One Station, Five Variants: Identifying, Measuring And Inspecting Cylinder Heads On A Mixed Line

Overview

Mixed-model lines are normal in engine plants now. A single line runs three, five or seven cylinder head variants, and the inspection station has to know which one it is looking at before it can decide whether the part is good.

Big Bird Industrial (formerly Shimada Big Bird Industrial) has delivered measurement and defect inspection stations for cylinder heads since 2001. This article covers the three jobs a mixed-line station must do in one cycle — identify, measure, inspect — and the sequencing that keeps them from interfering.

1. Identify Before You Measure

The identification step is not a formality. A station that measures against the wrong variant's drawing will reject good parts and accept bad ones, and it will do it consistently enough that the failure looks like a process problem rather than a configuration problem.

There are three practical identification channels. A data-matrix code read from the part is the most reliable where the code is present and legible. A physical feature — a rib, a boss, a machining mark — is the fallback where the code is missing. A carrier or pallet identifier is the weakest option, because it depends on the part staying with its fixture.

The design rule is to read the code and verify it against a physical feature. If the two disagree, the station should stop rather than pick one. A part whose identity is ambiguous is exactly the part that should not be measured.

2. Marking Is A Process Step, Not Decoration

Where the line requires a laser mark or a dot-peen code, the mark is a process feature with its own quality criteria: contrast, position, character height and legibility after paint. Inspecting the mark is part of the station's job, and the mark quality often degrades before the part does.

Two failure modes recur. A mark placed on a machined surface can be removed by a later operation, and a mark placed on a casting surface can be unreadable from the start. Both are caught by verifying the mark in the same image in which it is read, not by a separate downstream check.

3. The Measurement Subset That Matters

A cylinder head has dozens of dimensions and a production station cannot measure all of them in takt. The useful approach is to select the features that (a) are functionally critical and (b) drift with tool wear.

Everything else belongs in an offline audit, not in the station. The station exists to catch drift, and drift shows up first in the features that the tool touches most.

4. Defects: The Part That Needs Its Own Lighting

Surface defects on a head — sand holes, cold shuts, machining marks, scratches, residual chips — need a different illumination strategy from dimensional measurement. Dimensional work prefers a clean, controlled edge; defect work prefers grazing light that reveals texture and topography.

That is the practical argument for separate imaging stations within one cycle: one view for measurement, one for texture. Trying to serve both with a single light usually produces measurements that are slightly soft and defect images that are slightly flat, and both suffer.

Defect classification has to be variant-aware as well. A casting feature that is normal on one variant can look like a defect on another, and a model trained on a single variant will misclassify the moment the mix changes.

5. Sequencing And Takt

The order that works is identify, then measure, then inspect, with identification gating the other two. Measurement can start while the defect views are still being acquired, and the variant decision must complete before either result is evaluated.

On a mixed line, the takt problem is not the average part; it is the slowest variant. A station designed for the average will miss takt every time the complex variant appears, so the design basis should be the worst case with a margin.

6. Acceptance And Change Control

Mixed-line stations live or die on change control. A new variant added without a new recipe is a silent failure; a recipe changed without re-verification is a louder one.

7. Fixture Commonality Across Variants

The camera can be reprogrammed in an hour; the fixture cannot. This is the practical limit on how many variants a single station can serve, and it is usually discovered late, during commissioning, when the second variant arrives and does not sit in the first variant's nest.

The design rule is to locate on features that all variants share, typically the two main bearing bores or the deck face plus two dowel holes, and to let the variant-specific features be measured rather than located. A fixture that locates on a variant-specific boss will need a new nest for every model.

Where common features do not exist, the compromise is a family of interchangeable nests on a common base. That keeps the optical system and the calibration intact while the mechanical interface changes, which is a far cheaper form of flexibility than a second station.

8. What To Do When A Variant Is Missing

Mixed lines occasionally present a part whose identity cannot be resolved: a code that did not mark, a feature that is obscured, a carrier that arrived without its pallet record. The station's behaviour in that case is a specification item, not an implementation detail.

The wrong behaviour is to assume the most common variant. That produces measurements against the wrong drawing and a rejection stream that looks like a quality problem. The right behaviour is to stop, flag, and route the part to a manual decision point with the image attached.

A useful middle path is a quarantine lane: the part continues physically but is held logically, and the line keeps running while a person resolves the identity. That keeps takt without letting an unverified part into the flow.

9. Measuring The Station Itself

A mixed-line station has a failure mode that single-variant stations do not: it can be correct for every variant and still be wrong, because the variant switching logic is wrong. Verifying the optics does not verify the switching.

The practical check is a rotation test: run every variant in sequence, several times, and confirm that the recipe applied matches the part presented. A mis-identification that happens once in two hundred parts is invisible in production and obvious in a rotation test.

The same test exposes stale recipes. A variant whose drawing was revised but whose recipe was not will measure against the old nominal, and the error appears as a bias that no calibration can remove.

Frequently Asked Questions

What if the data-matrix code is unreadable?

Fall back to a physical feature and flag the part. Do not fall back to the pallet identity alone, because that is exactly how a wrong-variant measurement passes unnoticed.

Can one camera do measurement and defect inspection?

It can do both badly. Measurement wants controlled edges, defects want grazing texture. Two views with two light settings in the same station is the usual compromise.

How many variants can a station handle?

There is no hard limit on recipes; the limit is on fixture compatibility. If a variant needs a different locating scheme, it needs a different fixture, and that is a mechanical constraint rather than a software one.

Does a new variant require retraining the defect model?

It requires re-verification. Whether retraining is needed depends on whether the new variant introduces features the model has not seen; verification on a seeded defect set answers that question.

Conclusion

A mixed-line head station is a configuration problem before it is an optical one. Identify first, measure a small functional subset, inspect with dedicated lighting, and control the variant recipes like drawings. Get those four right and the station survives model changes; get them wrong and every new variant is a project.