Improving complex machines using multirate data and learning control
Max van Haren defended his PhD thesis at the Department of Mechanical Engineering on September 3rd.
A key challenge in Max van Haren鈥檚 research involved building models of systems that gather data at different intervals, so-called multirate systems. For example, some sensors might take measurements every millisecond, while others only update every few milliseconds, such as vision-based sensors. Traditional modeling methods often ignore these differences, which limits the accuracy of the resulting models. He developed methods to accurately model and identify the behavior of these multirate systems, allowing engineers better understand and control machines that operate on different time scales.
Another part of the research explored how machine learning can enhance control performance. While machine learning is widely used in applications such as text generation, its use in physical systems is still limited. Van Haren focused on combining machine learning with classical learning control approaches in a structured way. The results show that a combination of machine learning and classical control techniques is a key enabler for future machines.
Industrial Applications
The methods developed in Van Haren鈥檚 research help enhance machines in a wide range of applications. In the semiconductor industry, for instance, nanometer precision positioning accuracy is critical for the production of chips. In data storage systems like hard disk drives, vibration compensation and precise actuation directly impact performance. In healthcare, the automation of tasks such as tissue dissection relies heavily on the precision of medical equipment. Across all of these examples, the ability to build better models and more accurate controllers can lead to faster, cheaper, and more precise machines.
To validate his methods, the researcher applied them to realistic systems and in collaboration with companies through the IMOCO4.E project, such as a prototype wafer stage used in semiconductor manufacturing, a hard disk drive, and a belt-driven motion stage. These experiments demonstrate that it is possible to build accurate models and increase precision using the (multirate) data they already produce, reducing the need for expensive hardware upgrades.
By developing accurate multirate modeling and learning-driven motion control, Max van Haren鈥檚 research enables next鈥慻eneration machines that are faster, more precise, and more cost鈥慹ffective, which are essential for today鈥檚 most advanced technologies.
Title of PhD-thesis: . Supervisors: Tom Oomen (黑料福利网) and Lennart Blanken (Sioux and 黑料福利网).