☕️Cary Randazzo
👋 Hi! I’m Cary, a Simulation and AI Systems Engineer
Check out my resumé at the link and portfolio below 😎
3D Realtime Heat Conduction Simulation using the Finite Element Method in Unreal Engine
During my recent employment as a Simulation and AI Systems Engineer, I was tasked with developing a hyper realistic and performant heat conduction system. The system would be utilized in a VR environment developed on and hosted by the Unreal Engine. It is a 3D realtime solution of the Fourier Heat Conduction system and acts as Phase I of a three phase plan I had developed for achieving an optimally realistic and performant Welding simulation system for my employer. Prior to this, a prototype was created in Unreal Engine demonstrating heat conduction using the Finite Element Method. This project provides images, videos, and other shareable information, project files, and code. It is provided for demonstration.
Adaptive Mesh Refinement for Mesh Based Methods
There are many simulation algorithms that utilize mesh systems, this project focuses on a heavily researched technique that greatly improves both accuracy in refined regions and performance of mesh dependent calculations everywhere in the simulation. This project utilized the C programming language and could be adapted to various dimensions and objectives for simulation - it functions here as a demonstration.
FDM vs PINN
During the course of my simulation work, I had encountered both Finite Differencing Method(FDM) and their how accurate they tend to be and newer AI based methods particularly so-called Physics Informed Neural Networks(PINNs). I had reviewed research on PINNs briefly and had believed them to be inferior to mesh based methods, however I did not want to discount there potential usefullness. In this project, I tasked myself with determining the limitations of PINNs as as simulation method when compared to the FDM in the context of a relatively well known heat conduction system where the ground truth information for the comparison were the time of execution and the precise analytic solution of the system under specific conditions. The results turned out to be enlightning and gratifying and I intend to explore more complex tests in the future.
Simulating Lightning in the Atmosphere
During my time as a graduate student, my EM professor who is something of an expert in lightning tasked me with studying Lightning Pulses in the atmosphere. This project takes you through what that project was like including the historical data that laid the foundation and the Finite Difference Time Domain method that was studied for the followup software that the professor would later achieve after the results of this research concluded.
ML & SW Solutions for DQ Assessment in CERN’s ATLAS Experiment
Anomaly Detection for Data Quality Assessment in CERN’s ATLAS Experiment using Tensorflow, SQLite databases, synthetic datasets, ML Algorithms, Dash by Plotly, Python, pandas, etc.
ONNX based ML Integration System for CERN’s ATLAS Experiment
In CERN’s ATLAS Experiment, the Data Quality group has scientists working to constantly improve data quality recently using a host of Machine Learning techniques. This project, as part of my official qualification for authorship on ATLAS, tasks me with making a straightforward system to integrate new machine learning algorithms into the main framework ATLAS uses called ATHENA. It has been decided that the software which will facilitate this integration will be the ONNX framework. This project is showcased here.