If you have any problems related to the accessibility of any content (or if you want to request that a specific publication be accessible), please contact us at scholarworks@unr.edu.
SKI SENSE, A METHOD FOR CLASSIFYING SNOW
Date
2022Type
ThesisDepartment
Mechanical Engineering
Degree Level
Master's Degree
Abstract
The fields of snow science, snow sports, and surface classification rarely cross paths, however understanding the specific type of snow on a mountain has value to all of these industries. Understandingthe structure, quality, and type of snow is crucial for assessing avalanche safety, interpreting satellite imagery and choosing the right pair of skis for your weekend trip. Generally, the type of snow
is known before it is skied on and in the past, little research has been done on the classification of
snow while skiing on it.
This work explores the information that can be extracted from strain signals coming from an
alpine ski with the goal of classifying four snow types. Data was collected from powder, slushy,
groomer, and icy snow. Using the strain information and ski boot angular velocity I was able to
classify these snow types independently of the skiing style with a maximum success rate of 97%
by implementing a Na¨ıve Bayes classifier. Comparisons of classifier performance between strain
gauges, indicates that the optimal placement of strain gauges is halfway between the binding and
the tip/tail of the ski. Additionally, orienting the strain gauges parallel and perpendicular to the
length of the ski provides different, but complimentary information. Results from the principal
component cluster plots show greater separation of powder snow on perpendicular strain gauges
and parallel strain gauges show more defined clusters for hard snow types. The ability to classify
the snow being skied on opens the door for many applications in the snow sports industry, as well
as the robotics industry, in the context of surface classification.
To arrive at the results, I have implemented numerical methods of data analysis including singular value decomposition and uniform manifold approximation and projection. These tools have
allowed me to extract meaningful information from real world data. Additionally along the journey
I attempted using other techniques such as Dynamic Mode Decomposition to recreate a model
of the ski’s bending dynamics, however, due to the limited number of sensors this approach was
unsuccessful.
Permanent link
http://hdl.handle.net/11714/8201Additional Information
Committee Member | Tung, Ryan RT; Nolin, Anne AN |
---|