Application of machine learning in simulation of nonlinear behavior of hydrogel

• Main Author: Xin, Qianying
• Other Authors: Li Hua
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/158446
• Institution: Nanyang Technological University

Machine Learning (ML) increasingly become a popular technique to model and simulate the mechanical properties of solid materials in the present days. It works by learning hidden patterns of big data and mapping the nonlinear correlations between data sets. Previous most of works were done on ML based modelling of the mechanical behaviours of hard materials such as metals and alloys where ML showed merits in increasing the accuracy of modelling nonlinear mechanical behaviours and shorting the corresponding simulation time. However, there are less works on ML based modelling of soft material. The modelling of soft materials has stronger nonlinear characters when comparing with traditional metal and alloy materials where ML technique may be applied in. Specifically, hydrogels are a typical nonlinear soft solid material which are widely applied in soft robotics and medical devices. Modelling the nonlinear constitutive behaviour of hydrogels accurately is still considered as a challenge. The aim of this project is to applying ML techniques in modelling the non-linear behaviour of hydrogels such as swelling and tension behaviours. Firstly, we reviewed literature on experiments, simulation models of hydrogels swelling, tension and several commonly used ML approaches and their applications in simulating of solid materials. Secondly, we collect parts of training data for ML by reviewing literatures and generate more data by carrying on hydrogel simulations. To optimise the ML, 40000 datapoints were generated with the use of the simulation software, Abaqus CAE. Thirdly, we apply several ML methods such as Support Vector Machine (SVM), Ridge Regression, Decision Tree Learning, K-Mean Neighbour (KNN) and deep learning model Artificial Neural Network (ANN) to learn the training data to find a method with the highest prediction accuracy. In the end, we compare the ML model with the experiment and simulation results to prove the validity of our ML model. This project has proved that ML techniques can be effectively applied to model the nonlinear behaviour of hydrogels. Models such as KNN and Decision Tree Learning can be implemented successfully to achieve high goodness of fit. Moreover, ANN is most recommended as it is the model that is able to achieve among the highest score in both univariate and multi-variate study. Hence ANN has demonstrated its consistency and reliability in this study, and it is the best model to be used in the prediction of hydrogel.