10.1007 - 978 3 030 83765 5 - 6 - Ettb
10.1007 - 978 3 030 83765 5 - 6 - Ettb
10.1007 - 978 3 030 83765 5 - 6 - Ettb
Control
1 Introduction
Ultra-precision (UP) machining requires guiding the workpiece and tool with suffi-
ciently high accuracy and low vibration such that high form accuracy and low surface
roughness can be achieved. Often axis dynamics are limited deliberately in order to
reduce the excitation of vibration and dynamic path errors from tilting slides or
lacking trajectory tracking performance. Hence, a new motion system is required to
allow for UP-machining at significantly increased axis dynamics. Characteristics of
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 147
E. Brinksmeier and L. Schönemann (eds.), Ultra-precision High Performance Cutting,
Lecture Notes in Production Engineering, https://doi.org/10.1007/978-3-030-83765-5_6
148 P. Schreiber et al.
the feed motion are largely determined by trajectory planning, axis controls, and the
applied bearing and drive technology and its transmission characteristics. The Ultra-
Precision High Performance Cutting (UP-HPC) machine concept, which is described
in Sect. 2 and shown in Fig. 1 is a three-axis machining concept. It incorporates an
electromagnetic guide which adds five additional degrees of freedom (DoF). They are
partly redundant and can be utilized as additional fine positioning axes. The dynamic
properties of those redundant axes differ significantly from the other machine axes.
Also, axis dynamics are subjected to variations of the system setup such as changing
mass from the workpiece or the workpiece fixture. Thus, system models used for
instance for model based control or controller tuning not only need to be accurate
but also adaptable. Therefore, in order to obtain a model that fulfills the require-
ments of UP-machining, Sect. 3 of this work presents an approach to estimate actual
parameters of a state-space model during runtime. The proposed approach is able to
learn and adapt to varying conditions even if these conditions are not implemented
in the initial model. Moreover, the section addresses how substantial differences of
sub-system dynamics can be handled by the use of residual sub-band models.
Besides online parameter adaptation, another prerequisite for designing and opti-
mizing a complex axis control system is a feasible, external system to gain compa-
rable data on multi-axis positioning behavior [6]. Available systems such as laser
tracker and grid encoders lack either resolution or range. Therefore, in Sect. 4 a new
approach is introduced which combines the precision of laser speckle measurements
with a suitable measurement range. This contactless system operates without a cor-
poral solid measure and can easily be adapted to different kinematic set-ups and
machine tools.
Dynamic errors that are induced by the acceleration and jerk of axis feed motion
can affect machining quality. Thus, in UP-machining feed velocity, acceleration, and
jerk values are commonly limited to a large extent. However, in order to achieve
the targeted performance within the UP-HPC approach other measures to mitigate
dynamic errors are required. In Sect. 5, the capabilities of an active guide in combi-
nation with vibration avoidance in order to improve process performance and pro-
ductivity are shown. In order to mitigate the excitement of vibration, input shaping
is investigated for the feed axis system. Altintas and Khoshdarregi [1] and Sencer
et al. [14] showed that input shaping can effectively suppress vibration in conven-
tional machining. The use of this approach in ultra-precision machining has not been
investigated in depth. Secondly, the input shaping is investigated in combination with
a force feed forward control of acceleration loads. It is shown that both approaches
can improve productivity significantly. Further it is shown that the preferable choice
of compensation approaches in order to improve productivity depends on the aimed
cutting velocity and form deviation. To further support process and compensation
parameterization the capability of internal sensor data in order to predict workpiece
form deviations is evaluated. Finally, the results of this chapter are concluded and an
outlook on future work is given in Sect. 6.
Ultra Precision High Performance Axis Control 149
2 System Description
This section concerns the axis controls of the machine prototype which was developed
and set up in the project. The design of the prototype is shown in Fig. 1. The kinematic
design comprises three macro axes X , Y , Z . Another five fine-positioning DoF are
contributed by the electromagnetic guide of the X-axis. Those DoF are denoted with
their generalized coordinates q y and q z for the redundant linear DoF and q a , q b ,
and q c for the angular DoF. The X -axis is placed within the slide of the Z -axis as
a box-in-a-box design. Details on the design of the electromagnetic guide are given
in Chap. 4. The Z -axis is guided by conventional porous media aerostatic bearings.
Both, the X - and Z -axis are driven by two iron-less linear drives. The Y -axis utilizes
a conventional roller guide, driven by a rotary motor with a mechanical brake. The
Z -axis position is acquired by two linear encoders. The X - and Y -axis are equipped
with one encoder each.
In a machining process, the macro axes X , Z provide the motion for feed and
crossfeed and vice versa. Infeed is provided by the fine-positioning axis q y , while
the macro Y -axis is used for pre-adjustment of the spindle and tool position in advance
of the machining process. The Y -axis clamped and kept stationary with the brake
during the process. The Y -axis can carry a fly-cutting spindle or a stationary tool for
planing. Figure 1 shows the setup with the fly-cutting spindle.
The controllers of all motion axes as well as acquisition and processing of the sensor
data are executed on an industrial PC (IPC) system (Beckhoff). To this end, a modular
control framework was developed. The basic structure of the framework is shown
in Fig. 2. Modularity concerns flexible involvement of the prototypes’ nine DoF in
motion and machining as well as modularity of each of the axes’ controller and
compensation modules. Online change of axis configuration as well as change of
controller and compensation properties, was a fundamental requirement. This is
due to the fact that an offline change and recompilation of modules would require
shutting down the levitating axis. Shutting down the levitating axis however can
150 P. Schreiber et al.
lead to diverging from a thermostatic state and is thus not feasible for conducting
repeatable experiments. A state machine module handles system and module states
such as macro axis clearance depending on the levitation axis state or activation of
compensation parameters depending on axis states. Also base functionalities such
as homing are coordinated by the state machine. Set-point values for the axes are
generated by a G-code interpreter which also handles interpolation for up to three
linear axes. The modular system allows for flexible multi-axis interpolation with the
fine-positioning DoFs of the magnetic levitation guide in any combination with the
macro axes. Additional signals can be linked onto the output of the interpolation
system for compensation applications. The individual axis controller modules are
developed in Simulink and exported as IEC-61131 structured text using the Matlab
PLC-Coder extension. The controller modules, which control position and velocity,
are executed at 16 kHz on the IPC. The controller output, the drive current set-point
value is transferred to the servo drive controllers via an ethernet for control and
automation technology (EtherCAT) bus system. The current control is executed by
the servo drive controllers. They also connect the position encoders to the system
and transfer actual position and velocity to the IPC. For details regarding the control
structure of the magnetic guide system please consider Chap. 4, Sect. 4.4.
Cutting experiments were conducted, in order to evaluate the behavior of the axis
and control system and compensation approaches in terms of their impact on the
Ultra Precision High Performance Axis Control 151
machining quality. For the experiments a planing process with stationary tool attached
to the Y -axis was implemented. A planing process was chosen in order to evaluate the
axis control system. In contrast to a fly-cutting process planing allows to exclude cer-
tain effects from the machining result. Those effects are for instance vibrations from
spindle imbalances. Another disadvantage of fly-cutting in terms of investigating axis
controls is that it leads to a form of “undersampling”. For a typical fly-cutting spindle
speed of 1,500 min−1 and a single cutting edge there is a short tool-workpiece con-
tact at a frequency 25 Hz. Hence, axis and prototype behavior above 12.5 Hz would
already be distorted by aliasing effects in the machining result. The setup for planing
experiments is shown in Fig. 3. The stationary tool, is attached to the Y -axis such
that the tool center point (TCP) is close to the TCP for the fly-cutting set-up. Thus,
a part of the effects from structural dynamics of the Y -axis can be expected to be
apparent in both, the fly-cutting and planing set-up. A manual linear stage was used
for presetting the tool position while the depth of cut (a p ) is set by the fine posi-
tioning DoF q y of the electromagnetic axis. The direction of cut corresponds to the
Z -axis. For avoiding tool-workpiece contact during the retracting movement of the
Z -axis, the q y -axis was lowered to maximum extent while resetting the workpiece.
In advance of the actual cutting experiments workpieces were planed entirely in this
setup without re-clamping in order to achieve the best possible parallelism of the
workpiece surface to the machine axes.
This section gives an overview over the system capability of the described axis and
control system as a reference for further experiments. Figure 4 shows an exemplary
machining result of a planing process with the above described setup. The workpiece
(aluminum EN AW 5754) was machined in multi-axis interpolation with at a cutting
speed vc = 500 mm min−1 in Z -direction. Regarding axis control settings, the static
152 P. Schreiber et al.
Fig. 3 Setup for cutting experiments (left and center) with the shaping process (right)
Model based axes control for high speed UP-operations demands for high complexity
and associated computational costs to represent the machine behavior within the
required resolution and accuracy. To reduce the computational costs for realizing real
time application, the complexity can be reduced by in-process parameter estimation
and adaption that ensures the required precision [5].
Ultra Precision High Performance Axis Control 153
Fig. 4 Overview over system capability with exemplary surface and profile measurement
The model consists of two sub-models, the motion band sub-model (MBSM) and
the residual band sub-model (RBSM). While the MBSM calculates the larger scale
motion of the feed axes, vibrations and discrepancies are represented by RBSM. Due
to this reduction, the parameters of both sub-models can be estimated separately and
in individually suitable cycle times.
Each sub-model is represented as state-space model. In such models, the system is
expressed by the state vector x with n state-variables. n was chosen to 8 (MBSM) and
6 (RBSM). Each state variable behaves as energy storage and records the influence
of past inputs to the actual state. A state vector at a time t for an input vector u can
be calculated as
ẋ(t) = Ax(t) + Bu(t) (1)
Hereby, A is the state matrix which defines the single state variables interaction
and B is the input matrix that defines the input influence on the state variables.
The output y of the model can be calculated from the state vector with the output
matrix C.
y(t) = C x(t) (2)
The output of both sub-models must be phase corrected and added to the final
result. The complete process is shown in Fig. 5.
The parameters of the sub-models were separately estimated based on the predic-
tive error method (PEM) [11]. Therefore, the future model output is estimated with
154 P. Schreiber et al.
AL
u y
residual band sub-model (RBSM)
AH
u ymax +/-
identification I identification II
Eq. 2 and compared to the measurement at that time. The lowest deviation was then
found with the Levenberg-Marquardt method [10]. The separation of the estimated
and the measured data was realized with equiripple low-pass finite impulse response
filters due to their constant group delay. This feature ensures that the unavoidable
distortion of the motion data in time domain is independent of the velocity. For the
MBSM parameter estimation all frequencies 20 Hz while for the RBSM parameter
estimation all frequencies 200 Hz were suppressed. The procedure is shown in Fig. 6.
A window of the measurements with a fixed size of west samples and the input u is
used as input for the parameter estimation. The window lengths were heuristically
chosen as 83.33 ms (MBSM) and 50 ms (RBSM). To react to parameter variation
of the system, the parameters are estimated in each window individually. After this
estimation and a comparison with the previous parameter, the parameters with the
lower deviation are chosen as new model parameters. To get a robust system, the
parameter changes were limited to 15% of the actual value.
Ultra Precision High Performance Axis Control 155
50
µm
Position z
30
Load
20
10
Piezo
stage 0
0 1 2 3 s 5
Z Time t
Y
X Z-axis 51 2
µm µm
49 0
48 -1
4 s 5 3.5 s 4.5
Fig. 7 Model parameter identification a Setup b Z -position with set-point values (blue) and actual
values (red); cf. [5]
The identification algorithm was tested with a piezo stage “NanoCube” (P-611.3,
Physik Instumente) in open-loop control mode. After initial identification, an addi-
tional load on the NanoCube in z-direction (0, 22, 31, 52, 77 g) as variable parameter
simulated a change in the system and hence its behavior. Both the input to the piezo
stage (set-point values) and the internal position sensor of the piezo stage (actual
values) were used for the in process parameter estimation. As demanded motion in
Z -direction a step function with an amplitude of 50 µm was applied to emulate a
positioning task. This demanded trajectory was performed five times in a row with
each load. The waiting time between positioning was one second. An example of the
set-point trajectory and the corresponding actual trajectory is shown in Fig. 7.
The estimation error as difference between the measured actual position values
and the estimated position values calculated by the model is illustrated in Fig. 8. Mean
values were calculated during the stand still in the target position. The initial model
parameter estimation gains a quite good prediction with an estimation error less than
1 nm. Nevertheless, changing the physical system by additional load increases the
estimation error about several nanometers.
To judge the effects of parameter estimation and path control a measuring device
based on speckle pattern was developed. This sensor works without a corporal solid
156 P. Schreiber et al.
12
initial identification
Estimation error e nm
2 dynamic identification
-2
0 20 40 g 80
Load l
Fig. 8 Mean model estimation error during positioning tests with different loads with static and
dynamic parameter estimation
measure and can easily be adapted to different machine tools. The following sections
describe the principle of the measurement, the range extension procedure and results.
As general procedure, the first image is assumed to be the reference that is placed
in a known reference position. All displacements are calculated in respect to this
reference. The method to calculate the displacement consists of two steps. In the first
step, the position is roughly estimated by two-dimensional cross correlation of the
Ultra Precision High Performance Axis Control 157
highspeed-camera
camera Optronis CP80-4-M-500
500-80,000 fps
2304 x 1720 pixel
Laser-
half-mirror
source Laser- laser diode optics
beam ≈ 650 nm half-mirror
2 mW convex lenses
adjustable
test surface replaceable
Fig. 10 Displacement
between images
actual image (test pattern) with the reference. These results are highly affected by
the boundary values. Therefore, a rectangular window function was applied to set the
values at the window margins to zero. These margins were fit to be larger than the
maximum expected displacement between two consecutive captures to guarantee
an equal amount of summands in the cross correlation calculations. As a result,
the displacement is presented as a two-dimensional vector of integer values that
represent the number of acquired digital pixels. To enhance this limited resolution a
sub-pixel interpolation was applied in a second step. Therefore, a Gaussian function
is fitted onto the position of the maximum value of the correlation function and its
neighboring pixels. The maximum of the fitted curve represents a better estimation
of the displacement. This method of sub-pixel interpolation is performed for both
dimensions independently.
For higher displacements, the overlap of the images becomes too small for a
precise calculation. To handle this issue, the algorithm was extended with a function
for reference pattern changing. Therefore, a threshold value was introduced. In case
the displacement exceeds this value, the reference pattern is exchanged with the
158 P. Schreiber et al.
Fig. 11 Algorithm with reference pattern change, picture credit: [13], CC BY 4.0
Fig. 12 Test pattern (blue) traced with speckle sensor (orange) in 100 µm, 10 µm and 1 µm scale
(cf. [4]), picture credit: [13], License: CC BY 4.0
actual test pattern. For defining an adequate threshold value, divergent effects must
be considered. These are basically the effective pixel size that defines the ratio of
acquired pixel size and its real dimension [2] and the size of the images [3]. The
complete algorithm is shown in Fig. 11.
A test pattern was traced with a piezo stage “NanoCube” (P-611.3, Physik Instru-
mente) and measured with the speckle sensor in different scales see Fig. 12. Sufficient
path accuracy was verified by the internal measuring system. The speckle sensor per-
formed best in the middle scale with a moderate number of pattern changes. In the
large scale, the high number of pattern changes reduced the accuracy whereas in the
small sub-µm scale the whole system is limited by noise.
A significant reduction of the number of pattern changes could be achieved by
a cross-referencing algorithm. For that purpose, mapping of all images that were
used as reference patterns allowed for defining the shortest connection between each
actual pattern and the initial reference by Dijkstra algorithm for recalculation of the
measured shift. Especially for long paths in small areas, the tool center point crosses
Ultra Precision High Performance Axis Control 159
Fig. 13 Deviation of axis movement from test pattern. Left: without cross-referencing; Right: with
cross-referencing
its own trajectory several times and the cross-referencing significantly shortens the
path that is used in calculations. The effect is illustrated by tracking an edge of a
repeatedly covered rectangle, see Fig. 13.
The following section investigates the application of vibration avoidance and feed
forward compensation of dynamic deviations caused by axis motion. Emphasis of
the investigation is the impact of the proposed methods on productivity.
Dynamic and static path errors can influence workpiece quality in machining. Due
to finite stiffness of components and bearings, the acceleration of machine tool axes
lead to displacements and rotations in DoF that are supposed to be locked. In case
of the active electromagnetic workpiece slide, such displacements can be measured
continuously and actively compensated in order to increase accuracy. Here, position
offsets as well as force offsets can be imposed upon the electromagnetic guide set-
points. Figure 14 illustrates the basic structure therefore. Position offsets q pos can
be added to the set-point values qset of the magnetic guide in order to account for
static errors such as straightness deviations of the guides. A force feed forward Q f f
is added on the output of the electromagnetic guide control adding up to the set-
point vector of generalized forces Q set . Force feed-forward is utilized to counteract
160 P. Schreiber et al.
Fig. 14 Block diagram of the dynamic error feed forward, based on [9]
disturbance forces acting on the guide before they can cause a displacement. This
requires that these disturbance forces can either be measured with low delay or be
predicted. Here, force-feed forward is applied to loads from acceleration of the Z -
axis as these are the dominating disturbance forces on the workpiece slide and can
be derived from the acceleration set-point.
It was shown in simulation in [9] for the given setup that an acceleration force
feed-forward can significantly reduce the tilting error qa of the magnetic slide’s A-
axis for accelerated motion of the Z -axis. Empirical analysis showed that significant
deviations also appear in qz . Thus, for feed forward a force vector Q f f is set based
on the set-point acceleration z̈ set :
Fig. 15 Block representation of a zero vibration input shaper (left) and impulse response (right)
based on [15]
and
K
A2 = (6)
1+K
where,
√ζ π
K =e 1−ζ 2 (7)
In order to evaluate the effect of input shaping (IS) and acceleration feed forward (FF),
cutting tests with stationary tool according to the setup above were conducted. As
shown in Fig. 16 separate sections were planed with three different cutting velocities
vc , two jerk values j and three compensation settings. The acceleration profile of the
Z -axis was set as a trapezoidal jerk-limited acceleration profile with an upper accel-
eration limit of amax = 5 m s−2 . This value was however not reached for the jerk and
velocity values. As lowest cutting respectively feed velocity, 300 mm min−1 was cho-
sen as reference velocity for a typical UP-cutting process. 5, 000 mm min−1 was cho-
sen as a benchmark velocity for an UP-HPC-cutting process and 2, 000 mm min−1
as a medium velocity. For each of these velocities, sections were planed with jerk
values of 5 m s−3 , 50 m s−3 , and additionally 20 m s−3 for medium velocity. The
lower jerk values serve to benchmark the performance of IS and FF with the conven-
tional approach of limiting axis dynamics. To reduce the effect of cutting forces on
the resulting profile, a narrow diamond tool with rε = 0.011 mm was applied. For
each of the jerk-velocity combinations one section was planed with both, IS and FF
activated, both deactivated, and one section with only IS activated. A summary of
the process and trajectory parameters can be found in Table 1.
The planed sections were each measured with a tactile profilometer with a spher-
ical probe with 500 µm radius since the focus of this investigation is on the profile
deviation rather than roughness. For each section a median profile was calculated
from five profile measurements. Figure 17 shows an exemplary median profile for
vc = 5, 000 mm min−1 . It can be seen that the mentioned rigid vibration mode at
Ultra Precision High Performance Axis Control 163
Fig. 17 Exemplary profile measurements of the machined surface for vc = 5, 000 mm min−1
9.2 Hz results in the dominating waviness in the planed profile with ca. 9 mm wave-
length and over 300 nm in amplitude. The waviness is eliminated in the profiles with
activated IS, thus showing that the compensation is effective. Regarding FF, it can be
seen within the first millimeters of the profile that FF does reduce profile deviation
here, since it reduces the tilting of the X -axis. However, quantitative evaluation is not
conducted for FF as the spherical probe cannot probe the whole length of the profile
and some uncertainty is associated with the alignment of the tactile measurements
to each other in Z -direction.
Due to the limitations regarding the quantitative evaluation of FF, only IS is
considered below. The analysis focuses on the overall profile deviation. Therefore, the
range at the beginning and end of the profile, where the effect from axis acceleration
is visible, are cut from the profile measurements for the following analysis. For the
remaining part of the profile, the standard deviation Sstd is calculated as well as the
peak-to-peak value S p2 p of the profile after filtering with a linear Gaussian profile
filter (according to ISO 16610-21) with a cut-off wavelength of 2 mm. Sstd and S p2 p
are shown in Fig. 18 for both, profiles with and without IS. It can be seen that for all
velocity and jerk combinations, a S p2 p value of less than 100 nm is achieved with the
use of IS while for deactivated IS S p2 p exceeds 100 nm significantly for the higher
164 P. Schreiber et al.
Fig. 18 Profile standard deviation Sstd (left) and filtered profile peak-to-peak value S p2 p (right)
for different cutting velocities vc and jerk values j with and without input shaping
jerk and velocity combinations. This shows that IS can serve to allow for significantly
higher feed velocities in UP-cutting applications.
Fig. 19 Internal sensor values for deviation in qz (left) and roll angle qa (right) for with and without
input shaping and feed forward acceleration compensation for vc = 3, 000 mm min−1
per se as the Z -position is captured in the same measurement. The discussed effect
at 9.2 Hz is also strongly present in all three coordinates and reduced by the use of
IS to a large extent.
To the end of evaluating the feasibility of predicting profile deviation based
on internal sensor data, the standard deviation Sstd of the planed sections profiles
Sect. 5.3 is compared to the standard deviation values of internal guide sensor data.
Figure 20 shows how the Sstd values for internal guide coordinates qz and qa during
an air cut correlate with Sstd values of the measured profiles. The range in z, which is
directly affected from axis acceleration, was excluded from evaluation. It can be seen
that both, the standard deviation of the qz and the qa coordinate show a correlation
to the resulting workpiece profile with R 2 values of 0.86 and 0.89 respectively. As
expected, the correlation is clearer for higher deviations since the smaller values are
more affected by noise and measurement uncertainties. The result shows, that at least
for the given vibration mode, the sensory capability of the guide has the potential
to predict critical form deviations in the workpiece. The observation that deviations
in the guide coordinates are more than a magnitude higher than deviations in the
workpiece also highlights the potential of guide position data in terms of sensitiv-
ity. However, additional modeling and a higher number of experiments has to be
conducted in order to validate the correlation.
Fig. 20 Comparison of internal signal standard deviation vs profile measurement for qz (left) and
qa (right)
data can be used as a practical tool to predict workpiece form deviations. Thus, the
following investigation is based on internal guide coordinate data as they allow for a
better sensitivity and more comprehensive parameter combinations. Again, a shaping
or line milling process at constant vc is considered with a jerk limited trapezoidal
acceleration profile. Figure 21 shows an exemplary measurement of qz for a travel
range of 80 mm. For further consideration, only the range within which the axis
velocity deviates less than 2% from the set-point velocity vset is considered since a
constant velocity is assumed to be a process requirement. The path length required to
accelerate the slide to 98% vset is referred to as z acc and the corresponding duration
as tacc . Analogous, z dec and tdec denote the path length and duration for decelerating
the slide from vset to zero. It is assumed that in order to remain below a specified
profile deviation, qz should not exceed an allowance of q z_t ol . As the focus is on form
deviation, the allowance is applied after low-pass filtering qz with a linear Gaussian
filter at a cut-off wavelength of 2 mm analogously to the profile measurements in
the previous section. Based on that, z f ade and t f ade denote the axis path length and
duration required for the dynamic deviations induced by the slide acceleration to
decay and to keep within the allowance q z_t ol . For deceleration, the residual path
length and duration from leaving the allowance q z_t ol to v = 0 is denoted with zr es
and tr es .
Ultra Precision High Performance Axis Control 167
Fig. 22 Times and distances to reach q z_t ol for vc = 500 (left) and 5,000 mm min−1 (right). JL =
jerk limitation (10 m s−3 ), IS = input shaping, FF = feed forward
line amounts to 0.73 s at a required axis range of 43.2 mm (vc = 5, 000 mm min−1 ,
j = 10 m s−3 , without IS and FF). This corresponds to a 24.5% lower process time
and 15.7% reduction in required axis range. The results show that IS and FF can
allow for significantly increased axis feeds in UP-cutting processes and thus in syn-
ergy with the other UP-HPC approaches discussed in the previous chapters increase
productivity of UP-cutting processes by large extent.
Ultra Precision High Performance Axis Control 169
The increase in productivity with the UP-HPC approach requires the synergistic
interaction of the techniques discussed in the individual chapters. In this chapter,
approaches were discussed which allow for increased axis feeds while keeping
dynamic workpiece deviations within specified allowances. Firstly, an approach for
online parameter estimation of axis models was presented. The methods can be an
enabler to increase productivity as it can enable the use for more sophisticated control
approaches and reduce the effort to identify system parameters manually. To inves-
tigate two-axis motion behavior a laser speckle sensor was introduced that provides
a high accuracy and flexibility without the need for a containable grid-plate which
often prohibits application during cutting in a machine environment.
With increasing feed velocities, also dynamic errors and excitation of vibration
are intensified. Thus, it was investigated how acceleration force feed forward and
input shaping can reduce these effects and can thus allow for higher feed velocities
and productivity. It was shown that for high velocities above 2,000 mm min−1 process
times can be reduced by around 24% without increasing feed velocity. Taking into
account that such high velocities can only be achieved in combinations with the
other approaches developed within the UP-HPC research group, the combined gain
in productivity is even higher.
References
1. Altintas, Y., Khoshdarregi, M.R.: Contour error control of CNC machine tools with vibration
avoidance. CIRP Ann. 61(1), 335–338 (2012). https://doi.org/10.1016/j.cirp.2012.03.132
2. Bloem, A., Foremny, E., Schenck, C., Kuhfuss, B.: Using speckle patterns in a direct positioning
system. In: 4th International Conference on Nanomanufacturing (nanoMan 2014), Bremen,
Germany, 8–10 July 2014 (2014)
3. Bloem, A., Schenck, C., Kuhfuss, B.: 2D position sensor based on speckle correlation. In:
Annoni, M., Fassi, I., Wiens, G.J., Dimov, S. (eds.) 10th International Conference on Micro-
manufacturing (4M/ICOMM 2015), Milan, Italy, pp. 541–544 (2015)
4. Bloem, A., Wilhelmi, P., Schenck, C., Kuhfuss, B.: 2D position sensor based on speckle correla-
tion - experimental setup for high-speed measurements and in field tests. In: 11th International
Conference on Micro Manufacturing (ICOMM 2016), Paper No. 17, Irvine/CA, USA (2016)
5. Bloem, A., Schenck, C., Kuhfuss, B.: Adaptive state-space model for ultra-precision feed axis.
Procedia Technol. 26, 20–26 (2016). https://doi.org/10.1016/j.protcy.2016.08.004
6. Bloem, A., Schenck, C., Kuhfuss, B.: Dynamic behavior of ultra precision feed axis at increased
velocities. In: Wertheim, R., Ihlefeldt, S., Hochmuth, C., Putz, M. (eds.) Procedia CIRP 46:456-
459 (2016). https://doi.org/10.1016/j.procir.2016.04.040.
7. Charrett, T.O.H., Waugh, L., Tatam, R.P.: Speckle velocimetry for high accuracy odometry for
a Mars exploration rover. Meas. Sci. Technol. 21(2), 025301 (2009). https://doi.org/10.1088/
0957-0233/21/2/025301
8. Denkena, B., Dahlmann, D., Sassi, N.: Analysis of an ultra-precision positioning system and
parametrization of its structural model for error compensation. Procedia CIRP 62, 335–339
(2017). https://doi.org/10.1016/j.procir.2016.06.054
170 P. Schreiber et al.
9. Denkena, B., Bergmann, B., Schreiber, P.: Error compensation strategies for productivity
improvement in ultra-precision cutting. In: Euspen’s 19th International Conference & Exhibi-
tion, Bilbao, ES, June 2019, pp. 562-563 (2019)
10. Gill, P.E., Murray, W., Wright, M.H.: The Levenberg-Marquardt method. In: Practical Opti-
mization, chap. 4.7.3, pp. 136–137. Academic Press, London (1981)
11. Ljung, L.: System Identification: Theory for the User, 2nd edn. Prentice Hall, New Jersey
(1987)
12. Peters, W.H., Ranson, W.F.: Digital imaging techniques in experimental stress analysis. Opt.
Eng. 21(3) (1982). https://doi.org/10.1117/12.7972925.
13. Schönemann, L., Berger, D., Dörgeloh, T., Riemer, O., Brinksmeier, E., Krüger, R., Schreiber,
P., Denkena, B., Hochbein, J., Parsa, N., Schenck, C., Kuhfuss, B.: Synergistic approaches
to ultra-precision high performance cutting. CIRP J. Manuf. Sci. Technol. 28, 38–51 (2020).
https://doi.org/10.1016/j.cirpj.2019.12.001
14. Sencer, B., Dumanli, A., Yamada, Y.: Spline interpolation with optimal frequency spectrum
for vibration avoidance. CIRP Ann. (2018). https://doi.org/10.1016/j.cirp.2018.03.002
15. Singhose, W.: Command shaping for flexible systems: a review of the first 50 years. Int. J.
Precis. Eng. Manuf. 10(4), 153–168 (2009). https://doi.org/10.1007/s12541-009-0084-2
16. Tobiason, J.D.: Image correlation displacement sensor. U.S. pat. 8605291. Mitutoyo Corpora-
tion (2013)