TE Connectivity AI Cup 2024

I competeted in the TE Connectivity AI cup in 2024 representing Purdue University, winning first place in the US and 2nd place internationally. My team and I implemented a pipeline for collecting and labelling data and training a diffusion model (based on https://arxiv.org/pdf/2303.17959) to segment the various steps in the manufacturing process.

I trained a YOLO model using Ultralytics to detect the gloved hands of a worker during a video of a manufacturing process. Using a sliding window technique and a web app, we created a novel way of creating action segmentation datasets to identify the various stages of the manufacturing process. We then trained a diffusion action segmentation model to automatically cluster and classify various sections of a full video into the manufacturing stages. I optimized the hyperparameters to get a peak 88.7% accuracy on test datasets.

This scalable solution generates $170k per manufacturing line with over 50 deployment opportunities (so a total of $8.5 million a year if implemented everywhere) as it allows industrial engineers to optimize manufacturing processes and compare the productivity of workers.

Note: some aspects of this project are being filed for patents, I will update this project description with more information when I am allowed to.