Holding 0.02 mm on a 100 mm Casting With Telecentric Optics
The machined features on rail castings run beyond 100 mm, and an ordinary lens struggles to keep both sharpness and accuracy across a field that size. This project builds a measuring platform from a smart camera, a bi-telecentric lens and parallel light, solving resolution, distortion, shadow and edge blur at the optical level. Accuracy is 0.02 mm and the cycle is 45 seconds, after which the part moves on to the next operation, subject to site acceptance results.
Above 100 mm, an ordinary lens loses magnification first
As the field grows, each pixel covers more of the part and resolution falls. Worse, an ordinary lens has different magnification at the centre and at the edge: the same dimension measures a different number of pixels depending on where it sits in the frame. Distortion is not the picture looking bent; it is the measured value quietly shifting.
What bi-telecentric actually fixes
The entrance and exit pupils sit at infinity, chief rays run parallel to the axis, and magnification does not change with object distance. A workpiece moving within the depth of field images at constant magnification, and edge and centre share the same scale, so a measurement does not depend on where it lands. That is the substantive difference from an ordinary lens, and it is why the optic, not the algorithm, is the first decision on this kind of part.
Parallel light removes shadow and false edges
Under diffuse light a part edge carries shadow and blur, and a rough cast surface produces false edges that subpixel extraction mistakes for real contour. Parallel light arrives evenly and parallel, giving high edge contrast and a clean contour, and suppressing false edges on a surface that is not flat. The light is part of the measurement, not an accessory to brightness.
0.02 mm rests on three foundations
First calibration: a precise mapping between pixels and real dimensions, with working distance and field parameters fixed by it. Second the edge: a sharp contour under parallel light plus subpixel extraction. Third depth of field and fixturing: the telecentric path holds magnification within its depth of field, and the locating fixture constrains placement height inside that range so repeated measurement stays stable. Remove any one and the accuracy does not land.
45 seconds is the station cycle, not the algorithm
Placement, fixturing, triggered imaging, edge extraction, dimension calculation and verdict all complete inside 45 seconds, after which the casting moves on. The point of the cycle is that the station is not the line's bottleneck, not that the algorithm is fast.
Limits: field size is bound to lens aperture
A bi-telecentric field is limited by its aperture, so a very large part needs zoned measurement or stitching. Where thickness varies a lot, check that the depth of field covers the feature being measured, and image per measurement face if needed. The station should also control ambient light, since stray light added to the parallel beam degrades edge consistency.
Questions at selection
Why is an ordinary lens not enough above 100 mm?
Each pixel covers more real distance and resolution drops, while distortion and edge shadow add measurement error. A bi-telecentric lens on a large sensor keeps resolution across a bigger field.
Why does bi-telecentric remove distortion?
Magnification is constant with object distance, so centre and edge agree. What it removes is measurement error from position-dependent scale, not just visual bending.
Is there a limit on workpiece thickness?
Yes. Depth of field is finite and the measured feature must sit inside it. Where thickness varies, assess coverage or image in zones.