Simulation of current flowing through a nonlinear resistor network
❮
❯
Modal analysis of 4-story structure
The 4-story Aluminum structure pictured was modeled in CAD and a linear
eigenvalue modal analysis was performed to extract the natural
frequencies and mode shapes. Bayesian updating was performed to
minimize the error between the FEA-predicted frequencies and the
experimentally measured frequencies.
The novel aspect of this work was in the methodology for modeling the
bolted connections. Details can be found in the conference paper,
published in the proceedings of SEM IMAC XLI.
Mechanical hand CAD project
This CAD model was based on an artistic rendition of a mechanical hand.
The two bearings were imported from McMaster-Carr, but all other parts
were created in Solidworks with a combination of solid and surface
modeling. A design table was used to generate each of the finger
segments as configurations of the same model.
Nonlinear resistor network simulation software
The goal of this project was to create a neuromorphic circuit capable
of being trained as an artificial neural network. The chips were
comprised of Nickel nanostrands suspended in epoxy, and header pins
served as inputs and outputs. The strategy was to weaken undesirable
connections, leaving behind the desired input-output mapping.
My contribution to the project was software for simulating nonlinear
random resistor networks. The program populates a volume with
nanostrands (modeled optionally as points or as 1D fibers), creates
a connectivity graph based on the distances between neighboring fibers,
and solves Kirchoff's laws for the steady-state voltage at every node,
using the graph Laplacian matrix.
The program utilized PyTorch for GPU acceleration and was capable of
simulating tens of thousands of non-Ohmic resistors.