Using supervised machine learning methods to predict microfiber alignment and electrical conductivity of polymer matrix composite materials fabricated with ultrasound directed self-assembly and stereolithography

Abstract

Engineered polymer matrix composite materials with designer electrical properties are important for a myriad of engineering applications including flexible actuators and wearable sensors. We use stereolithography in combination with ultrasound directed self-assembly to align electrically conductive microfibers in a photopolymer matrix. We relate the fabrication process parameters to the resulting filler material alignment and corresponding electrical conductivity using supervised machine learning methods and quantify the prediction accuracy of data-driven models derived from different interpretable and non-interpretable algorithms. We determine that decision tree and artificial neural network algorithms result in data-driven models with R2 scores that are 79.8% and 83.2% higher, respectively, than a traditional multivariate regression analysis benchmark model in predicting the microfiber alignment. Similarly, random forest and artificial neural network algorithms result in data-driven models that predict composite material electrical conductivity 9.1% and 13.7% more accurately, respectively, than a logistic multivariate regression benchmark model. Relating the fabrication process parameters to the resulting electrical conductivity of the material is a crucial step towards fabricating polymer matrix composite materials with designer electrical properties for use in engineering applications.