Machine learning-based predictions for hydrogel damage induced by cyclic swelling and deswelling (C)

• Main Author: Chen, Yimin
• Other Authors: Li Hua
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering
• Online Access: https://hdl.handle.net/10356/176897
• Institution: Nanyang Technological University

Hydrogels are a type of soft polymer that exhibits non-linear mechanical behaviors and are often used in various industries for their adaptability. However, the most challenging feature is their fracture phenomena which is unpredictable due to countless random crosslinked network structures. Hence, this research looks more into this soft material. Deep Learning (DL) is widely used in prediction by learning the underlying patterns within the data. Convolutional Neural Networks (CNN) are known to deal with images and videos. Hence, it is being selected as the main DL model for this project. While many studies have researched on crack propagation in hard materials, there is limited research done using DL models on soft materials such as hydrogels. This report will start with an introduction to hydrogels, followed by a literature review on the fracture mechanics of hydrogel and various CNN models proposed to predict the crack propagation path on different materials. The data that will be used in the model are images from the simulations using Matrix Laboratory (MATLAB) and Abaqus software with a total of 300 sets of data. The focus will be on developing and applying four modified CNN modelsto predict the behavior of the crack propagation path. Finally, the performance of the four modified CNNs will be compared and the output image from the Abaqus simulation will be used to validate the accuracy of the prediction. In this project, two modified CNNs show a better performance among the four models. This project aims to demonstrate that the neural network model is a potential approach to predicting a crack propagation path in the hydrogel accurately while reducing the high computational cost when the randomized molecule density networks of simulated hydrogel samples become more complex.