In my research, I focus on learning from data for feedforward control of dynamic systems. Within feedforward control, both high performance as well as task flexibility are required. While model-based approaches are robust against varying tasks, their performance is limited by unmodelled dynamics that does not match the model. Learning control methods, such as iterative learning control, often lead to superior performance, but are only applicable to a single reference. My research focuses on learning and compensating the unmodelled effects through complementing model-based approaches with universal function approximators from the field of machine learning. Interpretability, data-efficiency, and safety are key topics within this research. The research is supported by and validated on industrial motion systems at ASML and Philips.
Johan Kon received his BSc Mechanical Engineering (cum laude) and MSc Systems and Control (cum laude) degree from Eindhoven University of Technology, Eindhoven, The Netherlands. BSc en MSc (cum laude). As part of his studies, he did an internship on model-free reinforcement learning at Sapienza University of Rome, and graduated in collaboration with Canon Production Printing on learning control.
Physics-Guided Neural Networks for Feedforward Control: An Orthogonal Projection-Based Approach2022 American Control Conference, ACC 2022 (2022)
Physics-Guided Neural Networks for Feedforward Control41st Benelux Meeting on Systems and Control 2022 (2022)
Intermittent Sampling in Repetitive Control60th IEEE Conference on Decision and Control, CDC 2021 (2022)
Neural Network Training Using Closed-Loop Data: Hazards and an Instrumental Variable (IVNN) SolutionIFAC-PapersOnLine (2022)
Intermittent sampling in repetitive control: exploiting time-varying measurementsBenelux Workshop on Systems and Control 2021 (2021)
Current Educational Activities
- Begeleiden van cursussen georganiseerd door Mechatronics Academy, Mechatronics Academy