Civic Score

About the Project — Mechanical Part Detection with AI

Inspiration

This project started from the challenge of recognizing simple mechanical parts from images. Since components like bolts, nuts, washers, and locating pins appear everywhere in mechanical designs, we thought it would be interesting to see if a machine learning model could identify and count them automatically.

What I Learned

During the hackathon, I learned how to work with image datasets and frame a problem as a multi-label prediction task. I also learned how small changes in training and preprocessing can affect model performance, especially when working with synthetic images.

How I Built the Project

The project was built using a computer vision model trained on the provided synthetic images. The model predicts how many of each part appear in an image—bolts, locating pins, nuts, and washers. The final output is generated in the required format: [ \text{Output} = [\text{bolt}, \text{locating pin}, \text{nut}, \text{washer}] ]

Challenges Faced

One of the main challenges was making sure the model could handle images with multiple parts at the same time. Another challenge was preventing overfitting, since the data was synthetic. Managing limited time during the hackathon and tuning the model efficiently was also a key learning experience.

Built With

• csv
• data
• jupyter-notebook
• numpy
• opencv
• python
• pytorch