Purdue University

A Purdue professor who uses data analysis to study city infrastructure such as roads, buildings, sewer pipelines and nuclear reactors and predict stresses before they happen, avoiding infrastructure failures, is using RCAC’s Gilbreth community cluster supercomputer to deliver the huge computational power needed for his research.

Mohammad Reza Jahanshahi, an associate professor of civil engineering, and his research team rely heavily on deep learning models to perform their data analysis of city infrastructure, which requires powerful GPUs like those available on Gilbreth.

In one project, Jahanshahi and his team have collected data from camera sensors on roads to identify the stresses and rate the quality of the roads. Using Gilbreth, they’ve trained a deep learning AI system to identify and classify road stresses.

Another project detects defects in pipelines around the points where they’re welded together. Using radiography data, Jahanshahi and his team again develop and train deep neural networks to identify and detect where the disturbances are in the welds. Gilbreth’s GPUs are essential for the large amounts of data needed to train the deep learning agent.

In a third project, Jahanshahi and his team are developing a tool to help guide a robot to detect cracks on the surface of a nuclear reactor. Rather than scanning the data and sending it home to be processed later, the goal is for the robot to inspect the reactor surface like a human would – essentially processing the data on the go, stopping to take a closer look at areas that look suspicious.

Before turning to RCAC’s clusters, Jahanshahi’s lab operated their own computers, which he says was a time-consuming and frustrating endeavor that often involved devoting a lot of graduate student time to fixing IT issues.

“Since moving our work to RCAC, we can do things much faster,” he says. “I don’t need to worry about devoting student time to server maintenance.”

“I’ve been very happy with not only the clusters, but also the staff support and storage available to us through RCAC,” adds Jahanshahi. “RCAC staff are available to troubleshoot anything that goes wrong with the clusters, and Integrated storage means we don’t have to worry about setting up a separate server to archive our data.”

To learn more about Gilbreth and other RCAC resources, contact rcac-help@purdue.edu and visit the RCAC website.