Vous êtes sur la page 1sur 20

600

Int. J. Mechatronics and Manufacturing Systems, Vol. 2, Nos. 5/6, 2009

Development of a metrology frame to improve


the positioning accuracy of Micro/Meso-scale
Machine Tools
Shawn Moylan*
National Institute of Standards and Technology,
100 Bureau Drive,
Gaithersburg, MD 20899, USA
E-mail: shawn.moylan@nist.gov
*Corresponding author

Daehie Hong
Department of Mechanical Engineering,
Korea University,
Seoul, 136-701, Korea
E-mail: dhhong@korea.ac.kr

Bradley N. Damazo, Johannes Soons


and Alkan Donmez
National Institute of Standards and Technology,
100 Bureau Drive, Gaithersburg, MD 20899, USA
E-mail: Bradley.damazo@nist.gov
E-mail: johannes.soons@nist.gov
E-mail: alkan.donmez@nist.gov
Abstract: The small work volumes of Micro/Meso-scale Machine Tools
(MMMTs) often present problems for calibration and error compensation,
but also allow solutions not practical on the traditional scale. Measuring tool
position with a separate metrology frame and compensating for error motions
is one such solution. The metrology frame design follows principles of
precision design and allows measurement of the position of the tool tip with
respect to the workpiece while minimising Abbe errors. Kinematic analysis
provides the relationship between metrology frame measurements and machine
tool coordinates. Error analysis reveals that sensor error has the only first order
influence on measurement accuracy.
Keywords: metrology frame; MMMTs; micro/meso-scale machine tools;
micromanufacturing; Abbe offset errors; Abbe alignment principle; kinematic
design.
Reference to this paper should be made as follows: Moylan, S., Hong, D.,
Damazo, B.N., Soons, J. and Donmez, A. (2009) Development of a
metrology frame to improve the positioning accuracy of a Micro/Meso-scale
Machine Tools, Int. J. Mechatronics and Manufacturing Systems, Vol. 2,
Nos. 5/6, pp.600619.

Copyright 2009 Inderscience Enterprises Ltd.

Development of a metrology frame to improve the positioning accuracy

601

Biographical notes: Shawn Moylan is a mechanical engineer in the


Manufacturing Engineering Laboratory of the National Institute of Standards
and Technology. He received his PhD in Industrial Engineering from Purdue
University in 2006. His research interests include micro-manufacturing,
on-machine part metrology, and machine tool metrology of five-axis machine
tools.
Daehie Hong received his PhD in the Department of Mechanical and
Aeronautical Engineering from the University of California at Davis, USA in
1994. He is a Professor of Mechanical Engineering at Korea University,
Seoul, Korea. His research interest includes machine tool control, micro/nano
manufacturing, mechatronics, and field robotics. He currently serves as an
Editor for Int. J. Precision Engineering and Manufacturing.
Bradley N. Damazo is a mechanical engineer with the Precision Engineering
Division, National Institute of Standards and Technology, Gaithersburg, MD
USA. He earned his PhD from Cranfield University, UK. His research areas of
interest include precision instrument design, precision motion control and
nano manufacturing. He has published in several journals and conferences in
Europe and USA.
Johannes Soons leads the Manufacturing Process Metrology Group of the
Manufacturing Engineering Laboratory at the National Institute of Standards
and Technology. He received his PhD at the Eindhoven University of
Technology in the Netherlands. His research interests include machining
processes, precision instrument design, machine metrology, and optics
fabrication and testing.
Alkan Donmez has been with NIST as a Mechanical Engineer for more than
20 years in conducting and supervising research in machine metrology and
high-precision manufacturing. He studied at Purdue University as a Fulbright
Scholar and received his MS and PhD Degrees (1985). During his carrier at
NIST, he received various awards, including R&D 100 Award, US Department
of Commerce Bronze and Silver Medals and NIST Applied Research Award.
He has published over 50 technical papers and reports. He has been an active
participant in the national and international standard committees related
to machine tool metrology and machine performance characterisation.

Introduction

MMMTs are special machine toolssignificantly smaller than traditional machine


toolsdesigned to manufacture mechanical components with features ranging from a few
micrometers to tens of millimeters. Increased industrial demand for small parts has led to
significant efforts toward developing MMMTs that can efficiently manufacture these
parts (Ehmann et al., 2005; Dornfeld et al., 2006; Weule and Hesselbach, 2002; Damazo
et al., 1999; Masuzawa and Toenshoff, 1997; Honeggar et al., 2006; Vogler et al., 2002;
Kussul et al., 2002; Subrahmanian and Ehmann, 2002; Kitahara et al., 1996).
These MMMTs typically have single-axis travel smaller than 50 mm, and are often
smaller than 300 mm 300 mm 300 mm in overall size. The smaller sizes make
MMMTs more energy efficient, more portable and potentially less expensive.

602

S. Moylan et al.

These are key concepts in micromanufacturing, and any complementary techniques or


equipment should comply with them.
In order to achieve their primary function of machining micro/meso-scale parts and
features, MMMTs must perform at a high level of accuracy. As parts scale down in size,
tolerances often follow suit, sometimes with tolerances at or below 1 m. As machine
tools scale down, however, their accuracies do not similarly scale. For example, many
traditional scale machine tools that have single axis travel of 500 mm can accomplish this
travel with less than 5 m positioning error. MMMTs are typically accurate to
approximately the same 5 m, even though their total travel may only be 25 mm.
The reason for the similar accuracies despite the smaller travel is that increasing the
mechanical accuracy of the machine tool components to achieve smaller error motions is
very expensive at any size scale. Thus, to machine small parts with the necessary
accuracy, some type of external accuracy enhancement is necessary.
In his book Precision Machine Design, Slocum (1992) offers three possible methods
to increase the accuracy of a machine tool:

specialised finishing processes on separate specialised machines

error mapping of the machine tool with software-based error correction algorithms

a metrology frame around the machine tool to measure and compensate for the
errors.

The small sizes of MMMTs and the parts machined on them make the first two methods
difficult to implement. Handling the tiny, fragile parts can lead to damage. Additionally,
with such small complex parts, re-establishing datum surfaces on a second finishing tool
can lead to a large increase in uncertainty. The tools used to perform error mapping
on traditional machine tools (laser interferometers, ballbars, differential levels, etc.) often
do not fit, or at least cannot properly function, in the very small work volumes of
MMMTs.
The third technique, the metrology frame, fits MMMTs very well. Slocum mentions
that metrology frames require the largest initial capital investment. This cost is greatly
reduced for the micro/meso-scale because the required parts and components are much
smaller and cost much less to manufacture. Additionally, because the total range of
motion for MMMTs is smaller, the sensors required to measure this motion do not
require a long operational range.
This paper discusses a metrology frame designed to improve the accuracy of an
existing, 3-axis benchtop milling machine. In doing so, the concept of metrology
frames how they improve measurement/positioning accuracy and where they have
previously proven successful is discussed. The specific design of a prototype metrology
frame for an MMMT is detailed. Finally, the kinematic and error analyses are presented
to further demonstrate the effectiveness of the metrology frame concept on MMMTs.
The result is a machine that better knows its errors and can compensate for them,
converting a typical MMMT into a Smart Machining System.

The concept of metrology frames

The purpose of metrology systems on a machine tool is to continually measure


the position of the cutting tool tip relative to the workpiece. A separate metrology

Development of a metrology frame to improve the positioning accuracy

603

frame is often able to accomplish this task more accurately than the typical embedded
measurement system. Two basic premises give metrology frames their power. The first
is that they separate the measurement loop from the structural loop. The second is that
they eliminate or account for Abbe offset errors.

2.1 Separation from the load bearing structure


The metrology frame mounts on or around the machine tool in such a fashion as to
isolate it from the machines structural deformations. Thermal, static, and dynamic loads
on the structural components in the machine lead to undesired deflections. Since the
machines intrinsic displacement measurement sensors (such as scales) are subject
to the same deformations, these deformations are undetected and therefore result in
positioning errors. Thus, separating the metrology frame from the structural loop keeps
the measurement sensors from deflecting with the machine, allowing these deflections
and errors to be detected by the measurement sensors. The result is a more accurate
measurement of the tools position.

2.2 Abbe offset errors


In traditional machine tools, displacement of a slide is measured by a position feedback
system usually aligned with the ballscrew (rotary encoder) or the guideway (linear scale)
of the slide. As such, both measurement strategies have the axis of displacement
measurement offset from the position of interest, namely the tool tip (Abbe offset).
This offset increases in machine designs that stack individual linear axes. As first
documented by Professor Ernst Abbe in 1890 (Evans 1989), displacement measurement
is influenced by angular error motion if the measurement is not carried out at the point of
interest. Therefore, in the presence of undetected angular errors of moving slides,
Abbe offset is a main source of linear displacement error observed in machine
tools. These observations led to the Abbe principle of alignment; stated simply,
the measurement system should be collinear with the displacement being measured.
In updating the Abbe principle, Bryan amends Abbes original statements, saying,
The path of the effective point of a displacement measuring system should
be collinear with the path of the functional point whose displacement is to be
measured. If this is not possible, either the slideways that transfer the
displacement must be free of angular motion or angular motion data must be
used to calculate the consequences of the offset. (Bryan, 1979a, p.2)

These statements allow two approaches to achieve the Abbe principle:

align all measurement sensors with the measurement point

measure and account for the angular errors of the motion.

The vast majority of metrology frames follow the second approach.

2.3 Examples of metrology frames on previous machines


The concept of the metrology frame is by no means a new one. The concept has been
used for a variety of measurements, including the measurement of a point in 3D space
(Cuttino et al., 1999; Schinstock and Cuttino, 2000), but the present discussion will

604

S. Moylan et al.

concentrate on its use for machine tools and measuring machines. The first documented
case illustrating use of a metrology frame is on the Rogers-Bond Universal Comparator
described in 1883 at the American Academy of Arts and Sciences (Evans, 1989). Large
diamond-turning machines were the first machine tools to benefit from position
measurement by a separate metrology frame (Bryan, 1979b, p.1; Donaldson and
Patterson, 1983). Ultra-high accuracy Coordinate-Measuring Machines (CMMs) require
metrology frame measurement of the probe position or of the workpiece position
(Teague 1989). More recently, metrology frames have helped convert traditional CMMs
to micro or meso-scale CMMs (Peggs et al., 1999; Brand and Kirchhoff, 2005). A key
distinction between metrology frame application on measuring machines and on machine
tools is the implementation of the error measurements. Accurate measurement of error
motions is not the entire task for machine tools. The measurements must be fed back
to the motion controller to allow for improvement in the positioning and thereby more
accurate parts and features.

Design of the metrology frame for an MMMT

The principles of precision machine design (Slocum, 1992; Schellekens and Rosielle,
1998) guided the decision making during the design of the current metrology frame.

3.1 Structural geometry


The process of applying the concept of the metrology frame to an existing machine starts
with establishing certain design constraints that consider operation, analysis, fabrication,
and assembly of the metrology frame. With no constraints, the metrology frame could
take nearly an infinite number of designs, but specifics of the application limit the
options. Keeping in mind that the metrology frame must isolate measurement from
the load-bearing structure and minimise Abbe offset errors, the subsections below discuss
the specific constraints that apply to the current metrology frame.

3.1.1 Number of sensors


The small sizes and work volumes of MMMTs lead to the desire to minimise the number
of sensors. With overall sizes of 300 mm 300 mm 300 mm and work volumes
of 50 mm 50 mm 50 mm, MMMTs do not have a lot of unused space to
accommodate the metrology frame components. Thus, accomplishing the goal with a
minimum number of components is important. Improving a 3-axis micro/meso-scale
milling machine requires a minimum of three sensors to measure the motion.

3.1.2 Abbes principle


Constraining the metrology frame to operate with only three sensors necessitates
that all sensors point directly to the tool tip. This follows the first approach to
satisfying Abbes principle. Most metrology frames, including all those referenced
earlier, to some extent follow path two of Abbes principle, measuring and accounting
for the angular error motions. However, measuring angular errors typically requires
multiple sensors per axis, resulting in six or more sensors in the metrology frame. This is

Development of a metrology frame to improve the positioning accuracy

605

less of a problem with large machines that have significant open space for these
sensors. Since the current metrology frame uses only three sensors, measurement of
angular errors is not an option. Thus, all sensors must point directly to the position being
measured.

3.1.3 Sensor configuration


A logical sensor arrangement that follows Abbes principle is one sensor aligned with
each Cartesian axis such that the sensor axes intersect at the tool tip. For MMMT
configurations that have a fixed tool position relative to the machine frame, such an
arrangement is easy to achieve by sensors mounted on the machine frame that measure
the displacement of flat targets on the workpiece table. For machine configurations
where tool tip moves, the sensors need to move with the tool tip in order to maintain
zero Abbe offset. Our machine has a spindle that moves in z-direction, requiring a
complicated bracket design to measure along the z-axis. An alternative design uses two
sensors mounted on the spindle head that are inclined at an angle of /4 with respect to
the x-axis in the x-z plane (Figure 1). This arrangement provides measurements in the
x-z plane, with a third sensor measuring in y-direction only, while keeping the
Abbe Offset zero. Disadvantages of this design are the lack of symmetry and inefficient
use of machine space. We chose the rotation symmetric design (Figure 3) where all three
sensors contribute equally in z-direction.
Figure 1

Early design of metrology frame for MMMTs: (a) front view; (b) side view
and (c) oblique view

(a)

(b)

(c)

606

S. Moylan et al.

3.1.4 Simplifying coordinate transformation


Three additional constraints stem from a desire to simplify analysis, specifically
conversion from local metrology frame coordinates to the Cartesian coordinates
of the machine. First, symmetrical distribution of the three sensors around the work
volume leads to some convenient trigonometric equalities in the sensor direction
vectors. Second, the z-direction components of each sensor vector should be equal.
Third, the three sensor vectors should form a right-handed orthogonal coordinate system,
making transformation from the metrology frame coordinate system to the machines
own Cartesian coordinate system defined by the motion of the machines slides a simple
multiplication by a rotation matrix.

3.1.5 Calculation of sensor vectors


A final, rather arbitrary constraint is required to pinpoint the final sensor-direction
vectors. It is designated that sensor number three be aligned within the y-z plane.
The sensor-direction vectors n1, n2, and n3 result as direction vectors along Cartesian
axes of the machine coordinate system after rotating the frame through a rotational
matrix. If a generic rotation operation is defined as rotation about the z-axis by an angle
of , followed by a rotation around the modified y-axis by and angle of , followed
by a rotation around the modified x-axis by an angle of (see Figure 2), this yields:
cos cos
cos sin + sin sin cos
cos sin

=
=
n1
n2 cos cos + sin sin sin
sin

sin cos
sin sin + cos sin cos
n3 = sin cos + cos sin sin

cos cos
Figure 2

(1)

Schematic of the rotations between the Cartesian coordinate frame and the metrology
frame coordinate frame. The order of rotation is first about the z-axis, then about the
modified y-axis, followed by rotation about the modified x-axis (see online
version for colours)

Using the constraint that all sensor vectors must have the same z-direction component,
the angle can be determined. Squaring the z-direction components of the vectors and
adding them together, yields
n12z + n22z + n32z = sin 2 + sin 2 cos 2 + cos 2 cos 2 .

(2)

Development of a metrology frame to improve the positioning accuracy

607

Re-arranging
sin 2 + cos 2 (sin 2 + cos 2 ) = 1.

(3)

Since n12z + n22z + n32z = 1 and n1z = n2 z = n3 z , one can see that 3n12z = 1 and thus,
3
= 0.6158 rad.
3

= sin 1

(4)

Substituting back into sensor 2, and solving for


2
= 0.7854 rad.
2

= sin 1

(5)

The final angle, emerges from the constraint that the third sensor be aligned in the
y-z plane. This means that the x-direction component of the third sensor must equal zero
n3 x = sin sin + cos sin cos = 0.

(6)

Processing this equation


sin sin = cos sin cos
sin tan = cos sin
tan tan = sin .

(7)

Substituting the angles already calculated

= tan 1 ( sin(0.6158)) = 0.5236 radians.

(8)

Thus, the final direction vectors are


0.7071
0.7071
0

n1 = 0.4082 , n2 = 0.4082 , n3 = 0.8165 .


0.5774
0.5774
0.5774

(9)

3.2 The final design


Figure 3(a) and (b) show the metrology frame mounted on a three-axis meso-scale
milling machine. The list of components making up the metrology frame are a spindle
interface adapter (orange in Figure, also see Figure 4(a)), a spindle mount (red in
Figure 3(a)), the sensor frame (see Figure 4(b)), three sensor attachment brackets
(lavender in Figure 3(a)), the sensors, and the workholding frame (see Figure 5).
Three linear sensors mounted symmetrically around the work volume point directly to the
tool tip. The sensors reside in the tool frame, affixed to the spindle and thereby the tool.
The sensors themselves are linear displacement sensors with 12 mm travel and 0.2 m
resolution (though similar sensors with different travel lengths could easily be
substituted). The target (pad) surfaces reside in the workholding frame, rigidly coupled to
the x-y carriage and thereby the workpiece. The linear sensors maintain orthogonal
contacts on inclined target surfaces. Such a displacement sensor and target arrangement
allows the measurement of tool position with respect to the workpiece.

608
Figure 3

S. Moylan et al.
(a) Isometric view of the final design of the metrology frame mounted on a 3-axis
meso-scale milling machine and (b) prototype of the metrology frame mounted
on a three-axis micro/meso-scale machine tool (see online version for colours)

(a)

(b)
Figure 4

(a) Spindle mount adapter; (b) sensor frame and (c) top view of metrology frame
mounted on machine. Note that the adapter allows the top portion of the frame
(in yellow) to pass through, therefore entirely surrounding the spindle (see online
version for colours)

Development of a metrology frame to improve the positioning accuracy


Figure 5

609

Isometric view of the workholding frame

3.3 Isolation of components from surrounding environment


It is important to minimise any influence the surrounding environment, such as
nearby heat sources, might have on the position measurement. To reduce the effects of
surrounding heat sources, both the sensor frame and the workholding frame were
fabricated from Invar, a material with extremely small coefficient of thermal expansion.
Additionally, the sensors connect to the spindle through flexures, and the workholding
frame mounts kinematically to the work table to further isolate the sensors and the pad
surfaces themselves from the effects of thermal and structural deformations.

3.3.1 Flexures in the sensor frame


In order for the sensors to continually point directly to the tool tip, the sensors must travel
with the tool. To best accomplish this, the connection should be as close as possible to the
tool itself. On the original machine, the spindle mount, which holds the spindle and
therefore the tool, connects directly to the z-axis carriage. To add the sensor frame
portion of the metrology frame, a spindle interface adapter between the spindle mount
and the z-axis carriage is necessary (see Figure 4). This adapter allows the sensor frame
to pass through, allowing the structurally sound frame holding the sensors to completely
surround the spindle (see Figure 4(c)). However, the sensor frame does not actually
connect to the adapter. Rather, the frame connects to the bottom of the spindle mount,
much closer to the tool.
Flexures separate the connecting collar at the bottom and the sensor mounts at the top
of the sensor frame (see Figure 4(b)). These flexures ensure that the sensor frame
connection to the tool is not over-constrained. If thermal deformations develop due to the
spindle heating, the deformations would cause an over-constrained component to warp.
However, the flexures prevent any warping, isolating the sensors from the effects of
thermal deformation and keeping the sensors pointed directly to the tool tip.

3.3.2 Kinematic mounting of the workholding frame


The advantages of kinematic mounting (see Slocum, 1992) are a high degree of stability
and minimistion of any distortions. Additionally, because the metrology frame is a
removable addition to the machine, kinematic mounting also provides repeatable
re-positioning.

610

S. Moylan et al.

A planar three-groove coupling is used to attach the workholding frame to the work
table. The workholding frame of the metrology frame connects to the work table through
a mounting plate. The mounting plate is aligned and bolted to the worktable. Beneath the
workholding frame are three v-grooves, symmetrically distributed, with the line of the
grooves intersecting at the centre. Atop the mounting plate are three balls. The balls fit
into the v-grooves in a unique fashion to ensure proper orientation.

3.4 Assembly and alignment


Proper alignment of the sensors is central to the concept of the metrology frame and,
for the current setup, requires the use of a special alignment tool. The first steps in the
assembly involved attaching and aligning with indicators the spindle mount adapter,
the sensor frame, and the spindle mount. Notice in Figure 4(b) that the sensor frame
does not have the sensor mounts built in. The reason for this is to allow adjustments to
ensure proper sensor alignment; alignment achieved using the sensor mounting station
shown in Figure 6. The mounting station is positioned on the work table (note the
workholding frame is not yet attached) beneath the spindle mount and sensor frame.
The z-axis is lowered until the bottom of the sensor frame touches the top contacts of the
mounting station, slightly compressing the spring. At this point, the sensor mounts are
loosely attached to the sensor frame. Extending the arms of the mounting station, the pins
at the ends are inserted into the holes in the sensor mounts. The axes of the rigid arms and
pins all intersect at the centre of the mounting station (where the tool tip will be when
inserted), and when inserted into the sensor mounts, ensure proper positioning
of the mounts. Finally, the sensor mount connections are tightened to the sensor frame,
the pins are removed and the sensors are inserted into the properly aligned mounting
holes. This completes the installation and alignment of the sensors and sensor frame.
Figure 6

Photograph of the sensor setting station (see online version for colours)

Installation of the workholding frame is comparatively simple. The baseplate is placed


on the worktable with balls facing upward and aligned then attached to the table.
The workholding frame mounts atop the baseplate such that the v-grooves beneath
the frame fit properly on top of the balls. All three balls will fit into the v-grooves only
if the workholding frame is in its proper orientation. Finally, the frame is bolted

Development of a metrology frame to improve the positioning accuracy

611

to the baseplate with enough force to achieve total restraint, but lightly enough to avoid
significantly distorting the frame.
To function properly, the tool must be at the proper length so that the tip is actually
at the point where the sensors intersect. Thus, when originally inserting the tool,
or whenever a tool change is necessary, a tool setting station must be used to ensure the
tool tip is in the correct position. The tool setting station is very similar to the sensor
setting station with a platform mounted atop a spring. A tool change starts by inserting
the tool into the collet, but not tightening the collet down. The z-axis is lowered until the
spindle mount contacts the top of the tool setting station, slightly compressing the spring.
At this point, the tool is drawn out of the collet until the tip contacts the platform, setting
the tool tip to the proper height. Finally, the collet is tightened, the z-axis raised, and the
setting station removed to complete the tool change.

Position measurement

To understand exactly how the sensor readings, in the sensor coordinate frame, describe
the position of the tool tip in the machine Cartesian coordinate frame, one must
understand the inverse and forward kinematics of the system as well as the contributions
of any misalignments or errors to the measurement. With the addition of the metrology
frame to the machine, a new coordinate frame is introduced. The sensor readings are in
the right handed coordinate frame prescribed by the sensor direction vectors, while the
machine maintains its Cartesian coordinates dictated by the motions of the slides.
Because these coordinate systems differ, a transformation matrix that allows conversion
from one to the other is necessary. It is the forward and inverse kinematics that lead to the
transformation matrix.

4.1 Inverse kinematics


Inverse kinematics explain the geometrical relationship between the metrology frame
components and Cartesian coordinates of the machine tool. For convenience,
the transformation from one frame to the other is derived in two dimensions before
directly extending to a three dimensional coordinate frame.
Figure 7 shows the vectors and parameters used to derive the kinematic relations of
the metrology frame in two dimensions. Two coordinate frames comprise the metrology
frame: a machine coordinate frame and a tool coordinate frame. The machine coordinate
frame contains the pad surfaces. The position of the tool is defined in the machine
coordinate frame. The tool coordinate frame contains the measurement sensors and its
origin is at the tool tip.
Following the inverse kinematics, one can determine the lengths of sensor legs given
the position of the tool tip p with respect to the machine coordinate frame. The vector
defined from the sensor base to a fixed reference point on the inclined pad (the centre of
the pad) is si. If the position of the tool tip p is known, the vector si is
si = p bi + ai

(10)

where ai and bi are the constant vectors for the sensors and pad positions, respectively.

612

S. Moylan et al.

Figure 7

Schematic of metrology frame simplified to two dimensions to show relative vectors


(see online version for colours)

The vector si can be decomposed as


si = li ni + mi ti

(11)

where ni are sensor direction vectors aligned to the sensor axes, the vectors derived in
Section 3.1.5, ti are tangent direction vectors aligned along the pad surfaces, and li and mi
are their magnitudes, respectively. Therefore, the values of li are the lengths of the sensor
arms, or the actual sensor readings.
Multiplying (dot product) both sides of equation (11) by ni gives the solution for the
sensor readings:
li = si ni .

(12)

4.2.1 Modified sensor vectors accounting for pad misalignment


Despite the rigorous assembly and alignment procedure discussed earlier, slight
geometrical deviations from the ideal design are unavoidable. Therefore, to maintain
accuracy, the sensor vectors must account for these deviations. Specifically, in the design,
the direction vector ni is orthogonal to ti. However, when realised, a slight deviation from
the right angle may exist. Figure 8 illustrates this situation. The sensor readings li are the
projections of the vector si on the sensor axes slightly modified due to the pad inclination.
Assuming the deviation is a small angle, equation (12) is modified to approximate the
magnitudes of the sensor vectors:
li si ni si ti (ni ti ) = si {ni ti (ni ti )}, i = 1, 2.

(13)

Similarly, the deviation affects the sensor vectors obtained from equation (9), resulting in
modified sensor vectors
ni = ni ti (ti ni ) ei (ei ni )

(14)

where ti and ei are two orthogonal direction tangent vectors on the ith pad surface,
which represent the pad surface orientations. If the pad surfaces are perfectly orthogonal
to the sensor vectors, the last two terms equal zero.

Development of a metrology frame to improve the positioning accuracy


Figure 8

613

Correction of sensor reading due to slight deviation from right angle

4.2.2 Coordinate Measuring Machine results


A Coordinate Measuring Machine (CMM) provided the actual geometry, and therefore
deviations from design, of the prototype. Table 1 shows the CMM measurement
results with the pad surface vectors as normal to the pad surface. CMM measurements
needed to be normalised to one point, so the i = 3 pad/sensor was chosen.
Table 1

Sensor vectors for the metrology frame components. A first order approximation
of the uncertainty in these results is 0.0002
N

CMM results

Design

0.7071 0.7071 0.0000


0.4082 0.4082 0.8165

0.5774 0.5774 0.5774

Pad surface

0.7083 0.7066 0.0000


0.4057 0.4095 0.8155

0.5776 0.5770 0.5787

Sensor axes

0.7097 0.7061 0.0000


0.4056 0.4093 0.8162

0.5760 0.5778 0.5777

The CMM measurements of both the pad surfaces and the sensors allow the calculation
of the sensor vectors (ni) and the real pad surface vectors (ti and ei). Sensor vectors result
directly from the CMM measurements for the sensors. The CMM measurements for the
pad surfaces, however, are reported as normal to the pad surface and therefore must be
rotated /2 radians about the modified x-axis and the modified y-axis. Substituting the
proper values into equation (14), the modified sensor vectors become
0.7085 0.7066 0.0000
n = 0.4055 0.4096 0.8155 .
0.5776 0.5770 0.5787

(15)

614

S. Moylan et al.

4.2.3 Orientation deviation between tool and workpiece frames


The pad surfaces and sensors were measured separately on the CMM and mounted
independently to the machine, therefore the CMM measurements do not account
for orientation deviations between the sensor frame and the workholding frame.
These orientation deviations were not measured directly, but were determined using
an optimisation procedure, and the sensor vectors were modified accordingly.
The optimisation procedure to determine orientation deviations involved determining
a rotation matrix
1

R = z
y

z
1

x .
1

(16)

R is multiplied by the modified sensor vectors, ni to minimise the difference between


measurement by the metrology frame sensors and externally mounted sensors. One face
of a precision square artifact was aligned to the x-axis, and fixed to the machine in the
workpiece position. Two linear displacement sensors mounted parallel with the machine
x- and y-axes respectively, but not connected to the machine or metrology frame in any
way, contacted the faces of the artifact and measured the x and y displacements of the
machine tool. These external sensors provided an independent assessment of
the machines positioning and allow comparison with internal, relative measurements.
The machine was programmed to execute a series of linear moves in the x-y plane
while the external sensors and the metrology frame sensors measured the motion.
The positions resulting from the metrology frame sensors and from the external sensors
are slightly different due to orientation deviations. The metrology frame sensor vectors
are multiplied by the rotation matrix, R, with initial guesses for x, y, and z, and the
position differences are recomputed. A non-linear least squares optimisation program
iterates this process, adjusting x, y, and z to minimise the position differences.
The rotation matrix that results from the optimisation is multiplied by the modified sensor
vectors to yield final sensor vectors
0.0056 0.0077
1
R = 0.0056
1
0.0170 ,
0.0077 0.0170
1
0.7327 0.6828 0.0297
nfinal = 0.3904 0.3848 0.8366
0.5593 0.6227 0.5495

(17)

4.3 Forward kinematics


Forward kinematics provide the relative position of the tool coordinate system, p,
with respect to the machine coordinate frame given the sensor readings. This concept
is the essence of the metrology frame, and therefore this kinematic analysis is of primary
importance. Combining equations (10) and (11) yields the following relation,
li ni + mi ti = p bi + ai .

(18)

Development of a metrology frame to improve the positioning accuracy

615

Assuming the sensor vectors are orthogonal (or very close to orthogonal) to the tangent
vectors, multiplying (dot product) both sides of equation (18) by the sensor vector, ni,
gives
li = ( p bi + ai ) ni .

(19)

Rearranging,
p ni = li + bi ni ai ni

(20)

If the coordinate frames are designed with the same origin (at the tool tip) and all the
sensors initialised at the same time, ai equals bi. The result is, the position vector in the
following simple form,
(21)

p = qN 1

where
n1x
p = ( px , p y , pz ), q = (l1 , l2 , l3 ), N = n1 y
n1z

n2 x
n2 y
n2 z

n3 x
n3 y .
n3 z

(22)

Note that the above N matrix stems from the ideal orthogonal relationship between the
sensor vectors and the tangent vectors. If deviation from orthogonal exists, equation (14)
provides the proper modified sensor vectors, and thereby N matrix. For the current setup,
nfinal is substituted for N.
In conclusion, the position of the tool relative to the workpiece, p, can be tracked
using the metrology frame sensor readings, q, assuming the exact geometry of the
metrology frame components is known.

Error sensitivity

In this section, the effects of various geometric error sources in the metrology frame are
modelled. To start, equation (18) is reordered, solving for the measurand, p
p = li ni + mi ti + bi ai .

(23)

This equation, however, does not account for the orientation deviation between the tool
coordinate frame and the machine coordinate frame. If the orientation deviation exists,
the vectors ai convert to
ai = Rait

(24)

and the sensor vectors convert to


ni = Rnit

(25)

where R is the rotational matrix to represent the orientation deviation and the superscript t
denotes the tool coordinate system (and the vectors without a superscript are with respect
to the machine coordinate system). Then, equation (23) becomes

616

S. Moylan et al.
p = li Rnit + mi ti + bi Rait .

(26)

The mathematical representation of the measurand has seven inputs, all of which have
their own uncertainties:

The uncertainties in the sensor measurements including the roughness of the pad
surfaces: li

The uncertainty in the orientation between the machine coordinate frame and tool
coordinate frame: R

The uncertainties in the directions of sensor axes: ni

The uncertainties of the pad surface orientation: ti, mi.

The uncertainties in the constant vectors: ai, bi.

However, an exploration of the combined standard uncertainty of the position


measurement reveals that the measurand is not necessarily sensitive to all of these inputs.
The combined standard uncertainty in position measurement starts with the partial
derivatives of equation (23):

p = li Rnit + li Rnit + li R nit + mi ti + mi ti + bi Rait R ait .

(27)

While the term R describes a constant misorientation between the tool frame and the
machine frame, the term R describes the uncertainty in this misorientation. The constant
misorientation has already been accounted for, as described in Section 4.2.3. However,
slide movement will result in some type of angular error motions. These angular errors
lead to a change in the orientation between the tool frame and the machine frame and R
captures these angular errors. The differential change of the rotational matrix can be also
represented as

R = R R

(28)

where
0

R = z
y

z
0

(29)

and x, y, z are small rotations about x, y, z, respectively. Then, equation (27) becomes

p = li Rnit + li R Rnit + li R nit + mi ti + mi ti + bi R Rait R ait .

(30)

Note that the products of the rotational matrix and the vectors in the tool coordinate
frame transform them into the machine coordinate frame. Additionally, R is a
skew-symmetric matrix, and the product of a skew symmetric matrix with a vector is the
same as the cross-product between two vectors. These facts change the first, second, and
seventh terms on the right in equation (30)

p = li ni + li ( ni ) + li R nit + mi ti + mi ti + bi ( ai ) R ait

(31)

Development of a metrology frame to improve the positioning accuracy

617

where

= ( x , y , z )T .

(32)

Multiplying (dot product) ni on both sides of equation (26),

p ni = li ni ni + li ( ni ) ni + li ( R nit ) ni + mi (ti ni ) + mi ( ti ni )
+ bi ni ( ai ) ni R ait ni .

(33)

Investigating equation (33) term by term reveals that not all of the terms are significant.
The first term on the right equals li because the dot product of the same vector is equal
and ni is
to one. The second term is equal to zero because the cross product of
orthogonal to both
and ni, and therefore the dot product with ni equals zero. The third
term possesses the dot product of a unit vector and the uncertainty in that same unit
vector. Because the infinitesimal change of a unit vector, nit, is essentially orthogonal to
the original vector, this term is negligibly small (because the dot product of orthogonal
vectors is zero) unless the sensor readings, li, are very large. However, the small work
volume of the micro/meso-scale machine tool ensures the sensor readings will always be
small and the term can be ignored. The fourth and fifth terms result from the orientation
error of the pad surfaces, and are already considered with the modified sensor vectors in
equation (14). The sixth and eighth terms, again, are uncertainty in the constant vectors
that cancel each other out if they are defined with the same origin and initialised at the
same time. The seventh term stays, but can be rewritten as below

p ni = li (ai ni ) .

(34)

Rearranging the above equation in terms of p in matrix form,


1

p = niT { li } niT (ai ni )T .

(35)

The first term on the right-hand side of equation (35) is the transformation of the sensor
uncertainty (including pad profile errors) to Cartesian coordinates. This uncertainty
can be minimised by selecting high quality sensors as well as by precisely machining the
pad surfaces. Also, care should be taken to properly align the sensor axes perpendicular
to the pad surface and to account for any small deviation by using modified sensor
vectors (equation (14)).
The second term on the right side of equation (35) is the transformation of the
Abbe errors to Cartesian coordinates. The cross product (ai ni) provides the deviation
of the tool tip (specified point) from the sensor axis ni (measurement axes), which is
by definition the Abbe offset. While the magnitude of this offset is second order, it can
become large depending on the magnitude of ai. Special care must be taken to accurately
align the sensor axes to point directly to the measurement point, i.e., making ni coincident
,
with ai. Note that this Abbe offset is multiplied by the frame orientation uncertainty
and that by minimising or eliminating Abbe offset, the metrology frame measurements
become more immune to angular error motions.

618

S. Moylan et al.

Conclusion

Measuring tool position with respect to the workpiece using a separate metrology frame
allows a machine to better track its position and to correct for its error motions, making
the machine more of a smart machining system. The metrology frame described above is
specific to the machine on which it is mounted. However, the concept of the metrology
frame can benefit many three-axis micro/meso-scale machine tools of different designs.

Acknowledgement
Official contribution of the National Institute of Standards and Technology (NIST);
not subject to copyright in the USA. The full descriptions of the procedures used in the
paper require the identification of certain commercial products and their suppliers.
The inclusion of such information should in no way be construed as indicating that such
products or suppliers are endorsed by NIST or are recommended by NIST or that they are
necessarily the best materials, instruments, software or suppliers for the purposes
described.

References
Brand, U. and Kirchhoff, J. (2005) A micro-CMM with metrology frame for low uncertainty
measurements, Measurement Science and Technology, Vol. 16, pp.24892497.
Bryan, J.B. (1979a) Design and construction of an ultra precision 84 inch diamond turning
machine, Precision Engineering, Vol. 1, No. 1, pp.1317.
Bryan, J.B. (1979b) The Abbe principle revisited: an updated interpretation, Precision
Engineering, Vol. 1, No. 3, pp.129132.
Cuttino, J.F., Schinstock, D.E. and Prather, M.J. (1999) Three-dimensional metrology frame for
precision applications, Precision Engineering, Vol. 23, pp.103112.
Damazo, B.N., Davies, M.A., Dutterer, B.S. and Kennedy, M.D. (1999) A summary of
micro-milling studies, Proceedings of the 1st International Conference and General Meeting
of EUSPEN, Bremen, Germany, pp.322324.
Donaldson, R.R. and Patterson, S. (1983) Design and construction of a large, vertical-axis
diamond turning machine, Proceedings of the SPIE, Vol. 433, pp.6267.
Dornfeld, D., Min, S. and Takeuchi, Y. (2006) Recent advances in mechanical micromachining,
CIRP Annals, Vol. 55, No. 2, pp.745768.
Ehmann, K.F., Bourell, D., Culpepper, M.L., Hodgson, T.J., Kurfess, T.R., Madou, M.,
Rajurkar, K. and DeVor, R.E. (2005) Final Report on WTEC Panel on International
Assessment of Research and Development in Micromanufacturing, World Technology
Evaluation Center, Inc., Baltimore, MD.
Evans, C. (1989) Precision Engineering: An Evolutionary View, Cranfield, Bedfordshire, UK.
Honegger, A.E., Langstaff, G.Q., Phillip, A.G., Vanravenswaay, T.D., Kapoor, S.G.,
Gauthier, G.W. and DeVor, R.E. (2006) Development of an automated microfactory:
Part 1 Microfactory architecture and sub-systems development, Transactions of the North
American Manufacturing Research Institute of SME, Vol. 34, pp.333340.
Kitahara, T., Ishikawa, Y., Terada, T., Nakajima, N. and Furuta, K. (1996) Development of
micro-lathe, Kikai Gijutsu Kenkyusho Shoho/Journal of Mechanical Engineering Laboratory,
Vol. 50, No. 5, pp.117123.

Development of a metrology frame to improve the positioning accuracy

619

Kussul, E., Baidyk, T., Ruiz-Huerta, L., Caballero-Ruiz, A., Velasco, G. and Kasatkina, L. (2002)
Development of micromachine tool prototypes for microfactories, Journal of
Micromechanics and Microengineering, Vol. 12, No. 6, pp.795812.
Masuzawa, T. and Tonshoff, H.K. (1997) Three-dimensional micromachining by machine tools,
CIRP Annals, Vol. 46, No. 2, pp.621628.
Peggs, G.N., Lewis, A.J. and Oldfield, S. (1999) Design for a compact high-accuracy CMM,
CIRP Annals, Vol. 48, No. 1, pp.417420.
Schellekens, P. and Rosielle, N. (1998) Design for precision: current status and trends,
CIRP Annals, Vol. 47, No. 2, pp.557586.
Schinstock, D.E. and Cuttino, J.F. (2000) Real time kinematic solutions of a non-contacting,
three dimensional metrology frame, Precision Engineering, Vol. 24, pp.7076.
Slocum, A.H. (1992) Precision Machine Design, Prentice-Hall, New Jersey, USA.
Subrahmanian, R. and Ehmann, K.F. (2002) Development of a meso-scale machine tool (mMT)
for micro-machining, Proceedings of 2002 Japan-USA Symposium on Flexible Automation,
1419 July, Hiroshima, Japan, Vol. 1, pp.163169.
Teague, E.C. (1989) The national institute of standards and technology molecular measuring
machine project: metrology and precision engineering design, Vacuum Science Technology,
Vol. B7, pp.18981902.
Vogler, M.P., Liu, X., Kapoor, S.G., DeVor, R.E. and Ehmann, K.F. (2002) Development
of meso-scale machine tool (mMT) systems, Technical Paper Society of Manufacturing
Engineers, Vol. MS02-181, pp.19.
Weule, H. and Hesselbach, J. (2002) MikroPro Investigation of the International State of the Art
of Micro Production Technology, Hesselbach, J. (Ed.), Vulkan-Verlag, Essen.

Vous aimerez peut-être aussi