时 间: 2019年12月20日上午9：00
报告二：Application of Deep Learning for Flow Visualization
主讲人：Kyung Chun Kim
Prof. Kyung Chun Kim (Distinguished Professor, Academician, School of Mechanical Engineering, Pusan National University, Busan, South Korea) received MS and Ph.D degree at KAIST, Korea in 1981 and 1987 respectively. Since 1983, he was hired as a faculty member in Pusan National University (PNU), Korea. He was invited as a visiting professor from Ottawa University in Canada for 1989-1990. He joined at the department of theoretical and applied mechanics in University of Illinois, Urbana-Champaign, USA as an invited professor for 1996-1997. He was invited as a special foreign professor from the University of Tokyo, Japan for 2009-2010. During 36 years at PNU, he has published about 605 refereed journal papers in Domestic and International Journals and supervised 203 MS and Ph.D students. On the basis of his research outcomes, he received the outstanding paper awards (1995, 2002) from KSME and KOSEF. In 2004, he was selected as a member of National Academy of Engineering. His research interests include: Turbulence, Two-Phase Flows, Flow Visualization, and Artificial Intelligence.
For a time-resolved turbulent flow reconstruction from discrete point measurements and non-time-resolved (non-TR) particle image velocimetry (PIV) measurements, an artificial intelligence framework based on long short-term memory (LSTM) was used. To this end, an LSTM-based proper orthogonal decomposition (POD) model is proposed to establish the relationship between velocity signals and time-varying POD coefficients obtained from non-TR-PIV measurements. Two different time-step configurations were employed to investigate the robustness and learning ability of the LSTM-based POD model: a single-time-step structure and a multi-time-step structure. The results demonstrate that the LSTM-based POD model has great potential for time-series reconstruction since it can successfully recover the temporal evolution of POD coefficients with remarkable accuracy, even in high-order POD modes. A general super-resolution reconstruction strategy was proposed for turbulent velocity fields using a generative adversarial network (GAN)-based artificial intelligence framework. Two advanced neural networks, i.e. super-resolution generative adversarial network (SRGAN) and enhanced-SRGAN (ESRGAN), were firstly applied in fluid mechanics to augment the spatial resolution of turbulent flow. The spatial resolution of the coarse flow field can be successfully augmented by 42 and 82 times with remarkable accuracy. The reconstruction performances of SRGAN and ESRGAN were comprehensively investigated and compared, including an analysis of the recovered instantaneous flow field, statistical flow quantities, and spatial correlations. The results convincingly demonstrated that both models can reconstruct the high-spatial-resolution flow field.
报告三：Push fluid mechanics to the epoch of non-ideality —— a perspective from laminar-turbulent transition
Dr. Jie Ren (University of Nottingham), UK, received B.Sc. degree in Mechanical Engineering from Tsinghua University in 2011, and the Ph.D. degree in Fluid Mechanics from Tsinghua University in 2016. His dissertation is published as a Springer Thesis and he wins the nomination award of Excellent Doctoral Dissertation of Chinese Society of Mechanics. He is currently a Research Fellow of Fluid Dynamics at the University of Nottingham.
Before joining University of Nottingham, he was a postdoc researcher at Delft University of Technology (TU Delft) and Tsinghua University. Recently, he was awarded a Humboldt Research Fellowship to carry out innovative research in Germany. Dr. Ren’s primary area of research is flow instability and turbulent flows. As the first author, he has published 10+ peer-reviewed journal articles including 3 in the Journal of Fluid Mechanic (the top journal in fluid mechanics). His recent work on non-ideal fluid dynamics was featured on “Focus of Fluids” of the Journal of Fluids Mechanics.
Non-ideal fluid has been deemed to be the key component of the next-generation power systems as it boosts productivity and efficiency beyond the limit of its idealized counterpart. For example, complex molecular fluids improve turbine efficiency significantly; power cycles operating with supercritical fluids lead to competitive utility-scale renewable electricity production. Dr. Ren’s study aims at delivering a step-change in the understanding of non-ideal fluid to inform engineering applications, targeting the grand challenge in fluid dynamics – the Laminar-Turbulent Transition (LTT). The over-arching objective is to unveil the secrets in different stages of LTT, through the establishment of a fledged analytical and numerical framework capable of adequately accounting for non-ideal effects. This study not only promptly brings current LTT prediction techniques and CFD (Computational Fluid Dynamics) systems to the epoch of non-ideal fluids, but also paves the way for future engineering applications of non-ideal fluids in high-efficiency energy generations.
报告四：Design, fabrication and analysis of mechanical metamaterials
Dr. Xiang LI（City University of Hong Kong, Shenzhen Research Institute, Shenzhen）, received PhD from Beihang university (BHU) in 2018. He is currently a Postdoc on Solid Mechanics at Shenzhen Research Institute of City University of Hong Kong, Shenzhen, leading projects in design, fabrication and analysis of metamaterials, composite and porous material. He has 15+ refereed journal papers and 5+ conference papers among which 8 papers were the first or corresponding Author. He has received 130+ cite in total and 40+ cite for the most cite paper.
Abstract：Metamaterials exhibit none occurring properties and functionalities that cannot be realized in natural materials. Examples of metamaterials include new and unusual optical properties, zero, or even negative refractive of sound, thermal cloak as well as ultralight metallic micro-lattices. Mechanical metamaterials, one important member of the metamaterials, realized unusual mechanical properties, such as negative Poisson ratio, negative linear compressibility, negative compressibility transitions or negative stiffness, ultralight density and tunable vibrational properties. Among the design of mechanical metamaterials, auxetic structures play an important role in the design of novel metamaterials. Metamaterials exhibit coexisting negative properties, such as combining negative Poisson’s ratio with negative compressibility, negative thermal expansion, or negative compressibility, are proposed based on auxetic structures. Here, we will discuss several novel mechanical metamaterials include novel auxetic materials with enhanced mechanical properties, materials with negative Poisson’s ratio and negative thermal expansion coefficient and materials with compression twist property.
报告五：Dynamic Remaining Useful Life Prediction by Joint Modeling of Degradation and Lifetime Data
Dr. Jiawen Hu is a Research Fellow with the department of Industrial Systems Engineering and Management at National University of Singapore. He received his B.S. degree in Mechanical Engineering from Shanghai Jiao Tong University in 2009, M.S. degree in Mechanical Engineering from Chinese Academy of Science in 2012, and Ph.D. degree in Industrial Engineering from Shanghai Jiao Tong University in 2017. His research interests are Reliability Engineering, Engineering Statistics, and Maintenance Planning. His work has been published on IISE Transactions, Reliability Engineering and System Safety, International Journal of Production Research, etc.
Abstract：Degradation is one of the major root causes of a system failure. In some applications, the degradation levels are different upon failure, in which the fixed failure threshold assumption commonly adopted in the degradation literature may not hold. This study tackles the difficulty by jointly analyzing the system degradation and the lifetime data, which enables the corresponding remaining useful life (RUL) prediction. We treat the degradation level as a multiplicative time-varying covariate of the system hazard rate, where a random-effects Wiener process is adopted to model the degradation process. All the model parameters are firstly obtained by the maximum a posteriori estimation. When new data are available, parameters related to the random effects are then updated by a particle filter method. The proposed model can realize online RUL prediction based on the in-situ degradation signals. Through case studies on lead-acid batteries and digital communication systems, the proposed model is shown to outperform existing methods in terms of a better RUL prediction accuracy.
报告六：Adaptive Observers for Multi-Agent Systems Requiring no Input Information
Dr. Wei Jiang received his B.S. degree in mechanical engineering and automation from Wuhan University of Technology, Wuhan, China, in 2011, and M.S. degree in automobile engineering from Beihang University, Beijing, China, in 2015 and Ph.D. degree in Automatic, Computer Engineering, Signal Processing and Images in CRIStAL, UMR CNRS 9189, Ecole Centrale de Lille, France, in 2018. He is now a postdoctoral researcher in Aalto University, Finland, leading projects in cooperative distributed control, robot control, machine learning, time-delay systems and intelligent transportation systems.
He has authored 3 journal papers including IEEE Transactions on Automatic Control and co-authored 1 journal paper about Motion Planning for a Humanoid Mobile Manipulator System (31 pages). He is an important partner for research work of UAV leader-follower formation tracking control with communication via Robot Operating System (ROS).
In a single-agent system, an observer like the Luenberger one needs input information for its own state estimation. In multi-agent systems, if other agents also need this agent’s state for controller design, sending and receiving input information continuously via communication links will consume tremendous amounts of energy and may as well become unavailable if links break. To deal with this issue, two new adaptive observers (AOs) are proposed in this paper without the requirement of input information if the input is bounded. First, a discontinuous AO with the asymptotic convergence of estimating error is designed by using a nonlinear function to compensate the input effect. Then, a continuous AO is further developed by adopting a new adaptive rule and a different nonlinear function with the estimating error being uniformly ultimately bounded. The convergence of the AOs is done by means of Lyapunov functions and linear matrix inequalities are derived to guarantee the stability of estimating error. Simulating examples are provided to validate proposed theories and factors influencing the stability and boundary of estimating error are analyzed.
This observer technique can be used for consensus, formation, containment and other cooperative distributed control for multi-agent/multi-robot systems with less communication energy consumption.
Some other works about communicating privacy, motion planning for a high redundant non-holonomic humanoid mobile dual-arm manipulator system will also be introduced.