© 1994 by Alexander H.

Slocum

Precision Machine Design

Topic 2

Assessment of errors in precision machines

Purpose:
This lecture defines types of errors and discusses methods for
modeling them and accounting for them (error budgets) when
predicting machine performance.

Outline:
• Introduction
• Definitions
• Homogeneous transformation matrices
• Introduction to error budgets
• Error types

"Mathematics is the language with which God has written the

universe"

Galileo

2-1
 © 1994 by Alexander H. Slocum

Introduction
• The key to precision machine design is predicting what the
errors (difficulty) will be and then designing the system to
minimize cost.

Metrology frame based

servomechanism
Cost

Mapped servomechanism

Simple servomechanism
Mechanical system

Difficulty = Environment × Load × Range × Speed

Accuracy

• Always keep this figure in your mind. It is the key to

success in precision machine design (e.g., mechatronics).
• Always try to think of new designs that can shift the curves
to the right.
• Prime examples of nested technology curves are:
• Sliding contact bearings • Ballscrews
• Rolling contact bearings • Rollerscrews
• Hydrostatic bearings • Linear motors

2-2
 © 1994 by Alexander H. Slocum

Definitions
Resolution

Positions achieved
while attempting to
position to target Repeatability
point plus finest
increment of motion Positions achieved
which can be while attempting to
programmed Accuracy position to target
Target point
point

• Accuracy is the ability to tell the truth or:

• The maximum translational or rotational error between
any two points in the machine's work volume.
• Linear, planar, and volumetric accuracy can all be
defined for a machine.
• REMEMBER mechanical accuracy is costly, whereas
repeatability is not expensive:

Metrology frame based

servomechanism
Cost

Mapped servomechanism

Simple servomechanism
Mechanical system

Difficulty = Environment × Load × Range × Speed

Accuracy

2-3
 © 1994 by Alexander H. Slocum

• Repeatability (precision) is the ability to tell the same story

over and over again or:
• The error between a number of successive attempts to
move the machine to the same position.
• Repeatability is often considered to be the most important
parameter of a computer controlled machine (or sensor)
• Often the intent is to map the errors and then
compensate for them.
• Minimize static friction and thermal variants to get better
repeatability.
• Resolution is how detailed your story is.
• Resolution is the larger of the smallest programmable
step or:
• The smallest mechanical step the machine can make
during point to point motion.
• Resolution gives a lower bound on the repeatability.
• Minimize static friction to get better resolution.

2-4
 © 1994 by Alexander H. Slocum

Abbe Errors
"If errors in parallax are to be avoided, the measuring system
must be placed coaxially with the axis along which displacement
is to be measured on the workpiece" . Dr. Ernst Abbe
• Perhaps the greatest design sin is amplifying an angular
error by a lever arm. This is known as an Abbe error:
• Abbe error illustrated through the use of a dial caliper and a
micrometer.

Abbe error
at caliper tips

(Bearings in-line) (Bearings displaced)

• The structural loop is the load path from the part, to the
tool, through the structure, into the fixture, and back into
the part.
• These structural loop concepts apply to many different
types of machines
• The principle of the Abbe error extends to locating bearing
surfaces far from the workpiece area of the machine tool.

2-5
 © 1994 by Alexander H. Slocum

The principle of reversal

• A method that is used to take out repeatable measuring
instrument errors from the measurement:

δ CMM (x)
δ part (x) before reversal

after reversal
Z probe before reversal(x) = δ CMM(x) - δ part(x)

Z probe after reversal(x) = δ CMM(x) + δ part(x)

-Z probe before reversal(x) + Z probe after reversal(x)

δ part(x) =
2
• One of the principal methods by which advances in accuracy
of mechanical components have been continually made.
• There are many variations for measurement and
manufacturing:
• Two bearings rails ground side-by-side can be installed
end-to-end.
• A carriage whose bearings are spaced one rail segment
apart will not pitch or roll.
• It shows how repeatability is often the most important
characteristic of the system.

2-6
 © 1994 by Alexander H. Slocum

Sensitive directions
• It is important to note which are the sensitive directions o f
the machine:

Nonsensitive
direction

ε
r
Workpiece
Sensitive
direction

Part
axis

Tool Sensitive
direction

• An error motion ε of a tool tangent to the surface of a round

part of radius r in a lathe results in:
• A radial error in the workpiece of magnitude ε 2 /r, which
is much smaller than ε.
• Sensitive directions can be fixed (e.g., a lathe), or rotating
(e.g., a jig borer).
• Sensitive directions can be controllable:
• Z and r directions in a lathe.
• Sensitive directions can be uncontrollable:
• Tool orientation errors in a Cartesian machine.
• Effort should not be expended to reduce errors that are
already inconsequential.

2-7
 © 1994 by Alexander H. Slocum

Error assessment and budgeting

• Given all the different types of errors that can affect all
different components:
• Keeping track of all the errors is such a daunting task:
• Most engineers don't bother and use "experience" to
guide the design.
• It is left up to manufacturing and service to work the
bugs out.
• This seems to be a major source of reliability and
performance problems.
• The solution to a successful project is a good budget:
• A project requires a good financial budget to make it
feasible.
• A project requires a good time budget to make it
feasible.
• A project requires a good error budget to make it
feasible.
• In order to make a good error budget for the system, a good
mathematical model is needed.

2-8
 © 1994 by Alexander H. Slocum

Homogeneous transformation matrices for error

assessment and budgeting
• Allows the designer to consider one part of the machine at a
time, and then link them all together like beads on a string.
• Based on rigid body model of a linear series (open chain) of
coordinate frames.
• Takes into account linear and angular offsets between
coordinate frames.
• Transforms XYZ coordinates of one frame into XYZ
coordinates of another frame:

Y1
a, b, c
YR

Z1 X1

ZR XR
• Used to transform locally referenced errors into errors
referenced with respect to the toolpoint and the workpiece.
• Coordinate frames are placed at bearings, joints, and areas
where other parameters are lumped.
• Closed chains (e.g. a five point bearing mount) need to be
modeled with the generation of constraint equations.

2-9
 © 1994 by Alexander H. Slocum

Structure of a homogeneous transformation matrix

• The first three columns are the direction cosines (unit
vectors i, j, k).
• They represent the orientation of the Xn , Yn , and Zn
axes with respect to an adjacent coordinate frame.
• The last column is the position of the rigid body's coordinate
system's origin with respect to the reference frame.
• The pre-superscript represents the reference frame in which
you want the result to be represented.
• The post-subscript represents the reference frame from
which you are transferring:
• The “O’s” are rotations (direction cosines” and the
“P’s” are translations:
Oix Oiy Oiz Px
Ojx Ojy Ojz Py
RT =
n Okx Oky Okz Pz
0 0 0 1

• The equivalent coordinates of a point in a reference frame n ,

in a reference frame R are:
XR Xn
YR Yn
= RT
n
ZR Zn
1 1

2-10
 © 1994 by Alexander H. Slocum

• Example: If the X1 Y 1 Z 1 coordinate system is translated by

an amount x along the X axis:
• The HTM that transforms a point's coordinates in the
X 1 Y 1 Z 1 frame into the XYZ reference frame is:

1 0 0 x
0 1 0 0
XYZT =
X1Y1Z1
0 0 1 0
0 0 0 1
• Example: If the X1 Y 1 Z 1 coordinate system is rotated by an
amount θ X about the X axis:

• The HTM that transforms the coordinates of a point in the

X 1 Y 1 Z 1 coordinate frame into the XYZ frame is:

1 0 0 0
0 cosθx -sinθx 0
XYZT =
X1Y1Z1
0 sinθx cosθx 0
0 0 0 1
• The results are similar for translations and rotations along
and about the Y and Z axes.
• Often it can be difficult to determine how a part modeled as
a rigid body actually moves.
• Care must be taken when evaluating the error terms in
the HTMs of systems with multiple contact points (see
example later).

2-11
 © 1994 by Alexander H. Slocum

• Transformation from the Nth axis to the reference system

will be the sequential product of all the HTMs:
N
RT
N = ∏ m-1T
m = 0T1 1T2 2T3 . . . .
m=1

• The relative error HTM Erel between the tool and workpiece
in the tool coordinate frame is:
-1
Erel = RTtool RTwork
• E rel is the transformation that must be done to the toolpoint
in order to be at the proper position on the workpiece.

Tool

R
Error
Work

2-12
 © 1994 by Alexander H. Slocum

• One must consider how the axes will be required to move

(order) in order to create the desired motion.
• For the general case of a machine with revolute and
translational axes.
• Most machine tools and CMMs have only translational axes.

• The error correction vector R P correction with respect to

the reference coordinate frame can be obtained from:
R Px R Px R Px

Py = Py - Py
Pz correction Pz work Pz tool
• Because of Abbe offsets and angular orientation errors of
the axes:

• R P correction will not necessarily be equal to the position

vector P component of Erel .

• R P correction gives the motions the X, Y, and Z axes must

make to compensate for toolpoint location errors.

Y axis: Can be used to

compensate for straightness
errors in the X axis.

X axis: Can be used to

compensate for straightness
errors in the Y axis.

• The controller multiplies the part height by the axis' pitch &
roll to yield the Abbe error for which orthogonal axes must
compensate.1

1 Two experts on error mapping and compensation are Prof. Robert Hocken,
University Of North Carolina, Charlotte, NC 28223and Dr. Alkan Donmez, NIST,
Building 233, Room B106, Gaithersburg, MD 20899.

2-13
 © 1994 by Alexander H. Slocum

Rigid body models of machine components

• The HTM for a linear motion carriage with small errors is:

1 -εZ εY a+δX
εZ 1 -εX b+δY
RT
nerr =
-εY εX 1 c+δZ
0 0 0 1
Vertical
straightness error δ y

Yn

Yaw ε y
YR

X axis servo
XR Roll ε x error δ x
Xn
ZR
Zn
Pitch ε z

Horizontal
straightness error δ z

• If a system is over constrained (non deterministic), then in

order to model system errors, one has to:
• Make an intelligent estimate.
• Rely on experience or direct measurement.
• Perform a sometimes-complicated analysis of the system
(model contact points as springs and solve the constraint
problem).

2-14
 © 1994 by Alexander H. Slocum

Estimating position errors from modular bearing

catalog straightness data
• The HTM method is powerful, but from where does one get
estimates of the errors?

• The HTM assumes that the errors occur at the center of

stiffness of the carriage.

• The center of stiffness is the point at which when a force

is applied to the system, no net angular motion results.

• The first step is to draw a sketch of the system in the ideal

and erred states:
YCS

Y∆CS
εy

δy3
L
δz3

δy δy2
δy4 δz
δz2
δz4

δy1
Z∆CS X∆CS
δz1 εx
εz
ZCS XCS

2-15
 © 1994 by Alexander H. Slocum

• Consider the elements of the HTM we are searching for:

1 -εZ εY a+δX
Xcs X∆cs
Ycs εZ 1 -εX b+δY Y∆cs
=
Zcs -εY εX 1 c+δZ Z∆cs
1 1
0 0 0 1
• The translational errors are based on the average of the
straightness errors experienced by the bearing blocks:

δx = δservo
δy1 + δy2 + δy3 + δy4
δy =
4
δ
δz = z1 + δ z2 + δz3 + δz4
4
• Bearing block straightness is a function of bed accuracy and
running parallelism of the bearing block to the bearing rail:

40

∆ Normal grade
Running parallelism (µm)

30

High grade
20
Precision grade

Super precision grade

10

Ultra precision grade

1000 2000 3000 4000

Rail length (mm)

2-16
 © 1994 by Alexander H. Slocum

• The angular errors are based on the differences in the

average straightness errors experienced by pairs of bearing
blocks acting across the carriage:

δy2 + δy3 δy1 + δy4

-
εx = 2 2
W
δz3 + δz4 δz1 + δz2
-
εy = 2 2
L
δy1 + δy2 δy3 + δy4
-
εz = 2 2
L
• Note we assumed for the starightness we assumed all the
errors are acting in the same direction.
• Here we assume one set of errors acts up, and the other acts
down.
• This is very conservative (makes up for other effects we
might miss).

2-17
 © 1994 by Alexander H. Slocum

Example
• Kinematic systems are by nature deterministic.
• Using an FFT, error motions can usually be traced to
their source without a great deal of difficulty.
• Kinematically supported (five rolling bearings on a vee and
flat) carriage's straightness error:

2.00E-06
1.50E-06
Carriage Straightness (m)

1.00E-06
5.00E-07
0.00E+00
-5.00E-07
-1.00E-06
-1.50E-06
-2.00E-06
0.00E+00 2.00E-01 4.00E-01 6.00E-01
Position (m)

• Once the machine is built, test it and go back and verify (or
correct) the model, so you will be better the next time.

2-18
 © 1994 by Alexander H. Slocum

• Fourier transform of the kinematic carriage's straightness

error:
Peaks likely due to rolling elements
2.50E-07 (ball and cam roller surface errors)
Overall bow in rail
Surface finish effects
2.00E-07
Amplitude (m)

1.50E-07

1.00E-07

5.00E-08

0.00E+00
1.00E-03 1.00E-02 1.00E-01
Wavelength (m)

• The Fourier transform, when plotted as error amplitude as a

function of wavelength, is an invaluable diagnostic tool.
• It can help identify the dominant sources of error, so
design attention can be properly allocated.
• In a kinematic system, once the source of error is identified,
it is more easily corrected.
• This is due to the fact that it is easier to see where the
error came from.
• Remember, for a rolling element:

• The center moves πD, while the element that rolls upon it
moves 2πD !

2-19
 © 1994 by Alexander H. Slocum

Rigid body models

• The HTM for motion and errors about an axis of rotation is:
8
RT
nerr
= ∏ Eq. 2.2.1 = (Eq. 2.2.3)(Eq. 2.2.4) ... (Eq. 2.2.8)
i=3

ZR Axis of rotation Zn

θZ Yn
Rotating body
δZ
εX δY δX εY
YR
XR
Xn

• With S = Sine and C = Cosine, the general result is:

CεYCθZ - CεYSθZ SεY δX
SεXSεYCθZ + CεXSθZ CεXCθZ - SεXSεYSθZ - SεXCεY δY
RT =
nerr
- CεXSεYCθZ + SεXSθZ SεXCθZ + CεXSεYSθZ CεXCεY δZ
0 0 0 1
• This general result may also be used for the case of a linear
motion carriage if ε Z is substituted for θ Z .

• Radial error motion conservatively is assumed to be the

ABEC bore tolerance.
• Tilt error motion is conservatively nassumed to be the ABEC
axial tolerance divided by the bearing OD.

2-20
 © 1994 by Alexander H. Slocum

• Many components of the total error motion must be

considered:

MRS center error motion value

PC center error motion value

PC center

Inner motion
Outer motion

MRS center

Total Error Motion

Average Error Motion Fundamental Error Motion

• The average error motion is indicative of the form error

that will be imparted to the part when held in a lath spindle.
• The asynchronous error motion is indicative of the surface
finish that will be obtained.

2-21
 © 1994 by Alexander H. Slocum

• The FFT is a vital tool for identifying the source of

errors in so the designer can seek to minimize them:2
500

400 Displacement due to

Displacement (nanometers)

machine deformation

300

200

100

0
0 50 100 150 200 250
Frequency (Hz)

• The spindle speed was 1680 rpm (28 Hz).

• The bearing inner diameter was 75 mm.
• The outer diameter was 105 mm.
• The number of balls was 20.
• The ball diameter was 10 mm.
• The contact angle was 15o .
• Was the 2x speed error from out-of-round bore or shaft, or
is it a balancing problem?
• The FFT gives you a hunting direction. Measure the
bore and the shaft.
• The subject of vibration-induced errors will be discussed
later on in greater detail.
• Bearings can be evaluated before they are mounted using an
ultra-precision rotary table and careful measurements.3

2 This figure is not related to the previous one.

3 Contact Steve Sanner at Professional Instruments Company, 7800 Powell Road,
Hopkins, MN 55343, (612) 933-1222

2-22
 © 1994 by Alexander H. Slocum

Error types
• Geometric errors:
• Errors in the form of individual machine components:
• Component straightness error due to machining
errors.
• Component straightness error due to gravity loading.
• Surface finish effects.
• Quasi-static accuracy of surfaces moving relative to each
other (e.g., linear or rotary motion axes):
• Linear motion axis:
• Pitch.
• Roll.
• Yaw.
• Straightness (2 components).
• Linear displacement.
• Rotary motion axis:
• Radial motion (2 components in fixed coordinate
frame, 1 component in rotating frame.)
• Axial motion.
• Tilt motion (2 components).
• Angular motion about axis of rotation.

2-23
 © 1994 by Alexander H. Slocum

• Kinematic errors:
• Errors in an axis's trajectory that are caused by
misaligned or improperly sized components.
• Orthogonality between axes.
• Parallelism between axes.
• Error motions in a closed kinematic chain.
• External load induced errors:
• Errors due to deformation of components:
• Gravity load induced errors.
• Cutting/probing force induced errors.
• Axis acceleration load induced errors.

2-24
 © 1994 by Alexander H. Slocum

• Thermal errors:
• Mean temperature other than 68 °F (20 °C).
• Gradients in environment's temperature.
• Errors caused by thermal expansion of elements:
• External heat sources:
• Mean temperature of the room.
• Sun shining through the window onto the machine.
• Nearby machine's hot air vent.
• Overhead lights.
• Operator's body heat.
• Internal heat sources:
• Motors.
• Bearings.
• Machining process.
• Pumps.
• Expansion of compressed fluids.

2-25
 © 1994 by Alexander H. Slocum

• Dynamic errors:
• Errors caused by vibration or control processes:
• Vibration:
• External environment.
• Cutting process.
• Rotating masses.
• Control system:
• Algorithm type (e.g., PID, adaptive, etc.).
• Stick-slip friction.
• Varying mass.
• Varying stiffness.
• Switching amplifiers
• One grinding machine was plagued by
“washboard” which had the same pitch regardless
of table speed.
• The problem was eliminated when the spindle
motor switching amplifier was changed from 10
kHz to 30 kHz!

2-26
 © 1994 by Alexander H. Slocum

• Instrumentation errors:
• Errors associated with sensors:
• Intrinsic accuracy.
• Interpolation.
• Mounting errors:
• Position.
• Mounting stress.
• Calibration (error associated with the mastering process).

2-27
 © 1994 by Alexander H. Slocum

• Computational errors:
• Error introduced in the analysis algorithms.
• Rounding off errors due to hardware.
• Additional sources of error (often very difficult to model):
• Humidity.
• Loose Joints.
• Dirt.
• Variations in supply systems:
• Electricity.
• Fluid pressure.
• Operator inattention.
• Fluid supply cleanliness.
• Operators (“52-2” factor)

2-28
 © 1994 by Alexander H. Slocum

Combinational rules for errors

• 3 Common types of errors which can be mixed in different
ways:
• Random - under apparently equal conditions at a given
position, errors that do not always have the same value,
and can only be expressed statistically.
• Systematic - which always have the same value and sign
at a given position and under given circumstances.
• Generally can be correlated with position along an
axis and can be corrected.
• If the relative accompanying random error is
small enough.
• Hysteresis - a systematic error which in this instance is
separated out for convenience.
• Usually repeatable, sign depends on the direction of
approach, and magnitude partly dependent on the travel.
• May be compensated for if the direction of approach is
known and an adequate pre-travel is made.

2-29
 © 1994 by Alexander H. Slocum

• 4 ways in which errors can be added together:

• Complete Detail - Generate a map of the resultant
displacement and orientation errors as a function of
slideway position, time, etc.
• Upper bound (implies all errors at a maximum
simultaneously):
N
Eworst case = ΣEsystematic + ΣEhysteresis + ∑ Erandom, PV
i=1
• Lower bound:
1
N
E best case = ΣE systematic + ΣE hysteresis + K 1 ∑ (E random, PV i) 2 2
Ni = 1

• K is a factor that depends on the probability distribution

of the error signal:
• Uniform distribution: K = 3.46.
• Gaussian distribution: K = 4.0 (typical).
• Expected Value - Estimate of composite error by averaging
the upper bound estimate with the lower bound estimate:

Eexpected = 1 Eworst case + Ebest case

2

• Next Generation: Dr. Dan Frey’s Monte Carlo simulation of

the toolpath motion.

2-30