Closing the Loop: What an SPC Report Needs From a One-Shot Measuring System

A measurement that is not attached to an identity is an observation, not a record. The gap between a working measuring cell and a usable SPC system is almost never the measurement — it is the field list, the identity convention and the gauge study that has to happen before any capability number means anything.

The Field List Comes Before the Software

Every record needs five things: which part, which feature, which value, which limit, and which moment. Software choice is downstream of agreeing that list. Most failed SPC rollouts we see had the measuring system working for months while the field list was still being negotiated between quality and production.

A workable minimum per feature: feature name as it appears on the drawing, nominal, upper and lower limit, measured value with its unit, pass/fail state, and the measurement recipe version. Without the recipe version, a change in lighting or edge threshold silently invalidates historical comparison.

Gauge Study Before Capability

A CPK computed from an unqualified measurement is a number, not a capability. The study that has to come first is repeatability and reproducibility on the actual fixture: the same part loaded repeatedly, by more than one operator if the cell is manual, with the spread recorded per gating feature.

The comparison that matters is the study spread against the tolerance band, not against the specification-sheet repeatability. The sheet figure is measured under controlled conditions; your fixture, your part surface and your operator are the conditions that decide the working number. A measurement system whose spread consumes a large share of the tolerance will produce a CPK that says more about the instrument than about the process.

Model range: which IMAGE 3 fits this part

Model range, as stated in the IMAGE 3 series specification sheet. Repeatability here is the figure printed in the sheet; it is not a guarantee for your own part, your fixture or your operator. Treat it as the floor you start from when you run a trial, not as the number you can quote to a customer on day one.

ModelPrecision-mode range (X×Y×Z)Wide-field range (X×Y×Z)Lens field of view
(precision / wide-field)
Repeatability
(precision / wide-field)
CameraHeight measurement
IMAGE 318 × 18 mm (fixed platform)Ø100 mm18 × 18 mm / Ø100 mm±0.5 µm / ±1 µmTwo 1" 20 MP B/WNot available
IMAGE 3 PLUS118 × 118 × 75 mm200 × 200 × 75 mm18 × 18 mm / Ø100 mm±0.5 µm / ±1 µmTwo 1" 20 MP B/WNot available
IMAGE 3 PRO218 × 118 × 75 mm300 × 200 × 75 mm18 × 18 mm / Ø100 mm±0.5 µm / ±1 µmTwo 1" 20 MP B/WNot available
IMAGE 3 PRO-H218 × 118 × 75 mm300 × 200 × 75 mm18 × 18 mm / Ø100 mm±0.5 µm / ±1 µmTwo 1" 20 MP B/WNot available
IMAGE 3 MAX205 × 104 × 75 mm223 × 118 × 75 mm6.4 × 4.8 mm / 18.8 × 14.1 mm±0.25 µm / ±0.5 µm1" 20 MP B/W8055 spectral confocal, WD 55 mm, range ±3 mm
IMAGE 3 MAX2206 × 106 × 75 mm225 × 125 × 75 mm6 × 6 mm / 25 × 25 mm±0.25 µm / ±0.5 µm1" 20 MP B/W8055 spectral confocal (optional)

The full specification sheet, including the camera and lighting options and the workpiece-to-model mapping, sits on the IMAGE 3 series product page. If you only want a first-pass sanity check on whether your part fits any model, the instant measurement selection tool takes the part envelope and returns the models that can cover it.

Traceability: The Convention Is the Deliverable

Three conventions decide whether a record survives an audit. First, part identity: a serial, a lot plus cavity, or a timestamp plus position — whichever the customer can actually map back. Second, time source: one clock for the cell, not the operator workstation and the measuring PC independently. Third, immutability: a record that can be edited after the fact is not a record.

The practical test is to pick a part from three months ago and reconstruct its measurement. If that takes more than a few minutes, the convention is not working, regardless of what the software brochure says.

What the Reports Should Show, and What They Should Not

Useful: a run chart per gating feature with limits drawn, a capability summary with the study spread stated alongside, an exception list of out-of-limit parts with their identity, and a change log of recipe versions. Not useful: a single CPK number with no study behind it, a report that mixes features judged by different instruments, and any chart whose sample size is not stated.

The sample-size point deserves emphasis. A capability index computed from a handful of parts is dominated by sampling noise, and the honest presentation is to state the sample size next to the number every time.

Where the Loop Usually Breaks

  1. The measurement is exported as a text file and re-keyed by hand, which introduces an error path nobody owns.
  2. Part identity is written on a bag rather than carried in the record.
  3. Recipe versions change without a log, so a trend is actually two datasets.
  4. NG parts are removed from the dataset before the report, which hides the very events the chart exists to show.

A Realistic Rollout Order

Agree the field list. Then run the gauge study on the fixture. Then fix the identity and time conventions. Then connect the export. Then compute the first capability number. Reversing the last two steps — connecting first and studying later — is the most common way a project spends a year producing charts that nobody trusts.

FAQ

What data fields does an SPC report need from a vision measuring system?

Per feature: drawing name, nominal, upper and lower limit, measured value with unit, pass/fail state, and the recipe version. Per record: part identity, lot or cavity, timestamp from a single cell clock, and operator if the cell is manual.

Can CPK be calculated directly from flash measurement results?

Only after a gauge study on the actual fixture. The specification-sheet repeatability is a controlled-condition figure; the working spread depends on the fixture, the part surface and the operator. A capability index computed without that study describes the instrument rather than the process.

How many samples are needed before quoting a capability index?

Enough that sampling noise does not dominate the result, and the number used has to be stated next to the index every time. A capability figure presented without its sample size is not interpretable, which is why the report should carry both.

How should parts be identified for traceability?

With a convention the customer can map back: a serial, lot plus cavity, or timestamp plus tray position. The test is whether a part from three months ago can be reconstructed in a few minutes. Identity written on a bag rather than carried in the record fails that test.

Why do SPC trends drift after a lighting or recipe change?

Because the measurement recipe changed while the historical data was kept in the same series. Recording a recipe version per record makes that visible; without it, a trend chart can silently combine two different measurements.

Should out-of-limit parts be excluded from SPC reports?

No. Removing them hides the events the chart exists to show. The report should carry an exception list with the identity of each out-of-limit part, alongside the run chart that includes them.

Related reading

Send Us One Part and We Will Tell You Which Route Fits

The fastest way to close this question is a real part, not a discussion. Send one sample with the drawing and the features you actually gate, and we will run it and report what the measurement returns — including the features that come back as a no. A short written feasibility note is more useful than a brochure.

Go Back Top
VK Message
WhatsApp

Scan QR Code

WhatsApp QR Code
Wechat

Scan QR Code

Wechat
Phone Number
+8618598102007
Copied!
Online Message

Online Message

Click to refresh