This PhD position at the Control Systems group of the Electrical Engineering department of TU/e is part of a research consortium aiming at the development of a radically new high-precision planar stage technology where the mechanical (ME) structure, the electromagnetic (EM) motor design, the metrology layout and the motion controllers (CT) design are simultaneously co-optimized. It is expected that taking these elements jointly into account during one combined AI-driven optimization process creates a significantly better end result than performing these optimizations individually and can lead to the discovery of unconventional ME, EM and CT designs and trade-offs that would have required decades of conventional design iterations to be discovered. The research is part of the AIDEAL project conducted together with academic partners (TU/e: Electromechanics and Power Electronics and TUDelft: Computational Design and Mechanics Group) and leading companies of the semiconductor industry (ASML, ASMPT, ITEC, MI-Partners, NTS-AM, Tecnotion, VAC Germany).
The objective of this PhD research is developing the high-precision motion control part of the co-design process for planar stage actuators. The research involves significant laboratory work in terms of implementation and experimental verification of the research results on a prototype stage developed at TU/e in collaboration with industrial partners.
The control performance of a planar motor, measured in nanometers of error during a scanning motion and is highly dependent on physical properties of the mover. Dynamic behavior in terms of resonant modes, position dependency, commutation strategy and of course feedback and feedforward controller design, are all critical elements. During optimization of the mechanical and electromagnetical structures, a fast assessment of control performance is required. However, the complexity of a full model and its associated complexity of the controller design is a limiting factor. Therefore, one research focus lies on constructing simplified (“surrogate”) models that are simple enough for an efficient performance assessment, but still capture the relevant mover properties. In the longer term, the availability of the control performance sensitivity to changes in the mechanical and electromagnetic design would make the co-optimization much more efficient.
Main research directions:
- Automated physics-driven surrogate modeling for co-design: establish learning-assisted automated approaches, capable of extracting the control specifications-relevant dynamics of planar motor designs and provide automatized trade-off selection between complexity of the model and the size of the uncertainty capturing the reduced dynamics. Key aspect is to learn a mapping that can re-cast the complex parametrization-based design variations to key low dimensional features on which the extracted surrogate dynamics depend on (reduced order modeling, co-parametrized model extractions, auto-encoders, reinforcement learning).
- Data-driven surrogate co-models: develop identification of surrogate representations from experimental data to allow their augmentation and adjustment to manufactured instances of the design based on physical measurements (physics-guided learning, deep recurrent neural networks, Gaussian processes).
- Automated design of optimal motion-controllers: development of automated design of advanced motion controllers based on the extracted surrogate models, capable to provide high-accuracy (sub-nm) point of interest (POI) motion of the mover of the planar motor by rejecting environmental disturbances and controlling the spatial deformation of the plate through the highly complex coil-magnet EM interactions. The focus is on the handling of a diverse set objective functions based on customer-specified motion behavior of the planar motor (e.g., guaranteed finite-time settling) under uncertain and nonlinear behavior of the system and on the computational efficiency of the optimization.