Item Details

title

Skeletal Microstructure Inverse Design: Spatially Variable Anisotropy Mimicry

author

Hezrony, Benjamin Shimon

abstract

With an ever-growing population, the burden of implant revision is dually increasing. To promote synergy between device and bone, this Dissertation presents the development of a multi-scale structural optimization approach to inverse design. The fundamental research question is posed: How can a printable construct be designed which mimics spatially variable anisotropy?To address this question, a bottom-up two-level structural optimization approach is devised. Broadly, the process from scan to object is: (1) map anisotropic elasticity in bone (perform AI Super Resolution if needed or if Low Resolution data is had, CHAPTER 2); (2) sequentially design a printable macrostructure featuring neighboring microstructures with synergistic anisotropic elasticity (CHAPTERS 3 – 5); (3) fabricate the design (CHAPTER 5). The developed 3D Super Resolution Neural Network (SRNN) has significantly improved homogenized mechanical accuracy relative to combinations of 2D and 3D interpolation, SRNNs, and Generative Adversarial Networks (GANs). Between the 3D SRNN and 3D GAN, significantly improved relative performance is observed in many metrics, indicating the importance of a volumetric data-driven process for skeletal Super Resolution. The developed Inverse Homogenization (IH) algorithm employs a convolutional abstraction to the Level Set Method for translation to 3D. In validation, second order optimization problem approximation yielded a symmetric Mean Absolute Percent Error (sMAPE) of 5.70%, indicating an excellent fit of anisotropic elasticity using an alternative material. On studying the variability of sMAPE and other performance metrics with respect to microstructural specimen and design material elasticity, it is found that design material elasticity yielded nonlinear, yet predictable, trends in performance. Finally, the sequential compound design process was developed and implemented. The compressive stiffness for five of the same identically prepared surrogates was experimentally determined (CHAPTER 5). Given the mimicry which: (1) is inclusive of both trabecular and cortical bone in a single model; (2) occasionally has 4X the targets of the validated IH approach; and (3) having a minimum sMAPE of 25.07%, this is considered a successful first-in-field investigation of whole bone 3D spatially-variable-triclinic-elasticity targeted bottom-up two-level inverse design. The design approach of 3D triclinic continua with Level Set accuracy extends to many fields and multiphysics problems (i.e., design of extreme aerospace constructs). The general sequential compound IH process directly translates to top-down two-level structural optimization, further extending the reach of this work and implications on the field of structural optimization. The implications of this work extend beyond biomimicry, bottom-up structural optimization, and even multi-scale structural optimization. In conclusion, this Dissertation contributes to our evolving understanding of skeletal Super Resolution, volumetric Level Set precision structural optimization, and skeletal inverse design.

subject

Additive Manufacturing

Deep Learning

Inverse Homogenization

Skeletal Inverse Design

Structural Optimization

Super Resolution

contributor

Brown, Philip J. (advisor)

Weaver, Ashley A. (committee member)

Weis, Jared A. (committee member)

Lopes, Pedro C. F. (committee member)

Pauca, Victor P. (committee member)

Canfield, Robert A. (committee member)

date

2025-03-12T08:36:46Z (accessioned)

2025 (issued)

degree

Biomedical Engineering (discipline)

embargo

2030-01-13 (terms)

2030-01-13 (liftdate)

identifier

http://hdl.handle.net/10339/110319 (uri)

language

en (iso)

publisher

Wake Forest University

type

Dissertation