Recycling Image Processing
The objective of our project was to investigate the plausibility of sorting recycled materials
using a convolutional neural network. Using Matlab we modified GoogLeNet to detect five
classes of recycling: glass, plastic, paper, cardboard, and metal. Our dataset contained
only 500 images, 100 for each class, and still provided accurate results. Our network
reached 86% validation accuracy which we believe to be a solid proof of concept for
sorting recycling with computation intelligence and automation. See our report at the link below.
Ski Composite Testing
The objective of this project was to compare simulation results to physical testing results
for bending of a ski composite. Rather than using a complete ski layup we decided to simplify
our model to just the wood core with a layer of 19 oz fiberglass above and below. We developed
two simulations of our bending load case using Matlab and Abaqus which we could compare to
the results of our physical testing. This project gave me experience with composites and developing
test procedures. See our memo at the link below.
Bicycle Geometry Study
The objective of this project was to design, build, and test a bicycle based on desired performance characteristics.
I ran simulations in Matlab based on single-track vehicle models to determine the geometry I would need
to increase the stability and rider height on an existing Trek bike. By changing the head tube angle
and the top tube length and angle I was able to exceed my desired stability. It turned out to be an
excellent commuter bike which was my goal for the project.