arXiv subjects

Max Bolderman

Publications and source records attributed to Max Bolderman.

8 recordsLinked to original sources

arXiv2025

Physics-guided gated recurrent units for inversion-based feedforward control

Inversion-based feedforward control relies on an accurate model that describes the inverse system dynamics. The gated recurrent unit (GRU), which is a recent architecture in recurrent neural networks, is a strong candidate for obtaining such a model from data. However, due to their black-box nature, GRUs face challenges such as limited interpretability and vulnerability to overfitting. Recently, physics-guided neural networks (PGNNs) have been introduced, which integrate the prior physical model structure into the prediction process. This approach not only improves training convergence, but also facilitates the learning of a physics-based model. In this work, we integrate a GRU in the PGNN framework to obtain a PG-GRU, based on which we adopt a two-step approach to feedforward control design. First, we adopt stable inversion techniques to design a stable linear model of the inverse dynamics. Then, a GRU trained on the residual is tailored to inverse system identification. The resulting PG-GRU feedforward controller is validated by means of real-life experiments on a two-mass spring-damper system, where it demonstrates roughly a two-fold improvement compared to the linear feedforward and a preview-based GRU feedforward in terms of the integral absolute error.

eess.SY↗

arXiv2024

Structured physics-guided neural networks for electromagnetic commutation applied to industrial linear motors

Mechatronic systems are described by an interconnection of the electromagnetic part, i.e., a static position-dependent nonlinear relation between currents and forces, and the mechanical part, i.e., a dynamic relation from forces to position. Commutation inverts a model of the electromagnetic part of the system, and thereby removes the electromagnetic part from the position control problem. Typical commutation algorithms rely on simplified models derived from physics-based knowledge, which do not take into account position dependent parasitic effects. In turn, these commutation related model errors translate into position tracking errors, which limit the system performance. Therefore, in this work, we develop a data-driven approach to commutation using physics-guided neural networks (PGNNs). A novel PGNN model is proposed which structures neural networks (NNs) to learn specific motor dependent parasitic effects. The PGNN is used to identify a model of the electromagnetic part using force measurements, after which it is analytically inverted to obtain a PGNN-based commutation algorithm. Motivated by industrial applications, we develop an input transformation to deal with systems with fixed commutation, i.e., when the currents cannot be controlled. Real-life experiments on an industrial coreless linear motor (CLM) demonstrate a factor 10 improvement in the commutation error in driving direction and a factor 4 improvement in the position error with respect to classical commutation in terms of the mean--squared error (MSE).

eess.SY↗

arXiv2023

Physics-guided neural networks for inversion-based feedforward control applied to hybrid stepper motors

Rotary motors, such as hybrid stepper motors (HSMs), are widely used in industries varying from printing applications to robotics. The increasing need for productivity and efficiency without increasing the manufacturing costs calls for innovative control design. Feedforward control is typically used in tracking control problems, where the desired reference is known in advance. In most applications, this is the case for HSMs, which need to track a periodic angular velocity and angular position reference. Performance achieved by feedforward control is limited by the accuracy of the available model describing the inverse system dynamics. In this work, we develop a physics-guided neural network (PGNN) feedforward controller for HSMs, which can learn the effect of parasitic forces from data and compensate for it, resulting in improved accuracy. Indeed, experimental results on an HSM used in printing industry show that the PGNN outperforms conventional benchmarks in terms of the mean-absolute tracking error.

eess.SY↗

arXiv2023

Data-driven feedforward control design for nonlinear systems: A control-oriented system identification approach

Feedforward controllers typically rely on accurately identified inverse models of the system dynamics to achieve high reference tracking performance. However, the impact of the (inverse) model identification error on the resulting tracking error is only analyzed a posteriori in experiments. Therefore, in this work, we develop an approach to feedforward control design that aims at minimizing the tracking error a priori. To achieve this, we present a model of the system in a lifted space of trajectories, based on which we derive an upperbound on the reference tracking performance. Minimization of this bound yields a feedforward control-oriented system identification cost function, and a finite-horizon optimization to compute the feedforward control signal. The nonlinear feedforward control design method is validated using physics-guided neural networks on a nonlinear, nonminimum phase mechatronic example, where it outperforms linear ILC.

eess.SY↗

arXiv2023

Physics-guided neural networks for feedforward control with input-to-state stability guarantees

The increasing demand on precision and throughput within high-precision mechatronics industries requires a new generation of feedforward controllers with higher accuracy than existing, physics-based feedforward controllers. As neural networks are universal approximators, they can in principle yield feedforward controllers with a higher accuracy, but suffer from bad extrapolation outside the training data set, which makes them unsafe for implementation in industry. Motivated by this, we develop a novel physics-guided neural network (PGNN) architecture that structurally merges a physics-based layer and a black-box neural layer in a single model. The parameters of the two layers are simultaneously identified, while a novel regularization cost function is used to prevent competition among layers and to preserve consistency of the physics-based parameters. Moreover, in order to ensure stability of PGNN feedforward controllers, we develop sufficient conditions for analyzing or imposing (during training) input-to-state stability of PGNNs, based on novel, less conservative Lipschitz bounds for neural networks. The developed PGNN feedforward control framework is validated on a real-life, high-precision industrial linear motor used in lithography machines, where it reaches a factor 2 improvement with respect to physics-based mass-friction feedforward and it significantly outperforms alternative neural network based feedforward controllers.

eess.SY↗

arXiv2022

Physics-guided neural networks for feedforward control: From consistent identification to feedforward controller design

Model-based feedforward control improves tracking performance of motion systems, provided that the model describing the inverse dynamics is of sufficient accuracy. Model sets, such as neural networks (NNs) and physics-guided neural networks (PGNNs) are typically used as flexible parametrizations that enable accurate identification of the inverse system dynamics. Currently, these (PG)NNs are used to identify the inverse dynamics directly. However, direct identification of the inverse dynamics is sensitive to noise that is present in the training data, and thereby results in biased parameter estimates which limit the achievable tracking performance. In order to push performance further, it is therefore crucial to account for noise when performing the identification. To address this problem, this paper proposes the use of a forward system identification using (PG)NNs from noisy data. Afterwards, two methods are proposed for inverting PGNNs to design a feedforward controller for high-precision motion control. The developed methodology is validated on a real-life industrial linear motor, where it showed significant improvements in tracking performance with respect to the direct inverse identification.

eess.SY↗

arXiv2022

On feedforward control using physics-guided neural networks: Training cost regularization and optimized initialization

Performance of model-based feedforward controllers is typically limited by the accuracy of the inverse system dynamics model. Physics-guided neural networks (PGNN), where a known physical model cooperates in parallel with a neural network, were recently proposed as a method to achieve high accuracy of the identified inverse dynamics. However, the flexible nature of neural networks can create overparameterization when employed in parallel with a physical model, which results in a parameter drift during training. This drift may result in parameters of the physical model not corresponding to their physical values, which increases vulnerability of the PGNN to operating conditions not present in the training data. To address this problem, this paper proposes a regularization method via identified physical parameters, in combination with an optimized training initialization that improves training convergence. The regularized PGNN framework is validated on a real-life industrial linear motor, where it delivers better tracking accuracy and extrapolation.

cs.LG↗

arXiv2021

Physics-Guided Neural Networks for Inversion-based Feedforward Control applied to Linear Motors

Ever-increasing throughput specifications in semiconductor manufacturing require operating high-precision mechatronics, such as linear motors, at higher accelerations. In turn this creates higher nonlinear parasitic forces that cannot be handled by industrial feedforward controllers. Motivated by this problem, in this paper we develop a general framework for inversion-based feedforward controller design using physics-guided neural networks (PGNNs). In contrast with black-box neural networks, the developed PGNNs embed prior physical knowledge in the input and hidden layers, which results in improved training convergence and learning of underlying physical laws. The PGNN inversion-based feedforward control framework is validated in simulation on an industrial linear motor, for which it achieves a mean average tracking error twenty times smaller than mass-acceleration feedforward in simulation.

eess.SY↗