Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Intelligent control for high-precision positioning
In this project we would like to combine LPV-methods with learning for adaptive control of motion stages, and compare their performance to our recent Bayesian-optimization based approach.
Keywords: linear parapmeter varying control (LPV), learning, motion stage
In high-precision positioning systems used in the semicondoctur industry, it is critical to maintain the highest possible productivity with elevated requirements for precision.Two main challenges call for improved diagnostics and control:
The systems are often tuned robustly considering the worst point of travel.
When a fault occurs, this leads to a degradation in performance and halt of production.
This project addresses mainly the first challenge by proposing the development of advanced control method leveraging real-time encoder data and signals from the closed loop controller and hierarchical predictive and learning control methods. For semiconductor applications the standard method of choice is repetitive control, exploring the occurrence of the same disturbances over identical paths. This approach can be further extended in several directions, interesting both from theoretical and from practical perspective. This project aims to enhance the motion performance of high-performance positioning systems by applying learning approaches to adaptively control the system. We would like to combine LPV-methods with learning, and compare their performance to the Bayesian-optimization approach that we demonstrated recently.
The resulting position-dependent adaptive control algorithm will be used for benchmarking the BO-based algorithm. The main goal of this project is to develop the LPV-controller, using an available model of the system, in combination with learned position dependence of the observed features in the system performance (positioning errors, etc). A high-performance motion stage with nanometer precision is available for the experimental realization of the approach.
In high-precision positioning systems used in the semicondoctur industry, it is critical to maintain the highest possible productivity with elevated requirements for precision.Two main challenges call for improved diagnostics and control: The systems are often tuned robustly considering the worst point of travel. When a fault occurs, this leads to a degradation in performance and halt of production. This project addresses mainly the first challenge by proposing the development of advanced control method leveraging real-time encoder data and signals from the closed loop controller and hierarchical predictive and learning control methods. For semiconductor applications the standard method of choice is repetitive control, exploring the occurrence of the same disturbances over identical paths. This approach can be further extended in several directions, interesting both from theoretical and from practical perspective. This project aims to enhance the motion performance of high-performance positioning systems by applying learning approaches to adaptively control the system. We would like to combine LPV-methods with learning, and compare their performance to the Bayesian-optimization approach that we demonstrated recently. The resulting position-dependent adaptive control algorithm will be used for benchmarking the BO-based algorithm. The main goal of this project is to develop the LPV-controller, using an available model of the system, in combination with learned position dependence of the observed features in the system performance (positioning errors, etc). A high-performance motion stage with nanometer precision is available for the experimental realization of the approach.
• Use the existing system model as a basis to develop a LPV controller incorporating the learned position and step-size dependence of the motion system low-level control parameters
• Analyse the resulting performance in simulation
• Implement the controller on the real system and compare with the available learning adaptive controller performance
• Use the existing system model as a basis to develop a LPV controller incorporating the learned position and step-size dependence of the motion system low-level control parameters • Analyse the resulting performance in simulation • Implement the controller on the real system and compare with the available learning adaptive controller performance
Send a short cv and transcripts to ralisa@ethz.ch and ebalta@ethz.ch
Send a short cv and transcripts to ralisa@ethz.ch and ebalta@ethz.ch