Tetranex recently had the privilege of sponsoring a University of Calgary 4th-year Engineering Capstone Project. Tetranex has interest in exploring how machine learning technology, specifically classification, can be integrated within the automation and controls industry. To that end, we provided the team with a NVIDIA Jetson Nano developer kit ($150), a Raspberry Pi CSI camera ($35) and an additional project budget of $1000.
Tetranex was looking for an early milestone; detection of a human and event notification to a control system via a dry-contact. The team was then asked to develop a further proof-of-concept of an industrial application executing solely on the Jetson Nano as an edge device. In addition, we required the team to develop a full engineering manual for their proof-of-concept.
The team chose to pursue an ambitious goal of automating the identification and sorting of refundable recyclables through a full working prototype which became known as the formidable Recyclops. A machine that could detect and sort four distinct refundable types and reject the rest.
After the team got their machine learning bearings with the early milestones, the Recyclops continued to demand a steep learning curve, especially considering the course load of a 4th year engineering student. To summarize some of the highlights of the required skills, drawn from the team’s documentation, they had to quickly get proficient with:
- Convolutional Neural Networks (CNN)
- Pre-trained Models and Transfer Learning
- SSD MobileNet V2
- NVIDIA JetPack SDK
- Python and TensorFlow
- Google Colab (Jupyter Notebooks)
- Data acquisition, labelling and augmentation
- Model training, tuning and tweaking
- 3D design and printing
- Servo control
- Procurement (Canadian Tire doesn’t sell application specific conveyor belts)
On top of these challenges, the team was hit with another when the University, and the world, locked-down due to the COVID-19 pandemic. The team persevered and pushed through to a full working prototype under a social-distancing mandate. Unfortunately, the team was unable to showcase their accomplishments at the Capstone Design Fair but they created a fantastic video of Recylops in action.
Working through programs like this offer real project experience to the students and yield tangible R&D value to Tetranex. Tetranex is very proud of the accomplishments of this team and we hope they had some fun along the way.
Credit where credit is due:
- Alen Mei
- Amanda Lee
- Chase Thomas
- Daniel Tablazon
- Joshua Flores
- Safa Syed
Academic Advisor: Jesse Chaulk
Teaching Assistant: Thomas Truong