Predictions of macroscopic mechanical properties and microscopic cracks of unidirectional fibre-reinforced polymer composites using deep neural network (DNN).

• Development of two deep neuron network models for predicting mechanical behaviours of UD FRP composite laminae. • All the data for deep learning predictive models obtained from an experimentally validated RVE based Discrete Element Method (DEM) model. • Consideration of fibre random distribution and fibre diameter in the microstructure of laminae in the development of deep learning models. • Discussion on accurate predictions for macroscopic mechanical properties and microscopic cracks of composite laminae. Fibre-reinforced polymer (FRP) composites have been widely used in different engineering sectors due to their excellent physical and mechanical properties. Therefore, fast, convenient and accurate prediction tools for both macroscopic mechanical properties and failure of the composites are highly demanded by industry and interested by academia. In this study, two back-propagation deep neural network (DNN) models are developed. The first model is a regression model for predicting macroscopic transverse mechanical properties of FRP laminae, which is based on a data set generated by Discrete Element Method (DEM) simulations of 2000 Representative Volume Element (RVE) with 200 different sets of fibre volume fractions and fibre radii. The second model, which is a classification model based on the results of 1600 DEM simulations of RVEs with a fixed 45 % fibre volume fraction and 3.3 μ m fibre radius, is developed for predicting microscopic crack patterns of the FRP laminae. The results show that the two developed DNN models are able to predict both the macroscopic transverse mechanical properties and the microscopic cracks of the RVE accurately. [ABSTRACT FROM AUTHOR]