Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Predict Solar Panel Hotspots Before Fires
Transcript
- Lucas: Back in February of 2025, a 50-megawatt solar farm in Riverside County, California, lost about eight hundred panels and started a small brush fire. The cause wasn't a grid fault or a lightning strike. It was a single hotspot on one panel that went undetected for three weeks. Luna: Three weeks. That’s a long time for a defect to just sit there cooking. How does a hotspot even start on a solar panel? Lucas: Most people think solar panels are just uniform slabs that soak up sunlight. But every panel is actually a series of individual photovoltaic cells wired together. If one cell gets shaded—maybe by a bird dropping, a leaf, or even a nearby module casting a shadow—it stops producing current and becomes a resistive load. The current from the other cells then flows through that shaded cell, heats it up, and if the temperature gets high enough—typically above 150 degrees Celsius—it can melt the backsheet, ignite the encapsulant, and start a fire. Luna: So it’s literally a hot spot. And in a big solar farm with tens of thousands of panels, how do you find one small hot zone before it becomes a catastrophe? Lucas: Right now, most operators rely on periodic drone flyovers with thermal cameras. A drone might sweep the whole field once a month or once a quarter. That’s what the Riverside farm was doing. But a hotspot can develop in a matter of days, especially after a weather event or a maintenance error. So there’s a big gap between inspections. And that’s where fixed IoT sensor arrays come in. Luna: Fixed sensors—so instead of waiting for the drone to come around, you have continuous monitoring. What kind of sensors are we talking about? Lucas: The most practical setup today uses arrays of infrared thermopile sensors. These are the same kind of low-cost sensors you find in non-contact thermometers. They measure the infrared radiation emitted by an object and convert that into a temperature reading. A single sensor costs maybe ten to fifteen dollars in volume. Deploy them on poles every few rows, and each sensor can monitor the surface temperature of dozens of panels within its field of view. Luna: So you get a temperature map, but you still need to distinguish a genuine hotspot from, say, a warm afternoon or a reflection. How does the system know the difference? Lucas: That’s the machine learning part. The sensor data goes to an edge gateway—a small computer right there in the field—that runs a model trained on thousands of hours of thermal data. The model learns the normal thermal profile of each panel under different conditions: time of day, season, cloud cover, soiling levels. When a panel’s temperature deviates from its own baseline by more than a certain threshold—say, 15 degrees above what’s expected for that moment—the system flags it. And it doesn’t just flag a single reading. It looks for trends over hours or days. A slow climb in temperature is more indicative of a developing hotspot than a one-off spike. Luna: So you’re predicting failure weeks in advance. That’s a huge shift from reactive maintenance. What’s the economic case for this? Installing sensors across a whole solar farm isn’t free. Lucas: Let’s run the numbers. A typical utility-scale solar farm in the U.S. costs about one to two million dollars per megawatt to build. So a 50-megawatt farm is a 50 to 100 million dollar asset. A single fire event—like the one in Riverside—can cause direct damage of several hundred thousand dollars in panels and wiring, plus the cost of firefighting, regulatory fines, and lost generation during repair. And if the fire spreads beyond the farm, liability can climb into the millions. A sensor network for that farm, covering maybe two thousand panels per sensor node, costs roughly fifty to a hundred thousand dollars installed. That’s a fraction of the cost of even a moderate incident. The return on investment is essentially guaranteed if you avoid just one fire over the twenty-five year life of the plant. Luna: And the sensors also catch non-fire issues—like panels that are degrading faster than expected—which helps with warranty claims and replacement planning. So it’s not just about fires. Lucas: Exactly. The same thermal data that catches hotspots can also identify panels with microcracks, delamination, or bypass diode failures. Those panels might not be fire risks, but they’re underperforming. The operator can schedule targeted replacements rather than swapping out whole strings. That drives down operations and maintenance costs significantly. I’ve seen estimates that predictive thermal monitoring can reduce O&M spend by 15 to 20 percent over the life of the plant. Luna: So the technology exists, the economics work. Why isn’t every solar farm doing this already? Lucas: A few reasons. First, the solar industry is still relatively young. Many farms were built before this sensor technology matured and don’t have the communication infrastructure—no power or data cabling out in the field. Retrofitting can be expensive. Second, there’s a data integration challenge. The sensor data needs to flow into the plant’s existing monitoring system, and not all SCADA platforms are set up to handle edge-processed thermal data. Third, there’s a trust issue. Operators are used to drone-based thermal surveys and aren’t sure they want to rely on a fixed sensor network that they can’t easily validate by eye. But that’s changing as more case studies come out showing that fixed sensors catch issues that drones miss. Luna: And what about the panel manufacturers? Do they see this as a threat to their own monitoring services? Lucas: Some do. A few large panel makers offer their own proprietary monitoring, but it’s usually module-level power electronics—things like microinverters or optimizers—that track electrical output, not temperature. So there’s a gap. The sensor companies argue that thermal monitoring is complementary. And I think the trend is toward open, interoperable systems. The industry is pushing for common data standards, which would make it easier for third-party sensor networks to plug into any farm. Luna: Let’s talk specifics. Who are the companies actually deploying these systems? I’ve heard of a few startups in this space. Lucas: One of the better-known players is a company called—let’s call it 'SunSight'—though there are several others. They use a combination of thermopile sensors and a small edge gateway that runs a lightweight neural network. They’ve deployed on about two gigawatts of solar capacity globally, mostly in the U.S. and Spain. Their pitch is that they can detect a hotspot an average of 18 days before it becomes a critical failure. Another company, 'ThermWave', uses a slightly different approach: instead of fixed sensors on poles, they embed a thin thermal sensor strip along the backsheet of each panel during manufacturing. That gives per-panel temperature data, but it’s more expensive and only works for new installations. Luna: So you’ve got a retrofit option and an embedded option. Which is winning? Lucas: Right now, the retrofit model is more common because the installed base of solar is huge—over 200 gigawatts in the U.S. alone. Most of those panels are already in the ground and won’t be replaced for another ten to fifteen years. So the addressable market for retrofit is enormous. But the embedded approach could become the standard for new builds, especially if the cost comes down and the reliability data piles up. Luna: And what about the regulatory side? Are utilities or fire codes starting to require this kind of monitoring? Lucas: Not yet, but it’s being discussed. After the Riverside fire and a few other high-profile incidents, the California Public Utilities Commission held a workshop on solar farm fire risk. One of the recommendations was to consider mandating continuous thermal monitoring for new installations above a certain size. I think we’ll see a voluntary standard emerge first, maybe from the Solar Energy Industries Association. Then, if adoption is slow, regulators might step in. The insurance industry is also pushing for it. Some insurers now offer premium discounts for farms with active thermal monitoring, which is a strong economic incentive. Luna: That’s a domino effect—insurers, then lenders, then owners—it’s how a lot of safety technology gets adopted in capital-intensive industries. So where does this go next? What’s the frontier? Lucas: Two frontiers, I think. One is combining thermal data with electrical data from inverters and string-level monitors to create a holistic health model for each panel. A panel that’s running hot but producing normally might have a different root cause than one that’s hot and underperforming. The second frontier is using the same sensor network for other things—like detecting wildlife intrusion, tracking soiling accumulation, or even monitoring vegetation growth around the panels. Once you have a communication backbone in the field, you can hang a lot of applications on it. Luna: That makes sense. And it brings us back to a broader point about IoT: sometimes the most impactful applications are the boring, mundane ones—not a flashy new gadget, but a sensor doing one thing really well, preventing a disaster nobody was thinking about. Lucas: Exactly. And frankly, that’s why we keep doing these episodes. We talk about sensors preventing all kinds of catastrophes, from grain elevator explosions to tunnel boring machine failures. The solar hotspot story is the same pattern: a cheap sensor, a smart algorithm, and a willingness to look at a familiar problem differently. Luna: And it’s ad-free, which is part of what lets us dig into these specific technical details without selling anything. Lucas: Right. That’s intentional. We keep the shows free of sponsors so we can focus on what’s actually interesting. If that model resonates with you, there’s a simple way to support it: buy me a coffee dot com slash fexingo. No pressure, just an option if you find value in these conversations. Luna: It helps us keep the lights on and the sensors buzzing. And speaking of sensors, let’s zoom back to that Riverside farm. After the fire, they installed a retrofit sensor array from one of the companies we mentioned. Within the first month, it flagged three developing hotspots that the monthly drone sweep would have missed. Lucas: Exactly. And those three panels were replaced during a scheduled maintenance window, not in an emergency. That’s the whole game: turning reactive firefighting into proactive maintenance. And as solar scales to terawatts globally, that kind of operational intelligence becomes critical. Luna: So next time you see a solar farm, remember there’s probably a lot more going on under the glass than just photons turning into electrons. Lucas: Absolutely. And maybe, someday soon, a little infrared eye watching for the one cell that’s about to cause trouble.