Computational Metrology for the

Design and Manufacture of

Product Geometry: A
Vijay Srinivasan Classification and Synthesis
Member ASME
IBM Corporation and Columbia University, The increasing use of advanced measurement tools and technology in industry over the
New York, NY past 30 years has ushered in a new set of challenging computational problems. These
and problems can be broadly classified as fitting and filtering of discrete geometric data
University of North Carolina at Charlotte, collected by measurements made on manufactured products. Collectively, they define the
Charlotte, NC field of computational metrology for the design specification, production, and verification
e-mail: [email protected] of product geometry. The fitting problems can be posed and solved as optimization prob-
lems; they involve both continuous and combinatorial optimization problems. The filter-
ing problems can be unified under convolution problems, which include convolutions of
functions as well as convolutions of sets. This paper presents the status of research and
standardization efforts in computational metrology, with an emphasis on its classification
and synthesis. 关DOI: 10.1115/1.2424246兴

Keywords: metrology, standards, fitting, filtering, optimization, convolution, tolerancing,

specification, production, verification, design, manufacturing

1 Introduction manufactured surfaces than smooth surfaces. So, at least

conceptually, a manufactured surface can be modeled as a
The past decade has seen the emergence of computational me-
fractal set and the only information we can obtain about it
trology as a separate discipline in computer-aided design and
comes from a discrete set of points sampled on that surface.
manufacturing. It deals with fitting and filtering discrete geometric
2. Axiom of measurement uncertainty: No measurement can be
data that are obtained by measurements made on manufactured
parts 关1兴. Thus, it plays an important role in manufacturing indus- absolutely accurate and with every measurement there is
try. A manufactured part may be measured to characterize the some finite uncertainty about the measured attribute or mea-
manufacturing process that produced it or to assess the conform- sured value 关5,6兴.
ance of the part to designer-specified tolerances. In either case, The measurement uncertainty axiom adds a sobering real-
measurements are made on the surface, or the volume, of the part ization to our computational endeavor. If we take the dis-
and the measured data are then reduced to a few numbers or crete set of measurements as input to our computations, it is
attributes by increasingly sophisticated computational techniques. good to remember that these input values can never be taken
Computational metrology, defined here as fitting and filtering of as absolutely accurate. It is equally important to remember
discrete geometric data, is a subset of coordinate and surface me- that results of our computations should be accompanied by
trology, which is a further subset of dimensional and geometric statements about their uncertainty, which is partly inherited
metrology practiced extensively in industry. It is in this sense the from the input uncertainty and partly attributable to the com-
phrase “computational metrology” was first coined by the author putational scheme itself.
in the early 1990s 关2,3兴. Metrology itself is the art and science of
It should be noted that these two axioms are independent. It is
measurements; it is practiced in all walks of life—in science,
possible to build a modeling framework where manufacturing is
commerce, industry, medicine, and agriculture just to name a few.
imprecise but measurements have negligible uncertainty, and vice
Before we proceed further, it is useful to establish some basic
versa. But to model reality more closely, both axioms should be
facts about manufacturing and measurement. These are best cap-
considered operative. Having established the fundamental axioms
tured in the following two axioms:
of manufacturing imprecision and measurement uncertainty, we
1. Axiom of manufacturing imprecision: All manufacturing pro- then ask how they impact product design specification, produc-
cesses are inherently imprecise and produce parts that vary. tion, and verification.
In fact, this axiom of manufacturing imprecision contains The inevitable variability in manufactured product geometry
a hidden primal fact that no man-made artifact has Platonic and the unavoidable uncertainty in its measurement are acknowl-
ideal form 关4兴. That is, no manufactured object can be per- edged up front and accommodated in the design specification of a
fectly cylindrical or perfectly spherical and so on. There is product using tolerances. The systematic use of tolerances in
increasing experimental evidence that the geometry of product geometric specification is relatively new. Government
manufactured surface behaves more like a fractal in the agencies in various countries mandated their use by issuing mili-
range of dimensional scale that is of interest in engineering. tary standards during and in the aftermath of the world wars of the
This means that fractals offer a better approximation to 20th century. The task of issuing and maintaining tolerancing
standards is now left to civilian control—in the United States, the
responsibility rests with the ASME 关7兴 and internationally it rests
Contributed by the Computer-Aided Product Development 共CAPD兲 Committee of
ASME for publication in the JOURNAL OF COMPUTING AND INFORMATION SCIENCE IN
with the ISO 关8兴.
ENGINEERING. Manuscript received October 6, 2006; final manuscript received May While tolerance specifications encode the allowable variation in
26, 2006. Guest Editors: Anath Fischer and Raffaello Levi. product geometry, how is it handled in production? There are at

Journal of Computing and Information Science in Engineering MARCH 2007, Vol. 7 / 3

Copyright © 2007 by ASME

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 discrete set of points sampled on a manufactured surface兲. Engi-
neers are interested in fitting for the following reasons:
• Datum establishment: Datum is a reference geometric object
of ideal form established on one or more nonideal geometric
forms on a manufactured part. Datums are used for relative
positioning of geometric objects in parts and assemblies of
parts. Figure 1 shows several datums indicated on a single
part. These datums have been used in various combinations
to specify perpendicularity, position, and runout of various
features in this part relative to other features in the same
part.
• Deviation assessment: It is often important to determine
how far a manufactured surface has deviated from its in-
tended ideal geometric form. Figure 1 shows an example of
flatness tolerance; it specifies how much a surface can devi-
ate from an ideal plane. Figure 2 explains how flatness tol-
erance can be specified and interpreted. As we will see soon,
this deviation can be quantified by fitting.

Historically such fitting was accomplished by the use of surface

Fig. 1 Example of an industrial drawing with dimensioning plates, collates, mandrels, and specialized measurement fixtures.
and tolerancing annotations on projected views of a part. More recently, manufacturing industry has started using modern
„courtesy Archie Anderson.… All dimensions are in millimeters.
measurement devices such as coordinate measuring machines
共CMM兲 and optical scanners. This has accelerated the use of fit-
ting by computation. Its initial success is placing increasingly
least two distinct ways in which manufacturing industry deals complex demands on our ability to compute.
with variability in production. The first is the assessment of manu-
facturing process capability, which is undertaken even before the 2.1 Deviation Assessment. Let us first look at the role of
specification of any product is received. The second is the manu- fitting in deviation assessment using a flatness tolerance example.
facturing process control, which monitors whether the manufac- Figure 2 shows how a flatness tolerance on a nominally flat fea-
turing process is producing parts whose variations are within the ture can be specified and interpreted. This figure illustrates the
capability of the process. In both cases, the produced parts are syntax and the semantics of flatness tolerance using an ISO stan-
measured and analyzed. dardized graphical language. ASME defines flatness tolerance
Finally, the manufactured parts are inspected to check if they similarly. How is such a specification of allowable variation in
meet the specified tolerances. This is also known as geometric form—here, how far a manufactured surface is allowed to deviate
verification of the product. Sometimes 100% of the produced parts from a mathematical plane—related to fitting? To answer this
are inspected. More often, especially in mass production, only a question, we need to consider some simple ideas from statistics on
carefully sampled set of parts are inspected. In either case, the how dispersion can be defined.
selected parts are measured and their conformance to specification Recall that if we have a set of measured data, then there are at
is verified. least two measures of its variability, which is also known as dis-
With these preliminaries, we can now proceed to address fitting persion. One is the standard deviation 共or some constant multiple
and filtering problems in the rest of the paper. Section 2 deals with of it兲, which can be obtained as the square root of the mean of the
a description of fitting problems that arise in product design and squares of the differences of each data from the mean of the data.
manufacturing, and Sec. 3 explains how these problems can be This is also the root mean square 共RMS兲 value of the deviation.
posed and solved as optimization problems. The problem of fil- The other is the range, which is the difference between the largest
tering is described in Sec. 4, which is followed by a unified math- and the smallest value in the data set. Both of these measures have
ematical treatment of filtering as convolution problems in Sec. 5. been used in applications of univariate statistics.
Section 6 summarizes and concludes the main body of the paper. To apply these simple ideas to the case of quantifying out-of-
flatness, let us assume that we can fit a mathematical plane to a set
of points measured on a nominally flat surface patch on a manu-
2 Fitting factured part. Since these points are not, in general, coplanar, we
We observed in the last section that no manufactured surface can find the perpendicular distance of each point from the fitted
can have the Platonic ideal form. Fitting is the task of associating plane and then compute the standard deviation or the range of
ideal geometric forms to nonideal forms 共such as, for example, these distances. Since these measures of deviation are dependent

Fig. 2 ISO definition of flatness tolerance †8‡. „a… The tolerance zone is limited by
two parallel planes a distance t apart. „b… The extracted „actual… surface shall be
contained between two parallel planes 0.08 units apart. No datum is needed. All
dimensions are in millimeters.

4 / Vol. 7, MARCH 2007 Transactions of the ASME

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 3 A comprehensive list of geometric tolerances defined by ISO †8‡

on the plane we fit, we can choose the fitted plane such that the only form tolerance in use. Figure 3 shows a comprehensive list-
standard deviation is minimized, or the range is minimized. The ing of geometric tolerances defined by the latest ISO standard 关8兴.
first will yield a plane that gives us an unbiased estimation of the These are the same as the ones defined by ASME. As the first
standard deviation and the second will yield a plane that gives us section in the list indicates, there are other types of form toler-
an unbiased estimation of the range. It is quite straightforward to ances. The general remarks we made above about flatness toler-
see that a plane so chosen that it minimizes the maximum distance ance apply to the rest of form tolerances as well. They can be
of the points from it minimizes the range, and this value for the assessed using fitting, which can be mathematically posed as op-
range can be used to check if it is less than the flatness tolerance timization problems.
specified per Fig. 2.
What we see from this example is the fact that current standard- 2.2 Datum Establishment. A quick look at the last column of
ized form tolerances use the measure of range to quantify allow- Fig. 3 reveals that datums are needed for most geometric toler-
able variation from Platonic ideal form. One may ask why the ances except for form. Datums are involved when we want to
range measure was chosen instead of, say, the standard deviation describe relative positions of geometric objects and allowable
共or, some constant multiple of standard deviation兲. The reason is variations in these relative positions. The role of datum can be
historic, as it is relatively easy to measure the range of variation illustrated using a simple example of parallelism, defined by ISO
using instruments such as a dial indicator. As more advanced mea- as in Fig. 4. 共ASME defines parallelism similarly.兲 If two nomi-
surement technologies become prevalent, it is time to re-examine nally flat surface patches are designed to be parallel, their realiza-
the definition of form tolerancing to see if other measures of tion in a manufactured part will not be exactly parallel. We then
variation can also be specified. In any case, we have established need to indicate how much deviation from parallelism we are
the fact that the fitting problem can be posed as an optimization willing to tolerate.
problem. We will explore this in some detail in Sec. 3. One obvious way to assess the out-of-parallelism between two
Before we close this section, we note that flatness is not the nominally flat surface patches is to fit a plane to points measured

Journal of Computing and Information Science in Engineering MARCH 2007, Vol. 7 / 5

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig. 4 ISO definition of parallelism tolerance †8‡. „a… The tolerance zone is limited
by two parallel planes a distance t apart and parallel to the datum plane. „b… The
extracted „actual… surface shall be contained between two parallel planes 0.01
units apart which are parallel to datum plane D. Clearly, a datum is needed here.
All dimensions are in millimeters.

on each of the manufactured surface patches, and then to check • Least-squares fitting: Here the objective is to find an ideal
what is the angle between these fitted planes. These planes should geometric object 共a smooth curve or surface兲 that minimizes
be fitted independently and the criterion used for fitting can be the sum of squared deviations of data points from this object
based on least squares or minimizing the maximum deviation, 关11,12兴. It includes linear least squares, total least squares,
depending on what makes the best sense from the design objec- and non-linear least squares techniques. These are continu-
tive. This, however, is not the standardized way to specify paral- ous optimization problems.
lelism. As Fig. 4 shows, one of the surface patches indicated as • Chebyshev fitting: Here the objective is to minimize the
datum D is chosen to establish a reference plane 共either by placing maximum deviation 关13兴. Some of these fitting problems
it on a surface plate or by fitting a plane to points measured on it兲 have been studied by discrete and computational geometers
and then demanding that points on the other surface patch should in the last 20 years. This has added some valuable insight in
be contained in the tolerance zone that is parallel to the datum designing algorithms to solve such problems. These are
共reference兲 plane. largely combinatorial optimization problems.
It is instructive to ask why standards have been using only such
a zone-based approach to define tolerances in the relative posi- National and international standards groups are actively working
tions of geometric objects. It is again due to historic reasons, on standardized definitions for the objective functions and con-
because specifications such as the one shown in Fig. 4 can be straints for the fitting problems. 共In the ISO terminology, fitting is
verified on actual parts using surface plates and dial indicators known as association 关14兴.兲 To get a feel for the types of optimi-
whereas fitting planes to measured points had to await the avail- zation problems they tackle, it is instructive to take one example
ability of CMMs 共coordinate measuring machines兲 and computers. and examine it in some detail.
Here again, we may want to reexamine the standards to allow Table 1 shows several plane fitting problems as optimization
alternate means of specifying tolerances on relative positions. problems, along with the objective functions and applicable con-
In any case, we see that fitting plays an important role for straints. If d1 , d2 , . . . , dn are the perpendicular 共Euclidean兲 dis-
datum establishment. Before we leave this section on fitting, a tances of n input data points from a plane P, then we can define
word of caution is in order. As Figs. 2 and 4 indicate, the toler- the distance between this set of points and the plane P using the
ances may be indicated on two-dimensional projected views but generic l p norm
their interpretations are always on three-dimensional parts. More
recently, even the tolerance specifications have been indicated on
three-dimensional geometric models, as shown in Fig. 5. Many
computer-aided design 共CAD兲 systems have been doing this for a
while and now there are ASME 关9兴 and ISO 关10兴 standards that
formalize such practices, as shown in Fig. 6.

3 Optimization
The computational scheme used for fitting is one of optimiza-
tion. As we saw in the last section, it may be of interest to fit a
plane to a set of points in space such that the sum of the squares
of the perpendicular distances of the points from the plane is
minimized. This is the total least-squares fitting problem. Such
problems have been studied in science over the past two centuries
and we can draw from this wealth of knowledge to find satisfac-
tory solutions. But there are also other seemingly simple fitting
problems that can tax our computational skills. For example, an
engineer may want to find the smallest cylinder 共that is, a cylinder
with the smallest diameter兲 that encloses a set of points in space
because it gives him some quantitative information about how a
part will fit in an assembly. This can be easily posed as a minimi-
zation problem, but computational methods to solve this problem
are not simple.
It is customary to divide the fitting problems broadly into the Fig. 5 Example of dimensioning and tolerancing an industrial
following two categories on the basis of the objective function part in a three-dimensional geometric model. „Courtesy Das-
that is optimized. sault Systemes.… All dimensions are in millimeters.

6 / Vol. 7, MARCH 2007 Transactions of the ASME

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 using an elegant algorithm that involves singular value decompo-
sition 关11,12兴. The Plane⬁ problem has received much attention in
the computational geometry literature 关15兴. This is closely related
to the topic of convex hulls and the width of a set. In contrast, the
Plane1C problem has received only scant attention. Since this
plane assures at least three points of contact, this is like to be the
preferred plane for fitting a datum plane.
A table similar to Table 1 can be defined for fitting each of the
other geometric objects such as straight lines, circles, cylinders,
spheres, cones, tori, and free-form curves and surfaces. Some of
these problems have been solved to varying degrees of satisfac-
tion 关11–13兴. But the problems are far from being solved com-
pletely, especially involving the measurement uncertainties in the
input. As we conclude this section on optimization, it may be of
interest to educators to note that fitting is one of the most direct
and practical means of educating engineering students in optimi-
zation and computational geometry.

4 Filtering
Filtering is the task of obtaining scale-dependent information
from measured data. At a more mundane level, filtering can be
used to remove noise and other unwanted information from the
measured data. In the context of engineering metrology, engineers
are interested in filtering mainly for the following two reasons.
Fig. 6 Example of standardized indications of dimensioning • Surface roughness: Many engineering functions depend on
and tolerancing an industrial part in a three-dimensional geo-
how rough or smooth a piece of surface is. Designers define
metric model. All dimensions are in millimeters.
bounds on certain roughness parameters obtained by obser-

再兺 冎
vation on a rather small scale to ensure functionality of
n 1/p parts. These small-scale variations have to be separated
兩di兩 p
共1兲 from medium and large scale variations before roughness,
i=1
waviness and form deviations are assessed.
• Manufacturing process diagnosis: Manufacturing processes
The individual distances can be signed, in the sense that points leave tool marks on surfaces. By measuring surfaces at fine
lying on one side of the plane can be assigned positive distances scale, it is possible to track the tool erosion and its effect on
and the points lying on the other side can be assigned negative the surface quality.
distances. The l1 norm is then just the sum of the absolute values
of the individual distances. The l2 norm is the square root of the In simple terms, the surface variations are separated into rough-
sum of the squares of the distances. The l⬁ norm is the maximum ness, waviness, and form deviations. These correspond to small,
of the absolute values of the distances; to see this we need to look medium, and large scale variations of the surface 关16,17兴.
at the definition of the l p norm as p tends to infinity. The objective of filtering is to decompose measured data into
Some of the problems in Table 1 are worth remarking. The different scales and subject them to further analysis. Some of the
Plane2 problem is the most celebrated problem of finding the total earliest filters used in industry were known as mean-line filters
least squares 共also known as orthogonal regression兲 plane. This because the output they produced is smoother than the input and
problem has been solved by numerical analysts and statisticians, the output profile ran through the input profile. These mean lines

Table 1 Several optimization problems stemming from fitting planes

Objective
function to be
Output Input optimized Constraints Comments Designation

Minimize l1 X lies to Minimum three- Plane1C

norm one side of points
P touching
plane
None Least squares Plane2
plane
Minimize l2 X lies to Constrained Plane2C
A set X of n norm
points one side of least squares
P plane
A plane None Minimax Plane⬁
P plane
Minimize l⬁ X lies to Constrained Plane⬁C
norm one side minimax
of P plane
Linearly Maximize the P linearly Maximin PlaneSep
separable sets minimum separates plane
X1 and X2 distance of X1 and X2
points in X1 艛 X2
from P

Journal of Computing and Information Science in Engineering MARCH 2007, Vol. 7 / 7

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Table 2 Convolutions of functions and sets

Convolution of functions Convolution of sets

y共x兲 = 兰−⬁
+⬁
z共s兲K共x − s兲ds C = A 丣 B = 兵a + b : a 苸 A , b 苸 B其.
z共x兲: Input function A: Input set
K共x − s兲: Kernel function that is symmetric B: Structuring element, symmetric about the origin
y共x兲: Output function C: Output set
y is the moving average of z weighted by the kernel C is the translational sweep of B by positioning on A

can be used as the reference about which small scale variations shows how the underlying mechanical surface can be computed
contained in the input can be assessed. This was followed by from such a data. This is called an erosion filter, in the terminol-
development of envelope filters, whose output is smoother than ogy of morphological filters 关19兴. The effect of another morpho-
the input but the output profile enveloped 共that is, rode over兲 the logical filter, called closing filter, is shown in Fig. 9. This is a true
input profile. These envelopes are better correlated with mechani- envelope filter—the output is an upper envelope of the input. It
cal functional behavior of such profiles. The mean-line filters and rides on the peaks and smoothes out the valleys at a chosen scale.
envelope filters were developed and promoted by different re-
search and industrial group. 6 Summary and Concluding Remarks
Historically, filtering techniques were pioneered by communi-
cation theorists. Developments in analog and digital signal pro- In this paper we saw that advances in measurement tools and
cessing strongly influenced how filtering was carried out in sur- technology over the last 30 years have ushered in a new set of
face metrology. More recently, developments in digital image computational problems that impact design and manufacture of
processing have been influencing computational surface metrol- product geometry. These problems involve fitting and filtering of
ogy. It has since been recognized that mean-line filters are convo-
lutions of functions while envelope filters are convolutions of sets.
This realization has made it possible to treat filtering for metrol-
ogy in a more unified setting.

5 Convolution
The computational scheme used for filtering is one of convolu-
tion. Engineers use the following two types of convolutions.
• Convolution of functions: Filtering is often implemented as
discrete convolution of functions 关17,18兴. In the most popu-
lar version, the measured data is convolved with the Gauss-
ian function. It has a smoothing effect on the surface data.
• Convolution of sets: Morphological filters are implemented
using Minkowski sums 关19兴. These can be regarded as con-
volutions where the input set is convolved with a circular or
flat structuring element.

Table 2 compares and contrasts the convolution of functions and

sets used in filtering. For the sake of simplicity, only the
Minkowski addition is shown in the convolution of sets. There are
other filters, called morphological filters, which involve Fig. 7 Output of the Gaussian filter superposed on an unfil-
tered input profile
Minkowski subtraction and repeated application of Minkowski
addition and subtraction. Also, the convolutions shown in Table 2
are continuous whereas we have to deal with discrete versions of
these convolutions in computational metrology.
Let us look briefly at convolution of functions first. Figure 7
shows the result of discrete convolution of an input function 关18兴.
Here the kernel used is a Gaussian weighting function. This figure
may explain why such Gaussian filters are known as mean-line
filters—they produce a “mean line” over which the roughness of
the profile is superposed. The mean-line filters are also linear fil-
ters. Although Gaussian filters are the most popular, standardized
linear filters, other weighting functions can also be used and these
are under development and standardization. Recent developments
in wavelets seem to indicate that the scale-dependent information
can be processed more effectively using wavelet filters. This is
still a topic of research.
Now, let us take a brief look at convolutions of sets. Figure 8
shows the result of applying a discrete morphological filter 共using
Minkowski subtraction that convolves sets兲 关19兴. Intuitively, it can
be explained as follows. Consider the sweep of a circular disk
whose center is positioned at every point of the upper, input curve. Fig. 8 Output of a morphological operation „erosion… shown
The lower envelope of this sweep is shown as the output curve. If as a lower curve. The input is the upper curve. The lower curve
a circular tactile probe is used to collect data on a surface and we is the mechanical surface corresponding to a 50 ␮m radius
only know the locus of the center of such a probe, then this figure disk, shown in three distinct places.

8 / Vol. 7, MARCH 2007 Transactions of the ASME

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 computed results. The field is likely to remain an active area of
research, while results of these researches continue to impact stan-
dardization and industrial implementation.
Acknowledgment
The author would like to acknowledge the help and support of
numerous colleagues in ASME and ISO standards committees.
However, opinions expressed in this paper are his own and do not
represent the official positions of these standards bodies.
References
关1兴 Srinivasan, V., 2005, “Elements of Computational Metrology,” DIMACS Book
Series, Vol. 67, R. Janardan, M. Smid, and D. Dutta, eds., Computer Aided
Design and Manufacturing, American Mathematical Society, R.I., pp. 79–116.
关2兴 Srinivasan, V., 1991, “A Geometer Grapples with Tolerancing Standards,” Pro-
ceedings of the CIRP International Working Seminar on Computer Aided Tol-
erancing, Penn. State University, College Station, PA.
关3兴 Srinivasan, V., 1996, “How Tall is the Pyramid of Cheops? . . . and Other
Problems in Computational Metrology,” SIAM News, April, Society for Indus-
trial and Applied Mathematics, Philadelphia, PA.
关4兴 Srinivasan, V., 1994, “Plato’s Factory,” ASME Manufacturing Review, Vol. 7,
No. 4.
Fig. 9 Output of an envelope filter, implemented as a morpho- 关5兴 ISO, 1993, Guide to the Expression of Uncertainty in Measurement, Interna-
logical filter called closing filter, is shown as the upper curve. tional Organization for Standardization, Geneva, Switzerland.
The input to the filter is the lower curve. 关6兴 Taylor, B. N., and Kuyatt, C. E., 1994, “Guidelines for Evaluating and Ex-
pressing the Uncertainty of NIST Measurement Results,” Technical Note 1297,
National Institute of Standards and Technology.
关7兴 ASME Y14.5, 1994, Dimensioning and Tolerancing, The American Society of
Mechanical Engineers, New York.
关8兴 ISO 1101, 2004, Geometrical Product Specifications (GPS)—Geometrical
discrete geometric data. Initially such problems were attacked in Tolerancing—Tolerances of Form, Orientation, Location and Run-out, 2nd,
ed., International Organization for Standardization, Geneva, Switzerland.
an ad hoc manner; each problem was tackled as it came along to 关9兴 ASME Y14.41, 2003, Digital Product Definition Data Practices, The Ameri-
satisfy the immediate industrial need. When these practices came can Society of Mechanical Engineers, New York.
up for standardization, it became clear that a more unified view 关10兴 ISO 16792, 2005, Technical Product Documentation–Digital Product Defini-
was warranted. This led to a careful classification of these com- tion Data Practices, International Organization for Standardization, Geneva,
Switzerland.
putational metrology problems into fitting problems that can be 关11兴 Shakarji, C. M., 1998, “Least-Squares Fitting Algorithms of the NIST Algo-
solved using optimization techniques and filtering problems that rithm Testing System,” J. Res. Natl. Inst. Stand. Technol., 103, pp. 633–641.
can be solved using convolution techniques, as described in the 关12兴 Forbes, A., 1991, Least-Squares Bets-fit Geometrical Elements, NPL Report
paper. DITC 140/89, National Physical Laboratory, U.K.
关13兴 Anthony, G. T., Anthony, H. M., Bittner, B., Butler, B. P., Cox, M. G., Dri-
Such classification of problems and unification of methods of eschner, R., Elligsen, R., Forbes, A. B., Gross, H., Hannaby, S. A., Harris, P.
their solution have several advantages. First, at a practical level, it M., and Kok, J., 1993, Chebyshev Best-fit Geometric Elements, NPL Report
is easier to consolidate the standardization process so that scare DITC 221/93, National Physical Laboratory, U.K.
resources are not wasted on duplicating efforts. Second, at a the- 关14兴 ISO/TS 17450-1, 2005, Geometrical Product Specifications (GPS)—General
Concepts—Part 1: Model for geometrical specification and verification, 1st
oretical level, they offer means for relating the problems of com- ed., International Organization for Standardization, Geneva, Switzerland.
putational metrology to those that have already been attacked and 关15兴 Houle, M. E., and Toussaint, G. T., 1988, “Computing the Width of a Set,”
solved in other areas of science and technology; this has a huge IEEE Trans. Pattern Anal. Mach. Intell., 10, pp. 761–765.
impact on the way robust computational techniques can be 关16兴 Whitehouse, D., 2002, Surfaces and their Measurements, Taylor & Francis,
New York.
quickly developed and implemented to solve these metrology 关17兴 Raja, J., Muralikrishnan, B., and Fu, S., 2002, “Recent Advances in Separation
problems. Finally, the clarity offered by such classification and of Roughness, Waviness and Form,” Precis. Eng., 26, pp. 222–235.
synthesis has an aesthetic, intellectual appeal that makes it suit- 关18兴 Krystek, M., Scott, P. J., and Srinivasan, V., 2000, “Discrete Linear Filters for
able for teaching computational metrology in higher educational Metrology,” Proceedings of the 16th IMEKO World Congress, Hofburg, Wien,
Austria.
institutions. 关19兴 Srinivasan, V., 1998, “Discrete Morphological Filters for Metrology,” Pro-
Several problems in computational metrology still remain open. ceedings of the 6th ISMQC Symposium on Metrology for Quality Control in
Of particular interest is the effect of measurement uncertainty on Production, TU Wien, Austria.

Journal of Computing and Information Science in Engineering MARCH 2007, Vol. 7 / 9

Downloaded From: http://computingengineering.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use