Explore

Thesis Research

My research is sponsored by the US Space Force under an SSTI Grant focused on In-Space Servicing and Assembly. My work is related to building an agnostic manufacturing platform to allow for the disassembly, inspection, and reassembly of electronic components for PCB repair applications.

Year One focus was on conceptual design and prototyping of an end-to-end servicing “lab” for in-space operation. For this, I constructed a physical platform from 80/20 with a UR5e robot mounted at the center, paired with a simulation running Gazebo built with ROS2 for future pick-and-place tasks.

Year 2 (2025-2026), my professor decided we would be moving in a different direction, so this work has been back-burnered for a full-time engineer to pick up. My new focus is on developing a machine-learning tool for a unique in-situ laser-based additive manufacturing machine for improving print reliability, reproducibility, and repeatability.

I tied this work to a class project, so you can find the full paper report here:

For brief understanding, here is the abstract:

The process for parameter discovery on a NanoPrintek 3D Printer to find a ‘recipe’ that prints silver onto FR4 is an extremely high-dimensional, low-efficiency, and high-cost process if following traditional methods, such as one-factor-at-a-time. To address this, I’ve designed, implemented, and worked towards the validation of a “product prototype” to solve this problem for a single output: bulk resis- tance. The prototype is a closed-loop optimization system designed to guide a manufacturing engineer towards optimal print parameters in the least number of trials possible. The system is architected around a Bayesian optimization frame- work, using a Support Vector Machine and Gaussian Process Regression (GPR) surrogate model ’team’. The program attempts to model the complex relationship between seven input print parameters, then requests a trial to refine its understand- ing of the underlying process or exploit promising regions of the parameter space, ultimately guiding an engineer towards identifying a set of parameters that mini- mizes trace resistance. Validation of this model showed promising results, but was unfortunately cut much shorter than desired due to machine malfunctions. Within the additional successful trials, the model discovered a few recipes that landed in the ”excellent” category, reducing resistance by 53%, however, due to the reduced dataset, the model fidelity error remained high at 23%, falling into the ”okay” performance band.

Want to print your doc?
This is not the way.

Try clicking the ··· in the right corner or using a keyboard shortcut (

CtrlP

) instead.