Graduate student earns second place in ASCE paper competition for research on UB's multiple-fan wind tunnel

Teng Wu, Shaopeng Li and Shivam Mishra stand in front of the wind tunnel.

Teng Wu stand with  Shaopeng Li and fellow graduate student Shivam Mishra in front of UB's wind tunnel. 

By Peter Murphy

Published June 17, 2020

The Engineering Mechanics Institute (EMI) Dynamics committee is one of the largest in the American Society of Civil Engineers (ASCE), and the committee awarded a UB civil engineering graduate student second place in its annual student competition. 

Machine learning in wind engineering

Print
“The research Shaopeng presents uses machine learning, specifically deep reinforcement learning, to intelligently control our wind tunnel so we can effectively generate the transient winds in extreme events. ”
Teng Wu, Associate Professor
Department of Civil, Structural and Environmental Engineering

Shaopeng Li earned second place in the competition for his paper titled Active Simulation of Transient Wind Fields in a Multiple-Fan Wind Tunnel via Deep Reinforcement Learning. According to Li, this work uses artificial intelligence (AI) to more accurately simulate critical wind events in a multiple-fan wind tunnel. 

“Downburst winds are responsible for large amounts of damages on civil structures. They have transient features and cannot be physically simulated in a conventional boundary wind tunnel,” Li says. 

Li and his advisor associate professor Teng Wu have utilized UB’s actively controlled multiple-fan wind tunnel to generate downburst wind fields, but these fields are difficult for the researchers to control. According to Wu, Li’s research will help develop ways to effectively generate downburst wind fields. 

“The research Shaopeng presents uses machine learning, specifically deep reinforcement learning, to intelligently control our wind tunnel so we can effectively generate the transient winds in extreme events,” says Wu. 

The researchers have already started some experimental work to incorporate machine learning into the wind tunnel. The researchers are proposing a novel control scheme based on deep reinforcement learning, which is automatic, efficient and accurate. 

Li, fellow graduate student Reda Snaiki and Wu propose a novel control scheme based on reinforcement learning that will be automatic, and efficiently generate downburst wind fields. 

Close to 20 students submitted papers to the EMI Dynamics Committee student paper competition this year. According to Wu, Li’s success in the competition is significant. “A second-place finish in this competition means Shaopeng’s research work is high quality, and gets his peers’ attention,” Wu says, “The feedback was very positive and this is a good sign for him.”

Li’s long-term research aims to combine numerical simulation tools with physical testing tool to enhance wind-sensitive structures, while incorporating state-of-the-art AI schemes into his systems.