July 25-30 & Aug 29-Sep 5, online
Zoom meeting link: Completed.
Bilibili live stream: Completed.
Youtube live stream recording (openly available): https://www.youtube.com/channel/UCX2dndkSsRZ1YrSLLH2Rvbg
(the rest are available on request from the contact.)
Note: The default timezone is in China, GMT+8
Venue: HITSZ, L0316
[非公开视频可邮件 Dr. DENG 获取]
Date: 07/25/2022, 09:00 (GMT +08:00)
Applied Mathematics and Engineering
Brown University; Also at MIT & PNNL, USA
Title: Discovering hidden fluid mechanics via physics-Informed neural networks
Abstract: We will review physics-informed neural networks (PINNs) and operator regression networks (DeepOnets) with emphasis on discovering missing physics and system identification in diverse applications in fluid mechanics. The diverse problems we consider are ill-posed and cannot be solved with any traditional methods. For example, we can obtain the drag and lift forces on bluff bodies from smoke visualizations or quantify the 3D velocity and pressure fields over an espresso cup based on Schlieren photography.
Bio: George Karniadakis is from Crete. He is a member of the USA National Academy of Engineering. He received his M.Sc. and Ph.D. from Massachusetts Institute of Technology (1984/87). He was appointed Lecturer in the Department of Mechanical Engineering at MIT and subsequently he joined the Center for Turbulence Research at Stanford / NASA Ames. He joined Princeton University as Assistant Professor in the Department of Mechanical and Aerospace Engineering and as Associate Faculty in the Program of Applied and Computational Mathematics. He was a Visiting Professor at Caltech in 1993 in the Aeronautics Department and joined Brown University as Associate Professor of Applied Mathematics in the Center for Fluid Mechanics in 1994. After becoming a full professor in 1996, he continued to be a Visiting Professor and Senior Lecturer of Ocean/Mechanical Engineering at MIT. He is a Vannevar Bush Faculty Fellow (2022), an AAAS Fellow (2018-), Fellow of the Society for Industrial and Applied Mathematics (SIAM, 2010-), Fellow of the American Physical Society (APS, 2004-), Fellow of the American Society of Mechanical Engineers (ASME, 2003-) and Associate Fellow of the American Institute of Aeronautics and Astronautics (AIAA, 2006-). He received the SIAM/ACM Prize on Computational Science & Engineering (2021), the Alexander von Humboldt award in 2017, the SIAM Ralf E Kleinman award (2015), the J. Tinsley Oden Medal (2013), and the CFD award (2007) by the US Association in Computational Mechanics. His h-index is 122 and he has been cited over 68,000 times.
Date: 07/26/2022, 09:00 (GMT +08:00)
Department of Mechanical Engineering
University of Washington, USA
Title: Machine learning for scientific discovery, with examples in fluid mechanics
Abstract: This work describes how machine learning may be used to develop accurate and efficient nonlinear dynamical systems models for complex natural and engineered systems. We explore the sparse identification of nonlinear dynamics (SINDy) algorithm, which identifies a minimal dynamical system model that balances model complexity with accuracy, avoiding overfitting. This approach tends to promote models that are interpretable and generalizable, capturing the essential “physics” of the system. We also discuss the importance of learning effective coordinate systems in which the dynamics may be expected to be sparse. This sparse modeling approach will be demonstrated on a range of challenging modeling problems in fluid dynamics, and we will discuss how to incorporate these models into existing model-based control efforts. Because fluid dynamics is central to transportation, health, and defense systems, we will emphasize the importance of machine learning solutions that are interpretable, explainable, generalizable, and that respect known physics.
Bio: Dr. Steven L. Brunton is a Professor of Mechanical Engineering at the University of Washington. He is also Adjunct Professor of Applied Mathematics and Computer science, and a Data Science Fellow at the eScience Institute. Steve received the B.S. in mathematics from Caltech in 2006 and the Ph.D. in mechanical and aerospace engineering from Princeton in 2012. His research combines machine learning with dynamical systems to model and control systems in fluid dynamics, biolocomotion, optics, energy systems, and manufacturing. He received the Army and Air Force Young Investigator Program (YIP) awards and the Presidential Early Career Award for Scientists and Engineers (PECASE). Steve is also passionate about teaching math to engineers as co-author of three textbooks and through his popular YouTube channel, under the moniker “eigensteve”.
Date: 07/26/2022, 15:00 (GMT +08:00)
Institut de Mathématiques de Bordeaux
Université de Bordeaux & Memphis Team, Inria, France
Abstract: The presentation focuses on examples of parametric problems governed by partial differential equations in which the linear representation of the reduced space fails. We introduce a nonlinear approximation technique based on a solution mapping as a function of the parameters via optimal transportation. We conclude by discussing the advantages and disadvantages in this framework of linear or nonlinear, intrusive or non-intrusive model reduction approaches.
Work in collaboration with S. Cucchiara, T. Taddei, H. Telib
Bio: Angelo Iollo graduated in Aerospace Engineering from Politecnico di Torino, Italy. After a PhD conducted at NASA Langley and granted from Politecnico di Torino, he was tenured researcher at the Politecnico di Torino and hence full professor of Applied Mathematics at Université de Bordeaux. He leads the research Team Memphis at Inria, the French national institute for Applied Mathematics and Computer Science.
Date: 07/27/2022, 09:00 (GMT +08:00)
Mechanical and Aerospace Engineering
Oklahoma State University, USA
Title: The prospects of hybrid analysis and modeling in fluid dynamics
Abstract: Advances in artificial intelligence and machine learning have led to a renaissance in learning and extracting patterns from complex data. The penetration of these learning approaches into the field of fluid dynamics has been further accelerated in recent years due to the ever-increasing availability of highly modular open-source packages and powerful accelerators. In this presentation, the use of machine learning for improving and/or accelerating coarse-grained models will be detailed to provide a basis to generate predictive technologies for a broad spectrum of closure modeling problems. In considering the examples from the speaker’s experience, a unifying theme that drives the discussion be the utilization of a hybrid modeling and analysis approach that combines physics-based modeling with the versatility of such data-driven approaches. The talk will therefore highlight various hybrid modeling mechanisms to utilize the strengths of different modeling strategies to overcome their individual weaknesses while preserving trends from high fidelity or full order model data.
Bio: Omer San has been a faculty member of Mechanical and Aerospace Engineering at Oklahoma State University, Stillwater, OK, USA, since 2015. He received his Ph.D. in Engineering Mechanics from Virginia Tech. His field of study is centered upon the development, analysis, and application of computational methods in science and engineering with a particular emphasis on fluid dynamics across a variety of spatial and temporal scales. He is a recipient of U.S. Department of Energy Early Career Research Program Award. So far, he graduated 5 M.Sc. and 3 Ph.D. students, and his current group includes 1 M.Sc. and 3 Ph.D. students. He organized several symposia and mini-symposia at AIAA, SIAM CSE and SASC meetings, and he is a member the American Physical Society.
Date: 07/28/2022, 09:00 (GMT +08:00)
Mechanical and Aerospace Engineering
Princeton University, USA
Abstract: This lecture provides an introduction to the Koopman operator and its use in understanding dynamical systems, and in determining data-driven models of them. We first discuss some concepts in ergodic theory, including ergodicity and mixing in dynamical systems, and their connection to the spectrum of the Koopman operator. We then describe various numerical methods for approximating the Koopman operator directly from data, including Extended Dynamical Mode Decomposition as well as methods based on manifold-learning.
Bio: Clancy Rowley is the Sin-I Cheng Professor of Engineering Science in the Department of Mechanical and Aerospace Engineering at Princeton, and is an associated faculty in the Program in Applied and Computational Mathematics. He received his undergraduate degree from Princeton and his doctoral degree from Caltech, both in Mechanical Engineering. He has received several awards, including an NSF CAREER Award and an AFOSR Young Investigator Award, and he is a Fellow of the American Physical Society. His research interests lie at the intersection of dynamical systems, control theory, and fluid mechanics, and focus on reduced-order models suitable for analysis and control design.
Date: 08/29/2022, 09:00 (GMT +08:00)
Mechanical and Ocean Engineering
Massachusetts Institute of Technology, USA
Title: Likelihood-weighted active learning with application to Bayesian optimization, uncertainty quantification, and decision making in high dimensions
Abstract: Analysis of physical and engineering systems is characterized by unique computational challenges associated with high dimensionality of parameter spaces, large cost of simulations or experiments, as well as existence of uncertainty. For a wide range of these problems the goal is to either quantify uncertainty and compute risk for critical events, optimize parameters or control strategies, and/or making decisions. Bayesian active learning provides a flexible framework for performing these tasks. However, Bayesian calculations are often prohibitively expensive in terms of the required simulations or experiments, even in the active learning setting.
In this talk, we introduce a new class of acquisition functions that utilize a likelihood-weighted ratio that accounts for the importance of the output relative to the input. This ratio acts essentially as a probabilistic sampling weight and guides the sampling algorithm towards regions of the input space where the objective function assumes abnormal values, resulting in significant savings of computational or experimental resources needed for convergence. We show that the adopted acquisition functions can be rigorously derived as the asymptotic limit of an optimal acquisition function that has a minimax form over a functional space. Subsequently, we demonstrate their favorable properties compared to standard methods on benchmark functions commonly used in the optimization community as well as real-world applications.
Bio: Dr. Sapsis is Professor of Mechanical and Ocean Engineering at MIT. He received a diploma in Ocean Engineering from Technical University of Athens, Greece, and a Ph.D. in Mechanical and Ocean Engineering from MIT. Before becoming a faculty at MIT he was appointed Research Scientist at the Courant Institute of Mathematical Sciences at New York University. He has also been a visiting faculty at ETH-Zurich. Prof. Sapsis’s work lies on the interface of nonlinear dynamical systems, probabilistic modeling, and data-driven methods. A particular emphasis of his work is the formulation of mathematical methods for the prediction, statistical quantification, and optimization of complex engineering and physical systems such as turbulent fluid flows, nonlinear waves in the ocean, and extreme ship motions. He has received numerous awards and recognitions including three Young Investigator Awards (Navy, Army, and Air-Force research office), the Alfred P. Sloan Foundation Award, and more recently the Bodossaki Award on Basic Sciences: Mathematics.
Date: 08/29/2022, 15:00 (GMT +08:00)
Mechanical Engineering Department
ENSTA Paris, Institut Polytechnique Paris, France
Title: The fluidic pinball : A simple yet challenging benchmark for model reduction and flow control
Abstract: The fluidic pinball has become a popular benchmark for model reduction and flow control using first principles and machine learning. We will present some important properties of this flow and some of the salient results recently obtained on this system.
Bio: Luc Pastur is Associate Professor in the Mechanical Engineering Department at ENSTA Paris. He has extensive experience in both experimental setups and metrology in fluid mechanics and approaches in the theoretical framework of nonlinear physics. He contributed to more than 55 scientific publications on topics dedicated to self-sustained oscillating flows, wake flows, flow instabilities, flow control, model reduction, Lagrangian mixing, transitions to space-time chaos, and fluid-structure interaction.
Date: 08/30/2022, 09:00 (GMT +08:00)
School of Engineering and Applied Science,
Harvard University, USA; also at ETHZ, Switzerland
Title: Alloys of Artificial Intelligence and Computational Science for modeling prediction and control of fluid flows.
Abstract: Over the last thirty years, we have experienced more than a billion-fold increase in hardware capabilities and a dizzying pace of acquiring and transmitting massive amounts of data. Artificial Intelligence (AI) has been the beneficiaries of these advances, and today it is increasingly embedded in technologies that touch every aspect of humanity. In this talk, I would offer a perspective on forming alloys of AI and simulations for the prediction and control of complex flow systems. I will present novel algorithms for learning the Effective Dynamics (LED) of complex flows and a fusion of multi-agent reinforcement learning and scientific computing (SciMARL) for modeling and control of complex flow-structure interactions. I will juxtapose successes and failures and argue that the proper fusion of fluid mechanics knowledge and AI expertise is essential to advance scientific frontiers.
Bio: Petros Koumoutsakos is Herbert S. Winokur, Jr. Professor of Engineering and Applied Sciences, Department Chair of Applied Mathematics, and Faculty Director of the Institute for Applied Computational Science (IACS) at Harvard University. He studied Naval Architecture (Diploma-NTU of Athens, M.Eng.-U. of Michigan), Aeronautics and Applied Mathematics (PhD-Caltech), and has served as the Chair of Computational Science at ETH Zurich (1997-2020). Petros is elected Fellow of the American Society of Mechanical Engineers (ASME), the American Physical Society (APS), and the Society of Industrial and Applied Mathematics (SIAM). He is recipient of the Advanced Investigator Award by the European Research Council and the ACM Gordon Bell prize in Supercomputing. He is elected International Member of the US National Academy of Engineering (NAE). His research interests are on the fundamentals and applications of computing and artificial intelligence to understand, predict and optimize fluid flows in engineering, nanotechnology, and medicine.
Date: 08/30/2022, 15:00 (GMT +08:00)
University of Oslo & Norwegian Meteorological Institute, Norway
Abstract: Deep Reinforcement Learning (DRL) has attracted a significant amount of interest for performing Active Flow Control (AFC) over the last couple of years. By contrast with traditional analytical or linearization-based methods that resort on a bottom-up approach and an analysis of the Navier Stokes equations, the DRL method belongs to the Machine Learning (ML) category of control methods. As a consequence, DRL works following a trial-and-error approach, by directly interacting with the problem to solve in a closed-loop fashion, without making any restrictive hypothesis on the structure or complexity of the underlying problem. DRL is a general approach, which is not specific to AFC problems, and was used, for example, for winning at the game of Go against the best human players. In the context of fluid mechanics, DRL potentially allows to find non-linear control strategies that are far away from the base flow, and to offer a route for performing model-free AFC.
In this talk, I will briefly introduce DRL in general, before detailing its application to AFC in particular and the state-of-the-art of DRL-based AFC. Finally, the perspectives for further development of DRL-based AFC will be discussed.
Bio: Dr. Jean Rabault received his Ph.D. in Fluid Mechanics from the University of Oslo (Oslo, Norway) in 2018, and his MSc as a double degree student in Mechanical Engineering and Fluid Mechanics from the KTH Royal Institute of Technology (Stockholm, Sweden) and the Ecole Polytechnique (Palaiseau, France) in 2015. In addition to a PostDoc position at the University of Oslo, Jean is now a permanent staff member at the Norwegian Meteorological Institute. Jean’s interests have been focused on two topics within Fluid Mechanics: the dynamics of sea ice in the Polar Regions, and the application of data-driven methods, in particular Deep Reinforcement Learning (DRL), to Active Flow Control (AFC) problems. On this second topic, Jean was among the early adopters of DRL to solve canonical AFC problems, and helped spark the interest of the community by offering reference open source implementations of simple benchmarks in 2019. Since then, Jean has been working on multiple projects and collaborating with teams from around the world with the goal of applying DRL to increasingly complex AFC problems, understanding the strengths and weaknesses of the DRL approach, and helping build up a strong open source ecosystem in this field.
Date: 09/05/2022, 15:00 (GMT +08:00)
Chair of Fluid Dynamics, Hermann-Föttinger-Institut,
Technische Universität Berlin, Germany
Title: Development of gas-turbine combustors with ultra-low emissions and dynamics using machine learning
Abstract: Lean premixed combustion is the state-of-the-art technology to achieve ultra-low NOX emissions in stationary gas turbines. However, lean premixed flames are susceptible to thermoacoustic instabilities, lean blowout, and flashback. The design of such a combustion system is thus always related to the balancing between the levels of emissions and flame stability. Often the objectives are conflicting, affecting the environment and the lifetime of the combustion chamber, respectively. Data-driven optimization methods and the adaptation of models through artificial intelligence have experienced a surge in development in the past years. The goal of this study is to show the potential of these methods for gas turbine burner development.
Different applications will be discussed. This includes the optimization of the operation of a swirl-stabilized premixed burner in lab experiments and the scaling of the results to elevated pressure and final gas-turbine application. Today, with the high fluctuation supply of wind and solar power plants, the balancing gas turbines must cope with the load fluctuations and need to be safely operated also under part load operation. This is a challenging task as emissions may increase and flashback margins are reduced. A pilot burner that features 61 different positions of fuel injection, manufactured by means of selective laser melting, is used to modify the gas mixture close to the flame anchoring position. Each of the injector lines is equipped with an individual valve, such that the distribution of fuel-air mixture can be modified variously. Installed into an industrial swirl combustor, a data-driven optimization method is used to find an optimal subset of injection locations by automated experiments. The method uses a surrogate model that is based on Gaussian Processes Regression. It is adopted for experimental optimization, keeping measurement efforts to a minimum. The optimizer controls the fuel valves and uses live measurements to find a distribution that generates minimal NOX emissions, while ensuring flame stability. The solutions found by the optimization scheme are analyzed and advantages and limitations of the approach are discussed.
Bio: Oliver Paschereit is heading the Chair of Fluid Dynamics, Hermann-Föttinger- Institut, TU Berlin, since 2003. His research and teaching cover a broad spectrum of topics related to fluid mechanics and combustion technology: flow and combustion control, vessel aerodynamics, gas turbine technology, ultra-low NOx combustion, thermoacoustics, pressure gain combustion, and wind turbine technology. Before moving to TU Berlin, Oliver Paschereit held an upper management position at ABB / ALSTOM in Switzerland, which he joined in 1994. From 1992 – 1994 he worked on high-speed train acoustics and helped to reduce train noise substantially. After his studies at TU Berlin and Ecole Centrale de Lyon, he received his Ph.D. in 1992 from TU Berlin and the University of Arizona. His work in the industry on clean, efficient, and reliable gas turbines had a major impact during the commissioning of several new gas turbine families. The developed methods and technologies paved the way for ultra-low emission gas turbine systems that will greatly extend state-of-the-art of nowadays power generation technology. Future technologies, like ultra-wet gas turbine cycles and the integration of constant volume combustion into gas turbines, complete his profile. His research activities on fluid dynamics have an impact on efficient trucks, cars, and trains and were used even in the America’s Cup. Innovative solutions are also applied to wind turbines and general fluid dynamics. They range from smart flow visualization techniques, custom-made power boost devices helping to raise the power output of wind turbines, and the development of smart blades for the wind turbines of the future. His work on fluidic nozzles helped to understand and to optimize the mixing process.
His scientific and technological achievements are demonstrated by more than 500 journal and conference publications. The research has not only academic interest but is also important for industrial applications documented in over 80 patent publications. A number of best paper awards, many research prices, the Silver Medal of the Combustion Institute, two ERC Advanced Grants – the highest European research price – and a Proof-of-Concept Grant underline his competence in combustion and fluid dynamics.
The Chair of Fluid Dynamics, Hermann-Föttinger-Institut, Technische Universität Berlin
The Chair of Fluid Dynamics – Hermann-Föttinger-Institut is active in research and development in fluid dynamics and combustion. About 100 employees are working on advanced solutions of flow and combustion control and efficiency increasing methods to reduce CO2 and other pollutant emissions. The Institute has access to high performance computer clusters and operates a wide range of combustion and gas turbine test rigs as well as a number of wind tunnels. Towing tanks are also available and are being used for marine development but also to achieve realistic test conditions for cars, trucks, trains and wind turbines.
Beside fundamental research, development work is being performed for major companies. The Institute collaborates closely with the Charité, the largest University hospital in Europe, on bio fluid mechanics.