Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Performance-based trajectory optimization for path following control using Bayesian optimization
We developed a method for trajectory optimization in precise positioning where parameters of the low-level and of the MPC-based predictive controller are jointly optimized using a data-driven approach. This project brings the method on a 2D positioning testbed for different contour geometries.
In machining, positioning systems need to be fast and precise to guarantee high productivity and quality. Such performance can be achieved by Model Predictive Control (MPC), however requiring precise tuning and good computational abilities of the associated hardware. We use the superior planning capabilities of MPC to generate offline an optimized trajectory which is provided as an input to a low-level cascade PID controller, to improve the tracking performance of the system without imposing any requirements for fast online computation. Tuning an MPC controller for path planning, coupled with a low-level controller for the controlling the internal behaviour of the associated mechanical system is not trivial and requires multiple iterations. We therefore apply Bayesian Optimization to tune automatically the parameters of the controllers used in the proposed control architecture. Constrained Bayesian Optimization (CBO) algorithm is used to jointly tune the MPC parameters (weights in the cost function) and the low-level controller gain parameters. We have already demonstrated that this approach achieves superior performance, both in terms of positioning accuracy and traversal time.
This project will bring this research forward, by transferring it on the 2-axis positioning testbed, and, if needed, adapting it for various contour geometries.
In machining, positioning systems need to be fast and precise to guarantee high productivity and quality. Such performance can be achieved by Model Predictive Control (MPC), however requiring precise tuning and good computational abilities of the associated hardware. We use the superior planning capabilities of MPC to generate offline an optimized trajectory which is provided as an input to a low-level cascade PID controller, to improve the tracking performance of the system without imposing any requirements for fast online computation. Tuning an MPC controller for path planning, coupled with a low-level controller for the controlling the internal behaviour of the associated mechanical system is not trivial and requires multiple iterations. We therefore apply Bayesian Optimization to tune automatically the parameters of the controllers used in the proposed control architecture. Constrained Bayesian Optimization (CBO) algorithm is used to jointly tune the MPC parameters (weights in the cost function) and the low-level controller gain parameters. We have already demonstrated that this approach achieves superior performance, both in terms of positioning accuracy and traversal time. This project will bring this research forward, by transferring it on the 2-axis positioning testbed, and, if needed, adapting it for various contour geometries.
Demonstrate the hierarchical MOC-based controller with jointly optimized controller parameters for path following on a real system.
Demonstrate the hierarchical MOC-based controller with jointly optimized controller parameters for path following on a real system.