3D printing technology has been widely used in many engineering areas due to its high-level design freedom and enhanced manufacturing capability and complexity. However, various imperfections exist in the printed parts such as wall thickness variations, which could degrade the final performance of the printed structure. The current quality control relies on the online monitoring and assessment, which can only detect the shape deviations of the printed part while fails to connect the deviated shapes to the final material performance during the manufacturing. To quantitatively determine material performance of the printed part, a multiscale modeling approach is needed as the imperfections usually exist at the subscale of the printed part. The major obstacle is the high computational efficiency requirement for the online multiscale performance prediction.In this paper, a novel physics-based data-driven multiscale modeling approach will be developed to enable an ultra-efficient yet accurate online performance evaluation. The hexagonal shape of the honeycomb cells will be printed by fused deposit modeling (FDM). In the offline stage, different profiles of the inner boundary of the cells with varied wall thickness will be created. The profiles will be used to create the digital twins of the unit cell of the printed honeycomb structure, and the digital twins will be meshed in the in-house code developed from Gmsh. The meshed digital twins will serve as structural genome (SG) to perform multiscale modeling using the mechanics of structure genome (MSG). Note that the SG only requires two-dimensional (2D) domain to compute the three-dimensional (3D) material properties instead of 3D domain in the commonly used representative volume element (RVE) models, and therefore, the MSG provides a better computational efficiency for generating simulation data. The MSG solid model will be performed to compute the elastic constants and initial failure strengths of the printed honeycomb materials with the imperfections. A convolutional neural networks (CNN) model will be trained with the inner profile images of the cells as inputs and elastic constants and failure strengths as outputs. This CNN model is developed based on the data from the physics-based model (i.e., the MSG model). In the online stage, the profile of the inner boundary of the cells will be captured by a point cloud from an online monitoring system. After processing the data, the image of the scanned inner boundary of the printed layer can be obtained. The image will be used as input in the developed CNN model to predict the elastic constants and initial failure strengths. The accuracy of the CNN model will be validated by the results predicted by the MSG solid model with the scanned data.The proposed physics-based data-driven multiscale modeling method will provide an ultra-efficient yet accurate approach to online performance evaluation during the manufacturing, which will enable process intervention to enhance the capability of fabricating good-quality products and reduce the cost associated with time, raw materials, and labor in the 3D printing.

Multiscale Online Performance Prediction of Additively Manufactured Honeycomb Materials by Convolutional Neural Networks and Mechanics of Structure Genome

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.