Predicting the Mechanical Behavior of Additively Manufactured Mechanical Metamaterials Using Point Cloud Representation Learning

Abstract

Abstract Additive manufacturing (AM) has been widely adopted to produce mechanical metamaterials for load bearing, energy absorption, and other applications in various industries such as aerospace, automotive, and healthcare. However, geometric imperfections largely exist in AM. Since the mechanical behavior of metamaterials is primarily dependent on their geometries, it is critical to evaluate how process-induced geometric imperfections affect the mechanical behavior of fabricated metamaterials. Most of the existing approaches for AM quality control concentrate on the detection of defects and are limited in their ability to assess defect-altered mechanical behavior of finished builds. Some studies leverage destructive tests or numerical methods for mechanical behavior assessment, which are costly and time-consuming, and impractical for high-throughput routine quality control. In this paper, a new machine learning framework is developed to predict the mechanical behavior of fabricated metamaterials based on their as-built geometries (represented as high-resolution point clouds). Specifically, the point cloud is first converted into an image profile, which preserves detailed geometric patterns. then, a deep neural network is constructed to encode salient features of the image profile and associate them with the load-deflection curve of the fabricated metamaterial. The effectiveness of the developed framework is experimentally validated through a case study with auxetic mechanical metamaterial. This work has great potential to be extended for in-process prediction of AM builds’ mechanical behavior based on layer-wise point cloud scanning.