Computational Optimization of Material Properties via a Multi-Field-Assisted Additive Manufacturing Process
Open Access
- Author:
- Widdowson, Denise
- Graduate Program:
- Mechanical Engineering (PHD)
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 13, 2023
- Committee Members:
- Robert Francis Kunz, Professor in Charge/Director of Graduate Studies
Paris Vonlockette, Chair & Dissertation Advisor
Zoubeida Ounaies, Major Field Member
Allison Beese, Outside Unit & Field Member
Mary Frecker, Major Field Member - Keywords:
- Multifield Processing
Multiobjective Optimization
Machine Learning
AI
Magnetic Particle Composite
Computational Homogenization
Support Vector Regression
Magnetic Polymer Composites
Magnetoactive Elastomer
Inverse Materials Design Problem
Additive Manufacturing
Electromagnetic
Dielectric
Microarchitecture - Abstract:
- Functional additively manufactured parts necessitate variable material properties such as conduction or insulation, stiffness or compliance, or a range of any material response. This range of material responses can be achieved by using multiple materials, mixing materials, or as will be shown in this work, by controllably varying the properties of a particulate filled polymer matrix composite material via electromagnetic processing conditions. Past experimental and computational research has shown that, in this fashion, for a fixed pairing of matrix and filler, composite material properties can be tailored not only by changing the constituents’ volume fractions, but also by manipulating the microarchitecture that the filler materials self-organize into during the processing event. Organization occurs because applying external electric and magnetic fields causes particles to orient and align themselves as they react both to the external applied field and the stray fields of other nearby particles. These distinct distributions of particles affect the composite’s material properties in all domains including elastic, magnetic, and dielectric. Having this range of material properties available within a single print is crucial for the development of a universal 3-D printer. This work will establish the previously undetermined relationship between processing parameters and effective properties to optimize the constituent set for the largest range of possible properties, building a framework for materials design applications. Consequently, it is of particular interest to materials design to understand the relationship between the applied field and the resulting microarchitectures and thereby the resulting effective properties. This dissertation focuses on the key components to creating a framework that predicts those composite properties from processing parameters and constituent sets, and determines the processing parameters for a desired set of material properties (the inverse problem). This identification of required processing parameters will be done by investigating the established homogenization techniques and their limits, developing a computational homogenization scheme, and incorporating modeling methods into optimization and data science structures, all to investigate the processing – constituent design space. In this work, particle dynamics simulations will form the basis for computational modeling methods by determining the particle microarchitectures formed from coupled electric and magnetic processing fields. This simulation will create the representative volume element (RVE), which with proper boundary conditions and sufficient size, will yield bulk material properties (magnetic, elastic, and dielectric) via computational homogenization using finite element analysis (FEA). With an established pathway from processing parameters to composite properties, the system can be optimized for the maximum and minimum properties and find the widest range possible, for any given set of constituents. With data collected, from experiments, computational simulations, and analytical modeling, data science techniques including regressions can be trained to augment the homogenization, and to solve the inverse problem of determining processing parameters for a selected material property. This work shows the path from processing parameters to effective properties and trains a model to effectively predict properties for use with an optimization scheme. The multi-objective optimization scheme presented here finds the constituent set for which the combined range of elastic, magnetic, and dielectric properties is largest. Finally, the trained model is used to further explore the design space, generate Pareto fronts, and address the inverse problem by determining the processing parameters that would achieve desired properties. It was shown that the volume fraction determines the possible ranges and tradeoffs between properties. The relative strength and directions of externally applied processing fields where shown to be critical for structures formed and therefore also resulting effective properties. It was determined that for use with optimization or materials design applications data science techniques are crucial for predicting system behavior for the feasibility of exploring the design space. This work presents the developed framework that models structures, determines properties, predicts composite behavior within the design space, and optimizes the constituent sets.