Quantitative Prediction of Fracture Toughness $(K_{{\rm I}c})$ of Polymer by Fractography Using Deep Neural Networks

TL;DR

This paper introduces a deep learning framework using transfer learning to estimate fracture toughness from 2D fracture surface images, enabling rapid, cost-effective analysis with small datasets in materials science.

Contribution

It presents a novel approach combining deep neural networks and transfer learning to predict fracture toughness directly from 2D images, overcoming data limitations.

Findings

Predicts $K_{Ic}$ within 1-5 MPa√m range

Estimation error standard deviation of ±0.37 MPa√m

Models can be built in a few hours

Abstract

Fracture surfaces provide various types of information about fracture. The fracture toughness $K_{{\rm I}c}$, which represents the resistance to fracture, can be estimated using the three-dimensional (3D) information of a fracture surface, i.e., its roughness. However, this is time-consuming and expensive to obtain the 3D information of a fracture surface; thus, it is desirable to estimate $K_{{\rm I}c}$ from a two-dimensional (2D) image, which can be easily obtained. In recent years, methods of estimating a 3D structure from its 2D image using deep learning have been rapidly developed. In this study, we propose a framework for fractography that directly estimates $K_{{\rm I}c}$ from a 2D fracture surface image using deep neural networks (DNNs). Typically, image recognition using a DNN requires a tremendous amount of image data, which is difficult to acquire for fractography owing to the high experimental cost. To compensate for the limited data, in this study, we used the transfer learning (TL) method, and constructed high-performance prediction models even with a small dataset by transferring machine learning models trained using other large datasets. We found that the regression model obtained using our proposed framework can predict $K_{{\rm I}c}$ in the range of approximately 1-5 [MPa$\sqrt{m}$] with a standard deviation of the estimation error of approximately $\pm$0.37 [MPa$\sqrt{m}$]. The present results demonstrate that the DNN trained with TL opens a new route for quantitative fractography by which parameters of fracture process can be estimated from a fracture surface even with a small dataset. The proposed framework also enables the building of regression models in a few hours. Therefore, our framework enables us to screen a large number of image datasets available in the field of materials science and find candidates that are worth expensive machine learning analysis.