-
Main page
-
Dashboard
-
Recognition accuracy
Inspiration
Most of our inspiration for this application comes from our university background. In classes, we had discussed the ways manufacturing is done and how certain cost cutting measures are taken during production.
What it does
This application uses machine vision to analyze production circuits and make sure that errors in production do not get forwarded to the end user. Components being implemented are put through a hand trained model to determine when production boards are missing essential components. This provides peace of mind for the customer along with providing a dataset of how the manufacturing company can make products.
How we built it
This system was built using PyTorch for the ML backend with React and Node.js for the front-end. The backend database and connections were accomplished using Django.
Challenges we ran into
A major challenge we ran into during the first day was the method of training our machine learning model. Initially we intended to use unsupervised learning, however, after doing more research, it was found that using unsupervised learning for the job we were intending would have been extremely difficult and using supervised learning would be a much more efficient and reasonable method for the time remaining.
Accomplishments that we're proud of
Being able to shift away from the primary system of our system to move to a different model requiring additional research and time to set up and implement was a difficult challenge, but we are very proud that we were able to do it within the time available.
What we learned
Something we have learned regarding the method of machine learning is the use cases each method would ideally be used for. However, a more important idea we learned was being flexible and being able to adjust to the challenges you face in the moment.
What's next for Assemb.ly
We would like to experiment with unsupervised learning again due to its ability to easily pick up on new designs for circuits without having some individuals manually label each component
Log in or sign up for Devpost to join the conversation.