Program Overview
SDCPS develops AI-based Co-Designers that fuse model-based design with machine learning to accelerate CPS design from years to months. Vanderbilt's contribution focuses on the Design Oracle, Bayesian optimization, and deep learning surrogates for UUV/UAV design.
Under the DARPA SDCPS program, I developed AI-driven approaches for autonomous engineering design optimization. My work focused on integrating Bayesian optimization, deep learning surrogates, and active learning with CAD-CFD toolchains (FreeCAD + OpenFOAM) to dramatically accelerate the UUV design process.
The Vanderbilt ISIS team’s Design.R project (PI: Péter Völgyesi) built the AI Co-Designer and Design Oracle — a system where automated designers submit component graphs and receive full multi-domain performance evaluations. The ATHENS project provided the Design Oracle infrastructure used across UUV and UAV design targets.
All research code is open-sourced on GitHub.
Research Works
Sample-Efficient and Surrogate-Based Design Optimization of UUV Hulls
Anvil: AI, Sampling Techniques, and a Combined CAD-CFD Tool
Constrained Bayesian Optimization for Automatic UUV Hull Design
Search for Universal Minimum Drag Resistance UUV Hull Using CFD
Machine Learning Assisted Propeller Design
Fusion of ML with Numerical Simulation for Optimized Propeller Design
Deep Learning-based FEA Surrogate for Sub-sea Pressure Vessel
Data Efficient Surrogate Modeling: Ensemble-free Batch Mode Deep Active Learning
