137091 research outputs found
Sort by
Designing High Performing Durable Bio-Based Coatings for Packaging Applications
No full textThis thesis focuses on developing new engineering knowledge essential for durable materials for sustainable packaging applications using renewable biopolymers. These include cellulose, extracted from plants, and chitin, sourced from mushrooms and food waste like crustacean exoskeletons. These natural polymers can form strong, high-performance films but are limited by moisture sensitivity and brittleness. Through chemical modification, compositional design, and controlled processing, this work demonstrates how their barrier and mechanical properties can be tuned to control and improve barrier and mechanical performance of renewable polymers.
Chapter 2 presents a fully biodegradable multilayer structure composed of a cellulose nanocrystal–chitin nanofiber (CNC–ChNF) coating on paper laminated with poly(hydroxyalkanoates) (PHA). The design leveraged the oxygen barrier of CNC–ChNF, the mechanical support of paper, and the moisture resistance of PHA. The most effective configuration, achieved through double-sided PHA lamination, protected the hydrophilic coating from humidity and achieved barrier performance comparable to petroleum-based plastics like poly(ethylene terephthalate) (PET) and poly(ethylene) PE.
Chapter 3 explores a crosslinked chitosan–citric acid–bentonite (Ch–CA–BNT) blend coated on PHA to improve barrier and mechanical performance without lamination. Crosslinking reduced water uptake, and the addition of BNT stabilized the coating’s strength, resulting in a bio-based material that maintained strong performance even after mechanical deformation.
Chapter 4 introduces a single-layer approach using microfibrillated cellulose – CNC (MFC–CNC) blends treated with atomic layer deposition (ALD) of aluminum oxide (AlOx). The ALD process rendered the films more hydrophobic without compromising mechanical integrity, producing a single-layer biodegradable barrier film that eliminates the need for complex multilayer systems.
In Chapter 5, poly(lactic acid) (PLA) and poly(ε-caprolactone) (PCL) were fabricated using 3D printing and evaluated under various printing conditions using a high-throughput mechanical characterization system (HTMECH) alongside conventional uniaxial testing. HTMECH enabled rapid screening of tensile properties and effectively captured trends related to material composition, layer thickness, and extrusion temperature, aligning well with traditional testing results. This chapter demonstrates HTMECH as a promising tool for accelerating the design and optimization of bio-based 3D-printed materials.
Overall, this thesis establishes a framework for developing sustainable, high-performance packaging materials by integrating bio-based polymers, nanomaterials, and scalable processing strategies. The findings highlight how careful design of structure, chemistry, and processing can produce materials that meet performance requirements. Collectively, these results contribute to advancing biodegradable alternatives to petroleum-based plastics and provide valuable insights for future development of renewable, recyclable, and compostable packaging systems
Flow Boiling in Meso-Scale Pin-Fin Coldplates
No full textColdplates are a critical component in various cooling applications, such as cooling of data centers, high-end central processing units (CPUs) and graphics processing units (GPUs), and the thermal management of power electronics. The rising demand for computational resources, high-power-density electronics, and compact-integrated systems is accompanied by thermal management challenges, necessitating the transition to more aggressive cooling technologies. This shift has led to a growing interest in two-phase coldplates/heat sinks as a promising solution due to their high heat transfer coefficients and improved temperature uniformity.
Recent flow boiling studies have focused on fin-enhanced silicon microgaps and microchannels, given their importance in cooling high-power-density chips operating at ultra-high heat fluxes. However, the majority of the research on flow boiling in mini- and macro-scale configurations has been limited to channels and tubes, overlooking crucial geometries such as pin-fin-enhanced mini-channels. This literature gap, particularly the study of flow boiling in pin-fin-enhanced channels in the meso-scale, has limited the availability of physical insights, data, flow regime maps, and correlations that assist in the design of two-phase coldplates.
Dielectric fluid flow boiling in a millimeter-scale coldplate was experimentally investigated under non-uniform heating conditions, representing realistic heat dissipation scenarios for high-power-density applications such as electronics cooling. Background heaters simulated low-dissipating-power components, while hotspot heaters mimicked high-heat-flux devices with heat fluxes reaching up to 1 kW/cm2. Flow visualization showed the role of localized nucleate boiling in enabling the two-phase coldplate to stably manage such high heat fluxes. High-speed visualizations provided detailed insights into the flow regimes and bubble dynamics, highlighting the effectiveness of two-phase coldplates in handling intense, localized heating.
A parametric study on dielectric flow boiling in meso-scale pin-fin coldplates using HFE-7200 dielectric fluid was conducted to develop an in-depth understanding of the effects of pin-fin structure and flow parameters on two-phase heat transfer, pressure drop, critical heat flux (CHF), and flow regime transitions. The study covered seven pin-fin-enhanced geometries with hydraulic diameters ranging from 880 µm to 4.25 mm and provided valuable insights through over 840 high-speed visualization videos, identifying two distinct CHF mechanisms and three main flow regimes. The primary observed flow regimes included bubbly, slug, and annular/stratified flows. In meso-scale coldplates, buoyancy, viscous, inertial, and capillary forces all play substantial roles in shaping flow patterns and boiling behavior, highlighting the need to understand how these regimes differ from micro-scale geometries, where surface tension forces dominate. The influence of these forces on flow boiling patterns and characteristics is discussed in detail.
A modeling framework was developed to assist in the design of two-phase pin-fin cold-plates by advancing the state-of-the-art Lee model. An enhanced phase-change Lee model, which accounts for both the degree of superheat required for the onset of nucleate boiling and the variation of saturation temperature with pressure, is presented. The computational predictions of this enhanced model were evaluated through experimental studies across scales, from dielectric fluid flow boiling in microgaps to millimeter-scale pin-fin-enhanced coldplates, with a focus on managing ultra-high heat fluxes and complex non-uniform heating conditions.
The enhanced model’s predictions were initially compared to those of the original Lee model for flow boiling behavior in enhanced microgap. The enhanced model demonstrated better agreement with experimental data in the subcooled section of the microgap compared to the original model. In the mm-scale coldplate, the enhanced Lee model accurately predicted flow boiling under non-uniform heating conditions, effectively capturing the mountain-valley trend created by the hotspots. Additionally, the model successfully predicted the flow regimes, capturing the transition from stratified flow at low flow rates to bubbly and slug-churn flow at higher flow rates, aligning closely with experimental observations. The enhanced Lee model demonstrated high efficacy in predicting thermal-hydraulic performance and flow regimes. Experimental validation confirmed the model’s value as a powerful tool for designing two-phase coldplates, supporting the development of advanced thermal management solutions for high-power-density applications.Ph.D.Mechanical Engineerin
Assessing Programmatic Variables and Uncertainties in Space Exploration Campaign Planning
No full textAll space exploration programs begin life as a set of requirements, driven by one or more scientific or technological objectives. Historically, the majority of those objectives were achieved using tailor-made spacecraft and systems to deliver their payloads, be they science experiments or astronaut crews, to their intended location. For example, the Apollo spacecraft were designed to deliver and return a specific amount of mass to and from a specific set of locations on the Moon; and most ISS-servicing vehicles were designed specifically to do just that. More recently, however, the number of vehicles that could service space exploration logistics needs has greatly increased. This is particularly true in the case of lunar exploration, largely due to the Artemis and CLPS programs, supported by government space agencies and private companies in the U.S. and across the world.
In the earliest stages of space program planning, the program or mission architect must find the most effective method by which to achieve the program objectives. ``Effective'', in this case, covers a wide range of needs, such as affordability, reliability, and robustness to uncertainties in the development cycle of the mission(s). It is the program planner's role to uncover the architecture that best satisfies these needs from among the mire of available options and associated uncertainties. The increasingly-large number of options of novel technologies and logistics vehicles, some that already exist but many that are still under development, coupled with the ambitious exploration objectives of programs like Artemis creates a large decision space for the program planner.
Space logistics, as a field, is concerned with the efficient planning of space missions. A wide number of classical logistics problems, when applied to space, offer methods for assessing the effectiveness of a range of mission types. Exploration missions can be modeled using network flow formulations, in which mixed-integer linear programming is used to find the most efficient flow of commodities through a network given a set of supplies, demands, and vehicles. However, in the early stages of program planning, many of the parameters of the network flow model are undefined, representing, for example, uncertain vehicle performances, payload masses, or launch schedules. Some of these are uncertainties inherent to the fact that some payloads and vehicles are still in their development cycles. Others represent decisions to be made by the program architect. Therefore, this thesis will explore methods by campaign model parameters can be analyzed, and their effect on the results of the model optimization can be assessed.
With a focus on launch schedule as an example of a programmatic variable, the methods developed here are used to find optimal launch schedules for a cislunar exploration campaign. In the deterministic case, the method is also used to study the impact of varying logistic provider availability on the optimal mission plan, and therefore make recommendations about logistics provider redundancies. In the stochastic case, the analyses are extended to identify the most-likely-optimal launch time for each payload in the campaign and most-likely-optimal overall schedule. The probability of a specific logistics vehicle be used in the campaign, and the number of times that the specific vehicle design is likely used, is also able to be calculated using these methods. This provides the campaign planner with some insights into the value of each available logistics vehicle in improving the robustness of the campaign under launch uncertainty.
Finally, a parametric programming method is employed to study how the optimal cislunar logistics plan varies as vehicle, payload, or commodity dynamics parameters change. The end result of this is the identification of regions within the parameter space for which the optimal plan remains feasible. These regions in parameter space can then be used to define requirements for the systems to which those parameters pertain. In the cislunar logistics case study, for example, the method is used to identify the bounds on in-situ-produced rocket propellant production rates and infrastructure maintenance requirements such that ISRU capability should, or should not, be incorporated into a logistically-optimal mission plan
Processing and Properties of High-Performance BNNT Fibers and other Macrostructures
No full textIn the last two decades, key innovations have been made in the synthesis and applications for boron nitride nanotubes (BNNTs). BNNTs are electrical insulators with a band gap of 5-6 eV, but also possess resistance to significant oxidation up to 900 °C, low-κ dielectric properties, high thermal conductivity, and high strength. This combination of properties makes BNNTs good candidates for applications requiring one or more of these attributes. Formation of BNNT fibers has been reported in the literature by two methods: wet spinning from BNNTs in a superacid solution and direct assembly from the as-grown BNNT material upon synthesis in the reactor. The wet spinning method to produce BNNT fiber involves a process of forming a liquid crystalline phase of BNNT in a superacid and produces moderately aligned BNNT fibers. To date, these literature-reported fibers have a tensile strength of 10-16 MPa and a modulus of 0.5-1.5 GPa. A new method to process BNNT fibers with a high degree of alignment, high modulus, and good tensile strength is presented in this thesis. The processing-structure-property relationship for these BNNT fibers is examined. New state-of-the-art tensile properties are achieved for BNNT fibers: tensile modulus of 396 GPa and strength of 500 MPa. Additionally, a new method is reported to disperse and process BNNTs using alcohols and glycerol. BNNT films from alcohol dispersions show thermal conductivity as high as 44 W/mK, dielectric breakdown strengths up to 160 kV/mm, and dielectric constants of 2.0-7.0 at 1-100 MHz. Factorial experimental design is used to study the impact of processing parameters on these properties. This work represents significant progress towards producing BNNT fibers and films with improved structure and properties.Ph.D.Materials Science and Engineerin
Development of the path-integral methodologies for non-adiabatic dynamics
No full textThis thesis presents the development of path-integral methodologies for non-adiabatic dynamics, which are crucial for accurately describing systems with energetically close electronic states. In such systems, the Born-Oppenheimer approximation often fails, necessitating a more comprehensive approach that includes both nuclear and electronic degrees of freedom will a complete description. The path-integral framework for nuclear degrees of freedom is employed due to its success in capturing nuclear quantum effects in ground-state problems.
Extending the path-integral treatment of nuclei to multi-state systems requires a semi-classical or mixed quantum-classical treatment for the electronic states. Two key methods are explored and tested in this thesis: Non-adiabatic Ring Polymer Molecular Dynamics (NRPMD), which combines path-integral nuclei with Meyer-Miller-Stock-Thoss (MMST) mapping for the electronic states, and the Ring Polymer Mapping Approach to Surface Hopping (RP-MASH), an extension of the deterministic quasi-classical surface hopping method. These methods allow for the treatment of nuclei and electronic states in a classically isomorphic manner, ensuring that all degrees of freedom are consistently represented. This uniform approach enables the use of classical computational techniques, such as molecular dynamics, to simulate electronically non-adiabatic dynamics while preserving important nuclear quantum effects. The performance of these methods on model systems, such as spin-boson and linear vibronic models, demonstrates their promise for more complex systems including realistic conical intersections.
However, while the methods show great promise, challenges remain. For example, the treatment of decoherence effects and the precise handling of frustrated hops during non-adiabatic transitions still require further refinement of RP-MASH. Future developments will focus on improving the accuracy of these methods, particularly by incorporating decoherence effects and refining the treatment of multi-state systems. There is also potential for extending these methods to account for quantum light-matter interactions, such as polaritons, which could open up new avenues in the study of complex quantum systems
Bioprinted Tri-leaflet Scaffold for Aortic Heart Valve Function and Repair
No full textHeart valve disease (HVD) is a prevalent and increasing clinical burden, with limited treatment options typically restricted to valvular repair or prosthetic replacement surgery. Existing alternatives, such as biological and mechanical valve replacements, have been recognized as inadequate for proper function in pediatric patients. The inability of these valves to grow or biologically respond to their environment presents ongoing challenges, often necessitating multiple valve-refitting surgeries and lifelong anticoagulation dependency. To this end, tissue- engineered heart valves (TEHVs) have emerged as an attractive therapeutic solution, offering patient-specific models that can self-repair and remodel. Advances in 3D bioprinting enable the recreation of native heterogeneity and anatomical fidelity, facilitating customizable designs. The proposed study focuses on constructing a TEHV using 3D bioprinting to recreate the three-layer leaflet structure of an aortic valve composed of poly-e-caprolactone (PCL) and a cell-laden gelatin- methacrylate (GelMA) and polyethylene glycol diacrylate (PEGDA) hydrogel scaffold, incorporating valvular interstitial- like (VIC-like) cells to promote regeneration and remodeling. The study is composed of two specific aims: (1) developing and assessing the feasibility of a 3D bioprinted, multilayered scaffold that emulates native valve structures and (2) evaluating the remodeling capabilities of the scaffold and extracellular matrix (ECM) production under dynamic bioreactor conditions to optimize valve performance. This project aims to describe the cellular and mechanical interactions between biomaterials and cells to overcome risks in pediatric populations, integrating autologous stem cells, biomaterials, and 3D bioprinted scaffolds to generate a valve leaflet capable of biological integration and mechanical function for optimal restoration.Ph.D.Biomedical Engineerin
Atmospheric Chemistry and Meteorological Controls on Air Quality Response to Emission Reductions: Insights from China and the United States
No full textThe effectiveness of emission control strategies for mitigating air pollution depends critically on the complex interactions between anthropogenic emissions, atmospheric chemistry, and meteorological conditions. This dissertation investigates these interactions through comprehensive observational and modeling studies in China and the United States, revealing fundamental constraints on the efficacy of emission reduction policies and providing new insights for air quality management.
Field measurements in the North China Plain reveal that springtime surface ozone concentrations exceed those in California by 50% despite similar meteorological conditions. Analysis of aromatic volatile organic compounds (VOCs) shows ARO2 and ARO1 concentrations 7.42 and 5.23 times higher than California levels, respectively. Box model calculations with the Master Chemical Mechanism demonstrate that aromatics contribute 73% of total ozone production, with ARO1 alone accounting for 62.8%. The implementation of observation-constrained emission redistribution based on industrial source patterns fundamentally shifts the regional chemical regime from VOC-limited to transitional conditions. The revised emission inventory correlates strongly with satellite-observed glyoxal columns, providing independent validation and highlighting the critical importance of accurate industrial emission characterization for ozone pollution assessment.
A decade-long analysis (2015-2024) of China's Spring Festival, characterized by consistent 30-50% NOₓ reductions, reveals an unexpected shift in atmospheric oxidant response. The oxidant response (ΔOx = ΔO₃ + ΔNO₂) to identical emission reduction patterns has transitioned from negative to positive values over this period. Chemical transport modeling successfully reproduces these trends with correlation coefficients of 0.65-0.80 across regions. Machine learning analysis identifies cloud cover and subsequent radiation changes as primary drivers, yielding robust correlations (R = 0.85-0.94) between normalized ΔOx and meteorological parameters. These findings demonstrate that meteorological variability can fundamentally override the expected chemical responses to emission reductions.
Long-term analysis of SO₂ emission controls in the United States (2004-2023) reveals declining wintertime control efficacy despite successful implementation of coal-to-gas transitions in power generation. While SO₂ and sulfate concentrations decreased significantly in both the Rust Belt and Southeast regions, wintertime sulfate fractions increased from ~40% to 55-60%, contrasting with stable summer values of ~65%. This seasonal divergence manifests as slower sulfate decreases during 2004-2013 compared to 2013-2023, despite greater SO₂ reductions in the earlier period. The analysis attributes this diminishing effectiveness to enhanced SO₂ oxidation efficiency as atmospheric conditions transition from SO₂-saturated to oxidant-limited regimes. This chemical damping effect, driven by limited H₂O₂ availability, represents a fundamental constraint on the benefits of continued emission reductions.
High-resolution inverse modeling of TROPOMI NO₂ observations during the COVID-19 lockdown reveals complex and counterintuitive emission patterns. Despite widespread reports of emission decreases and reduced economic activity, analysis shows 25% NOₓ emission increases over the Jiang-Han Plain region, specifically along supply routes to locked-down cities. Post-lockdown recovery patterns demonstrate rapid emission rebounds in northern Jiangsu and Fujian provinces, areas characterized by high concentrations of small-scale enterprises. These differential recovery rates between large and small businesses highlight the heterogeneous nature of emission sources and the importance of maintaining essential supply chains during disruption events.
This research advances atmospheric chemistry understanding through quantification of aromatic VOC contributions to ozone production, demonstration of meteorological controls on emission reduction effectiveness, identification of chemical limitations in long-term control efficacy, and revelation of unexpected emission behaviors during socioeconomic disruptions. The findings collectively challenge the conventional linear relationship assumed between emission reductions and air quality improvements. Successful air quality management requires dynamic, observation-constrained frameworks that account for chemical regime transitions, meteorological variability, and non-linear atmospheric responses.
The implications extend beyond regional air quality to global atmospheric chemistry and climate policy. As nations implement increasingly stringent emission controls, the diminishing returns identified in this work will become more prevalent, requiring adaptive strategies that consider both chemical saturation effects and meteorological influences. The research underscores the necessity of high-resolution monitoring networks capable of capturing fine-scale emission variations and atmospheric responses. Future air quality management must evolve from static, prescriptive approaches to dynamic systems that integrate real-time observations, advanced modeling capabilities, and machine learning techniques to predict and respond to changing atmospheric conditions. Only through such comprehensive, scientifically informed strategies can we effectively address the persistent challenge of air pollution in an era of rapid industrialization and climate change.Ph.D.Earth and Atmospheric Science
Decision-Driven Scenario Generation for Zero-Impact Aviation: A Multi-Stakeholder Collaborative Framework
No full textPresented at AIAA SciTech 2026, Orlando, FL.The aviation industry significantly supports global economic growth and connectivity, but its rising greenhouse gas emissions pose critical environmental challenges. With air travel projected to reach 10 billion passengers by 2050, emissions could escalate to 2000 megatons of CO2 without intervention. Achieving net-zero emissions demands coordinated advancements in aircraft technologies, sustainable energy, operational practices, and policy frameworks. This study explores the complex interdependencies among aviation stakeholders, including airlines, airports, energy producers, technology developers, OEMs, and policymakers, highlighting how collaborative and strategically timed investment decisions impact emission reduction pathways. Scenario-based simulations demonstrate the necessity of integrated planning, synchronized infrastructure and technological developments, and proactive management of uncertainties to successfully decarbonize aviation
Dynamical Origins of Warm-Season Precipitation Extremes over the Northern Extratropics: A Multiscale Diagnostic and Modeling Study
No full textExtreme precipitation events are increasing in frequency and intensity under climate change, causing on the order of hundreds of billions of US dollars in global annual economic losses. While many studies have focused on the thermodynamic drivers of these extremes, regional variability of extreme precipitation during the boreal warm season is often governed by complex multiscale atmospheric-flow interactions. These interactions range from planetary-scale forcing to regional meso- and micro-scale processes, and they remain inadequately understood and poorly represented in models. This thesis addresses these challenges by investigating the large-scale dynamical origins of warm-season precipitation extremes across the northern extratropics, with a particular focus on Mesoscale Convective Systems (MCSs) over the central United States and precipitation extremes at the margins of the Asian summer monsoon. Through a combination of observational analysis, model diagnosis, and idealized modeling, this work develops a hierarchical framework to disentangle the roles of large-scale forcing, synoptic variability, and upscale feedback in shaping the distribution and variability of regional extremes, with the ultimate goal of improving their simulation and prediction in climate models.
The first part of the study (Chapter 2) presents a comprehensive diagnostic analysis of MCSs over the U.S. Great Plains during boreal spring from 2000 to 2020. Using hierarchical clustering analysis, the study identifies five large-scale upper-tropospheric circulation patterns associated with MCS genesis. These clusters fall into two dynamical categories: “remotely forced” patterns such as downstream-propagating storm track eddies and “locally excited” patterns driven by regional instability. Among them, Cluster 2, linked with Pacific storm track disturbances, is the dominant contributor to the recent upward trend in MCS frequency. The phase transition of the Pacific Decadal Oscillation (PDO) emerges as a key climate mode contributing to this trend. This chapter also evaluates the performance of the NOAA GFDL AM4 model in simulating MCSs. While the model underestimates MCS frequency and shifts MCS activity eastward compared to observation, it successfully captures the observed large-scale forcing patterns of MCSs. The locational bias in MCS activity is ultimately traced to deficiencies in both seasonal mean circulation and synoptic biases, including weakened Great Plains low-level jets (LLJs) and misplaced surface fronts. The analysis shows that biases in the mean state and transient processes jointly contribute to the eastward shift of MCS activity in the model.
The second part of the study (Chapter 3) is a dynamical investigation of how MCS heating provides upscale feedback to large-scale atmospheric circulations. Using an idealized two-layer quasi-geostrophic (QG) model with empirical heating parameterizations, the chapter shows that latent heating from MCSs leads to stronger storm tracks, increased downstream eddy activity, and better organized synoptic wave propagation, as demonstrated through nonmodal instability (optimal mode) analysis. These effects also improve the representation of circumglobal teleconnection structures. For example, including MCS heating strengthens the feedback of North Atlantic synoptic eddies on the background Atlantic jet, with implications for the development of low- frequency modes such as the Northern Annular Mode (NAM). Although based on a highly simplified model, the study demonstrates the planetary-scale significance of MCS heating in modulating springtime extratropical multiscale variability.
The third part of the study (Chapter 4) combines methodologies from the previous two chapters, integrating multiscale diagnosis and dynamical analysis using idealized models to identify the fundamental causes of trends and variability in regional extremes. The focus is placed upon summertime extreme precipitation in Northeast China and Pakistan, two monsoon-margin zones with increasing vulnerability to extreme precipitation and flooding. Using clustering and a barotropic vorticity model, the study identifies distinct Rossby wave pathways associated with extreme precipitation in Northeast China. It then applies optimal mode analysis to background flows from two different periods (1982-2002 and 2003-2023) to assess how changes in the mean flow affect wave excitation and propagation and thus the occurrence of precipitation extremes. The results show that in the recent two decades the atmospheric waveguides feature an extension over Eurasia, strengthening within the Asian jet, and weakening at its flanks. These structural changes align with an increased frequency of a particular pathway of wave propagation. Further analysis confirms that optimal modes under the recent mean flow resemble this pathway of preferred wave propagation while correlations with other pathways (clusters) decrease. This provides compelling evidence that evolving seasonal mean flows are changing the frequency of regional precipitation extremes through their influence on the excitation and propagation of atmospheric disturbances.
The final chapter summarizes the findings and highlights the importance of understanding scale interactions in the climate system. Across all chapters, the thesis builds a multiscale framework that connects planetary-scale climate modes, synoptic forcing, and local processes to the genesis, modulation, and impact of regional extremes. It highlights how idealized models, when combined with observations, can provide fundamental insights into the mechanisms governing trends and variability of precipitation extremes. By linking biases in Earth System Models to their deficiencies in representing multiscale interactions, the thesis offers pathways for improving weather predictions and climate projections. This thesis advances our understanding of warm-season extreme precipitation by identifying the key dynamical processes that control their variability and trend. Ultimately, these insights contribute to better adaptation strategies and more reliable assessments of future climate and weather risks.Ph.D.Earth and Atmospheric Science
Synthesizing Execution Traces with Graph-Based Generative Models
No full textThe objective of this research is to investigate the use of graph-based generative models for synthesizing execution traces of machine learning workloads. Specifically, this work is guided by the following research question:
Can we develop a generative framework that produces synthetic execution traces which accurately mimic proprietary machine learning workloads, while preserving key performance characteristics such as latency and memory usage?
Execution traces are indispensable for benchmarking, system simulation, and hardware–software co-design; however, they also encode sensitive information that may reveal proprietary model architectures or training methodologies. Synthesizing realistic yet privacy-preserving traces thus presents a critical opportunity to enable secure data sharing and collaborative optimization at scale.
To address this challenge, we model execution traces as graphs and apply generative diffusion-based techniques to capture both structural and performance-related properties. The main contributions are threefold: (1) formalizing execution trace synthesis as a graph generation problem, (2) developing a generative framework that balances fidelity and privacy, and (3) performing comprehensive evaluations that quantify structural realism, performance relevance, and resistance to model-identification attacks.
By demonstrating that graph generative models can produce high-fidelity, privacy preserving execution traces, we advance the broader goal of secure, data-driven co-design for next-generation machine learning infrastructure