I should start with a confession: I'm not an engineer. I'm the person who reviews and signs the purchase orders. At a 180-person manufacturing company, that means managing roughly $300,000 a year in test and measurement orders and keeping eight or nine vendor relationships on track. I report to both operations and finance, so I see the request before the PO and the pain after the equipment doesn't meet expectations. I can't recite thermocouple formulas from memory. But after a few years of watching the same failure repeat, you don't need an engineering degree to spot the pattern.
The Request I Keep Seeing
Every few months, a request lands in my inbox that reads something like this: “Need National Instruments thermocouple measurement for the new temperature test stand. Quote us a DAQ device, an SCB-68A, a shielded cable, and accessories.” It sounds complete. There's a brand, a connector block, and a list of parts. I used to enter that into our system without asking a single question.
Three or four weeks later, a second email arrives. “Readings are drifting.” Or “Data doesn't match the vendor's calibrated probe.” Or “Should we have bought a thermocouple module instead?” The question is usually phrased as if the hardware is at fault. From where I sit, the hardware was doing exactly what it was built to do. The problem was that we'd chosen hardware before we'd defined the measurement.
Wiring Is Not Measurement
The first mistake is understandable. Thermocouple wiring is physical. You have a sensor, you need to connect it to something that has screws or pins, and the SCB-68A is a perfectly good object for that job. It's a shielded connector block for 68-pin National Instruments DAQ devices, and it gives you a clean place to land wires. That's its role.
But a connector block's job is connectivity, not interpretation. To me, that's an important boundary. A thermocouple generates a small voltage that changes with temperature. A K-type thermocouple, for example, produces roughly 40 microvolts per degree Celsius near room temperature (Source: NIST ITS-90 thermocouple database). If you're trying to hold an uncertainty of half a degree, you're working with microvolt-level signals. Every screw terminal, every connection, every bit of dissimilar metal in the signal path can create its own small thermal voltage.
That is why cold-junction compensation exists. A thermocouple measures the difference between the sensing tip and the terminal block where the wire lands. If the software assumes that junction is at 0 °C, or if the reference sensor itself is off by a degree, the reported temperature is off by roughly the same amount. This isn't a flaw in the thermocouple. It's physics.
A dedicated National Instruments thermocouple module handles a lot of that physics inside its design: a high-resolution front end, a cold-junction sensor placed close to the terminals, open-thermocouple detection, and calibration paths. If you use a general-purpose DAQ and an SCB-68A, someone on your team has to own all of that separately. The documentation may tell you how to connect thermocouple wires and read the on-board temperature sensor, but you're the one who has to make sure the sensor location matches the actual terminal temperature, that the wires are routed correctly, and that the correction is applied in software. None of that is impossible. But if it isn't assigned to anyone, it usually doesn't get done consistently.
Flexible Inputs Often Add Uncertainty
Here's the part that surprised me most. In the purchasing world, we're trained to appreciate flexibility. A general-purpose input board can measure lots of signal types. A shielded connector block supports many kinds of sensors. That sounds like a responsible purchase.
But in measurement, flexibility often adds uncertainty. The more general the front end, the more the rest of the chain has to compensate. The “we can wire anything to it” feature is not a bug; it's simply a different job. When you have a production test that needs trustworthy temperature data, you usually don't want a device that can do a hundred things with mediocre certainty. You want a specialist that does one thing with documented performance.
I'd argue the same principle applies to low-cost prototyping. You can log a thermocouple with a small microcontroller, and people often do. Some of the DIY debates I read sound like “NXP vs. whatever dev board is popular this month.” The chip was never the real issue. The analog front end, the reference junction, and the validation are the real issue. If you're building a one-off experiment, that work might be exactly the right learning experience. If you're validating a product for a customer, you need someone who has already worked out those details and will stand behind them.
The Cost of an Unmanaged Cold Junction
One project in 2023 still bothers me. We had an existing DAQ system with an SCB-68A in the rack, and adding thermocouple channels seemed like a no-brainer from a cost perspective. The hardware was already there; we just needed to “map the pins.” I'm not going to pretend I understood all the engineering details. I did understand the savings on the purchase order.
The chamber data looked stable at first. Then the customer's independent calibration disagreed with our readings by about two and a half degrees. Their quality engineer rejected the validation report. I knew I should have asked for a technical review before ordering, but the project was already late, and in my head I heard: “What are the odds that a couple of degrees matter?” The odds caught up with us when the report came back.
“Two degrees can look like a rounding error to an engineer and a failed audit to a customer.” An application engineer said that to me once, and it stuck.
We spent the next two weeks re-testing, re-checking wiring, and eventually ordering a dedicated thermocouple input module anyway. The financial impact was bigger than the module price. Between the customer's re-test requirement, expedited shipping, and engineering hours spent chasing what looked like a software issue but was really a wiring and reference-junction issue, the project cost us at least $10,000 that we didn't plan for. Don't hold me to an exact number. The softer cost—damaged confidence in our data—is harder to calculate.
That was also the third time we'd had a thermocouple requisition with no tolerance on it. That's when I finally fixed the process gap. I added a simple checklist to the internal request form: thermocouple type, expected temperature range, acceptable error, and lead length. The engineers rolled their eyes at first. Then the checklist caught a project asking for ±0.5 °C while specifying an analog input board with no cold-junction plan. The form didn't make the engineering decision, but it forced the conversation early, before the purchase order was cut.
What a Thermocouple Purchase Order Should Say Now
The solution is not “buy a fancier box.” The solution is to put the measurement definition on the paper before anyone quotes a part number. When I get a request for National Instruments thermocouple capability now, I push back with a few questions.
- What are you really measuring? “Type K, 0 to 200 °C, target uncertainty ±0.5 °C” tells everyone more than the phrase “thermocouple input.”
- Who owns the cold junction? If the plan relies on an SCB-68A and a general-purpose DAQ, the answer should name the specific sensor, the channel it's read on, and how it's verified. With a dedicated National Instruments thermocouple module, much of that is designed in. If you don't know which category you're in, stop and ask.
- Is the sensor grounded or ungrounded, and how far is the cable run? These details determine whether you need isolation and shielding, and they change the quote.
- Has a specialist reviewed the chain? National Instruments application engineers and their channel partners know the hardware boundaries. Ask them. An answer like “for that uncertainty, the dedicated module is the safer route” is not a sales pitch. It's a boundary being drawn honestly.
I've learned to respect those boundaries. The more complete the description, the less likely it is that someone will order a shielded connector block and expect it to carry the full weight of the measurement. The SCB-68A can be the right answer for some jobs. A dedicated thermocouple module is the right answer for others. The mistake is treating them as interchangeable just because both are in the National Instruments ecosystem and both have screw terminals.
Once our test racks are networked into a larger measurement network, it's tempting to think the network is the system. It isn't. You can stream perfect-looking data to a dashboard and still be wrong at the source. The network didn't create that problem, and the network can't fix it.
From where I sit, the most professional move is not to choose a product that claims to do everything. It's to choose a product that does the right job with a known uncertainty—and to get someone who understands the measurement boundary to sign off before the purchase order goes through. That's what I want on every National Instruments thermocouple order that crosses my desk.
Leave a Comment