The day a checkweigher lied to us
Last March, our checkweigher started rejecting good vials. The line stopped at 10:47 in the morning. I remember the time because the red stack light sits right above my desk.
First suspect? The load cell. Everyone said so.
For context, I'm the quality manager at a 140-person contract manufacturing site. I review every calibration report and certificate before it leaves our building—roughly 1,100 items a year. In Q1 2025, I rejected 9% of first submissions because of missing traceability or expired references. I don't like guessing, and I don't like being sold a story.
A checkweigher is a simple machine until it isn't. The load cell sends a low-level DC signal, a few tens of millivolts, to the controller. That signal says how much the vial weighs. If the signal is noisy, the controller thinks the weight is changing. It rejects the vial. The machine isn't broken. It's listening to bad data.
Our technician measured the load cell signal with a handheld multimeter. The readings jumped around 0.4 mV. The vendor's service engineer said, 'It's the load cell.' We had a quote for $3,800 plus travel before lunch.
'It's the load cell.' — our vendor's service engineer, who never saw the Keithley readings
Here's where communication failed. I told the production lead to 'check the load cell.' I meant measure the signal. He heard 'the load cell is bad.' Result: a purchase requisition, an overnight ship estimate, and a plan to tear down a perfectly good sensor.
I went to the spare bench and grabbed a Keithley 2110 digital multimeter. We connected to the same terminals, same wires, same load cell. The reading settled at 18.347 mV and held there. Then 18.345. Then 18.346. The old handheld was the source of the noise, not the scale.
People think a stable reading means a good meter. Actually, the relationship runs the other way. A good meter with documented calibration produces stable readings when the signal is stable. A meter that lost its calibration produces noise and makes a good machine look broken. The cheap meter didn't fail during the test. It failed before the test, on the day its calibration expired.
If you search for 'multimeter Keithley,' you'll see the 2110 described as a 5.5-digit bench meter. That doesn't explain why I trust it. I trust it because the reading didn't move when the signal wasn't moving. It gave me a number I could put in the investigation report without adding an asterisk.
So if you're comparing 'multimeter Keithley' options, don't start with digits alone. Start with the question: can I defend this reading in an audit six months from now?
Before we blamed the load cell, I ruled out another suspect: our 1260 HPLC. That's not a random detail. We had a grounding failure there once that caused pressure spikes, and those electrical pulses showed up in the checkweigher's readings. This time the 1260 HPLC pressure trace was flat for an hour. The pump was stable. The analyzer was innocent.
The real problem was a loose ground lug on the checkweigher's amplifier. Seven years of vibration had worked the nut loose. We cleaned the terminal, tightened it to torque, and the line came back. No load cell. No $3,800. No service visit.
Why did the loose ground cause random readings? Because a nearly tight ground lug acts like a loose connection. Vibration changes the contact pressure, and the resistance changes. The load cell's small signal rides on top of that fluctuating ground reference. A better meter doesn't remove the ground problem, but it reveals it. The handheld meter's own noise floor hid the difference.
Dodged a bullet? Yes. The new load cell was already in the cart. What stopped us was the meter.
I still kick myself for not pulling the handheld's calibration record before the panic started. If I had, I'd have seen it was 22 months past its 12-month calibration cycle. That's a lesson about my own checklist, not just the instrument.
While the line was down, our tech asked me a question I still think about: 'Where are Mitutoyo calipers made?' He had bought a set from an online reseller and suspected they were counterfeit. The problem wasn't the country stamped on the case. It was the lack of a serial number and a calibration document. For any measuring tool, the paper trail is part of the tool. The same logic applied to our load cell diagnosis: the stamped number on the sensor was irrelevant until a calibrated instrument confirmed it was bad.
We later found the same counterfeit batch offered on multiple marketplaces, all with the same stock photos. The tell wasn't the country of manufacture; it was the absence of traceability. That's a rule I now apply to every instrument in the building.
The total cost of one wrong reading
Let's do the math the way I should have done it years ago:
- Handheld multimeter: already in the drawer, so it looked free.
- Load cell replacement: $3,800 plus freight, if we had bought it.
- Vendor site visit: $1,900 estimate.
- Line downtime: about $18,000 in delayed shipments and overtime.
The Keithley 2110 digital multimeter cost less than the service visit. Not less than the whole cascade—less than one line item. The old buying logic that compares sticker prices was meaningless. On a total cost basis, the 2110 paid for itself in one morning.
I'm not saying every cheap meter is useless. I'm saying every measuring tool has a job, and the job description has to include calibration. If you can't prove the reference, you can't prove the reading.
When I hear people compare instruments by sticker price, I stop them. Total cost includes calibration schedule, traceability, repeatability, and the cost of being wrong. The cheapest meter in the drawer is the most expensive one when it tells you a good part is bad.
I justified the purchase with the DCV accuracy from Keithley's published specifications (keithley.com, accessed April 2025). The printed spec was tighter than anything we needed for a checkweigher. But the real reason was calibration documentation. An instrument without a documented reference is a guess with a display.
What I do now
We standardized the low-voltage test benches on the Keithley 2110. The reason is simple: it gives us a number we can defend in an audit. It also has a calibration certificate we can track. (Note to self: track calipers and the HPLC in the same system.)
The question is no longer, 'Which multimeter is cheaper?' It's, 'What happens when the reading has to stop a wrong decision?' For that question, the 2110 is worth every cent.
The meter didn't make a bad signal good. It made the good signal visible. That's the part no spec sheet tells you.
