Open access
Autor(in)
Datum
2020-05Typ
- Doctoral Thesis
ETH Bibliographie
yes
Altmetrics
Abstract
Precision manufacturing processes are strongly connected to the accuracy of machine tools.
There is an increasing demand for high precision workpieces, since the quality of the functional
surfaces can be linked to efficiencies of the parts during its operational phase. One
of the largest contributors to errors on machined workpieces are thermal influences of the used machine tools.
Therefore the following thesis deals with the development of an on-machine measurement
cycle, that captures the most dominant thermal errors in a timely manner and the design of
a self-learning adaptive thermal error modeling methodology, which is based on the system
identification theory, in order to resolve the most common problems of thermo-mechanical
error compensation, including model inaccuracy, non-robustness and long-term instability,
model calibration difficulties, lengthy experiments and model adaptation problems. The
developed model is then used to compensate the occurring thermal errors of a 5-axis machine
tool by correcting the axis movements utilizing the numerical control.
The developed on-machine measurement cycle is capable of measuring the five thermal
position and orientation errors of an axis of rotation, as well as the two thermal errors of
the functional table surface. As a measurement instrument a touch trigger probe is used.
Under the given circumstances an extended uncertainty of maximum 0.6 μm respectively
2.6 μm/m is achieved. The validation measurements showed, that the thermal behavior
can be captured and a minimal process intrusion is caused. The measurement cycle also
shows reliable results, when the machine tool is running with metal working
fluid and no signifi cant increase in measurement uncertainty is observed.
A common and effective method to model thermal errors on machine tools is the phenomenological
model analysis, which captures the correlation between the observed thermally
induced errors and the thermal and losses related information. Preferably, the
residual errors between the prediction model and the actual machine tool deviations will approach zero. However, the actual machining conditions may not be identical to the
machining conditions used to derive the model, which leads to model uncertainties. This
can cause complications, especially for small batch productions, where the sequence of
manufacturing processes changes repeatedly as do the direction and rate of change of
thermal effects. Due to statistical uncertainties, assumptions in the model and the everchanging
boundary conditions, the error models derived from pre-process calibration are
not necessarily accurate enough in the long term. They need to be verifi ed and updated
recurrently as the machine tool is continually used. Therefore an adaptive learning control
for thermal error compensation is developed, that combines the functionality of a dynamic
thermo-mechanical model, with fully automatized on-machine measurements of the thermal
errors. This enables self-learning and self-adaptation of the compensation model to
the current thermal state of the machine tool. Therefore a long-term stability and robustness
can be ensured and a minimum of machine downtime due to measurements is ensured.
Experiments on a 5-axis machine tool show, that over the period of 178 hours a reduction
of up to 80% of the thermal errors of a rotary axis C, are achieved. Furthermore is shown,
that the presented approach is capable of handling fast boundary condition changes, such
as fast fluctuations in the environment as well as switching conditions such as the metal
working fluid supply. In such an experiment a reduction of the occurring errors of up to
88% is achieved, which corresponds to an absolute reduction of almost 40 μm.
To demonstrate the reduction capabilities of the thermal adaptive learning control on an
actual machined workpiece a thermal test piece is developed. The thermal test piece offers
the possibility to evaluate the thermal deviations in all three spatial directions, one angular
deviation, as well as the distortion of the workpiece itself. The test piece is designed to
be measured directly on the machine tool, which enables a fully automatized evaluation
of the thermal errors of a machine tool and to monitor the thermal error compensation
quality. Two experiments are conducted, one with and one without active thermal adaptive
learning control. The deviations in the fi rst hour of the heating up phase are reduced by
up to 97% and a reduction of up to 91% over the whole 8 hours of the experiment is
achieved. Mehr anzeigen
Persistenter Link
https://doi.org/10.3929/ethz-b-000446451Publikationsstatus
publishedExterne Links
Printexemplar via ETH-Bibliothek suchen
Verlag
ETH ZurichOrganisationseinheit
03641 - Wegener, Konrad (emeritus) / Wegener, Konrad (emeritus)
ETH Bibliographie
yes
Altmetrics