Placing machine visions targets with high accuracy

My company is working on a machine vision application. One of our requirements is that the camera peer through a frame (like a picture frame - see here ), where the top and bottom border are etched with machine vision targets. The frame is mounted on the gray rectangle (one side of a box) in the photo. The frame is 1/2" aluminum and about 14" long and between 3" and 8" high. The targets are laser etched. The top and bottom targets are parallel. For various reasons, we'd like to vary the gap between the targets. Our original thought is that we'd need to create a difference size frame (see picture below - red frame)

One of our engineers suggested that instead of building one frame for each size, we could make a vertical slot on either size of the box and position two rows of targets (actually two long pieces of aluminum) accordingly, somehow fastening them in the slot. We could use a metal spacer, measured to a particular gap to insure that the top and bottom row of targets are correctly oriented. Correct orientation means that the targets rows are parallel and are spaced to a known distance. My question is as follows. Is it possible to achieve high tolerances (e.g. 1/100mm) as it relates to orientation using the second approach? How would we do it?

Disclaimer: I'm not a mechanical engineer. Apologies if I'm not including important information (or if I've included irrelevant information).

• This question would be greatly clarified by a diagram. Feb 25, 2016 at 9:54
• What is the actual resolving power of the camera? 0.01mm seems unrealistic -- that would be on the order of 32k pixels across a 14" (355.6mm) span. 4k pixels would have a resolving power of about 0.1mm at that scale. Feb 25, 2016 at 16:40
• How are you going to register the frame to the target objects, and register the angular as well as spatial alignment of the camera? This is one of those cases where you appear to be asking how to build a specific method, rather than asking how the root problem should be solved. So, what exactly do you need to measure and why isn't it sufficient to calibrate the camera's pixel -to-object plane relationship? Feb 26, 2016 at 17:22

0.01 of a mm is pretty tight by any standards, for example a typical jobing lathe might be graduated in divisions of 0.025mm. So 0.01mm is not impossible but certainly not trivial.

This level of accuracy is also getting into the realms where thermal expansion and cleanliness of surfaces become significant factors. To put this into perspective a 200mm length of aluminium will expand by around 0.05mm between 10 and 30 deg C.

A sensible approach would be to build the structure upwards from a very flat base, something like a calibrated surface plate using precisely machined spacers, probably cylindrical columns which you would need to be made to length for each gap spacing you want.

An alternative approach would be to use parallel high accuracy ball screws to adjust the gap, similar to the way the saddle on a lathe travels.

This method would be rather more complex than using calibrated spacers but gives you a continuous range of adjustment which allows you to manually compensate for things like thermal expansion and wear.

I would also question whether aluminium is the best material for this level of accuracy. Something like precision ground tool steel would give much better stiffness and dimensional accuracy as an off the shelf product. Steel is typically 3 times as stiff as aluminium alloys and typically has much better hardness and machinability.

In terms of practical implementation a good machinist should be able to provide you with a work around solution if you can tell them exactly what you need. A general purpose and 'foolproof' solution will probably require a bit more investment in design and manufacture.