Project Summary: Led Team RPM to first place in the 5th KAIST-POSTECH-UNIST AI & Data Science Competition, developing a defect rate prediction framework for tire pilot production among 83 teams with 238 participants, earning the Grand Prize and $2,300 as award with recognition from Hankook Tire & Technology.
The Manufacturing Challenge
Korea's top three science universities partnered with Hankook Tire & Technology to challenge teams with a critical industrial problem: predicting defect rates in pilot tire production to optimize decision-making. Our mission was to develop a machine learning system that could forecast quality outcomes and guide production choices based on limited pilot run data.
The Innovation
Fusion Input Architecture We developed a novel fusion framework that combines sequential sensor measurements with tabular manufacturing parameters. This hybrid approach captures both temporal patterns in the production process and static characteristics of each tire configuration, enabling comprehensive defect prediction.
Comprehensive Model Exploration Beyond our core LSTM-Transformer ensemble, we experimented with CNN for spatial feature extraction, GNN for relationship modeling, TabNet for tabular data, and MLE-STAR agent-based approaches. We also visualized sequential data as tire contour representations to explore geometric patterns in defect formation.
The Results
Team RPM achieved 1st place with a public leaderboard score of 0.59607 (1st place) and maintained competitive performance with a private score of 0.63260 (6th place). Our systematic approach and diverse model experimentation earned us the Grand Prize, $2,300 as award, and recognition for technical excellence.
As team leader and presenter, I coordinated our technical strategy across four team members, managed the model development pipeline, and delivered the final presentation showcasing our innovations to industry judges from Hankook Tire & Technology.
Industry Impact
The project received media coverage in News1 as part of Hankook & Company Group's sponsorship announcement. Our defect prediction framework demonstrates immediate applications for manufacturing quality control, pilot production optimization, and data-driven decision-making in tire production, contributing to the advancement of AI-powered manufacturing systems.