14-06-2012, 05:44 PM
Drafting robot
Drafting Robot.docx (Size: 23.73 KB / Downloads: 35)
Abstract
The modelling and control of an industrial SCARA robot is considered. T h e robot is used in the reticle haodling module of a wafer scarmer. During cperztion t h e robot has to perform repeated (straight-line) motions. A maximum straight-line position error of 0.2 mm is specified to prevent collision with other parts of the handling module.
The robot consists of two links and an end-effector and can move in the horizontal plane. During motion of the robot arm coupling forces and torques are present due to velocity and accelerations of the links. Each link is driven by a motor and between the motor and the link a gear train is present. On each motor an encoder is mounted for position measurement. The chain motor - transmission - load introduces dynamic effects. Furthermore, friction is dominantly present in the gear trains and the joints.
In this research an Iterative Learning Control (ILC) scheme is applied t o the robot. With this control algorithm it is possible t o eliminate reproducible errors of a motion on the basis of error signals. The powerful concept of ILC is illustrated by the achieved reduction of the position errors a t the encoders. The maximum error a t the encoders could be reduced a factor 10 to 25 when compared t o PID controllers. The maximum error of the end-effector could be reduced just within the specifications.
Introduction
For the reticle handling in a wafer scanner a SCARA robot is used. This is a two link robot with an end-effector. The robot has to perform different straight-line motions repeatedly &iring operation. Euring operation vibrations occur when tracking a (straight-line) trajectory. The allowed vibration and tracking error are specified within a narrow band to prevent collision of the end-effector with a submodule of the reticle handler.
This brings us to the problem definition of the project:
I s it possible t o increase t h e straight-line accuracy of t h e robot t o w i t h i n a n e m o r band of 0.2 m m ? From previous research [I] it was already clear tha t this straight-line accuracy could not be reached with only standard classical controllers, such as PID controllers. Additional friction feedforward and inverse model feedforward were necessary t o increase the
accuracy. A major drawback of these feedforwards is that they have to be finetuned for each robot individually. The maximum straight-line accuracy tha t was reached was approximately 0.25 mm. From this research it became also clear tha t the error signals show a large reproducible part. This opens a way for control strategies which aim a t 'learning' of the error signal.