Turning Light Direction Into Data on a Cast Block
An engine block is a large target: covering it needs a light source of the 400 mm class. Its appearance problems come in two typical forms. One is relief, such as a local dent or bump; the other is abnormal texture, such as uneven cast surface pattern. The two appear differently in an image. One shows as a brightness gradient, the other as the fine distribution of reflected detail, and one lighting direction struggles to explain both.
Bumps and texture were never one lighting problem
With a single direction, faces toward the light are bright and faces away are dark, so relief appears but texture detail is flattened by the overall brightness. Switch to soft, even light and texture becomes clear while relief goes flat. The scheme uses multi-directional time-division lighting: the source illuminates in sequence from different directions and several images are captured and analysed within a very short window, then combined into two images, one of relief form and one of texture pattern. This is not a post-processing trick; the two kinds of information are separated at capture.
Time-multiplexed capture and its cycle cost
More directions mean more shots, and the time cost is real. High-speed sequential capture keeps it inside the line cycle: at one workpiece position, several lighting sets fire in quick succession, the part does not have time to move, and the images share one moment. The cycle is 10 seconds, subject to site acceptance results, and it only holds if capture speed and light switching speed match. Whichever is slower sets the cycle.
A 400 mm source solves coverage, not brightness
If the source does not cover a large target like a block, defects in the dark zones simply cannot be imaged, whatever the brightness. The 400 mm source exists to light the whole target evenly and completely. The sequential source is a set of lights, and every direction has to meet the coverage requirement, not just the main one.
Why it tolerates ambient light and position error
Production lighting changes and workpieces sit slightly off position, and both shift the grey-level baseline of conventional single-source imaging. The redundancy of multi-directional capture does the work here: the same surface is captured repeatedly under several lighting sets, and the combined relief and texture images are insensitive to ambient change and small position error. The practical effect is that results do not depend on the workpiece deviation or the environment, and the maintenance burden of retuning parameters for one lighting condition disappears.
Where this is over-engineering
If the checks are mostly 2D marks such as barcodes, characters or flat position judgments, multi-directional time-division is overkill: more images and another light controller buy information nobody uses. Its value concentrates on cast parts with complex surface form where lighting direction is itself the criterion. The first question in selection is whether the defect is a colour problem or a shape problem.
The return shows up later
Stable imaging keeps appearance inspection consistent across shifts and operators. The return is rarely visible in the first year; it appears in long operation, when parameters do not need retuning for every lighting change and the criterion does not need re-confirming at every shift change. Inspection stops depending on a debugging state and starts depending on fixed imaging conditions.
What is worth the extra shots
Is this the same as photometric stereo?
The principle is related: surface form is inferred from brightness under different light directions. The engineering difference is whether it runs inline, inside the cycle, with a source large enough for the target.
Is 10 seconds enough for several captures?
Through high-speed capture and fast light switching, subject to site acceptance results, and dependent on part size, shot count and algorithm complexity.
How are the two images used?
The relief image judges dents and bumps, the texture image judges uneven surface pattern. Separate verdict logic on separate images is more stable than one image carrying both jobs.