Projects
Bipedal Humanoid Musician Robot
Created a humanoid robot with the ability to walk and dance. Components were 3D-printed using SolidWorks and assembled. Lewansoul LX16a motors were used, along with a raspberry pi 3b, power converter and control board. These components were put inside the torso, as seen to the right, where 8 motors were controlled. By motion planning these motors (4 in each leg), and utilizing python to implement the motion sequence, the robot was able to walk on flat surfaces. Afterwards, more parts were 3D printed to give the appearance of a musician, such as a guitar and a microphone.
Click on Image for Video
Click on Image for Video
Evolving Soft Robots
This project involved simulating and evolving soft robots using vpython (visual python). The robots were made of spring-mass systems, where masses were spheres connected by springs (helixes). By changing the spring constant k, and the spring's rest length as a function of 4 parameters, acceleration of the system was achieved. Soft robots were evolved by the magnitude of their velocity. At first, this involved only evolving the variable parameters a, b, c, and w in the equation Lo = a + b*sin(wt+c). Then, co-evolution between the spring parameters and the structure of the robot (positions and number of masses) was achieved. A video of one of these robots was captured, with its springs colored by the value of its spring constant.
Click on Image for Video
Automated Weaving Gantry
The project purpose was to automate the process of weaving a carbon fiber lattice, as done by a PhD candidate. Using Repetier-Host, a '.gcode' file was fed and executed. The '.gcode' file was made through a python script which lets the user input the commands for the gantry, allowing 3D movements, and to create a lattice pattern, as shown in this video. SolidWorks was used for the design of end-effector pieces (blue plate holding the latch hook) as well as other functional pieces (green endstops).
Click on Image for Video
Computer Vision & Deep Learning
Undertook several LinkedIn courses & Stanford's CS231N: Convolutional Neural Networks for Visual Recognition. Deployed TensorFlow training, validation, and testing models for predicting video game sales using a csv with 1000 entries. Visualized neural layers using TensorBoard (seen to the right) and exported models for production utilizing Google Cloud. Optimized memory cost by changing training epochs, layers nodes and learning rates through multiple runs. Developed a 3-layer CNN with SoftMax data loss and max pooling using the CIFAR10 image database. Generated labels and captions for 10,000 images and computed training-time loss.
Global Warming Parameters & Correlations Using Data Science
Led a group of 5 with the development of a python script which analyzed 6 different parameters and their correlations to global warming for 192 countries. The following parameters were used: forest coverage land area %, CO2 per person, dominant industry % of GDP, average income per person, and number of vehicles per 1000 people. A csv data sheet was imported for principal component analysis (PCA) using singular value decomposition (SVD), linear and polynomial regression. A loss function and gradient descent was performed on the number of vehicles parameter to gain further insight of the direct impact vehicles have on CO2 emissions. To the right is each country depicted by its land mass (size), population (y), population density (color), and CO2 emissions (x).
2-Part Plunger Syringe
The 2-part plunger syringe works similar to a pen-mechanism, where once the user presses the inner plunger all the way in, the needle hub rotates and retracts the needle. Once the needle is incapsulated, the syringe is no longer functional, preventing multiple people from using it. I specifically worked on the needle hub revisions (the mechanism which retracts the needle). I also worked on the trade study for the 3D printer resins used to print the prototype, the project website and GoFundMe, as well as scheduling team meetings and assignments.
Ballistics and Impact Testing
Maintained the structure and quality of lab apparatus and assisted in forming experimental specimens (metal plates). This involved sand blasting, curing, and polishing metal plates using various chemical compounds. Cooperated in a team of 5 to ensure specimens were properly administered in the wind tunnel chamber. Performed over 20 shock wave impact tests on pre-commercial adhesives, resetting the pressure diaphragms after each test. Compared analytical results to experimental by analyzing static and dynamic loads using MATLAB and a high speed camera.
Modeling & FEM Analysis of Scissors
This project involved modeling a pair of scissors, which bigger handle broke by cutting a piece of cardboard. 5 parts (bigger & smaller handle, 2 blades, and rivet/screw) were assembled together. 7 different case studies were performed with different constraints to replicate different loads and situations while using the scissors. Local (h-refinement) meshing was used to find stress convergence as a function of degrees of freedom (DOF). The study was used to showcase stress concentrations and compared to real life results, such as as the crack in the bigger handle. The blades had a material of Stainless Steel 316 and the handles a material of PS Plastic.
Air Crank (Engine) Model & CAM
Created a HDPE Air Engine in SolidWorks within a group of 6 and exported model to HSMWorks for the machining of parts. Applied 2D adaptive clearing, contours, pocketing, and drilling of wood stock to reach the desired shape of all parts. The model includes a vertically movable piston that is placed inside the hand pump which pushes air to the system when external force is applied to the piston. Utilized MATLAB to analyze the geometry and motion of the flywheel (#2 in the diagram) to demonstrate the physics behind the hand-powered engine. The real life machining of the parts using a CNC machine and assembly was disrupted by COVID-19.
