Assistant professor of mechanical engineering Anand Nellippallil is taking his research out of the university classroom and into the middle school classroom to teach the next generation of engineers the problem-solving skills they need for careers in that field.
Nellippallil had his first session with the Palm Bay Academy middle schoolers in September, where he introduced them to basic engineering principles of design optimization, tradeoffs, machine learning and AI-assisted decision-making.
He will meet monthly with the students to guide them in solving basic engineering problems and teach them how emerging technology can be used by engineers to make informed decisions.
“When I went there last time, I asked them, ‘How many of you want to become an engineer?’ And there were a couple of them who were very excited to become an aerospace engineer — some said computer science,” Nellippallil said. “So I told them, ‘Ok good, so now with that goal you have, let’s think of how you’ll be using this in your future.’”
Nellippallil’s PBA outreach is part of the first of two National Science Foundation grants he received focused on improving the decision-making process of engineers by implementing AI and machine-learning technology.
One of his goals is to strengthen STEM education by providing students with fundamental engineering knowledge, which prompted him to reach out to PBA to introduce these concepts to students at an early age. He began working with the students’ teacher Jose Perez, who has a Ph.D. in electrical engineering, to create an environment where students can learn from real-world experts.
“By connecting their classroom with the expertise of Florida Tech faculty, students gain a better understanding of what it means to pursue a career in engineering and related STEM fields,” Perez said.
In May 2023, Nellippallil received the first grant, an Engineering Research Initiation award titled “CMMI-2301808: Multilevel Inverse Robust Co-Design of Materials, Products, and Manufacturing Processes,” and started developing design methods and machine learning models to help find solutions to engineering problems.
There are three primary factors at play in Nellippallil’s work: manufacturing, material and product. In one part of his research, the product was a steel rod, manufacturing is the process of how it was made and the material is the microstructure of the steel that gives the product its physical properties.
Each of the three categories require different processes to be the most efficient, but improvements in one category often come with costs in the others. Managing all the variables at the same time and designing based on how they interact is known as co-design.
Nellippallil aims to use AI and machine learning models to simplify the co-design process and minimize the compromises that must be made.
He explained that each step of the process brings uncertainty and variables, which can create bigger problems down the line if they aren’t addressed.
“We use artificial intelligence to help us identify these robust regions, or robust solutions that are relatively insensitive to uncertainty,” Nellippallil said. A robust solution may not lead to the best outcome in any one single category, but it provides the fewest compromises throughout the entire process.
Nellippallil and his students don’t use physical experiments to find these solutions. Instead, they use simulations to mimic each step of the process, then collect data and measure what affects the material microstructure.
“Once we generate data from simulation, we then train a machine learning model to learn that process (so) we don’t have to use a simulation anymore,” Nellippallil said.
The machine learning models trained on the simulations can provide predictions for a robust solution area that leads to the best outcomes across the three engineering design categories.
The second NSF grant Nellippallil received in collaboration with the South Dakota School of Mines and Technology is titled, “POSE: Phase I: Open-Source Decision Support in the Design of Engineering Systems.” The grant started this year and is a part of the Pathways to Enable Open-Source Ecosystems program, which promotes the development of open-source engineering tools.
This grant allows Nellippallil to continue building the machine learning models and develop them into AI agents that can effectively make decisions and create robust solutions.
Agents are essentially one step above machine learning models. Machine learning models can predict outcomes based on given data, but the agents can be trained on what decisions humans make with that data, so they become capable of making those decisions themselves.
“The human decision maker only needs to be like a mediator. They can be an observer to how the agents are behaving, and as required, they can intervene,” Nellippallil said.
POSE provides Nellippallil and his team of researchers the opportunity to take the principles they developed in the first grant and use them in other fields as well.
Nellippallil said these same processes can be used to improve manufacturing networks and anything else that requires complex decision-making weighing multiple factors.
