You need to devise a method for the cmm to locally and temporarily 'find' the same alignment feature points on each part, every time.
Your variation is a result of up to 3 things.
1: Parts really are out of spec.
2: Your alignment is not robust.
3: Your dimension strategy is not robust.
Pursue one at a time.
1: I would get on a surface plate or a comparator and figure out the distance manually. If you use a sine plate and level out top plane, what's your length (with Pythagorean theorem applied)
2: If you have an analog scanning probe, like others have suggested, you could take a few hits on one of the planes, level/origin to the plane, then take a 4 hits spaced 90° apart from each other to generate a temporary centroid origin (construct 2 3d lines between each opposing point, then construct an intersect of the 2 3d lines for your origin).
Align/origin your level plane, then your origin (no rotation datum yet).
--Then, with this initial level plane and an XYZ origin, scan the perimeter of the part at a depth along the axis, which you feel is
most consistently formed.
--With this scan, rotate and origin about it to match your part. THEN re-measure your first plane with a scan (or series of scans).
Construct another alignment with the new plane and perimeter scan, to lock it down.
-Repeat on opposing side of part, entirely.
3: Your length can vary depending upon your centroid axis. The means to dimension can also vary wildly, depending upon how the print is defining the length. Communicate with customer or engineer to get a clear understanding of the dimension, and apply the strategy accordingly. Whatever you do, DO NOT dimension centroid of plane 1 to centroid of plane 2. You need to produce an axis line between your perimeter scans, and pierce the planes to get your distance consistently.
You need to devise a method for the cmm to locally and temporarily 'find' the same alignment feature points on each part, every time.
Your variation is a result of up to 3 things.
1: Parts really are out of spec.
2: Your alignment is not robust.
3: Your dimension strategy is not robust.
Pursue one at a time.
1: I would get on a surface plate or a comparator and figure out the distance manually. If you use a sine plate and level out top plane, what's your length (with Pythagorean theorem applied)
2: If you have an analog scanning probe, like others have suggested, you could take a few hits on one of the planes, level/origin to the plane, then take a 4 hits spaced 90° apart from each other to generate a temporary centroid origin (construct 2 3d lines between each opposing point, then construct an intersect of the 2 3d lines for your origin).
Align/origin your level plane, then your origin (no rotation datum yet).
--Then, with this initial level plane and an XYZ origin, scan the perimeter of the part at a depth along the axis, which you feel is
most consistently formed.
--With this scan, rotate and origin about it to match your part. THEN re-measure your first plane with a scan (or series of scans).
Construct another alignment with the new plane and perimeter scan, to lock it down.
-Repeat on opposing side of part, entirely.
3: Your length can vary depending upon your centroid axis. The means to dimension can also vary wildly, depending upon how the print is defining the length. Communicate with customer or engineer to get a clear understanding of the dimension, and apply the strategy accordingly. Whatever you do, DO NOT dimension centroid of plane 1 to centroid of plane 2. You need to produce an axis line between your perimeter scans, and pierce the planes to get your distance consistently.
