Automated measurement turns statistical process control (SPC) from a sampling exercise into continuous monitoring. When a vision system or data acquisition system measures every part, its results can feed control charts directly, so drift is visible within minutes instead of at the next manual sample. The data is only useful if the measurement system is capable and the charts are set up for the volume.
This article is part of our data acquisition guide. It is written for quality and continuous improvement engineers connecting automated measurements to SPC.
In short
- Prove the measurement system first: gage R&R before control charts.
- 100% measurement needs subgrouping, or the charts alarm on noise.
- Charts should watch the process, not just pass or fail parts.
- Store measured values with time, part and lot, or you can’t trace a shift.
What changes when every part is measured
| Manual SPC | Automated measurement | |
|---|---|---|
| Data | A few parts per hour, recorded by hand | Every part, recorded automatically |
| Detection of a shift | At the next sample, sometimes hours later | Within a few subgroups |
| Typical errors | Transcription, rounding, skipped samples | Measurement system drift, if not verified |
| Effort | Inspector time for every sample | Set-up and periodic verification |
Setting up control charts for high-volume data
- Subgroup the data. Charting every individual value on a fast line produces constant alarms from normal variation. Group consecutive parts, for example 5 at a time, and chart the subgroup mean and range.
- Base limits on stable data. Calculate control limits from a period when the process was running normally, not from specification limits.
- Watch the right characteristic. Chart the dimensions or signals that predict defects, not just the reject count.
- Use run rules carefully. With thousands of points a day, every extra rule adds false alarms. Start with points outside the limits and runs on one side of the center line.
Capability from automated data
Process capability compares the spread of the process with the specification. For a dimension with limits of 9.95 to 10.05 mm, a mean of 10.02 mm and a standard deviation of 0.01 mm:
- Distance to the upper limit: (10.05 − 10.02) ÷ (3 × 0.01) = 1.00
- Distance to the lower limit: (10.02 − 9.95) ÷ (3 × 0.01) = 2.33
- Cpk = 1.00, the smaller of the two
The process is capable on spread but sits off-center toward the upper limit. Re-centering, not tighter control, is the fix. Automated data makes this visible quickly because the sample is large.
Make sure the measurement can be trusted
If the measurement system contributes much of the variation, the control chart shows the gauge, not the process. Before using automated data for SPC:
- Run a gage R&R study on the measurement.
- Know the measurement uncertainty compared with the tolerance.
- Check a reference part each shift and chart it separately, so measurement drift isn’t mistaken for process drift.
Data you need to store
For each measurement: the value, timestamp, part or serial number, lot, station, recipe and software version. This lets you trace a shift to a material lot, a tool change or a station, and it is what auditors and customers ask for.
Want SPC from your automated measurements?
Sciotex builds data acquisition, vision and test systems that measure every part and store traceable results ready for SPC, on any hardware and language. Designed, programmed, built and supported in the US since 1997.
Frequently asked questions
Yes. Results from vision, test or data acquisition systems can feed control charts directly, as long as the measurement system is capable and the data is subgrouped for the volume.
Charting every individual value on a fast line exposes normal variation and makes run rules trigger often. Subgrouping consecutive parts and limiting run rules reduces false alarms.
No. Control limits come from the process’s own stable variation; specification limits define what the customer will accept.
Run a gage R&R study and compare measurement uncertainty with the tolerance. If the measurement system contributes much of the variation, the charts mostly show the gauge.
The measured value, timestamp, part or serial number, lot, station, recipe and software version, so a shift can be traced to its cause.