Repeatability of RP systems
From:
Ulrich Laible (Clemson University)
Date:
Thursday, April 13, 1995
From: Ulrich Laible (Clemson University)
To: RP-ML
Date: Thursday, April 13, 1995
Subject: Repeatability of RP systems
Hey,
I would like to get some views about something what I think is an
important issue for every RP-system: The REPEATABILITY of the system
I think there isn't a clear understanding what this means and how the
repeatability of a system should be measured. Any system which is not
accurate, but repeatable, can be corrected in some way to reach the
desired accuracy. But a system which is not accurate and not repeatable
seems impossible to be controlled.
A lot of studies mention some values for repeatability of the system.
But in most of the cases a mean-value or a standard deviation of a
variety of measurements was taken as basis for the repeatability-value.
In reviewing a number of studies I found at least 3 different ways to
get a value for repeatability:
1. Built a number of parts and take many measurements in x, y, and z
direction of each part (for different dimensions), then draw the errors
of all measurements on all parts together in a histogram and use the
distribution as a measurement for repeatability of the process
2. Built one part, take a number of measurements in any direction, group
the measurements in different dimension-groups, calculate the RMS
(standard deviation) for each dimension group and use this as a
measurement for repeatability. (f.e. all 5" dimensions have a standard
deviation of 0.01"). Actually this means the repeatability, how one
certain dimension is repeated in the same part.
3. Built a number of same parts in one built on the same platform, take
measurements in all directions, calculate the standard deviation for
measurements in x, y and z direction and compare the standard deviations
for the different parts (separate for each direction).
There are two things I would criticize on these procedures:
- comparing the standard deviation of several different measurements can
not be a proper way to measure repeatability. Let assume I have two
parts with two dimensions on each. Part #1 comes out with dimension1 to
short, dimension2 correct, part #2 comes out with dimension1 correct and
dimension2 to short, the standard deviation of both parts could be
almost the same, what refers to the assumption that the parts are
repeatable, but they are not, because they are not the same at all....
- Taking a lot of measurements on a complicated testpart always includes
a lot of interaction between the measurements. Therefore the error-
histogram as well as the standard deviation of all measurements together
reflects this interaction. Knowing that with a probability of 99.7% an
error will be under 0.007" don't tells me how repeatable is one certain
build-defect which causes an error on one certain dimension.
I think the only way to get a real value for the repeatability of a
process is to focus on one dimension, build a (not to small) number of
parts (after the same recipe) and observe the behavior of the error of
this dimension.
I would appreciate every comment or addition to this topic. I hope this
can help to get a better understanding of the repeatability of a RP system.
ULRICH LAIBLE |
Department of Mechanical Engineering | e-mail: ulaible@eng.clemson.edu
Box 340921 |
318 Riggs Hall | FAX: 803-656-4435
Clemson University |
Clemson, S.C. 29634-0921 |
USA |
Previous message
| Next message
Back to 1995 index