SPC & Cpk from Optical Metrology Data:
Turning Measurements Into Real Process Control

Pass/Fail Is Not SPC — It Just Feels Like It

Walk onto most shop floors and ask how they run statistical process control, and you'll get a version of: "every part gets measured, and if it's out of tolerance, we reject it." That is inspection. It is not SPC (statistical process control), and the difference matters more than the terminology suggests.

Pass/fail sorting tells you whether the one part in front of you meets spec right now. It tells you nothing about whether the process that made it is stable, whether it's drifting toward the tolerance edge, or whether the next fifty parts are about to follow the same drift off a cliff. By the time a pass/fail system catches the problem, you already have a reject in your hand — and probably several more upstream of it that haven't been measured yet.

Real SPC using optical metrology data means calculating process capability — Cp and Cpk — from every measurement, watching the trend as it happens, and getting an alarm when the process starts wandering, before a single part actually goes out of tolerance.

What Cp and Cpk Actually Tell You That Pass/Fail Doesn't

Cp and Cpk are both process capability indices, and they answer two different questions:

• Cp asks: if this process were perfectly centred on the nominal dimension, would its natural variation fit inside the tolerance band? It measures spread only, assuming ideal centring.
• Cpk asks the real-world question: given where the process is actually centred right now, how much room is left before it breaches tolerance? Cpk is always equal to or lower than Cp, and the gap between them tells you whether you have a centring problem, a variation problem, or both.

A process showing good Cp but poor Cpk is usually the easiest fix in metrology: the spread is fine, but something — a worn locator, a fixture offset, a tool that needs resetting — has pushed the average off-centre. That's a setup correction, not a redesign. You only know to look for it because you calculated Cpk in the first place.

Where the Data Actually Has to Come From

SPC software is only as good as the measurement feeding it. Manual gauging — callipers, height gauges, plug gauges — is slow enough that most plants sample rather than measure every part, which means the SPC calculation is built on a fraction of production, not the whole population. Optical instruments change that equation because the measurement itself is fast enough to run on 100% of parts.

Building an SPC Programme from Optical Metrology Data, Step by Step

Step 1 — Pick the Dimensions That Actually Matter

Not every feature on a part needs an SPC chart. Start with dimensions tied to fit, function, or a customer PPAP requirement — critical characteristics, in other words — rather than tracking Cpk on every measured feature by default. A QMM or VMM can measure 15–20+ features per cycle; tracking capability on all of them dilutes attention from the two or three that would actually stop production if they drifted.

Step 2 — Set Realistic Subgroup Sizes and Frequency

SPC references typically call for a minimum of 25–30 subgroups before a Cpk figure is statistically reliable. With 100% inspection running on a QMM, that sample size is reached within a shift rather than a week — which is exactly why moving from sampled manual gauging to full optical inspection changes what SPC can actually tell you, not just how fast you get the number.

Step 3 — Let the Software Set the Alarm, Not the Operator's Judgement

QMM and VMM CNC SPC modules calculate Cp/Cpk in real time and can flag process drift — a control limit breach or a run of points trending toward the tolerance edge — before any single part is actually out of spec. Configure these alarms deliberately, tied to control limits derived from the process's own variation, rather than leaving detection to whichever operator happens to notice a pattern on a printed chart.

Step 4 — Centralise the Data Instead of Trapping It on One Machine

A single QMM's SPC chart is useful. A shared SQL database across every QMM on the floor — supporting up to 50 machines and 5,000+ part templates in Optomech's implementation — is what lets a quality manager compare the same feature across machines, shifts, and operators from one workstation, and answer an audit question about a specific batch without walking the floor.

Step 5 — Close the Loop Back to the Process, Not Just the Report

A Cpk report that gets filed and not acted on delivers zero value over pass/fail sorting — it's just a more sophisticated way of documenting the same reactive behaviour. The point of real-time SPC is that a drift alarm should trigger a tool check, a fixture inspection, or a setup correction before the next batch is produced, not a root-cause investigation after a customer complaint.

Industries and Applications

Real-time SPC from optical metrology data has the biggest payoff wherever process drift is gradual, expensive to catch late, and tied to a formal capability requirement:

• Automotive component manufacturing: IATF 16949 PPAP submissions and ongoing production monitoring require documented Cp/Cpk against agreed thresholds, typically Cpk ≥ 1.33
• Plastics & injection moulding: Mould wear and shrinkage drift show up as a slow Cpk decline on gate vestige height, boss diameter, and snap-fit geometry long before a part is visibly out of tolerance
• Pharmaceutical packaging components: Dimensional consistency across long runs supports both regulatory audit readiness and downstream fit with capping and filling equipment
• Precision machined components: Tool wear on turned or milled features is a textbook case of gradual mean shift that Cpk trending catches well before a scrap event

Practical Takeaway

Measuring every part and rejecting the bad ones is quality control. Watching how the measurements move over time, calculating Cp/Cpk in real time, and acting on the trend before parts go bad is quality engineering — and it's the difference between a QA department that reacts to scrap and one that prevents it.

The instrument does the easy part automatically: a QMM or VMM CNC will calculate Cp/Cpk on every dimension you configure, without being asked twice. The harder part — deciding which dimensions matter, setting real control limits, and building an actual response when the alarm fires — is where the value of that data gets captured or wasted.

Get that right, and "we passed every part" turns into "we know exactly how much margin we have left, and on what."