As physical properties are a function of composition and processing, newly designed systems must be independently synthesized and characterized. This process can be swift or slow depending on the technique used, but when testing a host of distinct systems experimentally, there will often be a considerable time cost involved (Brinson et al., 2020). Thus, the use of continuum models or other mathematical representations which are quick to execute and easy to serialize becomes very appealing.
Existing studies have shown that the effective mechanical properties of composites can be extracted via FEA, although there is some model deviation for exceedingly small particles (Odegard et al., 2005). This method offers physically accurate predictions for microstructures of high complexity but incurs a large computational cost in doing so. We offer a method for reducing the cost of predicting the mechanical response of one class of microstructural materials, polymer nanocomposites (PNC), enabling fast search and prediction during materials discovery or optimization problems.
We constructed a response space of “ground truth” PNC performance by calculating the elastic modulus for PNCs of varying microstructural and constituent property values using 3D FEM. After training of a surrogate model on this set, we found significant time reduction when compared to FEM predictions while still maintaining high accuracy. Predictive strength of the new model was increased further via feature engineering.
Active Researcher on the Project:
Brinson, L.C., Deagen, M., Chen, W., McCusker, J., McGuinness, D.L., Schadler, L.S., Palmeri, M., Ghumman, U., Lin, A., Hu, B., 2020. Polymer Nanocomposite Data: Curation, Frameworks, Access, and Potential for Discovery and Design. ACS Macro Lett. 9, 1086–1094. https://doi.org/10.1021/acsmacrolett.0c00264
Odegard, G.M., Clancy, T.C., Gates, T.S., 2005. Modeling of the mechanical properties of nanoparticle/polymer composites. Polymer 46, 553–562. https://doi.org/10.1016/j.polymer.2004.11.022