Prediction of elastomer properties using Convolutional Neural Networks (CNN)

To predict the elastomer properties from its microstructures, material models in different scales have been developed ranging from the molecular level to the macro level. However, the application of these models is limited by either the computational costs or their accuracy. An efficient but accurate modelling is still missing. Due to its promising performances in prediction, machine learning provides a suitable alternative to traditional methods. In this study, the tensile properties of elastomers are predicted from the expanded TEM images using the Convolutional Neural Networks (CNN). The CNN model shows good accuracy and efficiency in prediction.