DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Classical and quantum spreading processes on disordered and complex networks
No full textMay 2022School of ScienceSpreading processes on networks provide a valuable, abstract framework to study a vast array of problems using a unified formalism, from the spread of wildfires in forests to opinion formation in social groups. This thesis presents three specific problems that relate to different spreading processes on complex networks. The first of these is a study of diffusive persistence on disordered lattices and complex networks to better understand the temporal characteristics and the lifetime of fluctuations in stochastic processes in networks. Diffusive persistence is defined as the probability that the diffusive field at a site (or node) has not changed sign up to a certain time (or in general, that the node remained active/inactive in discrete models). Applications of our research could help one better understand the lifetime and temporal dynamics of activity fluctuations and trends in social networks. We investigated disordered networks (characterized by the fraction of removed edges) and found that the behavior of the persistence depends on the topology of the network. In 2D networks we have found that above the percolation threshold diffusive persistence scales similarly as the original two-dimensional regular lattice, according to a power law with an exponent of 0.186 +- 1.4*10^(-4). At the percolation threshold, the scaling exponent changes to one with 0.141 +- 5.3*10^(-5), as the result of the interplay of diffusive persistence and the underlying structural transition in the disordered lattice at the percolation threshold. In contrast, we found that in random networks without a regular structure, such as Erdős–Rényi networks, no simple power-law scaling behavior exists above the percolation threshold. We also investigate finite-size effects for 2D lattices at the percolation threshold and find that the limiting value obeys a power-law with exponent zθ, where z = 2.56 +- 2.3 * 10^(-2) instead of the value of z=2 normally associated with finite-size effects on 2D lattices. Next, we discuss percolation on quantum networks. Quantum networks describe communication networks that are based on quantum entanglement. A concurrence percolation theory has been recently developed to determine the required entanglement to enable communication between two distant stations in an arbitrary quantum network. Unfortunately, concurrence percolation has been calculated only for very small networks or large networks without loops. Here, we develop a set of mathematical tools for approximating the concurrence percolation threshold for unprecedented large-scale quantum networks by estimating the path-length distribution, under the assumption that all paths between a given pair of nodes have no overlap. We show that our approximate method agrees closely with analytical results from concurrence percolation theory. The numerical results we present include 2D square lattices of 200^2 nodes and complex networks of up to 10^4 nodes. The entanglement percolation threshold of a quantum network is a crucial parameter for constructing a real-world communication network based on entanglement, and our method offers a significant speed-up for the intensive computations involved. Finally, we study how public transportation data can be employed in the modeling of the spread of infectious diseases based on SIR dynamics. We present a model where public transportation data is used as an indicator of broader mobility patterns within a city, including the use of private transportation, walking etc. The mobility parameter derived from this data is used to model the infection rate. As a test case, we study the impact of the usage of the New York City subway on the spread of COVID-19 within the city during 2020. We show that utilizing subway transport data as an indicator of the general mobility trends within the city, and therefore as an indicator of the effective infection rate, improves the quality of forecasting COVID-19 spread in New York City. Our model predicts the two peaks in the spread of COVID-19 cases in NYC in 2020, unlike a standard SIR model that misses the second peak entirely.Ph
Apparatus, methodology, and modeling for continuous in-situ impregnation of dry carbon fiber tows with thermoplastic polymers for use in additive manufacturing
No full textDecember 2018School of EngineeringA novel co-extrusion system for continuous fiber reinforced thermoplastic composites was designed, fabricated, and tested. This new process, called In-Situ Impregnation, is a pultrusion process that impregnates continuous dry fiber reinforcement tows in-situ with thermoplastic for applications ranging from additive manufacturing using robotic manipulation to automated fiber placement. The performance goal was to design a co-extrusion system that directly uses raw materials (thermoplastic pellets and rolls of carbon fiber tow) instead of ‘prepreg’ tow in an effort to streamline and cut costs in advanced composites manufacturing and deliver fully customizable fiber orientation. Analytical and computational modeling was conducted to describe the flow rate of polymer through the system, impregnation of the fibers with polymer, heat transfer and temperature uniformity of the die system, friction and fiber tensioning, and the interfacial shear strength between the polymer and the fibers. Multiple experiments were conducted on the working pultrusion system varying polymer, exit geometry, die temperature, and volumetric flow rate of polymer from a thermoplastic screw extruder to gather process-ability and validate the models. A new test method was developed to evaluate the relative interfacial shear strength of a carbon fiber tow (bundle) and thermoplastic polymer, which was tested experimentally and results were obtained for the three polymers used. Microscopy techniques were employed to estimate the degree of impregnation and fiber volume fraction based on cross sections of the resulting composite tows. Future work includes commercialization of the technology, automation of the manufacturing system, and publishing the work in peer reviewed academic journals.Ph
The mechanical response of materials at the nanoscale via simulations
No full textAugust 2023School of ScienceWe use both molecular dynamics (MD) simulations and density functional theory (DFT) calculations to investigate various nanomechanical properties of nanoparticles under compression, exfoliation of perovskite heterogeneous systems and epitaxial methods. The compression of nanoparticles is of fundamental importance both scientifically and to various other applications such as tribology, targeted drug delivery, and biosensors. We embark on an extensive investigation of the response of a nanoparticle to external compression as this field is still short on answering some lingering doubts and details. We find a size-dependent brittle to ductile transition. We formulate a Griffith based fracture model that calculates r_critical to the same order of magnitude. We report a tradeoff: On one hand compression strengthens the surface but on the other it accumulates more and earlier shear. That is critical as the advantages of material strength can then be weighed in design against change in mechanical properties and based on what a specific application is aimed for. We also apply this to more experimentally relevant systems as in exfoliation. We design build and run DFT calculations on heterogenous perovskite systems to investigate exfoliation. It is an invaluable means of detaching epitaxial layers from substrates to produce membranes that are essential in various applications such as optoelectronics and high-speed computing. We devise an exfoliation regime that matches experimental results in every case. We back that up further using MD simulations. This can open the door to experiment with and confirm many other systems that can now be more easily tested. Which makes the process more general, efficient, and cost effective. We demonstrate that the presence of a stressor is a necessary but not sufficient condition for exfoliation. As successful peeling is contingent on defect free film-interface-substrate, we use MD to simulate and confirm that graphene nanopatterning allows for great reduction of defects in freestanding single-crystalline membranes using deposition simulations of Germanium on Silicon. We show that as graphene coverage increases the dislocation density is greatly reduced. We successfully generalize epitaxy to include multiple layers. We simulate and show the effectiveness of growing and harvesting multilayered epitaxial systems through multiple graphene layers. This results in layer-by layer peeling culminating in multiple free-standing membranes. This offers a high throughput and low-cost production of single crystal membranes needed in many applications such as high-power electronics.Ph
Efficient and robust federated learning in heterogeneous networks
No full textMay 2023School of ScienceIn the modern day, data for machine learning tasks is generated by globally distributed devices. The data are often sensitive, containing private information that must not be shared. Federated learning (FL) algorithms were introduced as a means to learn from distributeddata in a privacy-preserving, communication-efficient manner. However, there are still a number of challenges that must be addressed. The globally distributed systems in which these algorithms run often have high communication latency. The network topology of devices does not always match the assumed topology of FL algorithms. Participants may have various computing speeds due to heterogeneous hardware or compute resources available. Data can be partitioned among parties by feature space, rather than the often assumed partitioning by sample space. Data can often be incomplete, e.g., missing labels and features. Data can also be spurious, containing irrelevant features that distract from the prediction task. In this thesis, we address these challenges to make FL efficient and robust. We first present a federated learning algorithm for multi-level networks: a set of disjoint sub-networks, each with a single hub and multiple workers. In our model, workers may run at different operating rates. We provide a unified mathematical framework and theoretical analysis that show the dependence of the convergence error on the worker node heterogeneity, hub network topology, and the number of local, sub-network, and global iterations. In the experiments, we find that our algorithm can converge up to 2× as fast as other federatedlearning algorithms for multi-level networks. We then present a vertical federated learning (VFL) algorithm for cases when parties store data with the same sample IDs, but different feature sets. This is known as vertically partitioned data. Our algorithm applies compression to intermediate values shared betweenparties. Our work provides the first theoretical analysis of the effect that message compression has on the convergence vertical federated learning algorithms. We experimentally show compression can reduce communication by over 90% without a significant decrease in accuracy compared to vertical federated learning without compression. Next, we present another vertical federated learning algorithm with the aim to accommodate device heterogeneity. We consider a system where the parties’ operating rates, local model architectures, and optimizers may be different from one another and, further, they may change over time. We provide theoretical convergence analysis and show that the convergence rate is constrained by the party operating rates and local optimizer parameters. We apply this analysis and extend our algorithm to adapt party learning rates in responseto changing operating rates and local optimizer parameters. In the experiments, we find that our algorithm can reach target accuracies up to 4× faster than other vertical federated learning algorithms, and that our adaptive extension can further provide an additional 30%
improvement in time-to-target accuracy. Next, we present three Self-Supervised Vertical Federated Learning (SS-VFL) algorithms. These algorithms learn from unlabeled and non-overlapping data in a setting with vertically partitioned data. The algorithms apply self-supervised and data imputation methods. We compare our algorithms against supervised VFL in a series of experiments. We show that SS-VFL algorithms can achieve up to twice the accuracy of supervised VFL when labeled data are scarce. We also show that these SS-VFL algorithms can greatly reduce communication cost to reach target accuracies over supervised VFL. Finally, we present a feature selection method for vertical federated learning. Our method removes spurious features from the dataset in order to improve generalization, efficiency, and explainability. Our method requires little communication between parties compared to other VFL feature selection methods. We analytically prove that our method removes spurious features from model training. We provide extensive empirical evidence that our method can achieve high accuracy and remove spurious features at a fraction of the communication cost of other feature selection approaches.Ph
The biosynthesis of chondroitin sulfate and its derivates using metabolic engineered escherichia coli strains
No full textDecember 2022School of ScienceWe metabolically engineered E. coli strains to produce chondroitin sulfate (CS). We then used these engineered strains to produce derivatives of CS through feeding experiments. We Synthesized N-glycolyl chondroitin (Gc-CN) and N-glycolyl chondroitin sulfate (Gc-CS). These derivatives have medical and evolutionary applications. Gc-CN/Gc-CS is hypothesized to be able to detect diseases like carcinomas, atherosclerosis, etc. because it is a metabolite of the sialic acids that contributes to these diseases in humans, N-glycolylneuraminic acid (Neu5Gc). Furthermore, the derivatives can help date the loss of the CMAH gene; a gene lost after a last common ancestor with the great apes. We successfully synthesized Gc-CN and Gc-CS in metabolically engineered E. coli K4 adapted for CS production. We fed the bacteria with a glucose carbon source supplemented with chemically synthesized N-glycolyl glucosamine (GlcNGc), which allowed the incorporation of the N-glycolyl into chondroitin. After exploring the pathway for chondroitin sulfate synthesis, we chose to increase N-glycolyl incorporation into chondroitin. To do this, we investigated the effect of downregulating some genes in order to shift the metabolic flux towards N-glycolyl incorporation. We investigated the effect of knocking the bifunctional N-acetylglucosamine-1-phosphate uridyltransferase and glucosamine-1-phosphate acetyltransferase gene (GlmU) and the Glutamine-fructose-6-phosphate aminotransferase gene (GlmS). We Further decided to synthesize other chondroitin derivatives by isotopically labelling them. Using the same metabolically engineered E. coli strain for CS production, we produced deuterated chondroitin and 13C labeled.Ph
Risk analysis with dynamic safety measure for pressurized water reactors with advanced safety features
No full textAugust 2023School of EngineeringAfter the Fukushima Daiichi Accident, the US Department of Energy initiated the LightWater Sustainability Program to investigate safety options for existing nuclear fleets, such as
accident-tolerant fuel and US diverse and flexible coping strategies. As these safety options are
relatively new, developing a framework that can assess their risk and benefits effectively is
essential. A dynamic probabilistic risk assessment framework is suggested to analyze the risk and
safety of these options during abnormal deviations or accidents. The framework can explicitly
highlight the risk and safety benefits of the existing and new safety systems, perform sensitivity
analysis of physical parameters and optimize the value of design variables of the system and
components.
The risk and sensitivity analysis has been demonstrated by leveraging the suggested
framework for analyzing two diverse accident scenarios: Station blackout and medium break loss
of coolant accident. A CDF- based benefit index and included in the framework can assess the risk
margin offered by the existing or the new safety options and provide a mathematical way of
comparing those. Additionally, a new concept design for forced safety injection tanks has been
suggested. The feasibility of this system has been demonstrated by modeling and integrating this
new system into the nuclear power plant model and analyzing the risk benefits. The presented
DPRA framework is further leveraged to optimize the FSIT's design parameters, such as actuation
set-point or the delays between the series operation of FSITs.
Since time is crucial during an accident evolution, a time-based dynamic event importance
index is introduced that supports the operator in deciding how much resources he should invest for
each component to recover the lost safety function. This measure provides reliability and timebased
component importance and ranks them for the operator to allocate resources to restore those
components. This measure is compared with the conventional measures. Then efforts are made to
estimate the cost savings from re-categorizing the safety components from RISC1 and RISC2
safety classification to RISC3 and RISC4, respectively.Ph
LOKE: Linked Open Knowledge Extraction for Automated Knowledge Graph Construction
No full textWhile the potential of Open Information Extraction (Open IE) for Knowledge Graph Construction (KGC) may seem promising, we find that the alignment of Open IE extraction results with existing knowledge graphs to be inadequate. The advent of Large Language Models (LLMs), especially the commercially available OpenAI models, have reset expectations for what is possible with deep learning models and have created a new field called prompt engineering. We investigate the use of GPT models and prompt engineering for knowledge graph construction with the Wikidata knowledge graph to address a similar problem to Open IE, which we call Open Knowledge Extraction (OKE) using an approach we call the Linked Open Knowledge Extractor (LOKE, pronounced like "Loki"). We consider the entity linking task essential to construction of real world knowledge graphs. We merge the CaRB benchmark scoring approach with data from the TekGen dataset for the LOKE task. We then show that a well engineered prompt, paired with a naive entity linking approach (which we call LOKE-GPT), outperforms AllenAI's OpenIE 4 implementation on the OKE task, although it over-generates triples compared to the reference set due to overall triple scarcity in the TekGen set. Through an analysis of entity linkability in the CaRB dataset, as well as outputs from OpenIE 4 and LOKE-GPT, we see that LOKE-GPT and the "silver" TekGen triples show that the task is significantly different in content from OIE, if not structure. Through this analysis and a qualitative analysis of sentence extractions via all methods, we found that LOKE-GPT extractions are of high utility for the KGC task and suitable for use in semi-automated extraction settings
Indoor climate control in multi-unit grid-interactive efficient buildings
No full textDecember 2022School of EngineeringIn this work, we develop a number of energy management policies for building heating,ventilation and cooling (HVAC) systems for satisfying various objectives. Control policies
whose objective is to improve building operations are primarily aimed at economizing indoor
climate control operations. On the other hand, occupant-centric control strategies prioritize
the comfort of individual occupants. In this dissertation, we evaluate the performance of
some of these policies in both simulated and physical setups.
The thermal inertia of buildings, along with the flexibility associated with thermostatically controlled loads (TCLs) allows HVAC systems to be used for grid demand response
(DR). The initial few chapters of this report develop control strategies aimed at minimizing
the operational costs of a building’s HVAC system. We first consider a hydronic HVAC
system that serves multiple units in a residential building to meet their space heating requirements. We determine the optimal power flow to each unit that minimizes the energy
costs (EC) incurred by the building while keeping in consideration the occupants’ thermal
comfort. The building is assumed to participate in a DR program which allows the building
temperatures to deviate from the set-points up to a maximum limit. Despite the complex, non-linear structure of the problem, we show how the optimal solutions can be obtained
efficiently using quadratic programming. Since HVAC systems can run on either electricity
or natural gas, we study the efficacy of the DR regime for both hourly electricity prices and
flat gas prices over the course of 24 hours. We also study the optimal thermal power and
the evolution of unit temperatures for various energy pricing schemes.
Subsequently, we expand the scope of our work to include TCLs in large commercial
buildings. The load profiles of most commercial and industrial consumers are characterized
by brief periods of very high power consumption followed by intervals of lower demand.
To encourage such consumers to flatten their load profiles, power utilities in around the
world often levy a monthly demand charge (DC) on the peak demand measured over brief
intervals. It was seen in the preceding study that a control policy that minimizes EC while
being agnostic to instantaneous power consumption can introduce significant spikes in the
building’s demand patterns. Therefore, we expand our study on hydronic HVAC systems
to consider the joint optimization of EC and the instantaneous peak power of a multi-unit
building that participates in a DR program. We study the power demand patterns resulting
from our proposed control strategy for TCLs, and evaluate its performance for various climate
zones in the US, under both typical and atypical weather conditions. The results show that
depending on the ambient conditions and the tariff structure, our control policy can result
in utility bill savings of up to nearly 19% compared to the baseline. Our power control
strategy was also seen to significantly reduce the instantaneous peak power consumption in
commercial TCLs.
The work summarized hitherto is primarily centered around developing theoretical
control frameworks for building HVAC systems as desired by building operators (BOs).
However, the next part of the research presented in this report develops multiple control
strategies for heating and cooling operations in buildings for meeting the objectives of both
the BO and the building occupants. Moreover, this part of our report develops a prototype
for autonomous temperature management schemes that can be readily deployed in a real-life
shared workspace. Specifically, we study the problem of indoor zone temperature control in
shared workspaces equipped with heterogeneous heating and cooling sources with the goal
of increased energy savings and environment personalization. Shared workspaces typically
witness distinct intervals when they are occupied or are unoccupied. Moreover, these intervals generally follow a fixed schedule which may be known in advance. In this work, we develop control strategies for space heating and cooling operations to achieve the various indoor conditioning objectives for each of these distinct intervals. Specifically, we consider two
contiguous intervals of equal duration where a shared workspace remains unoccupied prior to
hosting a scheduled event, such as a work meeting. For the first interval, when the workspace
is unoccupied, we propose multiple time-bound pre-cooling/pre-heating control strategies for
conditioning the workspace in preparation for a scheduled activity (Phase I). For the second interval, when the workspace is occupied, we propose a separate control strategy which
enhances the thermal comfort of the occupants by harnessing the spatial differentiation of
the thermal environment to satisfy the different temperature preferences of the individuals
(Phase II). Utilizing a physical test-bed, we use data-driven model learning to establish a
relationship between the HVAC control inputs of the indoor space, and the zone temperatures. Next, we present a simple control strategy to achieve the pre-conditioning objective in
Phase I and show that it is less computationally expensive than conventional model predictive control (MPC). For Phase II, we then use a simple, low complexity, quadratic program
to minimize the thermal discomfort experienced by individuals based on their temperature
preferences. The experimental results show that for Phase I, the proposed control policies
can save a significant amount of energy and achieve the desired mean temperature in the
space fairly accurately. We further note that for Phase II, the control scheme can achieve a
significant spatial differentiation in temperature towards satisfying the occupants’ thermal
preferences.
Occupant well-being requires not only efficient indoor thermal management but also
indoor air quality (IAQ) management. Control strategies aimed at improving the efficiency
of HVAC systems while enhancing occupant wellness must jointly optimize ventilation as
well as heating and cooling operations. The final direction of this work studies the problem
of minimizing the energy consumption of the HVAC system in a multi-unit building, while
meeting thermal comfort and IAQ requirements. Here, we use zonal carbon dioxide (CO2)
to be an indicator for IAQ in individual zones. We first perform a steady state analysis of
the zonal CO2 concentration and the temperature dynamics. The resulting expressions are
convex in the zonal mass flow rates and zonal temperatures. Guided by the steady state
solutions for meeting the thermal comfort constraints, we develop two control policies for
improving the energy efficiency of building HVAC systems while jointly satisfying indoor
temperature and IAQ constraints. We compare the performance of our proposed approaches with those of multiple baseline approaches which implement separate regimes for managing
zonal temperature and IAQ for a typical work-day in a multi-zone campus building. We have
evaluated the performance of our proposed approaches under varying levels of flexibility in
zonal temperatures. Our proposed approaches were seen to offer potential savings of nearly
29% compared to the baseline.
In the closing chapter of this report, we offer some remarks pertaining to the results
obtained from the aforementioned studies. We conclude this report by proposing possible
extensions to this work.Ph
Learning generalizable representations through compression
No full textDecember 2022School of Humanities, Arts, and Social SciencesThe ability of humans and other animals to generalize from their past experiences to novel situations is at the heart of intelligent behavior. Generalization is more likely to occur between objects that share some similarities, either perceptually or functionally. The functional- similarity-based generalization is often summarized as acquired equivalence (AE). AE has been ubiquitously documented in humans and animals, but its nature is still not well understood. Here, we propose an interpretation of the AE phenomenon by postulating that this generalization actually reflects the processes of representation compression. We formalize a representation compression framework on the basis of the rate-distortion theory (RD), a branch of information theory that characterizes a fundamental tradeoff between the complexity and accuracy of information processing. A reinforcement learning model derived from the RD theory was able to replicate human functional-similarity-based generalization. The model worked reasonably well in capturing human learning and generalization behaviors, even in an extended AE experiment paradigm where perceptual (visual) similarity was incorporated. We also identified from the model a set of low-level cognitive mechanisms (categorization and selective attention) proposed in the current AE theories to underlie the generalization in the AE task. We conclude that the representation compression framework provides a unified explanation of human AE.Ph
Examination of photophysical characteristics of perylene diimide chromophores and their potential application in electro-optical technologies
No full textMay 2017School of Science1,6-, 1,7-, and 1,6,7- derivatives of dodecylthio–N,N’– (2,4–diisopropylphenyl)-3,4,9,10-perylenetetracarboxylic diimide, and N,N’-Di(4-ethynylphenyl)-1,7-di(4-tert-butylphenoxy)-3,4:9,10-perylenebis-(dicarboximide) (PhO-Ph-PDI) were synthesized, isolated, and characterized. The three Thio-PDI derivatives, 1,6–Thio–PDI, 1,7–Thio–PDI, and 1,6,7–Thio–PDI, displayed noticeable differences in their photophysical properties including their absorption and emission spectra, fluorescence quantum yield, fluorescence excited state lifetimes, and excited state dipole moments as calculated by the Lippert-Mataga analysis. Additionally, the Thio-PDI derivatives exhibited different colors at neutral and reduced state as determined by chemical reduction and CIE calculations. These studies determine that different PDI derivatives can provide unique photophysical contributions as building blocks within the molecular assemblies which comprise new technologies such as electrochromic (EC) devices. Comparatively, PhO-Ph-PDI was utilized as the primary building block within a molecular assembly for potential use in EC materials. A molecular assembly, or thin-film, of PhO-Ph-PDI was fabricated Copper Azide-Alkyne Cycloaddition reactions. The PhO-Ph-PDI thin-film was studied with TBAPF6 and TMeAPF6 in various solvents to study electrolyte size penetration within the PhO-Ph-PDI thin-film. Different reduced states were probed with cyclic voltammetry studies, spectroelectrochemical studies, and potential step spectroelectrochemical studies. These studies revealed that a combination of a smaller TMeAPF6 electrolyte and a longer applied potential allowed for better cation penetration through the PhO-Ph-PDI thin-film. This better penetration of the thin-film led to higher conversion of reduced states which is an essential criteria of EC devices. While the PhO-Ph-PDI thin-film did not achieve full conversion to all reduced states with either electrolyte, analysis of the PhO-Ph-PDI thin-film provides better understanding of the criteria required to improve molecular assemblies for EC devices. This includes selection of appropriate sized electrolyte and synthetic design of molecular assembly channel size by appropriate selection of chromophores utilized within the molecular assembly.Ph