Envisioning a new era of design through artificial intelligence
Associate Professor Miguel Bessa focuses on the data-driven design and analysis of materials and structures.
Standing at his desk against a newly painted wall chosen for its resemblance to FC Porto colors, the newest Associate Professor of Brown Engineering radiates almost as much determination and team spirit as his hometown team from the top flight of Portuguese football. Miguel Bessa has comfortably settled into the dark blue office on the top floor of Barus and Holley where he welcomes the idea of initiating broad campus collaborations where artificial intelligence can help solve human challenges.
“I always knew I wanted to be an engineer, that was clear. My grandfather is an engineer, my father is an engineer, my mom is an engineer. In fact, most of them are mechanical engineers,” he said. “I actually wanted to do operations research – management engineering. I was thinking about that because it was a very selective program in Portugal. I met with faculty at the University of Porto, but it was very clear in that meeting that what I liked the most was mechanical engineering. Even though I didn’t want to do the same as my family, apparently it just runs in our blood,” he laughed. “And it was the right choice.
“Originally, I was thinking about renewable energies, thermal areas, nothing to do with structures. But then I met one of the most highly regarded professors in the field in Portugal; someone who has a worldwide reputation in composite materials. Pedro Camanho took me under his wing when I was just an undergrad. I started doing research on composites, but I remember telling him at the time that I was probably going to end up working on renewable energies … Truth be told, I got caught up in it. In fact, I discovered that what I really, really enjoyed was to do method development – developing new computational methods to solve problems.”
Still unsure of what the future would bring upon completion of his undergraduate degree from Porto, Bessa credits Camanho again for being the support system he needed to get him to the United States, not only for a doctoral degree, but to work with the best computational mechanics researchers he could find. That short list included Northwestern, Caltech and Brown. “As luck would have it, I ended up doing my Ph.D. at Northwestern, postdoc at Caltech, and now I’m faculty at Brown. It’s truly a dream come true and I’m not just saying that. I could not have planned for a better outcome. This is everything I dreamed of. I did my Ph.D. with some of the best people in the world in computational mechanics. It was very difficult – it was a very difficult period of my life and I think a lot of Ph.D. students can resonate with that – but it was very, very rich as well,” he said.
“We did some work with machine learning to actually solve solid mechanics problems,” Bessa said of his Ph.D. work. “It was fun, because at that time, the field was not doing artificial intelligence, certainly not for the problems we were looking into. Then in my postdoc, I wanted to work with experimentalists instead of method development people because I wanted to see if we could actually do something with the methods we created. I went to warm California, but it turned out to be a short postdoc because I got a job offer back in Europe. Still, Sergio Pellegrino’s group at Caltech brought my attention to the importance of using new methods to solve real engineering problems.”
It was at CalTech during his postdoc where Bessa noticed a structure in the corner of the Space Structures Lab that was able to open out large, expansive solar sails from within a very small storage space. He wondered if it would be possible to design a highly compressible yet strong material that could be compressed into a small fraction of its volume. "If this was possible, everyday objects such as bicycles, dinner tables and umbrellas could be folded into your pocket,” he said at the time.
Delft University of Technology (TU Delft) in the Netherlands would offer Bessa his first faculty position in 2017, and it was there that he would successfully invert the process by using machine learning for exploring new design possibilities while reducing experimentation to a minimum. Bessa and his team were able to find a design that transforms brittle polymers into lightweight, recoverable materials that are super-compressible. In collaboration with Richard Norte’s group, they also designed an optomechanical sensor with excellent properties. These two works are found in Advanced Materials (October 2019 - Bayesian Machine Learning in Metamaterial Design: Fragile Becomes Supercompressible and October 2021- Spiderweb Nanomechanical Resonators via Bayesian Optimization: Inspired by Nature and Guided by Machine Learning). “In both accounts we believe we found new properties,” he said. “In the first case, it was the first time to find reversible supercompressibility in a metamaterial, and in the second, at the time at least, we broke the world record for the highest quality factor for that sensor.
“So once again we were showing that machine learning can really help us design new things. There are a lot of things that are difficult to translate into practice when using AI, but with the right problems in the right context, you can really do great engineering with these techniques.
“I loved my time in Delft. Frankly, I was not looking for another job,” he said. “But when the opportunity to join Brown appeared, I did not hesitate to accept it. Brown is a special place for Solid Mechanics.”
Bessa now sits in Barus & Holley as part of the solid mechanics group: His focus remains on the data-driven design and analysis of materials through computational mechanics and machine learning. His expertise lies in the methodology and he believes Brown will give him exactly the opportunities he craves to develop new artificial intelligence techniques leading to new materials and new structures. “In the last two years,” he said, “I have pivoted to fundamental machine learning. I am almost doing computer science at the moment, because I want to develop new AI techniques so that I can go back and then design new materials and new structures. That’s the vision for my group – fundamental work that opens up doors to solve relevant engineering problems, and using the challenges found in engineering to motivate the fundamental work.
“The dream is actually to use artificial intelligence to do design from scratch without human intervention. We are far from being in that position, but the question is, ‘Can we use AI to discover new things autonomously?’ If we were close to getting that done, there would be ethical implications in doing this, but we are far from it. We want to find new algorithms that truly lead you to solutions you couldn’t get to easily by trial and error or human intuition. That’s the overarching goal and we try to make small steps toward this. For example, we developed an algorithm for cooperative data driven modeling that was just released where we can use machine learning cooperatively. If a lab develops a machine learning model, we can take that model, build upon it and don’t destroy what they did. This is something that was uncommon or inexistent in previous literature.
“Every code we develop, we put online on Github. We want people to find, use and improve them. That’s what helps science move forward. Our default policy is to put everything online once it is published, according to FAIR (Findable, Accessible, Interoperable and Reusable) principles,” he said.
Bessa will be teaching a graduate level course in Fall 2023: Data-driven Design and Analysis of Structures and Materials (written 3DASM and pronounced three-DAZZ-em), and his lab space will consist of his office and a dedicated allocation in Brown’s Center for Computation and Visualization supercomputer cluster. He is seeking to increase lab members beyond his postdoctoral researcher and a Ph.D. transfer from Delft, and has already been advising interested undergraduates.
“You just never know where the inspiration comes from,” he said. “It can be in the most random conversation, or random observation. That's what’s so beautiful about research, that’s why this never gets old. But it takes a lot of effort, that’s the catch. It never stops.”