Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Robust Data-driven Iterative Learning Control of Manufacturing Processes
Repetitive tasks are common in manufacturing, it is natural to try and exploit this repetition. In this project, we want to develop new methods for accomplishing this in a purely data-driven way while robustly enforcing operating constraints.
As systems are becoming more complex and data is becoming more readily available, control engineers are beginning to bypass classical model-based techniques in favour of data-driven methods. Data-driven methods are especially suitable for applications where first-principle models are difficult to obtain or when thorough modelling and parameter identification is too costly.
This project aims to develop a data-driven variant of iterative learning control (ILC) that can enforce state and input constraints. In ILC, an agent learns how to perform a task through repetition, i.e., by taking an action, observing the result, and adjusting accordingly. For example, a basketball player learning to shoot a free throw. In our case, the goal is to have a CNC machine quickly and accurately cut a specified part geometry despite disturbances such as friction, inertia and so on. These processes are subject to constraints, for example on the speed of the cutting head, geometric tolerances, or acceleration inputs to the machine that must be enforced.
Often, ILC controllers are designed using a model of the process. This model contains information about the ``sensitivity'' of the process and allows the controller to decide how to adjust its inputs in response to observed outputs, as well as predicting when an input would violate constraints. In this project, we want to develop a method that doesn't need a model, just data, and can robustly enforce constraints. To do this, we will apply an approach similar to that in DeePC [1] which borrows concepts from behavioural systems theory and distributionally robust optimization.
There is a precision machining testbed available for this project at the technopark. Data from the machine will be used to develop the initial simulation models and it is also probably possible to test the final ILC algorithm using the machine.
[1] J. Coulson, J. Lygeros, and F. Dörfler, “Distributionally robust chance constrained data-enabled predictive control,” arXiv preprint arXiv:2006.01702, 2020.
As systems are becoming more complex and data is becoming more readily available, control engineers are beginning to bypass classical model-based techniques in favour of data-driven methods. Data-driven methods are especially suitable for applications where first-principle models are difficult to obtain or when thorough modelling and parameter identification is too costly.
This project aims to develop a data-driven variant of iterative learning control (ILC) that can enforce state and input constraints. In ILC, an agent learns how to perform a task through repetition, i.e., by taking an action, observing the result, and adjusting accordingly. For example, a basketball player learning to shoot a free throw. In our case, the goal is to have a CNC machine quickly and accurately cut a specified part geometry despite disturbances such as friction, inertia and so on. These processes are subject to constraints, for example on the speed of the cutting head, geometric tolerances, or acceleration inputs to the machine that must be enforced.
Often, ILC controllers are designed using a model of the process. This model contains information about the ``sensitivity'' of the process and allows the controller to decide how to adjust its inputs in response to observed outputs, as well as predicting when an input would violate constraints. In this project, we want to develop a method that doesn't need a model, just data, and can robustly enforce constraints. To do this, we will apply an approach similar to that in DeePC [1] which borrows concepts from behavioural systems theory and distributionally robust optimization.
There is a precision machining testbed available for this project at the technopark. Data from the machine will be used to develop the initial simulation models and it is also probably possible to test the final ILC algorithm using the machine.
[1] J. Coulson, J. Lygeros, and F. Dörfler, “Distributionally robust chance constrained data-enabled predictive control,” arXiv preprint arXiv:2006.01702, 2020.
The goals of the project are as follows:
- Learn about ILC, distributionally robust optimization, behavioural systems theory;
- Using data, create a ``truth'' simulation model of the machining process (including noise and uncertainty);
- Define a control objective in terms of accuracy, cycle time etc. as well as constraints;
- Develop a suitable data-driven ILC formulation and test it in simulation using data obtained from your simulation model;
- (Stretch goal and corona willing) Test your algorithm on the precision machining testbed.
**Publications** If the final results are promising they can potentially be turned into a publication.
**Corona Disclaimer** This project can be done in person at the Automatic Control Lab, hybrid, or completely remotely. Most importantly, we can change between these forms as needed.
**Qualifications**
We are looking for a highly motivated student with background in control, optimization, and system identification.
- No specific experience with manufacturing systems is necessary;
- Some familiarity with: Convex Optimization, Probability and Statistics, System Identification, and Control Systems will be helpful;
- Enrolment in a masters program;
- Proficiency in English.
The goals of the project are as follows:
- Learn about ILC, distributionally robust optimization, behavioural systems theory;
- Using data, create a ``truth'' simulation model of the machining process (including noise and uncertainty);
- Define a control objective in terms of accuracy, cycle time etc. as well as constraints;
- Develop a suitable data-driven ILC formulation and test it in simulation using data obtained from your simulation model;
- (Stretch goal and corona willing) Test your algorithm on the precision machining testbed.
**Publications** If the final results are promising they can potentially be turned into a publication.
**Corona Disclaimer** This project can be done in person at the Automatic Control Lab, hybrid, or completely remotely. Most importantly, we can change between these forms as needed.
**Qualifications** We are looking for a highly motivated student with background in control, optimization, and system identification.
- No specific experience with manufacturing systems is necessary; - Some familiarity with: Convex Optimization, Probability and Statistics, System Identification, and Control Systems will be helpful; - Enrolment in a masters program; - Proficiency in English.
Please send your resume/CV (including lists of relevant publications/projects) and transcript of records in PDF format via email to dliaomc@ethz.ch.
Please send your resume/CV (including lists of relevant publications/projects) and transcript of records in PDF format via email to dliaomc@ethz.ch.