UC Irvine

Magnetorheological elastomer (MRE) is a rubbery composite material filled with micron-sized ferromagnetic particles whose mechanical properties can be tailored by the application of external magnetic fields. Due to its magnetic and mechanical coupling effect, MRE is increasingly used in the field of engineering. Capturing the responses of MRE is essential for materials modeling and can be reached either by the physics-based finite element modeling or data-based artificial intelligence modeling. In this thesis, machine learning-based data-driven models are built to discover the structure-property linkages of MRE. The proposed method employs a pre-trained Convolutional Neural Network (CNN) and also an artificial neural network (ANN) to evaluate the critical features of the material microstructures that lead to precise predictions for the critical mechanical properties of MRE. It has been proven that these approaches can make compelling predictions while dramatically reduce the time needed for the calculation process. With low computation cost, the machine learning models also exhibit great potential in microstructure optimization.