Session: 16-02-01: Poster Session: NSF Research Experience for Undergraduates (REU), NSF Posters

Paper Number: 100235

100235 - Rapid Creation of Metamaterials With Prescribed Mechanical Behavior 

Mechanical behavior of intrinsic bulk materials (including metals, ceramics, and polymers) is experimentally characterized by their stress-strain curves. For example, under a tensile load, the mechanical behavior of ceramics is represented by a linear-elastic region, followed by immediate fracture, represented by a sharp termination of the stress-strain curve; metals display a linear-elastic region, followed by yielding, strain hardening, necking, and fracture; elastomers display super-elasticity, characterized by a rapidly rising concave-up stress-strain curve without a noticeable linear-elastic region; plastics typically have a linear-elastic region, followed by yielding and a plateau stress and end with fracture. The mechanical responses of these bulk materials are dictated by their intrinsic crystal structure and atomic bonding in addition to microscopic defects. Manipulating chemistry and microstructure using thermo-mechanical processing is typically used to tailor the responses of these bulk materials. However, both these approaches have been exploited systematically leaving little room for further gains which are now incremental rather than step-like.

Recent advances in additive manufacturing (AM) allow the potential of tailoring material properties via designed three-dimensional (3D) micro-architectures. These materials, with their building blocks comprised of a network of designed micro-architectures, achieve properties that transcend their intrinsic counterparts, such as negative Poisson’s ratio, negative compressibility, ultralight, and ultrastiff, recoverable ceramics and metals, and multi-stabilities. When plotted on material selection charts, these materials achieve previously unattainable property pairs (density, Young’s modulus, strength, Poisson ratios). Architected materials that exhibit these properties are typically designed by forward design approaches, topology optimizations, and more recently machine learning (ML). While these methods are capable of searching designs that reach target peak property values, they have yet to be able to accurately capture and recreate the full mechanical behavior in manufactured products due to challenges in simultaneously dealing with a large number of variables within the entire history of loading/unloading, including its stress-strain evolutions, nonlinearity, negative stiffness, recoverability, strain hardening, and energy absorption. These design approaches are further complicated by manufacturing defects, process variabilities, and uncertainties, which require substantial calibrations to account for defects in additively manufactured samples with hundreds to millions of spatial strut members. The tested mechanical properties of fabricated samples often substantially deviate from designed extremal properties, which, if not considered, could lead to sub-optimal or catastrophic failure during application experimentation. 

Here, we present a rapid inverse design methodology for fully tailorable mechanical behavior by combining ML and AM while incorporating a given printing process and a range of polymeric base materials. The input data of our ML approach consists of (i) a target stress-strain curve described by its geometric features characterizing full mechanical fingerprints of the material {X} and (ii) a given printing process described by two printing parameters (i.e., the maximum build volume dimension and minimum printable feature size of a 3D printer). This input data is fed into our ML approach employing a sequential integrated prediction strategy, which outputs a digital 3D model that, once printed, will replicate the target stress-strain curve. To achieve this, we have developed a family of architectural genes with cubic symmetry, covering a wide range of mechanical behavior mimicking nearly all classic and tailorable behavior via distinct curve shapes. These genes serve as building blocks for creating a training dataset, from which our ML approach learns the relationship between mechanical behavior, architectural design, and process-dependent manufacturing errors and the 3D digital model replicating target stress-strain curve. We demonstrate the inverse design of arbitrary stress-strain curves that capture elasticity, nonlinearity, energy absorptions, and multiple peaks and valleys. These behaviors may easily be tailored via graphically modifying local geometric features of a stress-strain curve. In contrast to forward design approaches and topology optimizations, the reported methodology allows for rapid creation of materials with fully tailorable mechanical behavior while automatically learning manufacturing process errors and nonlinear behavior.

Presenting Author: Desheng Yao University of California, Los Angeles

Presenting Author Biography: Desheng Yao is a Ph.D. student in the CEE department at UCLA, working as a research assistant at the Additive Manufacturing and Metamaterials (AMML) led by Prof. Xiaoyu (Rayne) Zheng.

Authors:

Desheng Yao University of California, Los Angeles
Chansoo Ha University of California, Los Angeles