Advanced

Adaptation and Learning for Manipulators and Machining

Sörnmo, Olof LU (2015) In PhD Theses TFRT-1110.
Abstract
This thesis presents methods for improving the accuracy and efficiency of tasks performed using different kinds of industrial manipulators, with a focus on the application of machining. Industrial robots offer a flexible and cost-efficient alternative to machine tools for machining, but cannot achieve as high accuracy out of the box. This is mainly caused by non-ideal properties in the robot joints such as backlash and compliance, in combination with the strong process forces that affect the robot during machining operations.



In this thesis, three different approaches to improving the robotic machining accuracy are presented. First, a macro/micro-manipulator

approach is considered, where an external... (More)
This thesis presents methods for improving the accuracy and efficiency of tasks performed using different kinds of industrial manipulators, with a focus on the application of machining. Industrial robots offer a flexible and cost-efficient alternative to machine tools for machining, but cannot achieve as high accuracy out of the box. This is mainly caused by non-ideal properties in the robot joints such as backlash and compliance, in combination with the strong process forces that affect the robot during machining operations.



In this thesis, three different approaches to improving the robotic machining accuracy are presented. First, a macro/micro-manipulator

approach is considered, where an external compensation mechanism is used in combination with the robot, for compensation of high-frequency Cartesian errors. Two different milling scenarios are evaluated, where a significant increase in accuracy was obtained. The accuracy specification of 50 μm was reached for both scenarios. Because of the limited workspace and the higher bandwidth of the compensation mechanism compared to the robot, two different mid-ranging approaches for control of the relative position between the robot and the compensator are developed and evaluated. Second, modeling and identification of robot joints is considered. The proposed method relies on clamping the manipulator end effector and actuating the joints, while measuring joint motor torque and motor position. The joint stiffness and backlash can subsequently be extracted from the measurements, to be used for compensation of the deflections that occur during machining. Third, a model-based iterative learning control (ILC) approach is proposed, where feedback is provided from three different sensors of varying investment costs. Using position measurements from an optical tracking system, an error decrease of up to 84 % was obtained. Measurements of end-effector forces yielded an error decrease of 55 %, and a force-estimation method based on joint motor torques decreased the error by 38 %.



Further investigation of ILC methods is considered for a different kind of manipulator, a marine vibrator, for the application of marine seismic acquisition. A frequency-domain ILC strategy is proposed, in order to attenuate undesired overtones and improve the tracking accuracy. The harmonics were suppressed after approximately 20 iterations of the ILC algorithm, and the absolute tracking error was r educed by a factor of approximately 50.



The final problem considered in this thesis concerns increasing the efficiency of machining tasks, by minimizing cycle times. A force-control approach is proposed to maximize the feed rate, and a learning algorithm for path planning of the machining path is employed for the case of machining in non-isotropic materials, such as wood. The cycle time was decreased by 14 % with the use of force control, and on average an additional 28 % decrease was achieved by use of a learning algorithm. Furthermore, by means of reinforcement learning, the path-planning algorithm is refined to provide optimal solutions and to incorporate an increased number of machining directions. (Less)
Please use this url to cite or link to this publication:
author
supervisor
opponent
  • Prof. De Schutter, Joris, KU Leuven
organization
publishing date
type
Thesis
publication status
published
subject
in
PhD Theses
volume
TFRT-1110
publisher
Department of Automatic Control, Lund Institute of Technology, Lund University
defense location
Lecture hall M:B, building M, Ole Römers väg 1, Lund University, Faculty of Engineering LTH, Lund
defense date
2015-11-13 10:15
ISSN
0280-5316
ISBN
978-91-7623-493-8 (web)
978-91-7623-492-1 (print)
project
ProFlexa
COMET
SMErobotics
language
English
LU publication?
yes
id
fb143acb-10be-46dc-acf0-59435c88d16b (old id 8057993)
date added to LUP
2015-10-20 11:36:42
date last changed
2016-09-19 08:44:49
@phdthesis{fb143acb-10be-46dc-acf0-59435c88d16b,
  abstract     = {This thesis presents methods for improving the accuracy and efficiency of tasks performed using different kinds of industrial manipulators, with a focus on the application of machining. Industrial robots offer a flexible and cost-efficient alternative to machine tools for machining, but cannot achieve as high accuracy out of the box. This is mainly caused by non-ideal properties in the robot joints such as backlash and compliance, in combination with the strong process forces that affect the robot during machining operations. <br/><br>
<br/><br>
In this thesis, three different approaches to improving the robotic machining accuracy are presented. First, a macro/micro-manipulator<br/><br>
approach is considered, where an external compensation mechanism is used in combination with the robot, for compensation of high-frequency Cartesian errors. Two different milling scenarios are evaluated, where a significant increase in accuracy was obtained. The accuracy specification of 50 μm was reached for both scenarios. Because of the limited workspace and the higher bandwidth of the compensation mechanism compared to the robot, two different mid-ranging approaches for control of the relative position between the robot and the compensator are developed and evaluated. Second, modeling and identification of robot joints is considered. The proposed method relies on clamping the manipulator end effector and actuating the joints, while measuring joint motor torque and motor position. The joint stiffness and backlash can subsequently be extracted from the measurements, to be used for compensation of the deflections that occur during machining. Third, a model-based iterative learning control (ILC) approach is proposed, where feedback is provided from three different sensors of varying investment costs. Using position measurements from an optical tracking system, an error decrease of up to 84 % was obtained. Measurements of end-effector forces yielded an error decrease of 55 %, and a force-estimation method based on joint motor torques decreased the error by 38 %.<br/><br>
<br/><br>
Further investigation of ILC methods is considered for a different kind of manipulator, a marine vibrator, for the application of marine seismic acquisition. A frequency-domain ILC strategy is proposed, in order to attenuate undesired overtones and improve the tracking accuracy. The harmonics were suppressed after approximately 20 iterations of the ILC algorithm, and the absolute tracking error was r educed by a factor of approximately 50. <br/><br>
<br/><br>
The final problem considered in this thesis concerns increasing the efficiency of machining tasks, by minimizing cycle times. A force-control approach is proposed to maximize the feed rate, and a learning algorithm for path planning of the machining path is employed for the case of machining in non-isotropic materials, such as wood. The cycle time was decreased by 14 % with the use of force control, and on average an additional 28 % decrease was achieved by use of a learning algorithm. Furthermore, by means of reinforcement learning, the path-planning algorithm is refined to provide optimal solutions and to incorporate an increased number of machining directions.},
  author       = {Sörnmo, Olof},
  isbn         = {978-91-7623-493-8 (web)},
  issn         = {0280-5316},
  language     = {eng},
  publisher    = {Department of Automatic Control, Lund Institute of Technology, Lund University},
  school       = {Lund University},
  series       = {PhD Theses},
  title        = {Adaptation and Learning for Manipulators and Machining},
  volume       = {TFRT-1110},
  year         = {2015},
}