MAE Grad Seminars
Mechanical and Aerospace Engineering Graduate Seminars
The Mechanical and Aerospace Engineering Graduate Seminar provides a means for MAE graduate students to share their work and gain practice giving technical presentations in a formal (yet friendly) setting. Talks are generally organized in decreasing order of seniority, so that junior students can learn from the example set by the senior students.
The graduate seminars are open to the public. Notifications on the graduate seminars will be sent on the MAE-Seminar-L distribution list. To join the list, please send an email to firstname.lastname@example.org
February 4, 2021
Speaker 1: Bashir Alnajar, PhD
Title: Least Squares Regression Principal Component Analysis: a supervised dimensionality reduction method
Abstract: The coupled level set and volume-of-fluid (CLSVOF) method is an efficient approach used to simulate multiphase flows in which fluids of different phases are separated by a complex, evolving interface. This method leverages the advantages of both the level set (LS) and volume-of-fluid (VOF) approaches by combining the strong mass conservation properties of the VOF method, while retaining the accurate interface representation of the LS method. In this work, the flow field is discretized by a single-field, finite difference formulation of the weakly compressible Navier-Stokes equations on a stationary grid. A coupled second-order operator split algorithm is used to advect the volume fraction and level set function, and the interface is reconstructed using the least-squares volume-of-fluid interface reconstruction algorithm (LVIRA). A numerical code has been developed for 2D and axisymmetric cases, and its performance has been validated through a series of test cases, such as the simulation of a gas bubble rising with large density ratios, on the order of 1:1000 and the oscillation of a spherical bubble in response to changes in ambient pressure.
Speaker 2: Jared Strutton
Title: Exploiting Cure Kinetics in HTPB-Based Thermoset Composites to Engineer Interfaces for Traction-Separation Control
Abstract: Additive manufacturing (AM) is an effective processing tool implemented in a wide variety of disciplines and can be used to fabricate complex geometries that are difficult to impossible to produce using traditional subtractive techniques. Particulate composite feedstock options are severely limited due to the high viscosity implications when melt-processing thermoplastic binders with solid particles. The solid filler is incorporated inside a thermoplastic filament and undergoes multiple melt-processing steps prior to FDM. The polymer acts as a transport medium for solid particle extrusion and functions as the matrix that suspends the particles together. Controlling the cure kinetics during extrusion of hydroxyl-terminated polybutadiene (HTPB) composites during additive manufacturing (AM) allows for tuning of material properties via interface engineering. Understanding how the adhesive and cohesive properties can be altered to change bulk material properties, as well as location-based disparities, is needed to design functionally graded materials. These materials can be constructed as single structures with varying properties or to have location-specific property variations. This can be accomplished by controlling the material composition and cure kinetics while the material is undergoing 3D printing. Mechanical testing coupled with digital image correlation (DIC) will tie bulk material properties to strain field evolution in layered (and subsequently AM) materials.
March 4, 2021
Speaker 1: Fathia Arifi, PhD
Title: Optimal Control of Encapsulated Microbubbles for Biomedicine
Abstract: Encapsulated microbubbles (EMBs) are developed as contrast agents for ultrasound imaging, as vehicles for intravenous drug and gene delivery. Ultrasound can excite nonspherical oscillations or shape modes, that can enhance the acoustic signature of an EMB and incite rupture, which promotes drug and gene delivery at targeted sites. Therefore, the ability to control shape modes can improve the efficacy of both the diagnosis and treatment mediated by EMBs. This work uses optimal control theory to determine the ultrasound input that maximizes a desired nonspherical EMB response while minimizing the total acoustic input to enhance patient safety and reduce unwanted side effects. The optimal control problem is applied to models of both a free gas bubble and an EMB that account for small amplitude shape deformations. These models are solved subject to a cost function that maximizes the incidence of rupture and minimizes the acoustic energy input. The optimal control problem is solved numerically through pseudospectral collocation methods using commercial optimization software. Single-frequency and broadband acoustic forcing schemes are explored and compared. The results show that the encapsulation greatly increases the acoustic effort required to incite rupture. Furthermore, the acoustic effort required to incite rupture depends both on the form of the acoustic forcing (single frequency vs. broadband) and on the shape mode that is forced to become unstable.
Title: The Scaling Group of the 1-D Invisicid Euler Equations
Abstract: The one dimensional (1-D) compressible Euler equations in non-ideal media support scale invariant solutions under a variety of initial conditions. Famous scale invariant solutions include the Noh, Sedov, Guderley, and collapsing cavity hydrodynamic test problems. We unify many classical scale invariant solutions under a single scaling group analysis. The scaling symmetry group generator provides a framework for determining all scale invariant solutions emitted by the 1-D Euler equations for arbitrary geometry, initial conditions, and equation of state. We approach the Euler equations from a geometric standpoint, and conduct scaling analyses for a broad class of materials.
March 18, 2021
Title: Multiple Metrics Indicate that Deep Learning-Based Bone Segmentation of CT Data Outperforms Other Methods
Abstract: Separating bone from background is a crucial step in quantifying bone architecture in computed tomography (CT) data. Different approaches lead to variations in bone segmentation results, which lead to issues with reproducibility and interpretation in bone analyses. Our objectives were to evaluate the performance of fully convolutional neural networks (FCNNs) compared to other automatic methods for segmenting human vertebral body and femoral neck data and to investigate the performance of FCNNs trained and tested on similar, dissimilar, and combined image data. Using a cross-validation approach, 2D U-Net FCNNs were trained with image data for the femoral neck, the vertebral body, and a combined bone set and tested on separate unseen data. Global and local variance-based thresholding methods were also used to segment all image data. Analyses of five performance metrics indicated either higher or equal segmentation performance for FCNNS compared to threshold-based methods. FCNNs trained with diverse data could provide a standardized segmentation approach, ensuring that bone analyses are reproducible across research protocols. Although the FCNN approach was applied to high-resolution CT scans in this study, this technique may also provide an efficient means of accurately segmenting lower resolution data. Performance gains in segmenting image data of multiple types using FCNNs outweighed the initial, but limited, time required to train the networks.
Title: Modeling Plasticity in Materials with Crystal and Network Plasticity
Abstract: There are several mechanisms that cause plasticity in metals, all of which involve atomic displacements to relieve stress. The crystalline structure of a metal has a direct impact on how the material behaves when undergoing plastic deformation. Atoms will displace in the direction of most densely packed planes. This creates a slip system consisting of slip planes and directions for a given crystal structure. Models which consider these slip systems are called crystal plasticity and it provides a framework to combine the mathematical structure of continuum mechanics and the structure of crystals. On a larger scale within metals, there are grain boundaries, which are regions that have different rotations of the crystal structure. Under stress these grains grow or shrink to accommodate stress plastically. A new model called network plasticity is developed to describe the aggregate effects of grain boundary motion with a continuum framework. This model uses a directed graph to connect grains with their interface motion described by energy minimization. Four cases are considered: two vertically stacked grains, three vertically stacked grains, three grains arranged for a triple point, and hundreds of grains arranged randomly in a material. All these cases show the effects of rotating grain boundary interfaces with shear driven or strain driven cyclic motion. By combining both plastic models, the relationship between a material’s microstructure and strength can be better understood.
March 25, 2021
Pau Saldana Baque
Title: Diffusion phenomena in the fused deposition modeling process of fluorinated thermoplastic binders
Abstract: Motivated by the lack of polymer welding information in the additive manufacturing (AM) field, this work studies the effect of diffusion phenomena in fused deposition modeling (FDM) of fluorinated thermoplastic binders. In particular, this study compares the anisotropic response of 3D printed binary blends of poly(vinylidene fluoride) (PVDF) and poly(methyl methacrylate) (PMMA) with the isotropic response of these blends fabricated via molding techniques. PVDF/PMMA filaments were produced by twin screw extrusion and, subsequently, injection-molded or 3D printed into dog-bone shapes. Specimen mechanical and thermal properties were evaluated by tensile testing and differential scanning calorimetry, respectively. Results show that higher PMMA concentration not only improved the processability and increased specimen fragility but prevented the crystallization. Blends with a higher PMMA concentration demonstrated a better mechanical response than those with a lower amount of PMMA. Similarly, injection molded samples reveled better mechanical properties compared to 3D printed specimens. The present study provides new data to improve the description of the effect of PMMA in processed PVDF/PMMA blends. This will be useful to optimize the mixture composition when producing energetic thermoplastics through FDM.
April 8, 2021
Title: A Preliminary Review on Hypersonic flow
Abstract: Hypersonic flow is one of the most challenging areas in the field of aerospace engineering. In this talk, a preliminary review will be given on the hypersonic flight environment, the aerothermodynamics, and the principal aerodynamic phenomena in hypersonic flight, such as viscous effects, high-temperature effects, flow instability, transition, and turbulence. Numerical simulations on hypersonic flow past over canonical bodies will also be discussed.
Title: Continuum phase field modeling to determine burn rates in AP/HTPB
Abstract: Ammonium perchlorate (AP)-based composite propellants have been a fuel of choice in the field of solid rocket propulsion for several decades. Simplicity, reliability, and stability during long-time storage make them favorable for use in tactical missions (such as space launches and missiles) over the higher performance and the more controllable liquid-propellant formulations. This type of propellant typically contains a multi-modal distribution of AP (NH4ClO4) grains (~20 to 200 mm) embedded in the hydroxyl-terminated polybutadiene (HTPB) matrix. The geometry of the solid model changes as the material burns. This variable geometry of the solid model, along with a wide range of AP particle sizes, renders solid phase modeling a challenging task. Solid phase computations are performed using adaptive mesh refinement (AMR) using the AMReX package (from Lawrence Berkeley Laboratory) increasing the efficiency of computation. In the simulation, burn behavior has been determined as a function of particle size, concentration, and morphology. The preliminary findings show that the simulation values closely match experimental values. This will serve as a design tool for optimizing composition and act as a predictive tool for additive manufacturing of AP/HTPB
April 22, 2021
Maycon Meier dos Santos
Title: Effects of Molecular Descriptors on Atomization Energy Modeling with Machine Learning
Abstract: Using Machine Learning to solve Quantum-Mechanics variables can be an efficient alternative to replace traditional methods where numeric accuracy is obtained at the cost of computational efficiency. Molecular descriptors are numeric representations of molecules that enable one to train a machine learning algorithm to predict different properties of molecules. Molecular descriptors can take different forms: some descriptors account for bond and ring structures of the molecules, some uses only the atomic positions of the atoms in the molecules. In this talk, we examine whether different molecular descriptors will lead to different machine learning outcomes. The descriptors we will compare in this talk are Morgan fingerprints, which uses an array of binary digits to encode bond structures in a molecule; and Coulomb matrix, which is matrix constructed based on the distances between the atoms in the molecule. The machine learning method we will use is Gaussian process regression, which is a method that relies on a stochastic perspective on regression. We find that for predicting the atomic energies of the molecule, the Coulomb matrix descriptor outperforms the Morgan fingerprints.
Title: Koopman Analysis And Control Of Nonlinear Bubble Dynamics
Abstract: Volume and shape oscillations of gas bubbles in liquids form a central area of study in multiphase fluid mechanics, with important applications to intravenous drug delivery, contrast-enhanced ultrasound imaging, and cavitation-induced flow instabilities and erosion in turbomachinery. In this study, we use emerging tools from Koopman operator theory to analyze the Rayleigh-Plesset equation governing spherical bubble oscillations. Koopman theory is a framework that provides a globally linear representation of even strongly nonlinear dynamical systems, and that can extract coherent spatio-temporal structures from data. Such a Koopman embedding allows for future state prediction and the application of classical control techniques, including optimal control. Through combination with data-driven and machine learning methods, a controller can be trained on both numerically simulated data and experimentally obtained time series. Here we use algorithms called Hankel-DMD (dynamic mode decomposition) and SINDy (sparse identification of nonlinear dynamics) to extract eigenfunctions of the Koopman operator for the RPE. These nonlinear functions then provide a basis, analogous to Fourier modes, for a linear embedding of the nonlinear dynamics. Fundamental frequencies and harmonics emerge naturally. Finally, these eigenfunctions are used as coordinates to develop optimal linear controllers.
May 6, 2021
Title: Nucleate void analysis using machine learning techniques.
Abstract: While microstructure represents the overall uniform description of a material, it is the imperfections or defects like grain boundaries, the weak links, which dictate some material properties, especially its propensity to fail. In the case of pure ductile metals, like tantalum, it is known that all grain boundaries are not equal in their propensity for either strengthening or weakening a material. However, the reason for this selectiveness is not well understood. To develop next generation predictive capabilities for dynamic failure in engineering materials and to design future damage tolerant materials, it is imperative to not just qualitatively understand the reasons behind why all grain boundaries are not equal but also to quantify this selectiveness in a manner that can be useful for both design and prediction. In this work, we focus on developing a qualitative and quantitative understanding of failure at grain boundaries after shock loading of high-purity tantalum plate samples. This study is developed from input data consist of a reconstructed grain boundary data file and grey scale image of the EBSD scan provided by Edax OIM Analysis software and treated through using machine learning techniques for the purpose of data analysis, process of inspecting, cleansing, transforming, and modeling data with the goal of discovering useful information.
Hector Pascual Herrero
Title: Least Squares Regression Principal Component Analysis: a supervised dimensionality reduction method
Abstract: Dimension reduction is an important technique in surrogate modeling and machine learning. In this talk, we present a supervised dimensionality reduction method, "least squares regression principal component analysis" (LSR-PCA), applicable to both classiffication and regression problems. To show the efficacy of this method, we present different examples in visualization, classiffication and regression problems, comparing it to several state-of-the-art dimensionality reduction methods. Finally, we present a kernel version of LSR-PCA for problems where the input are correlated non-linearly. The examples demonstrate that LSR-PCA can be a competitive dimensionality reduction method.