Optimizing Carbon Nanotube Composite Performance via Novel Multiscale Modeling and Machine Learning Methods

A new lightweight structural material would help achieve NASA’s goal of extending human presence to the Moon and Mars. It has been estimated that each pound of payload costs 300 pounds of propellant. Reducing the mass of structural elements will help by reducing the mass of the propellant. Alternatively, the mass savings can be used to increase safety margins or bring a larger payload. Carbon nanotubes (CNTs) have a truly unparalleled strength-to-weight ratio, that is the strength per kilogram of material. In order to get the same structural strength as steel, you only need about one one thousandth of the mass in CNTs, assuming you have a perfect material, wherein lies our problem. A material that fully leverages the immense strength of CNTs is yet to be created. Material development is a slow and expensive process, one often done through trial and error. My goal is to speed this process up by creating computational models of the material and finding out what their material properties are through simulation, then efficiently finding optimal, yet feasibly creatable composites. The first step is to create a model. CNT composites naturally create complex structures with features spanning a large range of length scales. For example, the diameter of a single CNT can be as small as a few carbon atoms wide, and the composite can have multiple layers that are on the order of 100,000 carbon atoms wide. This is why a multiscale model is needed. The multiscale model is broken down into several models, each at a different length scale. The smallest model simulates individual atoms, while the largest model captures the entire composite. This series of models is simulated starting with the smallest. Once the smallest model’s mechanical properties are quantified, they are used to drive the mechanical behavior of the next model, all the way up to the macroscale where the results will be compared to their experimental counterparts for validation. This model will then be used to perform simulated experiments on different combinations of properties. Although this is easier than actually creating the materials, this method will still run into limitations in the form of computation time. This is where machine learning is used to help. A machine learning model that predicts a composite’s mechanical properties will be created. Multiple models will be produced, and their accuracy will be evaluated by testing them on data that it was not trained on. Since a material’s processing parameters are related to its performance by altering the composite's physical features, the predictive capability of the machine learning model allows for the co-optimization of the composite’s processing parameters. This will not yield a super ideal, impossible to create, material. It will instead lead to a best guess of what processing parameters will create the best composite. If there is a discrepancy between experimental results and the multiscale or machine learning models, that data will be used to create even better models, which will, in turn, give a new best set of processing parameters. Iterating on this will eventually converge to an optimal material.

This is early/mid-stage (TRL 2) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.