A machine learning approach to modeling the effects of fiber shape and interphase on the thermoelastic properties of composites

This study proposes an innovative micromechanics-based deep neural network method to efficiently investigate the effects of fiber shape and interphase on the thermoelastic properties of unidirectional composites. Firstly, this work establishes a micromechanical finite element approach by simulating the internal microstructure of the composite and verifies its rationality by comparing it with experimental results. Subsequently, using DOE sampling method based on global arrangement, data groups for training are obtained through the finite element simulation, and the machine learning model is further constructed utilizing deep neural network algorithm. The effectiveness of the machine learning model is validated by comparing the true values from the finite element simulation with the predicted values from the machine learning. Finally, a comprehensive investigation is conducted to elucidate the effects of fiber concentration and morphology, interphase concentration and characteristics on the thermoelastic behavior of composites. The results show that the established machine learning model provides a fast and accurate prediction for the thermoelastic properties of composites considering microstructural features.