137091 research outputs found
Sort by
Dense Interconnect and Power Delivery Technologies for 3-D Heterogeneous Integration Architectures
No full textHeterogeneous integration (HI) has emerged as a potential system-level scaling solution to meet the exponential growth in artificial intelligence (AI) applications. However, to fully leverage the advantages of HI systems, several critical challenges remain. In this work, we address two major challenges faced by emerging 3-D heterogeneous integration (3-D HI) architectures: (1) difficulty of scaling interconnect technologies and (2) the impact of 3-D stacking on the power delivery network (PDN). First, we propose and demonstrate the use of selective Cobalt atomic layer deposition (Co ALD) for high-density die-to-wafer (D2W) bonding on a 2-D lateral gap and 3-D gap testbed. Second, we demonstrate inverse hybrid bonding (IHB) on a 2.5-D and 3.5-D multi-chip testbed and measure the electrical performance and reliability to assess process feasibility. Finally, we present a MATLAB-based PDN framework for emerging 3-D heterogeneous integration architectures and perform a design space exploration for early predictive modeling and finding the limitations of emerging HI architectures. The framework is also used to design an integrated methodology for early PDN design space exploration for 3-D compute-in-memory (CIM) architectures to help bridge interconnect design and CIM inference accuracy for rapid prototyping
Design And Fabrication of MR-Safe Core Needle Breast Biopsy Robot
No full textBreast cancer is one of the most prevalent cancers affecting millions of women worldwide. Excisional biopsy, especially MRI-guided percutaneous core-needle biopsy, remains a crucial diagnostic tool for breast cancer. However, MRI-guided biopsies present significant challenges, including the restricted space within the scanner bore and the necessity for multiple imaging steps, all of which extend the procedure and heighten patient discomfort. Unlike ultrasound-guided biopsies that allow for real-time needle positioning, MRI-guided procedures also lack dynamic visual feedback, complicating needle insertion. Advances in medical robotics offer a promising solution to these challenges. Robots can maneuver within the constrained MRI environment, enabling biopsies to be performed with real-time image feedback, thereby enhancing precision and reducing procedure time. This robotic approach has the potential to greatly improve the accuracy and efficiency of MRI-guided breast biopsies, ultimately reducing procedure duration and minimizing patient discomfort.M.S.Biomedical Engineerin
Designing a Multimodal, Environmental Interface to Enhance Time-Based Urgency Perception of Takeover Requests and Understanding of Takeover Reasons in Semi-Autonomous Vehicles
No full textThe emergence of Semi-Autonomous Vehicles (SAVs), also known as Level 3 Autonomous Vehicles (AVs), creates a need for timely intervention from humans where the drivers need to take over the vehicle control when necessary. However, current SAVs are limited in providing evolving context or urgency level due to their static alerts. The driver may, therefore, misjudge the situation and lead to undesired outcomes: underestimation can result in a delay of action, whereas overestimation can cause panic and trigger erratic or reckless behavior. This paper investigates the design of a multimodal TOR interface to counter this problem by improving human drivers’ perception of time-based urgency and the understanding of the TOR reasons. Two different designs were established for comparison – the baseline design was designed based on current state-of-the-art TOR interfaces (i.e., Autopilot of Tesla, Pilot Assist of Volvo, Super Cruise of Cadillac, etc.), which include static visual cues (i.e., static icons) and auditory cues (i.e., acoustic alerts), whereas the Context-Aware Adaptive (CAA) design has more dynamic visualizations (i.e., animated icons and ambient lighting) and auditory signals (i.e., informative speech and acoustic alerts). Four relevant driving scenarios, later simulated in a Virtual Reality (VR) environment, were identified through a focus group study as the settings for the summative evaluations with 25 participants. Our results show that our CAA design significantly improved people’s time-based urgency perception and the understanding of the TOR alert reasons for some scenarios. This work adds to the literature on Human-Machine Interface (HMI) design for SAVs, specifically on designing a dynamic, multimodal, and environmental interface that improves vehicle-to-driver communication in TOR situations.M.S.Industrial Desig
Modeling and Control Methods for Adaptive Thermal and Energy Management for Bio-implants and Human-in-the-loop HVAC Operations
No full textThis dissertation proposes novel modeling and control methods for thermal and energy management applications, specifically focusing on Implantable Medical Devices (IMDs) and Heating, Ventilation, and Air Conditioning (HVAC) systems. Both applications consider energy consumption with additional constraints: IMDs aim to maximize power consumption while avoiding tissue damage from overheating, whereas HVAC applications focus on saving energy while maintaining occupants' thermal comfort.
This thesis proposes a novel online prediction algorithm to model the thermal dynamics of IMDs with multiple heat sources. Additionally, a Model Predictive Control (MPC) scheme is proposed to achieve real-time thermal and power management of neural IMDs. Furthermore, the dissertation develops a feature importance analysis method utilizing meta-learning to evaluate the importance of features used in modeling the thermal comfort of occupants in HVAC applications.Ph.D.Electrical and Computer Engineerin
High-Explosive Detonation-Driven Simulant Decomposition in Confined Environments
No full textThis work presents a numerical study focusing on thermal decomposition of a chemical simulant in an confined domain, subject to a high-explosive detonation. A two-step modeling framework is employed, in which a spherical free-field detonation is first simulated and subsequently interpolated into a three-dimensional confined cylindrical domain. The methodology is intended to reduce computational cost while preserving key flow features necessary for capturing early blast behavior and post-detonation wave reflections. Grid refinement and verification studies suggest that the framework is capable of reproducing primary and reflected shock interactions with reasonable accuracy, showing agreement with experimental overpressure data and the free-field Kingery–Bulmash semi-empirical relation.
The verified framework is then used to explore the decomposition behavior of a gas-phase simulant introduced into the confined chamber. Parametric studies examining both spatial placement and initial geometry indicate that proximity to the blast origin has a strong influence on decomposition efficiency, likely due to the intensity of thermal exposure during the initial milliseconds following detonation. Geometric effects, particularly differences in surface area-to-volume ratio, appear to affect thermal retention and product formation, especially in lower-energy regions of the domain. The formation of primary and secondary decomposition products is found to depend on both thermal history and activation energy thresholds
Adaptive Digital Twins: Continuous Subspace Learning for Dynamic Domains
No full textPresented at AIAA SCITECH 2026 forumDigital twins can produce unreliable predictions during deployment when their operating environment is uncertain and dynamic. Models are commonly validated for specific environments and conditions, so their training data may not fully represent the test data encountered in real-world scenarios, especially when their operating envelope is unknown or shifting over time. In these situations, extrapolation on out-of-distribution data significantly erodes model accuracy. An example of such a scenario is in predicting atmospheric flight dynamics of a reentry spacecraft, in which the vehicle is subject to unpredictable atmospheric conditions that introduce high uncertainty in flight behavior. To address this, we frame the problem within unsupervised domain adaptation and propose a method to maintain digital twin accuracy in uncertain, evolving environments. We treat the subspace of the test data encountered during deployment as a non-stationary, continuously evolving domain, and adapt a model from training to test data by finding a time-evolving latent space representation of the data. We then evaluate the effectiveness of our method on a synthetic example of a spacecraft digital twin during an aerobraking campaign, where the digital twin must accurately predict thermal loading even when encountering out-of-distribution, extrapolatory atmospheric data
Proprioceptive State Estimation for Legged Robots with Probabilistic and Hybrid Kinodynamics
No full textLegged robots hold immense promise for robot navigation within both human-centric and unstructured environments, due to their ability to perform dynamic motions, giving them high potential for widespread use and a myriad of applications.
To be effective though, legged robot controllers require estimates of the robot's state at significantly high frequencies in order to plan and execute dynamic motions.
Exteroceptive sensors, such as cameras and LiDAR, are unable to provide measurements at these desired rates, thus becoming a limiting factor.
On the other hand, proprioceptive sensors, such as the Inertial Measurement Unit (IMU), are capable of providing measurements at immense frequencies and are not affected by changing environmental factors.
However, low cost versions of these sensors are encumbered with significant and varied types of noise, yielding poor results when used directly.
In this thesis, we propose algorithms and techniques for improving state estimation of legged robots using only proprioceptive sensing.
We leverage and improve upon existing works on legged robot state estimation and probabilistic modeling to tackle various sources of noise and inaccuracies, improving the accuracy and robustness of the state estimates.
Our first contribution is a novel factor graph model for performing smoothing-based Maximum A Posteriori estimation, which takes advantage of the legged robot form factor.
Building upon IMU preintegration theory and the use of environmental contact during leg stance phases for constraining sensor noise, we probabilistically model the kinematic chain of each leg within the robot, which compensates for nonlinear deformations and leads to improved accuracy.
Furthermore, we show how leveraging M-estimators allows for automatic slip rejection, providing our estimator with greater robustness.
We then introduce a new group-theoretic metric for measuring the performance of state estimators. Our metric combines the evaluation of the pose and the unobservable linear velocity of the robot into a singular form, providing a unified way to measure state estimator performance. This contrasts with existing metrics which measure the accuracy of the pose and the velocity estimates individually, thus leading to potential trade-offs and, consequently, subpar results.
We further demonstrate the use of Chebyshev polynomial-based differentiation-via-interpolation to compute the true velocity from ground truth pose values, alleviating the need for expensive systems to measure the ground truth linear velocity.
Our third contribution is the development of a hybrid factor graph framework for modeling discrete and continuous estimation problems simultaneously, and a novel variable elimination algorithm for converting the hybrid factor graph into a hybrid Bayesian network.
The proposed elimination algorithm results in exact posterior probabilities, overcoming the drawbacks of existing approximation based approaches.
We further propose novel methods for constraining the computational and runtime complexities, making large scale deployment of our framework viable.
This is demonstrated through the development of a hybrid legged robot state estimator which is capable of simultaneously estimating continuous robot states and discrete leg contact events using only kinematic information and measurements.
We present results on both simulated robots and real-world hardware, showcasing the efficacy of our hybrid factor graph framework
Hydrogen Embrittlement in Precipitation-Hardenable Nickel-Based Alloys
No full textHydrogen embrittlement (HE) in precipitation-hardenable (PH) nickel-based alloys poses critical risks for energy infrastructure, including subsea systems, petrochemical processing, and hydrogen production technologies. This thesis quantifies the interplay between hydrogen transport, microstructural trapping, and mechanical degradation across representative PH Ni alloys (UNS N07725, N09925, N09945, N09946, N09955) using coordinated Devanathan–Stachurski electrochemical permeation (ASTM G148) and in-situ slow-strain-rate tensile (SSRT) testing under cathodic charging, supported by scanning electron microscopy fractography and XRF analysis.
A clear HE-susceptibility ranking was established (725 > 945X/945 > 925), consistently supported by ductility-loss ratios (epR), brittle-area fractions, and hydrogen-transport metrics. HE susceptibility did not correlate with any single transport parameter (e.g., total hydrogen inventory or lattice diffusivity) in isolation. The most robust, mechanistically grounded predictor of ductility was the composite mobility-to-storage ratio, Ddecay/Cobs, which showed a strong correlation with epR (r=0.96,). Furthermore, in Alloy 945X–140, susceptibility increased from 25–55 °C, attributed to thermal activation of reversible traps that increased the trap bias (TB) ~5.8× and facilitated trap-assisted hydrogen transport.
Overall, HE susceptibility is governed by the mode of hydrogen transport, controlled by the accessible reversible-trap population. Alloys with high reversible-trap accessibility (e.g., 725) exhibit strong kinetic asymmetry (ΔD) and high HE susceptibility. In contrast, HE-resistant alloys (e.g., 945) mitigate embrittlement by minimizing reversible-trap participation, enforcing a slower, lattice-dominated diffusion mode
Input-State Estimation of Inelastic Structural Systems: Theoretical Framework and Experimental Validation
No full textSystem identification through online estimation algorithms allows for a comprehensive understanding, prediction, and assessment of the intricate behaviors exhibited by complex in-situ systems in a variety of applications. This model-based technique leverages the system's noisy output data and integrates existing mathematical physics-based models to infer the unknown inputs or unobservable dynamic states. The adoption of these online techniques in real-world settings has gained momentum over the past several years, owing to their ease of implementation, the robustness and reliability of the results, and the continuous advancements in sensing technologies. Furthermore, when structural systems are loaded beyond their elastic limit, they exhibit inelastic behavior which necessitates different methods to accommodate this complex nonlinear phenomenon. This dissertation contributes to this research area by developing a robust framework that integrates hysteretic models with nonlinear stochastic filtering methods to quantify the input characteristics of systems exhibiting inelastic behavior due to material plasticity.
Towards this goal, an input-state estimator for linear systems is first established. The estimator is designed to reduce the dependency on heuristically chosen input statistics by incorporating an online input covariance updating routine. Numerical and experimental validation is conducted, the results of which highlighted the robustness of the estimator in successfully tracking the input and state time history for various initialized input statistics. The estimator is then extended to nonlinear systems using an Extended Kalman framework. To efficiently model the system dynamics in the presence of hysteresis or plastic deformation, two modeling approaches of the continuum system are explored: an equivalent single degree of freedom formulation combined with a uniaxial Bouc-Wen model and a planar multiaxial hysteretic beam model. A comprehensive numerical validation of the proposed framework is conducted to gain insight into the performance of the inelastic models and the estimation algorithm. The results underscored the effectiveness of the proposed estimator and integrated models in successfully characterizing the input and states in the presence of nonlinearities in the system. Finally, the framework is experimentally validated using data collected from a beam subjected to an impact at its midspan. The novel estimator, along with the integrated models, adequately tracked the impulsive load. As such, these efforts represent an important contribution to the experimental validation of joint input-state estimation methods for inelastic continuum structures subjected to high-rate dynamic inputs.Ph.D.Civil Engineerin
Human-Centered Specification and Explanation for Mixed-Initiative Interactions
No full textAs Artificial Intelligence systems become commonplace in society, we require bidirectional, human-centered, mechanisms of communication.
Particularly in a mixed-initiative setting, wherein humans and AI systems are collaborating towards a shared goal, humans should be able to specify their intentions to AI-systems and also interpret the intentions of an AI-collaborator.
Ill-fitting methods can undermine the human-AI interaction, leading to a downgrade in performance and an increase in mistrust of autonomous systems.
In this thesis, I build and study human-centered methodologies which enable humans to specify their desired behavior of an AI system, as well as receive suitable explanations which optimize their ability to perform the shared task.
First, I build two machine learning frameworks to enable humans to specify their desired behavior of an autonomous system via unstructured natural language.
In my first contribution, I develop a machine learning framework that translates a user's unstructured description of their desired policy into a decision-tree, which is then utilized to initialize a differentiable decision tree (DDT) policy. The use of a ``white-box'' framework such as a DDT, enables users to interpret the final learned policy of the agent.
Next, I expand this method towards interpreting the high-level strategies of a user rather than a description of a specific policy.
For this task, I train a computational interface, powered by a large-language-model, to translate language descriptions of strategic intent into actionable ``Commander's Intent'' in the form of goals and constraints.
The second half of my thesis pertains to explainable-AI, specifically, understanding the factors which influence a user's interaction with an explanation and developing personalizable explanation methods which consider the specific user and the context of the interaction.
First, I conduct an human-subjects experiment aimed at understanding the differences, wherein we observe a counter-intuitive phenomenon; participants performed significantly worse with explanations they preferred at a significant level.
To address this finding, I developed a first-of-its-kind personalized explainable AI framework to adaptively balance a participants preference and task-performance.
This method is comprised of two components. First, I build a federated, personalization model which predicts when a participant is likely to make a mistake. Second, I utilize this prediction to provide a suitable explanation to a user, i.e. if a user is likely to be incorrect, they are more likely to be given an explanation that optimizes their performance and if they are likely to be correct, they are more likely to be given an explanation which they prefer.
Finally, I conclude this thesis with a human-subjects experiment which compares different forms of personalization for explainable AI.
We compare adaptive personalization with adaptable personalization to study the impact of varying the approach to personalizing an explanation received by a user.Ph.D.Computer Scienc