Establishing Methods to Track Surface Avalanches on a Bead Pile

Benjamin Harris


I studied surface avalanches on a granular pile model using two experimental apparatuses and a computer simulation. I vetted a three dimensional Newton's law bead pile simulation against results from previously conducted experiments. The simulation allowed for mass data collection of beads that fell off the pile during successive avalanches. Data collection was performed in the range of 1000 to 10 000 bead drops. I fit the data to a power law with an exponent τ = 1.5±0.1, compared to 1.5 observed in previous research for bead piles with small drop heights. Validating realistic behavior of the simulation established the potential to extend tracking to the movement of all beads and surface avalanche behavior. I constructed two experimental apparatuses to observe surface avalanches. I utilized gumballs in developing experimental methods because they were large, bright, colorful, and cheap. The gumballs used in the first and second apparatus were manufactured to be 0.92 ± 0.05 inches and 0.76 ± 0.05 inches respectively. The first model was a quarter conical pile composed of 1000 gumballs with a camera and microphone mounted to record the visual and aural qualities of avalanches. The second apparatus used automated bead delivery to drive a full conical pile. I established video tracking as the ground truth method to track avalanches by the total displacement of beads on the surface. The total displacement was the fundamental metric by which other methods were compared to. I expected to see larger displacements for larger avalanches. The disadvantage of video tracking was the tedium of the procedure. Manual tracking was slow and quickly encumbered the software used. My second method was subtracting preand post-avalanche images to measure the resulting difference in visual characteristics. I expected to see larger visual changes on the surface for larger avalanches. Image subtraction was advantageous because I developed a procedure to automate the process. The disadvantage was that data only correlated to the ground truth with typical correlation coefficients of 0.61 and 0.71 for the two subsets of image property channels tested. The last method was audio tracking. Data collection was conducted in a quiet place with a microphone to record avalanches. I expected to see that louder avalanches were larger in magnitude. Audio tracking was advantageous because it established a method of comparison to ground truth that could be automated with different equipment. The disadvantage was that when comparing audio data to ground truth, I obtained a typical correlation coefficient of 0.86.