By Danielle Fretwell

The Francis College of Engineering, Department of Mechanical Engineering, invites you to attend a doctoral dissertation defense by Weidi Wang on ``REDUCED ORDER MODELING AND DESIGN OF 3D PRINTABLE MECHANICAL METAMATERIALS’’.

Candidate Name: Weidi Wang

Degree: Doctoral

Defense Date: Friday March 31, 2023

Time: 9:00 AM to 11:00 AM Eastern time

Location: Dandeneau 220, north campus. The virtual defense will be hold via Zoom. Those interested in attending should contact the student weidi_wang@student.uml.edu and committee advisor Alireza_amirkhizi@uml.edu at least 24 hours prior to the defense to request access to the meeting.


Advisor: Alireza Amirkhizi, Associate Professor, Mechanical Engineering, UMass Lowell

Committee Members*

Christopher Hansen, Associate Professor, Mechanical Engineering, UMass Lowell

Scott Stapleton, Associate Professor, Mechanical Engineering, UMass Lowell

Thomas Plaisted, Materials Engineer, CCDC Army Research Laboratory,

Ankit Srivastava, Associate Professor, Department of Mechanical, Materials, and Aerospace Engineering,

Reza Abedi, Associate Professor, Department of Mechanical Aerospace and Biomedical Engineering, University of Tennessee Space Institute

Brief Abstract:

Mechanical metamaterials (MMs) are artificial media primarily composed of periodic micro-structures, designed for manipulating the propagation and distribution of stress and deformation.

This work focuses on the dynamic behavior of 3D printable MMs and utilizes reduced order models (ROMs) to investigate the exotic properties of MMs, reduce the computational effort, and further facilitate design optimizations. The ROM approach in this work is a discrete yet analytical representation of the continuum system, developed based on beam theories and optimization methods. With accurate reproduction of eigen-analysis results, the ROM effectively preserves the salient physics of the continuum unit cell in a parameterized fashion. Physical phenomena such as band gaps, level repulsion, and exceptional points are investigated through the discrete setups.

The ROM approach also allows for the generation of large-sized training data of graded MM arrays, with a significant saving in computational effort. Then, a generative inverse design method is developed using the variational autoencoder (VAE) network. The VAE approximates the design distribution through statistical learning and extracts the design-performance relationship from the ROM-generated data. This combined ROM-VAE approach, therefore, leads to efficient exploration in a vast design space. Given a desired functionality, the generative neural network can almost immediately provide the design configurations associated with the performance metrics of interest, significantly simplifying the design optimization process.

Furthermore, a vast collection of samples can be retrieved with consistently targeted performance and randomly varied latent parameters. The feature distribution of this generated set is then subjected to statistical analysis to reveal the underlying design principles. By determining the features with high probability, the primary design factors that contribute to the desired performance are identified.

Practical MM applications are exploited in this work, based on the physical understanding of metamaterial dynamics and the developed design tool. We propose a new impact mitigation approach using graded MM arrays. The studied designs, generated by the VAE, can steer and spread the loading pulse into transverse directions, leaving a minimal amount of energy to be directly transmitted. In addition, we also examine the application related to exceptional points and develop a new source localization method based on the angle-dependent properties of metamaterials, which is presented as an independent chapter.

In summary, the purpose of this work is to develop a versatile toolset for efficient modeling and design of 3D printable MMs.

The specific goals of this effort are:

- To develop a reduced order modeling method that can (a) provide fast computation of band structures and finite-sized array responses; (b) preserve the fundamental physics; (c) serve as an infrastructure for data-driven design optimization;

- To establish a deep learning design approach in corporation with the ROMs, allowing for on-demand retrieval of inverse designs and identification of design principles.

This research contributes to the modeling/designing of mechanical metamaterials by reducing the computational effort and providing an efficient method to design and exploit novel applications.

All interested students and faculty members are invited to attend the online defense via remote access.