Last March, I almost signed off on a $3,800 boiler valve replacement that wasn’t needed. That same week, our drum filling line quietly gave away about 1.5% of everything it filled. Two different “process failures,” one underlying theme: the measuring instruments were lying, and we didn’t catch them early enough.
I’m the quality and compliance manager at a mid-sized food manufacturer. I review every lot release — roughly 200 batches a year — and every calibration record before it’s archived. I’ve rejected maybe 1% of first deliveries in four years, and none of them were boiler-related. So I’m not someone who usually blames the machine first. But this time, I did.
Here’s the thing: both problems looked like maintenance failures. The boiler was misbehaving, and the scale was drifting. It took me three days to remember my own rule — when the data stops making sense, verify the measurement before you touch the process.
The boiler wasn’t broken. The O2 sensor was.
Our plant runs on a 300 hp natural gas boiler that supplies steam for cooking kettles and clean-in-place systems. Every Monday, our maintenance tech pulls out a combustion analyzer, checks the stack, and adjusts the burner for the week. We’ve standardized most of our portable temperature and combustion kit on testo measuring instruments, so the analyzer in his hand is a testo 320.
That Monday, the steam header started swinging by ±4 psi. The burner flame looked ragged to the operator. By Tuesday, the working theory was a failing control valve. We drafted a purchase order for a replacement — around $3,800 with labor.
Before cutting the PO, I asked the tech to do one more combustion test. He ran the boiler up to steady load and watched the analyzer. The O2 reading wandered from 5.1% down to 3.4% and back up to 6.2% over two minutes, with no change in burner position. That’s not a process story. A healthy electrochemical O2 cell should hold within roughly 0.2% in stable flue gas.
We ran a fresh-air check. It eventually settled near 20.9%, but it was slow and sticky. The cell was at the end of its service life. I placed the order for a testo replacement o2 sensor for model 320 before lunch. The sensor cost somewhere around $170 — maybe a little more, I’d have to check the invoice — and the swap took about ten minutes. After a fresh-air calibration cycle, the readings steadied out. The steam header stopped swinging by midday. The valve went back to stores.
Then the filler started giving away product
That same week, a warehouse spot check found our 55-gallon drums running heavy. Not by much — about 1.5% above the fill target — but consistently. At first, production blamed the operator. The lab spun retained samples in the centrifuge and confirmed the product itself was within formula spec. The problem wasn’t the recipe. It was the scale telling the filler to stop too late.
The drum filler uses a platform scale with a Rice Lake load cell under the deck. The controller keeps filling until the scale reads the target weight. If the scale under-reports, every drum leaves a little overfilled. That’s not a quality defect for the customer — they get more than they paid for — but at 1.5% overfill, it becomes a serious profitability leak.
Before I called the scale service company, I walked through the basics myself. I checked the mechanical path first: no debris under the platform, no binding hoses, nothing leaning against the frame. I used the 62 mini IR thermometer that lives on my maintenance cart to scan the junction box and load cell area — no abnormal heat, so we weren’t looking at thermal drift.
Then I opened the junction box. Water from washdown had gotten past the cable gland. One of the connectors was corroded, and the signal from that cell was dropping intermittently. That alone explained the drift. But when we tested each load cell channel with a multimeter, we found a second problem: one cell wasn’t returning to zero after load was removed. It would sit at 0.6 mV instead of 0 mV, which is enough to shift the final fill weight.
How to troubleshoot a Rice Lake load cell before calling a technician
If you’re wondering how to troubleshoot a Rice Lake load cell system, or any strain-gauge scale system, this is the path that worked for us:
- Check the mechanical path first. Look for product buildup, jammed debris, binding hoses or pipes, and anything that touches the platform or vessel. A healthy load cell will still lie if something is leaning on it.
- Watch the zero. Put the indicator in live weight mode and see if the zero returns to the same point after each load cycle. A slow or wandering zero is a red flag.
- Open the junction box. Look for moisture, corrosion, loose terminations and damaged cable glands. In food plants, washdown water finds every weak seal.
- Measure each cell directly. At the junction box, check the excitation voltage — usually 10 VDC — then measure the signal between SIG+ and SIG-. For a typical 1,000 lb, 2 mV/V load cell with 10 VDC excitation, full-scale output is 20 mV. Put 100 lb on the deck and you’d expect roughly 2 mV from that cell. If the reading doesn’t follow the weight, or doesn’t return to zero, you’ve found your suspect.
- Apply certified test weights. After any repair, verify the system against known weights across the range, not just at zero. That’s the step most people skip, and it’s the one that catches the rest of the problem.
In our case, the corroded connector explained most of the drift, and the slow-to-zero cell explained the rest. We replaced the connector and the bad cell, resealed the junction box, and recalibrated with certified weights. The filler returned to target on the next shift.
The part I keep replaying
What bothers me isn’t the repair cost. It’s that we were ready to spend $3,800 on a boiler valve that was perfectly fine, and we were ready to accept a 1.5% giveaway on the drum line as “one of those things.” Both were measurement problems hiding inside process problems.
The frustrating part is that soft sensor failures are the hardest to catch. A total failure is obvious. A dying O2 sensor just gets a little slow and a little wander-y. A corroded load cell connection doesn’t stop the scale; it just makes it dishonest.
I still kick myself for not checking the instruments on Monday morning instead of Wednesday afternoon. If we had run the analyzer fresh-air test and the scale zero test on day one, we would have saved three days of investigation and a lot of unnecessary stress.
The fundamentals haven’t changed — you still need to verify the instrument before you trust the process. But the execution has improved. Since we moved to digital maintenance logs, our newer testo measuring instruments remind us when sensors are due for replacement, and the calibration history is right there in the file. That doesn’t replace good judgment. It just makes it easier to use it.
A sensor failure that looks like a machine failure is the most expensive kind.
Last March convinced me of an old quality rule: when a process changes overnight, check the instruments first. The machine might be fine. The measurement might be the thing that broke.
