Our program recognizes 6 defects on printed circuit boards such as:
- Missing hole
- Mouse bite
- Open circuit
- Short
- Spur
- Spurious copper
- Technologies
- System Requirements
- Architecture
- Deploy
- Program Demonstration
- Limitations
- Success Metrics
- Contributing
- FAQ
- Project Team
- Links
You will need ~10 GB of free disk space to deploy the container on your machine
The system consists of three services:
- Django - a web application that provides a way for the user to upload a picture of the chip
- RabbitMQ - as a task broker to organize a message queue between the model and the application
- Fast API - program interface for sending the results of the YOLO model work
Using Docker compose, the system can be started using the following steps:
- Clone the given repository
- Run docker on your machine
sudo docker-compose buildsudo docker-compose up -d.
The following things can be emphasized as limitations at the moment:
- Frame rate per second - 4 (since the response time of the model is 200-250 ms).
As a business metric, we use the reduction in enterprise costs and board survey time by replacing controllers REA by the service being developed.
We used mAP50-95 (mean Average Precision in the range [.50: .05: .95]: 0.5 to 0.95 with a step size of 0.05 as the metric for evaluating the success of the experiments. The original goal was to achieve a value of at least 0.9, which was achieved in experiment 8 - we achieved mAP50-95 = 0.9289
If you would like to participate in the project development, give feedback or complain about errors - write to someone from the project team (below).
We will fill it in as errors occur while using the project.
- DE + ML + PM - Elizaveta Talynkova
- ML + Back-end - Mulham Shahin
- DE + ML - Siniaev Viacheslav