State estimation when only a partial model of a considered system is available remains a major challenge in many engineering fields. This work proposes a joint, square-root unscented Kalman filter to estimate states and model uncertainties simultaneously by linear combinations of physics-motivated library functions. Using a sparsity promoting approach, a selection of those linear combinations is chosen and thus an interpretable model can be extracted. Results indicate a small estimation error compared to a traditional square-root unscented Kalman filter and exhibit the enhancement of physically meaningful models.

Low-quality models that miss relevant dynamics lead to major challenges in modelbased state estimation. We address this issue by simultaneously estimating the system’s states and its model inaccuracies by a square root unscented Kalman filter (SRUKF). Concretely, we augment the state with the parameter vector of a linear combination containing suitable functions that approximate the lacking dynamics. Presuming that only a few dynamical terms are relevant, the parameter vector is claimed to be sparse. In Bayesian setting, properties like sparsity are expressed by a prior distribution. One common choice for sparsity is a Laplace distribution. However, due to disadvantages of a Laplacian prior in regards to the SRUKF, the regularized horseshoe distribution, a Gaussian that approximately features sparsity, is applied instead. Results exhibit small estimation errors with model improvements detected by an automated model reduction technique.

Within this work, we investigate how data-driven numerical approximation methods of the Koopman operator can be used in practical control engineering applications. We refer to the method Extended Dynamic Mode Decomposition (EDMD), which approximates a nonlinear dynamical system as a linear model. This makes the method ideal for control engineering applications, because a linear system description is often assumed for this purpose. Using academic examples, we simulatively analyze the prediction performance of the learned EDMD models and show how relevant system properties like stability, controllability, and observability are reflected by the EDMD model, which is a critical requirement for a successful control design process. Subsequently, we present our experimental results on a mechatronic test bench and evaluate the applicability to the control engineering design process. As a result, the investigated methods are suitable as a low-effort alternative for the design steps of model building and adaptation in the classical model-based controller design method.

The design of control engineering applications usually requires a model that accurately represents the dynamics of the real system. In addition to classical physical modeling, powerful data-driven approaches are increasingly used. However, the resulting models are not necessarily in a form that is advantageous for controller design. In the control engineering domain, it is highly beneficial if the system dynamics is given in PCHD form (Port-Controlled Hamiltonian Systems with Dissipation) because globally stable control laws can be easily realized while physical interpretability is guaranteed. In this work, we exploit the advantages of both strategies and present a new framework to obtain nonlinear high accurate system models in a data-driven way that are directly in PCHD form. We demonstrate the success of our method by model-based application on an academic example, as well as experimentally on a test bed.

In control design most control strategies are model-based and require accurate models to be applied successfully. Due to simplifications and the model-reality-gap physics-derived models frequently exhibit deviations from real-world-systems. Likewise, purely data-driven methods often do not generalise well enough and may violate physical laws. Recently Physics-Guided Neural Networks (PGNN) and physics-inspired loss functions separately have shown promising results to conquer these drawbacks. In this contribution we extend existing methods towards the identification of non-autonomous systems and propose a combined approach PGNN-L, which uses a PGNN and a physics-inspired loss term (-L) to successfully identify the system's dynamics, while maintaining the consistency with physical laws. The proposed method is demonstrated on two real-world nonlinear systems and outperforms existing techniques regarding complexity and reliability.

Ultrasonic wire bonding is a solid-state joining process used to form electrical interconnections in micro and power electronics and batteries. A high frequency oscillation causes a metallurgical bond deformation in the contact area. Due to the numerous physical influencing factors, it is very difficult to accurately capture this process in a model. Therefore, our goal is to determine a suitable feed-forward control strategy for the bonding process even without detailed model knowledge. We propose the use of batch constrained Bayesian optimization for the control design. Hence, Bayesian optimization is precisely adapted to the application of bonding: the constraint is used to check one quality feature of the process and the use of batches leads to more efficient experiments. Our approach is suitable to determine a feed-forward control for the bonding process that provides very high quality bonds without using a physical model. We also show that the quality of the Bayesian optimization based control outperforms random search as well as manual search by a user. Using a simple prior knowledge model derived from data further improves the quality of the connection. The Bayesian optimization approach offers the possibility to perform a sensitivity analysis of the control parameters, which allows to evaluate the influence of each control parameter on the bond quality. In summary, Bayesian optimization applied to the bonding process provides an excellent opportunity to develop a feedforward control without full modeling of the underlying physical processes.

While trade-offs between modeling effort and model accuracy remain a major concern with system identification, resorting to data-driven methods often leads to a complete disregard for physical plausibility. To address this issue, we propose a physics-guided hybrid approach for modeling non-autonomous systems under control. Starting from a traditional physics-based model, this is extended by a recurrent neural network and trained using a sophisticated multi-objective strategy yielding physically plausible models. While purely data-driven methods fail to produce satisfying results, experiments conducted on real data reveal substantial accuracy improvements by our approach compared to a physics-based model.