Production test
Line-side test: the measurement is easy, the record is what fails an audit
A station that measures correctly but cannot prove which limit applied in March is a station that will fail its next customer audit. The engineering that prevents that is mostly about levels, timing and capability — not about the converter.
Versioned limits as the first traceability defect, 5 V logic against 24 V field wiring, hardware-timed digital words, poll loops that miss short faults, and what a Gage R&R study actually asks you to prove.
Ask an engineer what is hard about a production test station and the answer is usually about accuracy. Ask an auditor and the answer is different. The recurring findings are not measurement errors — they are records that cannot be reconstructed: a limit that was changed without a version, a part that passed against the wrong revision of the specification, a fault that lasted 5 ms and was never inside a polling window.
Version the limits, or the data means nothing later
A test record is a claim that a specific part was compared against a specific requirement at a specific time. If the requirement is a number embedded in a script, that claim cannot be verified a year later, because the script has been edited several times since and there is no way to know which version was running.
- Keep limits in a data file, not in code, with an explicit revision identifier on every revision.
- Write the limit revision into every record, alongside the measured value and the verdict.
- Record the instrument identity too — module serial number, and firmware or driver version where it can affect the result.
- Never overwrite a limit revision. Supersede it with a new one and a reason, so the old records stay interpretable.
5 V logic against 24 V field wiring
A DAQ module's digital lines are 5 V logic. Machine control cabinets are 24 V. Connecting the two directly is the single most common wiring mistake on a test station, and it fails in both directions: 24 V into a 5 V input damages it, and a 5 V output into a 24 V input either reads as permanently low or is ignored entirely.
| Module | Digital lines | Drive capability | Counters |
|---|---|---|---|
| USB-6001 / 6002 / 6003 | 13 | 4 mA sink per line | 1 |
| USB-6501 | 24 | 5 V logic; check the module datasheet for per-line sink current | 1 |
| USB-6509 | 96 | 24 mA sink per line | None |
| USB-6525 | 8 inputs + 8 relay outputs | Inputs accept ±60 VDC; relays switch ±60 VDC at 500 mA with channel-to-channel isolation | None |
| USB-6212 / 6229 | 32 / 48 | Logic-level only — needs an interposer for field wiring | 2 × 32-bit at 80 MHz |
Swipe the table sideways to see all columns
The USB-6525 is worth singling out because it is the module that removes the interface hardware from the panel. Its eight inputs read 24 V limit switches and proximity sensors directly, and its eight outputs are solid-state relays isolated channel-to-channel, so a lamp, a solenoid valve or a small contactor coil can hang off it without a relay board in between. On every other module in the table, driving a real field load means adding an interposer.
Land the digital word on the analog sample clock
Many test specifications are really about correlation: this analog value was out of limit while that digital state was true. Reading the analog channels in one call and the digital lines in another call produces two different time bases, and under load they can be tens of milliseconds apart.
Modules that support hardware-timed digital input solve this by clocking the digital lines from the same sample clock as the analog inputs. The digital word is then captured at the same instant as the analog sample, and the correlation in the record is real rather than assumed.
A 5 ms fault is invisible to a 50 Hz poll
A software loop that reads a status line every 20 ms samples the machine at 50 Hz. A contact that bounces open for 5 ms has, at best, a one-in-four chance of being inside any given polling window — and if the fault is synchronised to the machine cycle, the chance is either always or never, which is worse, because the test appears to work until a specific product variant is run.
- Move any event that must be counted to a hardware counter. The 62xx modules carry two 32-bit counters clocked at 80 MHz; at that rate a 5 ms pulse is 400,000 counts, so no event is lost and its width is measured too.
- Where the counter is unavailable — the USB-6509 has none, and neither does the USB-6525 — the answer is a latch in the fixture: a set-reset relay or a monostable that stretches the fault until the next poll. That is a hardware fix, and it is the correct one.
- Do not solve it by polling faster in software. You will shrink the window without closing it, and you will add host load that makes the jitter worse.
Capability: what a Gage R&R study actually asks
A station can be accurate and still be useless as a test, if its variation is a large fraction of the tolerance it is judging against. That is what a measurement systems analysis establishes, and the AIAG MSA manual is the reference most automotive customers point at.
| %GRR of tolerance | Verdict |
|---|---|
| Below 10% | Capable. Accept the measurement system. |
| 10% to 30% | Conditional. Acceptable depending on the criticality of the characteristic and the cost of the alternative — the decision has to be argued, not assumed. |
| Above 30% | Not acceptable. Improve the system before using it to judge product. |
Swipe the table sideways to see all columns
The standard study design is ten parts spanning the process range, three operators, and two or three repeated trials per operator per part — which also gives you the operator-by-part interaction that a two-operator study cannot see. Where a single instrument is being qualified rather than a whole system, the corresponding check is a Type 1 study reporting Cg and Cgk, with 1.33 as the common acceptance floor.
Running for months without an operator
Line-side logging is usually specified as 24/7 for the life of the product. A USB module is a reasonable fit for that, mainly because it has no moving parts and no separate power supply, and because a spare can be swapped and re-identified in minutes. What needs designing is everything around it.
- Buffer on the device, not in the host. Hardware-timed acquisition with onboard FIFO means a busy host causes a late read, not a lost sample.
- Plan for re-enumeration. Any USB device will eventually be unplugged, reset or power-cycled. Detect it, log the gap, and resume — an operator should be able to fix it without engineering support.
- Rotate files and check free space. The most common cause of a lost month of data is a full disk discovered afterwards.
- Include a self-check channel: a shorted input or a fixed reference, recorded with everything else. A drifting channel then becomes provable, and a station can be shown to have been healthy across a production run.
Keep the safety function off the measurement channel
A test station sometimes ends up guarding a robot cell or a press, because it is the computer that is already there. That is a category error. Functional safety is a separate discipline with its own architecture, redundancy and diagnostic coverage requirements, defined by ISO 13849-1 for performance levels and IEC 62061 for SIL.
The dividing line is simple enough to state and worth stating explicitly in the station documentation: the DAQ measures, records and decides pass or fail. The safety circuit stops the machine. They can share a cabinet and share a sensor, but they cannot share a signal path — a safety input must reach the safety relay independently, and it must be possible to lose the DAQ entirely without losing the protective function.
A station that is mostly counting and switching
| The job | Part | Why this one |
|---|---|---|
| Many indicator lamps, relay coils and status lines; no analog | USB-6509 | 96 lines at 24 mA sink drives panel loads directly and removes the driver board from the cabinet. |
| 24 V field wiring in and out, with isolation between channels | USB-6525 | Inputs take ±60 VDC and the eight relay outputs are isolated channel-to-channel, so no interposer is needed on either side. |
| Mixed analog signatures plus a high line count | USB-6229 | 32 analog inputs at 250 kS/s, 48 digital lines and four simultaneous analog outputs if the fixture needs an excitation or a setpoint. |
| Fast analog plus events that must be counted | USB-6212 | 400 kS/s on a single channel, 32 digital lines and two 80 MHz counters on one USB cable. |
Swipe the table sideways to see all columns
Errors that show up as audit findings
- Limits written into the test script, with no revision recorded in the data.
- Wiring 24 V field signals into 5 V digital inputs, and discovering it when a module fails.
- Assuming the analog read and the digital read happened at the same moment.
- Relying on a software poll loop to catch a contact that bounces for a few milliseconds.
- Never running a Gage R&R, and discovering the station cannot separate good from marginal parts.
- Letting the test PC double as the machine's safety interlock because it happened to be nearby.
Every figure in this note is either arithmetic from published resolution and sample-rate specifications, or a value stated by the standard it cites. Where a number depends on a particular module, confirm it against that module datasheet before design freeze.