Prediction and validation of the transverse mechanical behavior of unidirectional composites considering interfacial debonding through convolutional neural networks

In this work, we propose a prediction model of the transverse mechanical behavior of unidirectional (UD) composites containing complex microstructure with the help of a convolutional neural network (CNN). For this prediction, a total of 900 representative volume elements (RVE) samples were generated by constructing 300 RVEs for each V-f of 40%, 50%, and 60% with the random sequential expansion (RSE) algorithm. The stress-strain (S-S) curves in terms of transverse elastic modulus, transverse tensile strength, and toughness considering interphase debonding were obtained by a finite element (FE) simulation with the RVE samples. After converting FE models with 900 RVE samples to corresponding microstructural binary images, CNN modeling was employed to construct a prediction model on the microstructural images. To demonstrate the performance of the proposed CNN model, we predicted the transverse mechanical behavior in terms of the S-S curves on various test datasets. Prediction accuracy was verified in terms of the loss functions and the error of the S-S curve. The prediction results were in excellent agreement with the test datasets, and the transverse mechanical behavior was quickly predicted for other microstructures. This confirmed that the proposed CNN model is simple and powerful and can efficiently clarify the relationship between the microstructure and transverse mechanical behavior of UD composites.