<p>We are employing machine learning techniques to predict crack growth path during brittle fracture in polycrystalline&nbsp;graphene&nbsp;sheet. Computational modeling of crack propagation of graphene requires&nbsp;expensive atomic&nbsp;level modeling techniques such as molecular dynamic simulations. The high cost of such techniques hinder rapid prediction of material behavior. Machine learning models, on the other hand, are able to use the data sets generated using traditional computational and experimental techniques to make predictions about the material behavior at a low computational cost. Due to their capabilities in accelerating the prediction of material and system behavior, machine learning techniques have found applications in a wide range of industries. Recently, researchers are employing them to predict material and fracture properties. &nbsp;We employ machine learning models for rapid prediction of the fracture process in polycrystalline&nbsp;graphene&nbsp;sheet.&nbsp;&nbsp;</p><p>The polycrystalline&nbsp;graphene&nbsp;sheet considered is of dimensions 20 x&nbsp;40 nm with an edge crack of 0.7 x&nbsp;4 nm. We have conducted simulations on 700 polycyrstalline sheets composed of randomly oriented grains. The grain sizes considered are in the range of 3 – 9 nm. We have studied the crack growth mechanism and path by applying tensile loading to the polycrystalline sheet. The molecular dynamic simulations are conducted using&nbsp;LAMMPS&nbsp;molecular dynamics package. Our simulation results show that polycrystalline&nbsp;graphene&nbsp;sheet fails through the brittle crack propagation. Careful observation of crack propagation path shows that the crack path is dependent on&nbsp;the orientation of&nbsp;the grain and the grain boundary atomic structure. We build machine learning model that can learn features like grain orientation and the grain boundary and predict crack propagation path. We are employing a machine learning architecture&nbsp;which is&nbsp;a combination of&nbsp;convolutional&nbsp;neural network (CNN), bi-directional recurrent neural network (Bi-RNN), and fully-connected neural network. One of the challenging tasks is to build the training set to train the machine learning model. Our training set&nbsp;is composed&nbsp;of images of dimension 128 x&nbsp;256 pixels. Input image is the initial configuration of polycrystalline&nbsp;graphene&nbsp;sheet with edge crack. Each grain in the image&nbsp;is colored&nbsp;based on its orientation. The corresponding output image is the one which shows the&nbsp;fully grown&nbsp;crack in the&nbsp;graphene&nbsp;sheet. We have &nbsp;prepared 700 sets of input and output images from the simulation data. Data is divided into training, validation, and test sets. Training data consists of 560, validation data consists of 70, and test data consists of 70 sets of images. We use training and validation sets to train the machine learning model. Our machine learning model can predict the crack propagation path in the polycrystalline&nbsp;graphene&nbsp;sheet. Crack path predicted by the machine learning model is in a good agreement with the molecular dynamics simulations and performs well on training, validation, and test sets.</p>

Applied Machine Learning Method to Predict Crack Propagation Path in Polycrystalline Graphene Sheet

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.