Deep learning based nanoindentation method for evaluating mechanical properties of polymers

• Published in: International journal of mechanical sciences 2023-05, Vol.246, p.108162, Article 108162
• Main Authors: Park, Soowan, Marimuthu, Karuppasamy Pandian, Han, Giyeol, Lee, Hyungyil
• Format: Article
• Language: English
• Subjects: Deep neural network; Drucker-Prager model; FEA; Nanoindentation; Polymer
• ISSN: 0020-7403; 1879-2162
• DOI: 10.1016/j.ijmecsci.2023.108162

•A deep neural network (DNN) based nanoindentation method was proposed for polymers.•A material database was generated via finite element (FE) simulations to train DNN.•Hyperparameter tuning process was employed to derive optimum hyperparameters.•Material parameters of polymers were extracted from nanoindentation load-depth curves.•Trained DNN model was validated by performing nanoindentation tests on PC and PMMA. In this study, a deep learning based nanoindentation method is proposed to reduce the complexities in evaluating mechanical properties of polymers. To uniquely identify the material parameters, a set of nanoindentation simulations are performed by employing spherical and Berkovich tips. A database that represents the material behavior of polymers under nanoindentation is generated for a set of Drucker-Prager model parameters. A deep neural network (DNN) is trained based on optimized hyper-parameters identified through Bayesian hyperparameter tuning process. The performance of trained DNN model is experimentally validated by performing nanoindentation tests on PC and PMMA. From nanoindentation load-depth (P-h) data, the trained DNN model accurately predicts the material parameters, which are in good agreement with those in the literature. [Display omitted]