NRT Soft AE Immersive Experience – Trainee Reflection

Bridging Polymer Science, Electrochemistry, and Machine Learning at ENGINE2025

By Justin Hughes 
Dept. of Materials Science & Engineering, University of Pennsylvania
NRT Soft AE Trainees
Published March 11, 2025

 

For my Immersive Experience, I attended the Grenoble Energy Conversion & Storage (ENGINE2025) Winter School at the Grenoble Institute of Technology in Grenoble, France. Given that developing energy production and storage technologies are integral to energy, this was an interesting and crucial topic for instruction, especially for graduate students working on topics related to this field. This workshop was also a  unique experience since the organizers facilitated a mix of content: lectures on fundamentals of electrochemistry and interfacial science, selected research applications such as all-solid-state batteries, tours of world-class international research facilities such as the Institut Laue-Langevin (ILL) and European Synchrotron Radiation Facility (ESRF), and poster sessions where we had the chance to both present our own research, as well as discuss our topics with other students, staff scientists, and others. I presented a poster, “Bayesian Optimization-Driven Autonomous Exploration of Polymer Phase Boundaries” and focused on my recent developments in establishing experimental workflows to increase resource utilization efficiency and time efficiency in the study polymer-based materials by employing Gaussian process (GP)-informed Bayesian optimization (BO) methods.

group of people standing in front of the EPN Science Campus sign

This entire experience in Grenoble was unique because it helped me gain perspective on both the scientific content and machine learning-related aspects of my research on developing polymer phase diagrams. On the science side, I eventually want to move towards using these Bayesian methods to develop solid polymer electrolytes (SPEs) for solid state batteries, but out of all of the workshop attendees, I probably had the least extensive background in electrochemistry. However, through the earlier fundamentals-oriented presentations, I began to learn the vocabulary and basic mechanisms of how chemical processes such as lithiation during battery cycling affected the structure and properties of a battery. Additionally, I developed an appreciation for how my specific research would fit into the context of applied science: solid state batteries are generally useful since they have a higher energy density than liquid electrolyte-based batteries, as well as having more favorable safety metrics. But one major drawback is volumetric expansion and contraction from lithiation and delithiation during battery cycling causing ceramic-based electrolytes to either delaminate or have otherwise adverse interfacial interactions with the electrodes. Alternatively, SPEs have more favorable flexibility and toughness which help promote better interfacial contact, even during cycling.

On the machine learning side, it was interesting to see that I don’t recall any of the student-presented posters mentioning machine learning methods, and few of the presentations did either. For a field where many of the materials used in complex, multicomponent applications are composites themselves, there is a lot of unexplored potential in leveraging machine learning processes. For instance, ML could be used to help navigate high-dimensional material selection, sample preparation, and characterization parameter spaces to make the best material possible for a given application.

On the other hand, it was encouraging that during discussions with beamline scientists at the ILL and ESRF, they were either already starting to or at least open to implementing data science-oriented methods, such as BO, to help with data analysis and sample preparation. This makes sense given not only the complexity of the science at hand, but also the complexity of data analysis: the data output from an experimental session of neutron or X-ray scattering could reach gigabytes or more of total storage space, so having either parallel analysis running during data acquisition, or at least a robust framework laid out beforehand, would be useful. Since beamtime at either of these facilities is such a scarce resource, it would be wise to leverage methods such as BO that employ acquisition functions to recommend only the most useful experiments to be conducted.

Lastly, it was really encouraging to see the level of collaboration that was present even on the other students’ research topics. Many of the other students that presented their posters indicated that several of their collaborators were from foreign countries, so I got to see how frequently and easily different scientists and engineers from different backgrounds and perspectives were able to come together to help address common research goals.

Attending this workshop was an invaluable experience. It taught me how my research in soft matter fits into the bigger picture of energy storage applications, how large-scale research facilities are run, and how collaborative research is fostered across even international borders. It also helped elucidate the underexplored potential of machine learning-based methods for improving scientific workflows, especially in the context of large-scale data processing. In sum, attending the ENGINE2025 Winter School helped me develop a broader understanding and appreciation for how research is conducted – lessons learned that I will apply in my research moving forward.