The requirements for high accuracy and throughput in next-generation data-intensive motion systems lead to situations where position-dependent feedforward is essential. This article aims to develop a framework for interpretable and task-flexible position-dependent feedforward through systematic learning with automated experimental design. A data-driven and interpretable framework is developed by employing Gaussian process (GP) regression, enabling accurate modeling of feedforward parameters as a continuous function of position. The data is efficiently collected and illustrated through an iterative learning control (ILC) algorithm. Moreover, a framework for experimental design in the sense of automatically determining the training positions is presented by exploiting the uncertainty estimates of the GP and the specified first-principles knowledge. Two relevant case studies show the importance and significant performance improvement of the approach for position-dependent snap feedforward for a simplified 1-D wafer stage simulation and experimental application to position-dependent motor force constant compensation in an industrial wirebonder.

In this work, the proportional Clegg integrator (PCI), a resetting proportional-integrator (PI) element, is studied with the aim of improving the performance of an industrial motion stage currently controlled by a linear controller. A novel parallel continuous reset (CR) architecture, based on the PI, is presented, along with frequency-based tuning guidelines, similar to linear time-invariant (LTI) loopshaping techniques. Open-loop higher order sinusoidal input describing functions (HOSIDFs) and pseudo-sensitivities computed through analytically derived approximate closed-loop HOSIDFs were effectively applied to predict steady-state performance. The experimental results, obtained on a wire bonding machine, confirmed that resonance-induced vibrations of the machine's base frame can be suppressed more effectively by adopting a PCI-PID controller compared to the currently used linear controller. The novel structure does not only reduce unwanted excitation of higher order harmonics of the base frame resonance, such as the series CR architecture recently introduced in literature, but also avoids amplification of noise when implemented in practice. With the novel parallel structure, a significant (32%) decrease in the root mean square (rms) of the settling error could be achieved when compared to the linear controller currently used and the series CR reset structure.

The increasing demands on throughput and accuracy of semiconductor manufacturing equipment necessitates accurate feedforward motion control that includes compensation of input nonlinearities. The aim of this paper is to develop a data-driven feedforward approach consisting of a Wiener feedforward, i.e., linear parameterization with an output nonlinearity, to achieve high tracking accuracy and task flexibility for a class of Hammerstein systems. The developed approach exploits iterative learning control to learn a feedforward signal from data that minimizes the error and utilizes a control-relevant cost function to learn the parameters of a Wiener feedforward parameterization. Experimental validation on a wirebonder shows that the developed approach enables high tracking accuracy and task flexibility.

Mechatronic systems have increasingly stringent performance requirements for motion control, leading to a situation where many factors, such as position-dependency, cannot be neglected in feedforward control. The aim of this paper is to compensate for position-dependent effects by modeling feedforward parameters as a function of position. A framework to model and identify feedforward parameters as a continuous function of position is developed by combining Gaussian processes and feedforward parameter learning techniques. The framework results in a fully data-driven approach, which can be readily implemented for industrial control applications. The framework is experimentally validated and shows a significant performance increase on a commercial wire bonder.

Advanced feedforward control methods enable mechatronic systems to perform varying motion tasks with extreme accuracy and throughput. The aim of this paper is to develop a data-driven feedforward controller that addresses input nonlinearities, which are common in typical applications such as semiconductor back-end equipment. The developed method consists of parametric inverse-model feedforward that is optimized for tracking error reduction by exploiting ideas from iterative learning control. Results on a simulated set-up indicate improved performance over existing identification methods for systems with nonlinearities at the input.