Software tool for in-space printing of reliable parts

R. Bhowmik, S. Jha
Polaron Analytics,
United States

Keywords: machine learning, additive manufacturing, data analytics, planform, images

Summary:

We are customizing our in-house cloud-based data analytics platform, MatVerse, for automatic inspection of the additive manufacturing (AM) process both in one and zero gravity conditions to enhance reliability. The MatVerse platform utilizes machine learning algorithms to analyze structured datasets and provide helpful information in the form of plots and graphs. The users also have the flexibility to download their datasets, calculations, and all the learned models. We are in the process of incorporating a module named "CLADMA" (short form for "cloud additive manufacturing") in the MatVerse platform for analyzing both the structured and unstructured datasets from AM printers. The module will include all the learned models developed for printing high-quality parts. "CLADMA" will be capable of automatically differentiating flaw-free components from defective components using machine learning methods. We aim to make MatVerse extremely user-friendly such that users with little or almost no experience in data extraction, cleaning, and analysis can easily use it. We recently developed a preliminary deep neural network model using in-situ thermal images generated during printing of coupons on a Fabrisonic printer. The initial model can classify images to differentiate between defect-free and defect-containing parts. We are in the process of capturing more datasets to increase our model's accuracy. For developing a model for the in-space AM process, we are generating datasets in vacuum conditions, replicating a partial space environment. Furthermore, we are conducting initial conversations with various launch and satellite providers to successfully place a miniature Fabrisonic machine in an on-orbit (in-space) condition with the goal to generate additional datasets for optimizing the learned models for printing reliable parts in zero gravity conditions. The final deep learning model will be beneficial for printing defect-free components for satellites or spacecrafts. We will implement the learned model in the "CLADMA" module for easy use. The on-orbit printer with the MatVerse platform will also be helpful for on-demand printing of nano- and micro-satellites.