Engine Block Surfaces: Holes, Machining Marks And Chips In One Inspection Pass

Overview

An engine block leaves machining with three kinds of problem on its surfaces: holes that should be there and are not, surfaces that should be smooth and are not, and chips that should have been washed away and were not. Each has a different consequence and each needs a different imaging approach, yet they are usually inspected in one place.

Big Bird Industrial (formerly Shimada Big Bird Industrial) has built cleaning and inspection systems for engine blocks since 2001. This article looks at why the three problems cannot share a single light, how the staging decision affects cycle time, and what an acceptance criterion for a surface inspection station should contain.

The unifying idea is that a block surface is not a single inspection task. It is a set of tasks that happen to be located on the same casting.

1. Missing And Extra Holes

Hole verification is a presence and position problem, and it is the easiest of the three to solve well. A backlit silhouette or a coaxial front light produces a clean edge, and a classical detector compares the pattern against the variant template.

The difficulty is not detection; it is variant handling. A block family may have the same casting with different drilled patterns, and a template for one variant will report false holes on another. The station must know the variant before it compares, which means identification is upstream of inspection, not beside it.

A second difficulty is the partially obscured hole: a chip or a burr sitting on the rim changes the silhouette subtly and can make a present hole look absent. The classifier should distinguish missing from obscured, because the corrective action is different.

2. Machining Marks And Surface Defects

A machined surface carries a texture that is normal: tool marks, a feed pattern, a slight colour variation from coolant. Defects are deviations from that texture, and the deviation is often small in contrast and large in shape.

Grazing light is the standard answer because it converts topography into contrast. A scratch catches light on one edge and shadows on the other; a dent shows a shadow on one side. Diffuse light flattens both and makes the defect nearly invisible.

The second standard answer is a second viewpoint. A defect that is invisible from directly above may be obvious at thirty degrees, and a station that only looks straight down will have a blind spot that no model can compensate for.

3. Chips: The Cleaning Interface

Chips on a machined surface are a cleaning defect, not a machining defect, and they are the most common finding on a block line. They are also the finding most likely to be misclassified, because a chip and a dent have similar size and similar contrast.

The reliable discriminator is height. A chip sits above the surface, a dent below it. A single 2D image cannot separate them, which is why chip detection benefits from either a low-angle light that casts a shadow behind the chip, or a 3D measurement that reports the sign of the deviation.

The other discriminator is location. Chips collect in pockets, at hole rims and along the parting line; dents occur where something struck the surface. A classifier that uses position as a feature performs better than one that uses appearance alone.

There is also a process point worth making: chip detection at inspection is a late signal. If the station finds chips consistently in the same place, the fix is upstream in the washer, usually a nozzle that has drifted or a fixture that shields a pocket.

4. One Pass Or Several

The temptation is to build one station that does everything in a single view. In practice a block needs at least three view types: a top view for the deck face, a side view for the flank surfaces, and a tilted view for the pockets and rims.

Four views on one part is normal. The design question is whether they are four fixed cameras, one robot with four stops, or a mix. Fixed cameras are faster and less flexible; a robot is the reverse. Most block lines end up with fixed cameras for the deck face and a robot for the flanks and pockets.

5. Acceptance Criteria And The Escape Rate

A surface inspection station is judged on escapes, not on detections. The specification should therefore be written around a defect set that includes real escapes, not only synthetic examples.

Per-class numbers are the important part. A station that catches every scratch and half the chips has a chip problem, and a single overall accuracy figure will hide it.

6. The Cleaning Loop

Surface inspection closes the loop with cleaning in a way that dimensional inspection does not. Chips found at inspection are direct evidence about the washer: which nozzles are working, which pockets are shielded, whether the filtration is loading up.

That makes the inspection data useful beyond pass and fail. A weekly map of chip locations, aggregated by position on the block, will point at the specific nozzle or fixture element responsible. Stations that log positions rather than counts get this for free; stations that log only a rejection rate do not.

7. Wet Versus Dry Parts

A block that arrives at inspection still wet behaves differently from one that has dried. A water film changes contrast, fills scratches and makes a chip look like a discolouration. The same threshold that works on a dry block will produce false rejects on a wet one.

There are two honest solutions. Either the part is dried before inspection, which costs energy and time but stabilises the image, or the inspection is designed for a wet surface with its own thresholds and its own baseline images. Mixing the two in one recipe is the mistake.

In practice most lines inspect after drying because the residue question and the surface question are then answered in the same state. Lines that inspect wet do so to save a handling step, and they pay for it in classifier tuning.

8. Verifying The Station

A surface station is verified with seeded defects, and the seeding has to be physical. Projected or synthetic defects test the algorithm, not the optics, the lighting or the fixture, and those three are where the real failures live.

A workable protocol uses a set of parts carrying known defects at known positions, cycled repeatedly across shifts, with results reported per defect class and per position on the block. Repetition matters because the interesting failures are intermittent: a reflection that only appears at one angle, a shadow that depends on how the part sits.

The protocol should also include good parts with challenging appearances: a stained surface, a coolant mark, a repaired area. Those are the parts that drive the false rate, and the false rate is what determines whether the line trusts the station.

Frequently Asked Questions

Can one camera see the deck face and the flanks?

Only at reduced resolution on the flanks, because the angle is oblique. Most lines use a dedicated top camera and a separate side arrangement.

How do you distinguish a chip from a dent?

By height sign, indirectly through shadow direction in 2D or directly in 3D. Appearance alone is not reliable enough for a process decision.

Does the block need to stop for inspection?

Usually yes for a full surface pass. In-motion inspection is possible for a single face but not for a set of views that includes tilted angles.

What if the coolant makes the surface look wet?

A wet film changes contrast and can create false edges. Either dry the part or accept a different threshold, but do not mix wet and dry parts in the same recipe.

Conclusion

A block surface is three inspection problems wearing one casting. Holes want silhouette, defects want grazing light, chips want height or shadow, and no single view serves all three.

The station that works is a small set of dedicated views with per-class acceptance criteria, tied back to the washer so that findings become process improvements rather than rejection statistics.