Title: Machine Learning and Artificial Intelligence in PEM Fuel Cell Development

Speaker: Prof. Yun Wang, ASME Fellow and RSC Fellow; Mechanical and Aerospace Engineering, University of California, Irvine, CA, 92697-3975

Abstract:Polymer electrolyte membrane (PEM) fuel cells are electrochemical devices that directly convert the chemical energy stored in fuel into electrical energy with a practical conversion efficiency as high as 65%. They are widely regarded as a potential power source for portable, transportation and stationary applications. Though several fuel cell electric vehicles such as Toyota Mirai and Hyundai Nexo were successfully commercialized in recent years, technological advancements are still highly needed to further improve fuel cell performance and durability and to reduce cost. In this talk, I will first present the current status of PEM fuel cell technology, including hydrogen refueling station (HRF) development. The operation of PEM fuel cells involves multiphysics processes, including fluid flow, phase change, species transport, energy conservation, and proton/electron conduction, which determine fuel cell performance and durability. Some fundamental and material aspects critical to PEM fuel cell development will be reviewed and discussed. Machine learning and artificial intelligence (AI) have received increasing attention in material/energy development. This talk will also discuss their applications and potential in the development of fundamental knowledge and correlations, material selection and improvement, cell design and optimization, system control, and monitoring of operation health for PEM fuel cells. Two optimization examples of material design and operational condition using machine learning and genetic algorithms (GA) will be presented and explained in detail.

Bio: Yun Wang received his B.S. and M.S. degrees in Mechanics and Engineering Science from Peking University in 1998 and 2001, respectively. He went to the Pennsylvania State University where he earned his Ph.D degree in Mechanical Engineering in 2006. Dr. Wang joined the MAE (Mechanical and Aerospace Engineering) faculty at the University of California, Irvine in 2006. He has produced over 80 publications in PEM fuel cell, Li-air battery, and other energy systems, including a book on PEM Fuel Cell Water and Thermal Management Fundamentals, Practical Handbook of Thermal Fluid Science (in press), and a PEM fuel cell review paper that has been cited over 2,690 times (googlescholar). He has received a number of awards, including the prestigious President's Award and Outstanding Educator Award from Orange County Engineering Council and the 2011-2012 Applied Energy Certificate of Excellence: Most Downloaded Authors. Several of his seminal works are highly cited in the major fuel cell journals. Dr. Wang served as Track chair/co-chair, session chair/co-chair, conference chair and committee member for many international conferences on fuel cell, thermal energy, and engineering. Dr. Wang received 2018 Reviewer of The Year from the Journal of Electrochemical Energy Conversion and Storage. Dr. Wang is currently Professor at the UC Irvine, ASME fellow, and RSC fellow.

Zhiguo Qu

Xi’an Jiaotong University

Dr. Qu is a professor of Energy & Power Engineering at Xi’an Jiaotong University. He obtained the Ph. D degree in engineering thermophysics from Xi’an Jiaotong University in 2005. He worked as a visiting scholar at Advanced Heat Transfer, LLC, USA and Pennsylvania State University in 2006 and 2013, respectively. His main research interests include multiscale transport phenomena, energy conversion and storage, micro-nanofluidics, and hydrogen & fuel cell technologies. He has published 191 SCI indexed papers in peer-reviewed journals. These papers are cited by more than 4600 times from google scholar with h-index of 41. He serves as the editorial board member for several journals. Dr. Qu is a recipient of China National Funds for Distinguished Young Scholars.

Prediction and control of heat and mass transfer in porous media using a deep learning method

Abstract:Thermo-mass transfer in porous media are ubiquitous in both natural and industrial processes. Accurate prediction and control of the thermo-mass transport properties in porous media is still hot issue. Therefore, in this keynote lecture, four parts are involved. A deep learning method based on a convolutional neural network (CNN) with sample structure information self-amplification is firstly proposed to predict the effective diffusivity of porous media, which can achieve a fast and accurate prediction of the effective diffusivity in a wide porosity range. Then, a three-dimensional fully CNN with the encoder-decoder architecture based on sample structure information self-amplification is proposed to predict the diffusivity, flow and temperature field and corresponding coefficients in different type topological porous media. The comprehensive shape of pore morphology is nothing with the diffusivity and temperature. While the comprehensive porosity in cross section of porous media determines the form of vector, and the shape of pore morphology determines the value of all the porosities. Later, in order to control the heat and mass transfer in porous media, a reinforcement learning method is built to optimize the mass transfer in porous media. The optimized diffusion coefficient can be obtained once the particle size becomes homogeneity. Lastly, a comprehensive model combining genetic algorithm with deep learning method is developed to optimize the distribution of high thermal conductivity carbon fiber bundles and pore structures in the carbon/carbon composite to achieve a better heat transfer performance. The structure with carbon fiber bundles near the heated surface and pore distribution in middle of the structure can achieve a better heat transfer performance. An increase in porosity increases the temperature of the heated surface. Above findings indicate that the proposed deep learning method has a powerful learning ability and can serve as a promising tool to predict and control the thermo-mass transfer process in complex porous media.

Key words: porous media, machine learning, heat and mass transfer, pore morphology

Raffaella Ocone

Heriot-Watt University, UK

Raffaella Ocone obtained her first degree in Chemical Engineering from the Università di Napoli, Italy and her MA and PhD in Chemical Engineering from Princeton University, USA. She holds the Chair of Chemical Engineering in the School of Engineering and Physical Sciences at Heriot-Watt University (HWU) since 1999. She is a Fellow of the Royal Academy of Engineering, the Royal Society of Edinburgh, the Institution of Chemical Engineers, and the Royal Society of Chemistry. In 2007 she was appointed Cavaliere (Knight) of the Order of the Star of Italian Solidarity by the President of the Italian Republic. In The Queen’s 2019 New Year Honours she was appointed OBE for services to Engineering. Recently she has been announced as one of the top 100 Most Influential Women in the Engineering Sector.

At HWU, she is the Head of the Multiphase Multiscale Engineering Modelling (MMEM) research group. Raffaella has worked in a number of highly recognised international Institutions such as the Università di Napoli (Italy); Claude Bérnard Université, Lyon (France); Louisiana State University (USA); Princeton University (USA). She was the first engineering “Caroline Herschel Visiting Professor” at RUHR Universität, Bochum, Germany (July-November 2017) and the recipient of a Visiting Research Fellowship from the Institute for Advanced Studies, Università di Bologna, Italy (March-April 2018). Raffaella’s main area of research is in the field of modelling complex (multi-phase) reactive systems. Raffaella has taken a leading role in debating the role that ethics plays in engineering. Currently she is the EPSRC Established Career Fellow in Particle Technology.

Responsible Technology: are we ready for it?

Abstract:Technology is at the heart of the world where we live providing, among other things, energy solutions, assuring food and drinking water, generating electricity, goods and services. Emerging technologies rise fast carrying the potential to deliver economic and social benefits to a world that is challenged to sustain 10 billion people. Technological and scientific achievements pose challenges and opportunities. The exponential growth of computers, communication and artificial intelligence, for example, is changing the way we work and think, impacting on human activities and ways of living. This talk will explore how global responsibility is embedded in technological solutions and how the ethical dimension affects the way scientists and engineers work and operate.

Professor Jianguo Liu

College of Engineering and Applied Sciences,

Nanjing University, China

Jianguo Liu is a professor of College of Engineering and Applied Science, Nanjing University. His research interest is mainly related to hydrogen production and proton exchange membrane fuel cell. A total of 121 SCI papers have been published in Angewandte Chemie International Edition, JACS, Energy & Environmental Science, Nano Energy and other journals, which have been cited 5800 times, and the H factor is 41.

New research paradigm of proton exchange membrane fuel cells led by artificial intelligence

Highlights

Machine learning models could intelligently provide with valuable decision suggestions and also reveal hidden key factors based on big data for material optimization.

• Interpretation methods applied on as-trained artificial intelligence (AI) models could analyze the influence trends of numerous coupling variables with conciseness and find out the core conflict in a complex system.

• Compared with trail-and-error process, AI guided experimental exploration and model retraining formed a much more efficient developing loop.

Abstract: Development of electrochemical devices and materials was still performed by inefficient trial-and-error processes guided by researchers’ intuition which would not keep up with the increasing demand from renewable energy. For complex systems like proton exchange membrane fuel cells (PEMFCs), chemical reactions at the microscale, mass transfer processes at the mesoscale, and complex thermoelectric coupling fields at the macroscale are simultaneously involved. Therefore, manual analysis and optimization with numerous variables would be extremely strenuous. Therefore, we introduced AI aided models to determine key parameters and give reliable decision aids on strategy worth-taking to fabricate an ideal membrane electrode assembly (MEA). Moreover, different machine learning algorithms were chosen to perform regression task that directly map experimental input to performance output. With an error less than 10% when predicting maximum power density, AI prediction could greatly boost the experimental works as reliable reference.[1] Moreover, we standardized the research paradigm and set up a close-looped workflow to instruct on how the potential of machine learning could be comprehensively exploited to optimize the performance. Four modules: feature selection, decision modelling, regression modelling and online optimization were proposed to combine with efficient AI guided experimental exploration. Both material performance and AI prediction accuracy could mutually benefit each other. Besides electrochemical materials, this paradigm can be widely used as tools in chemistry, biology, medicine, engineering and other fields where the traditional experimental data can be digitized.[2] Moreover, we have discovered the ability of AI to dig out hidden key factor which were long ignored by human researchers. And further provide insights based on big data to reveal the structure-activity change mechanism cooperating with characterization/experimental results. AI could also serve as surrogate model of expensive multiphysics models to greatly reduce the computing time cost as well as to meet the lightweight deployment in practical applications. Beyond that, we introduced interpretation methods on AI surrogate models to analyze the influence trends of numerous coupling variables with conciseness and find out the core conflict in a complex system. AI-guided optimization suggestions were fully supported by the experimental results. Furthermore, the final product achieved 3.2 times the Pt utilization of commercial products, but with a time cost orders of magnitude smaller. Thereby proving the AI-participated decision-prediction-experiment workflow would be revolutionary new research paradigm for the development of electrochemical materials and other labor-intensive research fields.

[1] R. Ding, R. Wang, Y. Q. Ding, W. J. Yin, Y. D. Liu, J. Li, J. G. Liu, Angew. Chem.-Int. Edit. 2020, 59, 19175-19183.

[2] R. Ding, Y. Ding, H. Zhang, R. Wang, Z. Xu, Y. Liu, W. Yin, J. Wang, J. Li, J. Liu, Journal of Materials Chemistry A 2021, 9, 6841-6850.

Markus Kraft

University of Cambridge, UK

Prof Markus Kraft is a Fellow of Churchill College Cambridge and Professor in the Department of Chemical Engineering and Biotechnology. He is the director of CARES, the Singapore-Cambridge CREATE Research Centre, and Principle Investigator of C4T the “Cambridge Centre for Carbon Reduction in Chemical Technology”, which is a CARES research programme. Professor Kraft obtained the academic degree 'Diplom Technomathematiker' at the University of Kaiserslautern in 1992 and completed his Doctor rerum naturalium in Chemistry at the same University in 1997. Subsequently, he worked at the University of Karlsruhe and the Weierstrass Institute for Applied Analysis and Stochastics in Berlin. In 1999 he became a lecturer in the Department of Chemical Engineering, University of Cambridge. In 2012 he obtained a ScD form the same University. He has a strong interest in the area of computational modelling and optimisation targeted towards developing CO2 abatement and emissions reduction technologies for the automotive, power and chemical industries.

Intelligent Decarbonisation

Abstract:Global warming caused by greenhouse gases have caused great concern for a number of reasons. It is clear that drastic changes have to be implemented in the near future to reduce or stop the increase of average temperature and the many negative consequences that go with it. In my talk I shall concentrate on AI-based Cyberphysical systems and knowledge graphs. The decarbonisation of energy provision is key to managing global greenhouse gas emissions and hence mitigating climate change. Digital technologies such as big data, machine learning, and the Internet of Things are receiving more and more attention as they can aid the decarbonisation process while requiring limited investments. The orchestration of these novel technologies, so-called cyber-physical systems (CPS), provides further, synergetic effects that increase efficiency of energy provision and industrial production, thereby optimising economic feasibility and environmental impact. This comprehensive review article assesses the current as well as the potential impact of digital technologies within CPS on the decarbonisation of energy systems. Ad-hoc calculation for selected applications of CPS and its subsystems estimates not only the economic impact but also the emission reduction potential. This assessment clearly shows that digitalisation of energy systems using CPS completely alters the marginal abatement cost curve (MACC) and creates novel pathways for the transition to a low-carbon energy system. Moreover, the assessment concludes that when CPS are combined with artificial intelligence (AI), decarbonisation could potentially progress at an unforeseeable pace while introducing unpredictable and potentially existential risks. The cyber-physical system we are currently developing is called J-Park Simulator (JPS) which is the signature project in the C4T programme of CARES at the University of Cambridge and part of the http://www.theworldavatar.com/ project. JPS consists of a network of IRIs comprising domain ontologies, a knowledge base and different types of agents. One important application is the modelling and optimisation of eco-industrial parks. This includes the electrical grid, various networks of materials, for example, waste heat network along with a detailed model of each industrial process. In my talk I shall explain how JPS works and show a couple of examples.