In April 2023, I pointed my FLIR E5 Pro thermal camera at a 480-volt switchgear cubicle. The screen showed 58°C on a bus bar—warm, but inside the limit I had in my head. I signed the inspection checklist, logged the reading, and walked to the next panel.
Three days later, that cubicle failed. The bus bar had reached a temperature high enough to melt the insulation. Production line down for six hours. Repair invoice: $3,200.
My first thought was: the FLIR camera is lying to me. It wasn't. That's the part I had to learn the hard way. In my eleven years as a process engineer, this was the most expensive mistake that didn't end with someone in the hospital.
The Surface Problem: The Camera Must Be Off
It's easy to blame the tool. The FLIR E5 Pro is a solid entry-level professional thermal camera. It has a real 180 by 135 pixel detector, MSX image overlay, and a price that makes it accessible to smaller shops like ours. When the reading didn't match what I expected, I assumed the camera had drifted or the detector was bad.
What I mean is, I assumed the problem existed outside my own head. That assumption cost me a week.
Our tool room also has a FLIR smartphone thermal camera for quick walk-arounds. It's a great screening tool for obvious hot spots, but it is not the same class of instrument as the E5 Pro. I knew that. I still found myself using it to make judgement calls it wasn't designed for. Same pattern, different tool.
Here's something vendors won't tell you: a thermal camera does not measure temperature. It measures infrared radiation and converts it to a temperature using a set of parameters. The most important parameter is emissivity.
The Real Problem: I Asked the Camera the Wrong Question
Emissivity is a number between 0 and 1 that describes how efficiently a surface radiates infrared energy. Bare copper is around 0.6 to 0.7, depending on oxidation. My FLIR E5 Pro was set to 0.95, a decent default for electrical tape and paint, a terrible assumption for bare bus bar.
The camera did the calculation with the wrong emissivity and displayed a temperature that was lower than reality. The 58°C I logged was probably closer to 75 or 80°C on the bus bar. That's not a camera failure. That's operator error, politely wrapped in a $3,200 invoice.
(Note to self: if you haven't verified the emissivity setting before you look at the screen, you haven't taken a measurement yet.)
Calibration Is a Spec, Not a Suggestion
Even after I fixed the emissivity setting, I still had a nagging doubt. Was the E5 Pro accurate? FLIR's support documentation (flir.com/support) recommends annual calibration for professional thermal cameras. I had no idea when ours was last calibrated. I checked the tool room log. Nothing. The camera was a year and a half old, and calibration had never crossed anyone's mind.
Per FTC guidelines (ftc.gov), accuracy claims in advertising must be substantiated, which is why FLIR can state a figure like ±2°C or ±2 percent with confidence. But that spec assumes a properly calibrated unit. Drift happens. Lenses get dinged. Internal references age.
I don't have hard data on how many inspections we did with a potentially drifting camera. What I can say anecdotally is that at least one was wrong enough to start a fire.
The Same Pattern Ate My Multimeter, Flowmeter, and HPLC Column
As I dug through our records, the same pattern appeared everywhere. We trusted instruments because they turned on, not because we had proof they were still accurate.
My 87 multimeter, for example. I used it to confirm that the voltage downstream looked right. Then I checked the fuses and found the 10-amp fuse was blown. That meant the current channel would have been producing garbage readings. I never used that channel that day, but the fact that I didn't know was a red flag. Oh, and the battery was low, which added another layer of doubt to every reading I took that week.
The flowmeter on the cooling water line was reading about 25 percent low. That threw off the thermal balance calculation for the entire system. We only caught it because a batch failed and someone did a meter comparison. My best guess is the transducer diaphragm got damaged by a pressure spike. Honestly, I'm not sure. Nobody had logged the event.
And in the quality lab, we faced the classic question: how often should you change an Agilent HPLC column? The vendor guidance that came with ours said typical lifetime: 500 injections. But peak shape still looked okay, so nobody wanted to stop and change it. The result was two rejected batches and a revalidation delay. That lab failure was actually what triggered the wider investigation, not the switchgear issue.
What That $3,200 Mistake Actually Cost
The repair invoice covered the switchgear. It didn't cover the six hours of downtime, the overtime, the paperwork, or the credibility hit when I presented a signed inspection report that said no issues found three days before a failure.
The worst part wasn't the money. It was losing confidence in every reading I'd taken over the past year. How many other bus bars were hotter than I thought? How many batch results were quietly skewed by the flowmeter? How many more injections could that HPLC column have pushed through before a routine QC run turned into something much worse?
I still don't have solid answers. That uncertainty is what keeps me up at night, more than the dollar figure. At least, that's how it played out in our plant, with its mix of old switchgear and rushed maintenance windows.
The Short Version: How I Prevent This Now
I'm a process engineer with eleven years of plant experience, not a calibration lab. I can't verify the absolute accuracy of a thermal camera, a multimeter, or a flowmeter. But I can make sure they get calibrated by people who can, and I can stop treating a working power light as proof of accuracy.
Here's the short list, hard-won by experience:
- FLIR E5 Pro and any thermal camera: annual calibration, with the date written on a label on the tool. Before every job, verify the emissivity setting and adjust it for the surface you're inspecting. I keep a laminated card of common emissivity values taped to the camera case.
- FLIR smartphone thermal camera: use it for screening, not for sign-offs. It's a fine tool for finding a hot breaker before you open the panel, but it's not a substitute for a professional instrument when your name is on the report.
- 87 multimeter: check fuses and battery before critical measurements. A blown fuse produces readings that look fine until you compare them with a known-good reference.
- Flowmeters: verify at the manufacturer's recommended interval, and after any event that can stress them, like a pressure spike or an unexpected shutdown.
- Agilent HPLC columns: log injection count on a physical tag. Use the vendor guidance as a starting point, for example 500 injections in our case, and change the column when you hit the limit or when pressure and peak shape tell you it's done. Pick a trigger and commit to it before the column fails in the middle of a run.
I keep a running log of every reading I've signed off on that later turned out to be wrong. It's not a comfortable document to maintain, but it makes the next checklist better. The vendor who told us that calibration wasn't their core service and pointed us to a specialist earned our trust more than the one who promised a software fix for everything. There's no shame in admitting the boundaries of your own measurement skill. The shame is signing off on a reading you never actually verified.
I still trust the FLIR E5 Pro. I just trust my pre-fight checklist more. The camera wasn't lying. I just didn't know what I didn't know. That checklist costs about ten minutes. The alternative cost me $3,200, a production shutdown, and one very uncomfortable phone call.
