Abstract: Physics-informed machine learning (PIML) is a set of methods and tools that systematically integrate machine learning (ML) algorithms with physical constraints and abstract mathematical models developed in scientific and engineering domains. The basic premise of PIML is that the integration of ML and physics can yield more effective, physically consistent, and data-efficient models. In this talk, we will discuss the advantages and disadvantages of PIML methods for control of dynamical systems.

Bio: Andrea Carron is a Senior Lecturer at ETH Zürich. He received the bachelor's, masters, and Ph.D. degrees in control engineering from the University of Padova. During his master and Ph.D. studies, he spent three stays abroad as a Visiting Researcher: the first at the University of California Riverside, the second at the Max Planck Institute, and the third at the University of California Santa Barbara. From 2016 to 2019, he was a Post-Doctoral Fellow with the Intelligent Control Systems Group at ETH Zürich. His is research interests include safe-learning, learning-based control, multiagent systems, and robotics.

Abstract: Sequential Monte Carlo methods (including the particle filters and smoothers) allows us to compute probabilistic representations of the unknown objects in models used to represent for example nonlinear dynamical systems. Physical knowledge of a system can be used in the identification process to improve the predictive performance by restricting the space of possible mappings from the input to the output. Typically, the physical models contain unknown parameters that must be learned from data. Sequential Monte Carlo methods enable learning for a more general class of models. In this talk I will show how this can be done. Towards the end I will also hint at how the Gaussian process can be tailored to correctly take known linear constraints into account. These constraints can come from an underlying physical model, such as Maxwell's equations for magnetic fields.

Bio: Thomas B. Schön is the Beijer Professor in Artificial Intelligence in the Department of Information Technology at Uppsala University. He received the PhD degree in Automatic Control in Feb. 2006, the MSc degree in Applied Physics and Electrical Engineering in Sep. 2001,  the BSc degree in Business Administration and Economics in Jan. 2001, all from Linköping University. He has held visiting positions with the University of Cambridge (UK), the University of Newcastle (Australia) and Universidad Técnica Federico Santa María (Valparaíso, Chile). In 2018, he was elected to The Royal Swedish Academy of Engineering Sciences (IVA) and The Royal Society of Sciences at Uppsala. He received the Tage Erlander prize for natural sciences and technology in 2017 and the Arnberg prize in 2016, both awarded by the Royal Swedish Academy of Sciences (KVA). He was awarded the Automatica Best Paper Prize in 2014, and in 2013 he received the best PhD thesis award by The European Association for Signal Processing. He received the best teacher award at the Institute of Technology, Linköping University in 2009. He is a Senior member of the IEEE and a fellow of the ELLIS society. Schön has a broad interest in developing new algorithms and mathematical models capable of learning and acting based on data. His  main scientific field is Machine Learning, but he also regularly publishes in other fields such as Statistics, Automatic Control, Signal Processing and Computer Vision. He pursues both basic research and applied research, where the latter is typically carried out in collaboration with industry or applied research groups.

Abstract: Data-driven approaches achieve remarkable results for modeling nonlinear systems based on collected data. However, these models often neglect basic physical principles which determine the behavior of any real-world system. This omission is unfavorable in two ways: The models are not as data-efficient as they could be by incorporating physical prior knowledge, and the model itself might not be physically consistent. In this talk, I will present our results on physics-constrained Gaussian processes for learning of dynamical system with a focus on the class of electromechanical systems. I will propose Gaussian Process Port-Hamiltonian systems (GP-PHS) as a physics-constrained, nonparametric Bayesian learning approach with uncertainty quantification. In contrast to many physics-informed techniques that impose physics by penalty, the proposed data-driven model is physically correct by design. The framework is in particular suitable for composable learning as its structure can be preserved under interconnection. Finally, I demonstrate the application of the model within a robust control framework to enable safe learning-based control.

Bio: Thomas Beckers is an Assistant Professor of Computer Science and Mechanical Engineering at Vanderbilt University. Before joining Vanderbilt, he was a postdoctoral researcher at the Department of Electrical and Systems Engineering, University of Pennsylvania, where he was member of the GRASP Lab, PRECISE Center and ASSET Center. In 2020, he earned his doctorate in Electrical Engineering at the Technical University of Munich (TUM), Germany. He received the B.Sc. and M.Sc. degree in Electrical Engineering in 2010 and 2013, respectively, from the Technical University of Braunschweig, Germany. In 2018, he was a visiting researcher at the University of California, Berkeley. He is a DAAD AInet fellow and was awarded with the Rhode & Schwarz Outstanding Dissertation prize. His research interests include physics-enhanced learning, nonparametric models, and safe learning-based control. He has organized several invited sessions and workshop on Gaussian process based control and physics-informed learning.

Abstract: A dynamical system can have different representations; some are parametric like the ubiquitous state-space representation, but others represent system dynamics in a non-parametric fashion. These non-parametric representations typically require the availability of consecutive historical data collected from the system. Under certain excitation conditions on the data, such representations can be exact, i.e., they can explain the complete behavior of a system. However, if only irregular samples of the measured system trajectories are available, obtaining such representations becomes a difficult task. Moreover, it becomes unclear how to impose the desired excitation conditions on the data. This talk presents a computational method to obtain the complete behavior of a linear, time-invariant system from potentially irregularly measured data. For the special case of periodically missing output samples, suitable excitation conditions on the input are provided such that this method is guaranteed to succeed. Once such representations are obtained, it is shown how to exploit them to provide computationally more efficient methods for (i) low-rank matrix completion and (ii) data-driven predictive control problems.

Bio: Matthias A. Müller received a Diploma degree in Engineering Cybernetics from the University of Stuttgart, Germany, an M.Sc. in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign, US (both in 2009), and a Ph.D. from the University of Stuttgart in 2014. Since 2019, he is director of the Institute of Automatic Control and full professor at the Leibniz University Hannover, Germany. His research interests include nonlinear control and estimation, model predictive control, and data- and learning-based control, with applications in different fields including biomedical engineering and robotics. He has received various awards for his work, including the 2015 European Systems & Control PhD Thesis Award, the inaugural Brockett-Willems Outstanding Paper Award for the best paper published in Systems & Control Letters in the period 2014-2018, an ERC starting grant in 2020, the IEEE CSS George S. Axelby Outstanding Paper Award 2022, and the Journal of Process Control Paper Award 2023. He serves as associate editor for Automatica, editor of the International Journal of Robust and Nonlinear Control, and as a member of the Conference Editorial Board of the IEEE Control Systems Society.

Abstract: Recent developments in physics-informed neural networks (PINNs) and Deep Operator Networks (DeepONet) has provided an impetus to many new discoveries in fluid dynamics, material sciences, non-destructive evaluation, shape optimization, subsurface imaging and many more. In nutshell, PINNs incorporate the governing physics of underlying physical processes as soft-constraint in the loss function. PINNs seamlessly integrate the multi-fidelity data (e.g., numerical and experimental data along with PDEs and ODEs, to solve the forward and inverse problems in a given computational domain. As an example, I will present the accuracy and efficiency of PINN in locating the defect in a material by using the ultrasound data acquired using the laser vibrometry. In second part of my talk, I will discuss about the DeepONet which learn the operator (deferential, integral, Laplace, etc.) between infinite dimensional functional space. I will first discuss the implementation of DeepONet as CFD surrogate for a shape optimization problem. Second, I will explain the integration of DeepONet in accelerating Mesoscale Multi-phase Phase field Simulator (MEMPHIS-developed by Sandia National Lab) solver for Phase-field simulation in manufacturing thin films.

Bio: Khemraj Shukla is Assistant Professor of Applied Mathematics at Brown University. He held an appointment as a Research Scientist at Hewlett-Packard (HP) Labs, CA, BP America, TX and Halliburton, CO. His research focuses on the development of scalable codes on heterogeneous computing architecture. For high order numerical methods. He has also worked as a Computational Scientist at University of Chicago. As a doctoral student, he developed high order numerical methods for wave propagation in a fluid saturated porous medium under the guidance of Prof. Maarten V. de Hoop and Prof. Jesse Chan of Rice University.

Abstract: The electric power grid is a fundamental part of the transition to a sustainable future society. Power transmission infrastructure also need to be designed, manufactured and operated considering sustainability values such as circularity and energy efficiency. An important aspect is the ability to predict the remaining useful life based on stresses caused by electric current and overall temperature distribution. The requirement for investment in new projects could be reduced by reusing power system components either fully or by selecting the best-preserved parts for repair of other units.

As the data is limited and complex in the field of components' ageing, Physics-Informed Neural Networks (PINNs) can help to overcome the problem. PINNs are a novel method in machine learning that exploits the prior knowledge stored in partial differential equations (PDEs) or ordinary differential equations (ODEs) modelling the involved systems. This prior knowledge becomes a regularization agent, constraining the space of available solutions and consequently reducing the training data needed.

In this talk we will present how PINNs can be used to integrate PDEs governing the thermal behavior of power components and their degradation. We will show two use cases comprising temperature distribution and degradation of the insulation system inside power transformers.

Bio (Tor Laneryd): Tor Laneryd is Senior Principal Scientist at Hitachi Energy. He obtained a Master Degree in Engineering Physics at Chalmers University of Technology in Göteborg, Sweden in 2003, and was conferred the degree of Doctor of Engineering from Kyoto University Graduate School of Engineering, Department of Aeronautics and Astronautics in Kyoto, Japan in 2007. His area of technical responsibility is Fluid Dynamics, Heat and Mass Transfer in Power Devices. Since 2020 he is affiliated faculty at KTH Royal Institute of Technology, School of Electrical Engineering and Computer Science, in Stockholm, Sweden, and since 2022 he is the Swedish representative of Cigre Study Committee C3 "Power system sustainability and environmental performance".

Bio (Federica Bragone): Federica Bragone obtained her BSc Degree in Mathematics at the University of Aberdeen in Aberdeen, Scotland, in 2019. She pursued her MSc Degree in Applied and Computational Mathematics at KTH Royal Institute of Technology in Stockholm, Sweden, in 2021. Since January 2022, she has been pursuing a doctoral degree titled “Physics-Informed Neural Networks Applied to the Circular Use of Power Components” under the supervision of Prof. Stefano Markidis, Dr. Kateryna Morozovska, Dr. Tor Laneryd and Dr. Michele Luvisotto at the Division of Computational Science and Technology, Department of Computer Science, KTH Royal Institute of Technology.

Abstract: This talk will introduce the decision-making framework for controlling and observing different components and states in the system. The example application would focus on the power system and the problem of sensor allocation for observing and management of various devices in the system to increase system flexibility and resilience when facing additional uncertainties. The solution is compared with the traditional methods for power system optimization and management in order to highlight the responsiveness and adaptivity of the PINN solution.

Bio: Kateryna Morozovska is a Postdoctoral Researcher at the Department of Intelligent Systems at KTH Royal Institute of Technology. She has received PhD in Electrical Engineering from KTH Royal Institute of Technology and has been actively engaging with academia, industry, and entrepreneurship. Kateryna has been a Guest Research Affiliate at Berkeley Lab, the Technical University of Denmark and MINES ParisTech. She was a main organizer of the 2023 PhD Summer School on Physics Informed Neural Networks and Application with G.Em.Karniadakis. Both in her vision as an innovator and founder of AI and engineering-focused start-up and in her academic work, Kateryna aims to bridge the gap between theory and applicability of newly emerging hybrid-AI methods, specifically PINNs.