Advanced materials manufacturing suffers from several challenges including inherent stochasticity in the manufacturing process. Inherent stochasticity means that even with the same inputs and processes, the resulting material outputs follow a large distribution of performance and safety standards. Large distributions of material outputs mean that significant amounts of low-to-mid performing material are scrapped leading to millions of dollars in waste and overproduction.
Up until now, there have been only a few characterization tools that provide rich and meaningful insights in real-time for advanced materials manufacturing. Beta gauges are the industry standard for inline characterization, but they are limited to low coverage (less than 1%) and primarily one measurement, thickness. Vision systems are quickly advancing to uncover a limited set of surface defects, but are limited to what is optically visible, i.e., they cannot characterize sub-surface conditions, delamination, and/or micro- and nano-scale defects. Terahertz technology poses a multivariable option but is limited in its penetration depth and large-scale adoption.
To address some of these challenges that plague existing technologies, physics-rich, multimodal data combined with data-driven modeling will be a crucial component in next-generation materials characterization. Cold atmospheric plasma provides multi-modal excitation that generates information-dense data with several thousands of features. Machine learning provides a means to fill the data processing gap between these high-dimensional raw data features to interpretable metrics of interest, including a variety of material properties. We demonstrate this combination of plasma-generated data with machine learning-based processing to provide a flexible, real-time solution to next-generation metrology.