerate. As sensor costs drop and analytics improve, the 'unmonitored' boiler will start to look like a liability. The question is how fast the installed base turns over. Luna: All right, I'm going to check the boiler in my building's basement next time I'm there. Thanks for the deep dive, Lucas. Lucas: And if it's running at 82 percent instead of 88, you'll know why. Thanks, Luna.