Martha Baldwin

I am currently a Ph.D. Candidate at Carnegie-Mellon University researching the application of Machine Learning within Additive Manufacturing Design. I am a member of the Design Research Collective lab administrated by Dr. Chris McComb.

I specialize in leveraging Machine Learning techniques to enhance Design for Additive Manufacturing, specifically focusing on the creation of multi-lattice structures. The objective of my research is to extend the design possibilities for users in additive manufacturing by enabling the utilization of a broader range of unit lattice cells within a component.

My expertise lies in geometry augmentation, supported by a strong foundation in mechanical applications and a deep understanding of machine learning and data science. I bring a unique blend of analytical and technical skills, capable of implementing effective data-driven solutions and tools, where I bridge the gap between research and data engineering in the field of additive manufacturing.

I have developed a well-rounded skillset developing custom datasets, designing mechanically inclined architectures, and implementing existing architectures for unique applications. My experience is valuable for advancing both theoretical and applied aspects of AI-driven design.

Research

Multi-lattice Topology Optimization with Lattice Representation Learned by Generative Models

This research presents a multi-scale topology optimization framework for multi-lattice structures, leveraging neural networks to optimize both macroscale topology and mesoscale lattice heterogeneity. By embedding a generative lattice model and using Fourier Neural Operators, the framework enhances structural performance, achieving a higher stiffness-to-weight ratio under complex loading conditions compared to traditional methods.

Ultimately, this work proved that the generative design space developed by VAEs can be effectively utilized for optimization tasks.

Smooth Like Butter: Evaluating Multi-lattice Transitions in Property-Augmented Latent Spaces

This research delves into the latent space of variational autoencoders (VAEs) trained on synthetic unit cells, examining both geometric and physical properties. It also explores how hybrid geometry/property VAEs enhance transition regions in multi-lattice structures, driving innovation in additive manufacturing design.

Ultimately finding that embedding physical property data improves the embedding of unit cells, which will better serve optimization approaches.

Smoothing the Rough Edges: Evaluating Automatically Generated Multi-Lattice Transitions

In this work, I harness the generative capabilities of Variational Autoencoders and their corresponding latent spaces to facilitate the creation of seamless interpolations between unit lattice cells. This research further explores the relationship between the performance of interpolations in the latent space and the intrinsic properties of the latent space itself. Ultimately finding that distance in the latent space primarily controls the geometric smoothness of multi-lattice transition regions.

A Data-Driven Approach for Multi-Topology Lattice Transitions

In this work, I harness the generative capabilities of Variational Autoencoders and their corresponding latent spaces to facilitate the creation of seamless interpolations between unit lattice cells. This research explores the relationship between the performance of interpolations in the latent space and the intrinsic properties of the latent space itself.

As part of my investigation, I have developed a novel measure to quantify the smoothness of image interpolations. This measure serves as a crucial tool in evaluating and analyzing the quality and cohesiveness of the created interpolations.

Past Projects

2023 Computer Vision Project

This project utilized many of my skills including machine learning, computer vision, and data processing. The goal of the project was to transform information to a user that did not have a visual of an object using an object that could see an object. The motivation was to utilize drone video footage to feedback information to the user relative to their coordinates.

The project demonstrated that even advanced tasks can be achieved with affordable hardware given you have the expertise to understand how to address its limitations.

2023 Mentorship Pairing Algorithm

This project was completed as part of the Mentorship Committee in Carnegie Mellon's Mechanical Engineering Graduate Student Organization (MEGSO). The mentorship program matches junior graduate students with senior graduate students to foster relationships and help students adjust to graduate school.

The matching algorithm was designed to reduce the workload of the mentorship committee, which had to manually match 100s of pairs. The algorithm uses the responses of individuals to determine their compatibility. The output is a list of pairs that match the mass email format, which allows all the pairs to be automatically emailed with a custom email.

Interp Imaging - a Hugging Face Space by cmudrc

2023 Interactive Research GUI

This project aims to make cutting-edge research on Multi-lattice transition regions using Variational Autoencoders (VAEs) more accessible to everyone. With our user-friendly interface, you can effortlessly create your own transition regions and dive into the possibilities of applying this groundbreaking work to various fields. Explore, experiment, and expand your understanding of this innovative technology today!

2021 Designing and Deploying ML Systems Project

This project was completed for the Designing and Deploying AI/ML Systems course at Carnegie Mellon University. The goal of this project was to develop an application that could convert the color of items using a clear user interface. The prototype utilized Streamlit to develop a front-end user interface where the user could select the object to recolor. The backend of the code utilized machine learning to identify the object and perform the color changing operation.

2021 Mechatronics Project

This project was completed for the Mechatronics course at Central Michigan University. The goal of this project was to 3D print a scale model of a FANUC robotic arm. Click the image to view more information about this project!

Pictured: Dr. Brian DeJong

Central Michigan University Cardboard Boat Race

Each fall, mechanical engineering students have the option to compete in the cardboard boat competition. I competed in this race for 3 years in a row, and it was always a great way to start the year!