End-to-end deep learning method to predict complete strain and stress tensors for complex hierarchical composite microstructures

•A deep learning approach predicts the complete strain and stress tensors from composite geometries.•The predictions of the model reach high accuracy and observe rigorous mechanical principles.•A small amount of data is used to train the model with fine-tuned data statistics.•Enhanced model is able to handle material microstructures with varying constituent ratios and hierarchical geometries.•A powerful tool is provided for design problems including multifunctionalities design and hierarchical structures optimization.