This project demonstrates how we can reduce CO2 emissions from maritime freight traffic by leveraging a hub-and-spoke model to aggregate vessel cargo. An array of shipping network models were created via a clustering algorithm varying the clustering hyperparameter. The analysis demonstrated an optimal shipping network is possible and could theoretically reduce CO2 output by 38%.
Dashboard  | Video Walkthrough  | Full Report  | Repository

These visualiations demonstrate the flexibility of D3 for creating static and interactive visualizations.
Repository

Here is a study meant to diagnose the convergence robustness of a given FEA model. The dataset includes multiple runs of a single model with different loads and time steps, and results of the solution's convergence. Using Support Vector Machine (SVM), this study demsontrates that convergence can be predicted before time is spent on a simulation.
Repository

Part of team of MIT aerospace engineers that participated in Red Bull Flugtag. Designed and built a 30-foot wingspan human carrying glider.
Video

sEMG signal data is used to predict the type of grabbing motion of a subject. This study evaluates many state-of-the-art machine learning algorithms as well as data processsing and transformation techniques. An maximum accuracy of around 75% was achieved, which is high considering the number of classification possibilites.
Data  |  Report  | Repository

Every year, thousands of scientific papers are peer reviewed and published in verified academic journals focused on a wide range of different industries. This python notebook and script automates the process of analyzing an entire journal of academic papers by web scraping and word processing to gain insight to an industry.
ScienceDirect  |  Report  | Repository

Finite element analysis (FEA) has greatly assisted engineers in innovating across all industries. However, there is a computational ceiling that results from models containing too many elements. Deep learning may be a solution to reduce the run time and memory of large models. In this study, four types of graph convolution networks (GCNs) are trained and tested on a set of FEA results.
GCN Review  |  Report  | Repository