Aryan Deshwal Wins Innovative Deployed Application Award at IAAI Conference

Department of Computer Science & Engineering Assistant Professor Aryan Deshwal won the Innovative Deployed Application Award at Association for the Advancement of Artificial Intelligence (AAAI) Innovative Applications of Artificial Intelligence (IAAI) Conference in January. The winning paper, “Discovery of Feasible 3D Printing Configurations for Metal Alloys via AI-driven Adaptive Experimental Design,” showcases how AI can expedite the manufacturing process for certain metal alloys by narrowing down the feasible configurations to 3D print a substance.

Collaborating with the additive manufacturing team at Washington State University, Deshwal created an algorithm to help 3D print the alloy GRCop-42, which is particularly useful in aerospace and biomedical applications. Manufacturing these types of alloys is typically very expensive due to cost of physical resources and lab time. Using AI, Deshwal was able to narrow 100 million feasible configurations to print this alloy down to 40, all while keeping the laser power level below 1000-watts. 

“This project falls broadly under the umbrella of adaptive experimental design problems,” Deshwal said. “The main challenge is that the number of feasible configurations to print this alloy are roughly around 100 million, and trying every option is very expensive and time consuming. My collaborators had tried a number of trial and error-type methods, but it wasn’t working. That’s where AI comes. My expertise is developing AI algorithms to enhance and accelerate these search processes.”

Deshwal’s research develops what he calls “AI-driven adaptive experiential design.” This work requires a two-step process that can be applied across multiple domains. When trying to design an experiment in the manufacturing space, scientists do not know the relationship between the configuration and the quality of output. Deshwal’s first step is to use machine learning to model this relationship and quantify the uncertainty of output for each remaining choice. The second step is to use formal reasoning under uncertainty to figure out which remaining options should be pursued to reach the end goal. It combines machine learning with reasoning under uncertainty and human expertise in order to accelerate the experimental process and use fewer resources.

“A big part of the reason we earned this award is because of the ‘lab in the loop’ framework we used,” Deshwal said. “Every time AI suggests a design, domain experts actually carry out the experiment. Experiments can take time and often this is a key bottleneck in real world science and engineering problems. We worked to trim down the time of this loop and completed our experiments in three months.” 

This project was the first time that someone discovered feasible configurations for GRCop-42 under the laser power limitations set by the lab (1000 watts). Deshwal’s team actually found solutions with power requirements lower than that suggested limit. This type of work enables scientists to design alloys in a resource-efficient manner, and proves how AI-driven acceleration of scientific discovery can be helpful.

“At the University of Minnesota, we have a lot of momentum around this research direction of AI-accelerated scientific discovery with leadership from Vipin Kumar and Shashi Shekhar,” Deshwal said. “We have done a number of widely-attended workshops on this topic and have more planned in the coming months.”

Learn more about Deshwal’s work and read the full paper.