Delving into data to solve public works and manufacturing problems
May 8, 2020 / by Ryley McGinnis
From the time of the Roman aqueducts to the age of superhighways, engineers have solved problems and devised solutions to improve people’s lives. Today’s engineers use algorithms, coding, and find high tech solutions to solve contemporary public works and manufacturing problems.
In their last semester, two capstone project teams of data analytics engineering master’s students demonstrated this premise by applying their knowledge and engineering solutions. Both teams are hand-coding their solutions using the You Only Look Once (YOLO) algorithm.
“These students understood the business problem, wrangled complex data, wrote detailed code that used a powerful algorithm, and deployed an anomaly detection model with exceptional visualization,” says adjunct professor and the teams' project advisor Thomas Ferleman, an artificial intelligence/machine learning business development manager at Amazon.com.
One team examined metals like steel to detect flaws that may occur during the manufacturing process. “Defects in steel can’t be seen with the naked eye,” says team lead Shouryasimha Addepalli. “If there is a defect while manufacturing steel, you can have hundreds of sheets with these defects. But if you can identify a defect while manufacturing, you can save a lot of time and money, and your steel will be the best quality,” he says. Faulty metals can lead to architectural issues in bridges, planes, cars, and more.
The second team adapted the algorithm to find and locate potholes, which can cause accidents and damage cars. “Our industry sponsor went back to his home country, and he noticed potholes in the roads that go unnoticed by the authorities,” says team lead Jiyad Ur Rehman. “So, we wanted to make a program that can automate this process of finding potholes and marking them while driving down the road.”
The road to finding their solutions wasn’t always easy, and they encountered a few potholes of their own that they had to overcome. Both teams had to adapt to changes in the computer system they were using to find their solutions. After they worked with one type of model, a single-shot detection (SSD) model, for about a month, their sponsor, Shamshad Anasri, Chief Executive Officer of Accure AI Inc., requested a change to the YOLO algorithm, and the teams had to start over.
However, whether the teams had to change their scope or solve technical problems, they always rolled with the punches and joined forces with their sponsor and instructor. “When we were testing, our model wasn’t running, and we later discovered it was because of one comma in the code that it wasn’t working,” says Addepalli. “We hadn’t worked with these models before, and this taught us that minute details play a crucial role in the success of a project.” Their code will be available on GitHub so that people can use their code to learn and for their own projects.
Communication and goal setting between everyone was also essential. “We meet with our industry sponsor and Professor Ferleman every week, and even with COVID-19 we are still keeping that communication,” says Rehman.
The teams say that this project and the guidance has helped them push themselves. “Professor Ferleman has been there to guide us and provide support. He pushes us to draw up new ideas, reach new goals, and get better solutions,” says Addepalli.
“My role is to ensure each student succeeds. If they fail, it's because I failed them,” says Ferleman.