Browsing by Author "Kong, Zhenyu"

Now showing 1 - 20 of 43
  • Thumbnail Image
    3D Massive MIMO and Artificial Intelligence for Next Generation Wireless Networks
    Shafin, Rubayet (Virginia Tech, 2020-04-13)
    3-dimensional (3D) massive multiple-input-multiple-output (MIMO)/full dimensional (FD) MIMO and application of artificial intelligence are two main driving forces for next generation wireless systems. This dissertation focuses on aspects of channel estimation and precoding for 3D massive MIMO systems and application of deep reinforcement learning (DRL) for MIMO broadcast beam synthesis. To be specific, downlink (DL) precoding and power allocation strategies are identified for a time-division-duplex (TDD) multi-cell multi-user massive FD-MIMO network. Utilizing channel reciprocity, DL channel state information (CSI) feedback is eliminated and the DL multi-user MIMO precoding is linked to the uplink (UL) direction of arrival (DoA) estimation through estimation of signal parameters via rotational invariance technique (ESPRIT). Assuming non-orthogonal/non-ideal spreading sequences of the UL pilots, the performance of the UL DoA estimation is analytically characterized and the characterized DoA estimation error is incorporated into the corresponding DL precoding and power allocation strategy. Simulation results verify the accuracy of our analytical characterization of the DoA estimation and demonstrate that the introduced multi-user MIMO precoding and power allocation strategy outperforms existing zero-forcing based massive MIMO strategies. In 3D massive MIMO systems, especially in TDD mode, a base station (BS) relies on the uplink sounding signals from mobile stations to obtain the spatial information for downlink MIMO processing. Accordingly, multi-dimensional parameter estimation of MIMO channel becomes crucial for such systems to realize the predicted capacity gains. In this work, we also study the joint estimation of elevation and azimuth angles as well as the delay parameters for 3D massive MIMO orthogonal frequency division multiplexing (OFDM) systems under a parametric channel modeling. We introduce a matrix-based joint parameter estimation method, and analytically characterize its performance for massive MIMO OFDM systems. Results show that antenna array configuration at the BS plays a critical role in determining the underlying channel estimation performance, and the characterized MSEs match well with the simulated ones. Also, the joint parametric channel estimation outperforms the MMSEbased channel estimation in terms of the correlation between the estimated channel and the real channel. Beamforming in MIMO systems is one of the key technologies for modern wireless communication. Creating wide common beams are essential for enhancing the coverage of cellular network and for improving the broadcast operation for control signals. However, in order to maximize the coverage, patterns for broadcast beams need to be adapted based on the users' movement over time. In this dissertation, we present a MIMO broadcast beam optimization framework using deep reinforcement learning. Our proposed solution can autonomously and dynamically adapt the MIMO broadcast beam parameters based on user' distribution in the network. Extensive simulation results show that the introduced algorithm can achieve the optimal coverage, and converge to the oracle solution for both single cell and multiple cell environment and for both periodic and Markov mobility patterns.
  • Thumbnail Image
    Advanced Data Analytics for Quality Assurance of Smart Additive Manufacturing
    Shen, Bo (Virginia Tech, 2022-07-07)
    Additive manufacturing (AM) is a powerful emerging technology for fabricating components with complex geometries using a variety of materials. However, despite the promising potential, due to the complexity of the process dynamics, how to ensure product quality and consistency of AM parts efficiently during the process remains challenging. Therefore, this dissertation aims to develop advanced machine learning methods for online process monitoring and quality assurance of smart additive manufacturing. Driven by edge computing, the Industrial Internet of Things (IIoT), sensors and other smart technologies, data collection, communication, analytics, and control are infiltrating every aspect of manufacturing. The data provides excellent opportunities to improve and revolutionize manufacturing for both quality and productivity. Despite the massive volume of data generated during a very short time, approximately 90 percent of data gets wasted or unused. The goal of sensing and data analytics for advanced manufacturing is to capture the full insight that data and analytics can discover to help address the most pressing problems. To achieve the above goal, several data-driven approaches have been developed in this dissertation to achieve effective data preprocessing, feature extraction, and inverse design. We also develop related theories for these data-driven approaches to guarantee their performance. The performances have been validated using sensor data from AM processes. Specifically, four new methodologies are proposed and implemented as listed below: 1. To make the unqualified thermal data meet the spatial and temporal resolution requirement of microstructure prediction, a super resolution for multi-sources image stream data using smooth and sparse tensor completion is proposed and applied to data acquisition of additive manufacturing. The qualified thermal data is able to extract useful information like boundary velocity, thermal gradient, etc. 2. To effectively extract features for high dimensional data with limited samples, a clustered discriminant regression is created for classification problems in healthcare and additive manufacturing. The proposed feature extraction method together with classic classifiers can achieve better classification performance than the convolutional neural network for image classification. 3. To extract the melt pool information from the processed X-ray video in metal AM process, a smooth sparse Robust Tensor Decomposition model is devised to decompose the data into the static background, smooth foreground, and noise, respectively. The proposed method exhibits superior performance in extracting the melt pool information on X-ray data. 4. To learn the material property for different printing settings, a multi-task Gaussian process upper confidence bound is developed for the sequential experiment design, where a no-regret algorithm is implemented. The proposed algorithm aims to learn the optimal material property for different printing settings. By fully utilizing the sensor data with innovative data analytics, the above-proposed methodologies are used to perform interdisciplinary research, promote technical innovations, and achieve balanced theoretical/practical advancements. In addition, these methodologies are inherently integrated into a generic framework. Thus, they can be easily extended to other manufacturing processes, systems, or even other application areas such as healthcare systems.
  • Thumbnail Image
    Advanced Machine Learning for Surrogate Modeling in Complex Engineering Systems
    Lee, Cheol Hei (Virginia Tech, 2023-08-02)
    Surrogate models are indispensable in the analysis of engineering systems. The quality of surrogate models is determined by the data quality and the model class but achieving a high standard of them is challenging in complex engineering systems. Heterogeneity, implicit constraints, and extreme events are typical examples of the factors that complicate systems, yet they have been underestimated or disregarded in machine learning. This dissertation is dedicated to tackling the challenges in surrogate modeling of complex engineering systems by developing the following machine learning methodologies. (i) Partitioned active learning partitions the design space according to heterogeneity in response features, thereby exploiting localized models to measure the informativeness of unlabeled data. (ii) For the systems with implicit constraints, failure-averse active learning incorporates constraint outputs to estimate the safe region and avoid undesirable failures in learning the target function. (iii) The multi-output extreme spatial learning enables modeling and simulating extreme events in composite fuselage assembly. The proposed methods were applied to real-world case studies and outperformed benchmark methods.
  • Thumbnail Image
    Advancing the Utility of Manufacturing Data for Modeling, Monitoring, and Securing Machining Processes
    Shafae, Mohammed Saeed Abuelmakarm (Virginia Tech, 2018-08-23)
    The growing adoption of smart manufacturing systems and its related technologies (e.g., embedded sensing, internet-of-things, cyber-physical systems, big data analytics, and cloud computing) is promising a paradigm shift in the manufacturing industry. Such systems enable extracting and exchanging actionable knowledge across the different entities of the manufacturing cyber-physical system and beyond. From a quality control perspective, this allows for more opportunities to realize proactive product design; real-time process monitoring, diagnosis, prognosis, and control; and better product quality characterization. However, a multitude of challenges are arising, with the growing adoption of smart manufacturing, including industrial data characterized by increasing volume, velocity, variety, and veracity, as well as the security of the manufacturing system in the presence of growing connectivity. Taking advantage of these emerging opportunities and tackling the upcoming challenges require creating novel quality control and data analytics methods, which not only push the boundaries of the current state-of-the-art research, but discover new ways to analyze the data and utilize it. One of the key pillars of smart manufacturing systems is real-time automated process monitoring, diagnosis, and control methods for process/product anomalies. For machining applications, traditionally, deterioration in quality measures may occur due to a variety of assignable causes of variation such as poor cutting tool replacement decisions and inappropriate choice cutting parameters. Additionally, due to increased connectivity in modern manufacturing systems, process/product anomalies intentionally induced through malicious cyber-attacks -- aiming at degrading the process performance and/or the part quality -- is becoming a growing concern in the manufacturing industry. Current methods for detecting and diagnosing traditional causes of anomalies are primarily lab-based and require experts to perform initial set-ups and continual fine-tuning, reducing the applicability in industrial shop-floor applications. As for efforts accounting for process/product anomalies due cyber-attacks, these efforts are in early stages. Therefore, more foundational research is needed to develop a clear understanding of this new type of cyber-attacks and their effects on machining processes, to ensure smart manufacturing security both on the cyber and the physical levels. With primary focus on machining processes, the overarching goal of this dissertation work is to explore new ways to expand the use and value of manufacturing data-driven methods for better applicability in industrial shop-floors and increased security of smart manufacturing systems. As a first step toward achieving this goal, the work in this dissertation focuses on adopting this goal in three distinct areas of interest: (1) Statistical Process Monitoring of Time-Between-Events Data (e.g., failure-time data); (2) Defending against Product-Oriented Cyber-Physical Attacks on Intelligent Machining Systems; and (3) Modeling Machining Process Data: Time Series vs. Spatial Point Cloud Data Structures.
  • Thumbnail Image
    Analysis of Information Diffusion through Social Media
    Khalili, Nastaran (Virginia Tech, 2021-06-16)
    The changes in the course of communication changed the world from different perspectives. Public participation on social media means the generation, diffusion, and exposure to a tremendous amount of user-generated content without supervision. This four-essay dissertation analyzes information diffusion through social media and its opportunities and challenges through management systems engineering and data analytics. First, we evaluate how information can be shared to reach maximum exposure for the case on online petitions. We use system dynamics modeling and propose policies for campaign managers to schedule the reminders they send to have the highest number of petition signatures. We find that sending reminders is more effective in the case of increasing the signature rate. In the second essay, we investigate how people build trust/ mistrust in science during an emergency. We use data analytics methods on more than 700,000 tweets containing keywords of Hydroxychloroquine and chloroquine, two candidate medicines, to prevent and cure patients infected with COVID-19. We show that people's opinions are concentrated in the case of polarity and spread out in the case of subjectivity. Also, they tend to share subjective tweets than objective ones. In the third essay, building on the same dataset as essay two, we study the changes in science communication during the coronavirus pandemic. We used topic modeling and clustered the tweets into seven different groups. Our analysis suggests that a highly scientific and health-related subject can become political in the case of an emergency. We found that the groups of medical information and research and study have fewer tweets than the political one. Fourth, we investigated fake news diffusion as one of the main challenges of user-generated content. We built a system dynamics model and analyzed the effects of competition and correction in combating fake news. We show that correction of misinformation and competition in fake news needs a high percentage of participation to be effective enough to deal with fake news.
  • Thumbnail Image
    Applications of Motor Variability for Assessing Repetitive Occupational Tasks
    Sedighi, Alireza (Virginia Tech, 2017-06-07)
    The human body has substantial kinetic and kinematic degrees-of-freedoms, so redundant solutions are available for the central nervous system (CNS) to perform a repetitive task. Due to these redundancies, inherent variations exist in human movement, called motor variability (MV). Current evidence suggests that MV can be beneficial, and that there is an inverse association between MV and risk of injury. To better understand how the CNS manipulates MV to reduce injury risks, we investigated the effects of individual differences, task-relevant aspects, and psychological factors as modifiers of MV. Earlier work found that experienced workers adapted more stable movements than novices in repetitive lifting tasks. To expand on this, we quantified how MV differs between experienced workers and novices in different lifting conditions (i.e., lifting asymmetry and fatigue). Three different measures (cycle-to-cycle SD, sample entropy, and the goal equivalent manifold) were used to quantify MV. In a symmetric lifting task, experienced workers had more constrained movement than novices, and experienced workers exhibited more consistent behavior in the asymmetric condition. Novices constrained their movements, and could not maintain the same level of variability in the asymmetric condition. We concluded that experienced workers adapt stable or flexible strategies depending on task difficulty. In a prolonged lifting task, both groups increased their MV to adapt to fatigue; they particularly increased variability in a direction that had no effects on their main task goal. Developing fatigue also makes it difficult for individuals maintain the main goal. Based on these results, we conclude that increasing variability is an adaptive strategy in response to fatigue. We also assessed variability in gait parameters to compare gait adaptability using a head-worn display (HWD) compared with head-down displays for visual information presentation. An effective strategy we observed for performing a cognitive task successfully during walking was to increase gait variability in the goal direction. In addition, we found that head-up walking had smaller effects on MV, suggesting that HWDs are a promising technology to reduce adverse events during gait (e.g., falls). In summary, these results suggest that MV can be a useful indicator for evaluating some occupational injury risks.
  • Thumbnail Image
    Behavioral Monitoring to Identify Self-Injurious Behavior among Children with Autism Spectrum Disorder
    Garside, Kristine Dianne Cantin (Virginia Tech, 2019-03-25)
    Self-injurious behavior (SIB) is one of the most dangerous behavioral responses among individuals with autism spectrum disorder (ASD), often leading to injury and hospitalization. There is an ongoing need to measure the triggers of SIB to inform management and prevention. These triggers are determined traditionally through clinical observations of the child with SIB, often involving a functional assessment (FA), which is methodologically documenting responses to stimuli (e.g., environmental or social) and recording episodes of SIB. While FA has been a "gold standard" for many years, it is costly, tedious, and often artificial (e.g., in controlled environments). If performed in a naturalistic environment, such as the school or home, caregivers are responsible for tracking behaviors. FA in naturalistic environments relies on caregiver and patient compliance, such as responding to prompts or recalling past events. Recent technological developments paired with classification methods may help decrease the required tracking efforts and support management plans. However, the needs of caregivers and individuals with ASD and SIB should be considered before integrating technology into daily routines, particularly to encourage technology acceptance and adoption. To address this, the perspectives of SIB management and technology were first collected to support future technology design considerations (Chapter 2). Accelerometers were then selected as a specific technology, based on caregiver preferences and reported preferences of individuals with ASD, and were used to collect movement data for classification (Chapter 3). Machine learning algorithms with featureless data were explored, resulting in individual-level models that demonstrated high accuracy (up to 99%) in detecting and classifying SIB. Group-level classifiers could provide more generalizable models for efficient SIB monitoring, though the highly variable nature of both ASD and SIB can preclude accurate detection. A multi-level regression model (MLR) was implemented to consider such individual variability (Chapter 4). Both linear and nonlinear measures of motor variability were assessed as potential predictors in the model. Diverse classification methods were used (as in Chapter 3), and MLR outperformed other group level classifiers (accuracy ~75%). Findings from this research provide groundwork for a future smart SIB monitoring system. There are clear implications for such monitoring methods in prevention and treatment, though additional research is required to expand the developed models. Such models can contribute to the goal of alerting caregivers and children before SIB occurs, and teaching children to perform another behavior when alerted.
  • Thumbnail Image
  • Thumbnail Image
    Closed-loop Tool Path Planning for Non-planar Additive Manufacturing and Sensor-based Inspection on Stationary and Moving Freeform Objects
    Kucukdeger, Ezgi (Virginia Tech, 2022-06-03)
    Additive manufacturing (AM) has received much attention from researchers over the past decades because of its diverse applications in various industries. AM is an advanced manufacturing process that facilitates the fabrication of complex geometries represented by computer-aided design (CAD) models. Traditionally, designed parts are fabricated by extruding material layer-by-layer using a tool path planning obtained from slicing programs by using CAD models as an input. Recently, there has been a growing interest in non-planar AM technologies, which offer the ability to fabricate multilayer constructs conforming to freeform surfaces. Non-planar AM processes have been utilized in various applications and involved objects of varying material properties and geometric characteristics. Although the current state of the art suggests AM can provide novel opportunities in conformal manufacturing, several challenges remain to be addressed. The identified challenges in non-planar AM fall into three categories: 1) conformal 3D printing on substrates with complex topography of which CAD model representation is not readily available, 2) understanding the relationship between the tool path planning and the quality of the 3D-printed construct, and 3) conformal 3D printing in the presence of mechanical disturbances. An open-loop non-planar tool path planning algorithm based on point cloud representations of object geometry and a closed-loop non-planar tool path planning algorithm based on position sensing were proposed to address these limitations and enable conformal 3D printing and spatiotemporal 3D sensing on objects of near-arbitrary organic shape. Three complementary studies have been completed towards the goal of improving the conformal tool path planning capabilities in various applications including fabrication of conformal electronics, in situ bioprinting, and spatiotemporal biosensing: i. A non-planar tool path planning algorithm for conformal microextrusion 3D printing based on point cloud data representations of object geometry was presented. Also, new insights into the origin of common conformal 3D printing defects, including tool-surface contact, were provided. The impact and utility of the proposed conformal microextrusion 3D printing process was demonstrated by the fabrication of 3D spiral and Hilbert-curve loop antennas on various non-planar substrates, including wrinkled and folded Kapton films and origami. ii. A new method for closed-loop controlled 3D printing on moving substrates, objects, and unconstrained human anatomy via real-time object position sensing was proposed. Monitoring of the tool position via real-time sensing of nozzle-surface offset using 1D laser displacement sensors enabled conformal 3D printing on moving substrates and objects. The proposed control strategy was demonstrated by microextrusion 3D printing on oscillating substrates and in situ bioprinting on an unconstrained human hand. iii. A reverse engineering-driven collision-free path planning program for automated inspection of macroscale biological specimens, such as tissue-based products and organs, was proposed. The path planning program for impedance-based spatiotemporal biosensing was demonstrated by the characterization of meat and fruit tissues using two impedimetric sensors: a cantilever sensor and a multifunctional fiber sensor.
  • Thumbnail Image
    Compressive Sensing Approaches for Sensor based Predictive Analytics in Manufacturing and Service Systems
    Bastani, Kaveh (Virginia Tech, 2016-03-14)
    Recent advancements in sensing technologies offer new opportunities for quality improvement and assurance in manufacturing and service systems. The sensor advances provide a vast amount of data, accommodating quality improvement decisions such as fault diagnosis (root cause analysis), and real-time process monitoring. These quality improvement decisions are typically made based on the predictive analysis of the sensor data, so called sensor-based predictive analytics. Sensor-based predictive analytics encompasses a variety of statistical, machine learning, and data mining techniques to identify patterns between the sensor data and historical facts. Given these patterns, predictions are made about the quality state of the process, and corrective actions are taken accordingly. Although the recent advances in sensing technologies have facilitated the quality improvement decisions, they typically result in high dimensional sensor data, making the use of sensor-based predictive analytics challenging due to their inherently intensive computation. This research begins in Chapter 1 by raising an interesting question, whether all these sensor data are required for making effective quality improvement decisions, and if not, is there any way to systematically reduce the number of sensors without affecting the performance of the predictive analytics? Chapter 2 attempts to address this question by reviewing the related research in the area of signal processing, namely, compressive sensing (CS), which is a novel sampling paradigm as opposed to the traditional sampling strategy following the Shannon Nyquist rate. By CS theory, a signal can be reconstructed from a reduced number of samples, hence, this motivates developing CS based approaches to facilitate predictive analytics using a reduced number of sensors. The proposed research methodology in this dissertation encompasses CS approaches developed to deliver the following two major contributions, (1) CS sensing to reduce the number of sensors while capturing the most relevant information, and (2) CS predictive analytics to conduct predictive analysis on the reduced number of sensor data. The proposed methodology has a generic framework which can be utilized for numerous real-world applications. However, for the sake of brevity, the validity of the proposed methodology has been verified with real sensor data associated with multi-station assembly processes (Chapters 3 and 4), additive manufacturing (Chapter 5), and wearable sensing systems (Chapter 6). Chapter 7 summarizes the contribution of the research and expresses the potential future research directions with applications to big data analytics.
  • Thumbnail Image
    Computational Simulation and Machine Learning for Quality Improvement in Composites Assembly
    Lutz, Oliver Tim (Virginia Tech, 2023-08-22)
    In applications spanning across aerospace, marine, automotive, energy, and space travel domains, composite materials have become ubiquitous because of their superior stiffness-to-weight ratios as well as corrosion and fatigue resistance. However, from a manufacturing perspective, these advanced materials have introduced new challenges that demand the development of new tools. Due to the complex anisotropic and nonlinear material properties, composite materials are more difficult to model than conventional materials such as metals and plastics. Furthermore, there exist ultra-high precision requirements in safety critical applications that are yet to be reliably met in production. Towards developing new tools addressing these challenges, this dissertation aims to (i) build high-fidelity numerical simulations of composite assembly processes, (ii) bridge these simulations to machine learning tools, and (iii) apply data-driven solutions to process control problems while identifying and overcoming their shortcomings. This is accomplished in case studies that model the fixturing, shape control, and fastening of composite fuselage components. Therein, simulation environments are created that interact with novel implementations of modified proximal policy optimization, based on a newly developed reinforcement learning algorithm. The resulting reinforcement learning agents are able to successfully address the underlying optimization problems that underpin the process and quality requirements.
  • Thumbnail Image
    Conformal Additive Manufacturing for Organ Interface
    Singh, Manjot (Virginia Tech, 2017-06-08)
    The inability to monitor the molecular trajectories of whole organs throughout the clinically relevant ischemic interval is a critical problem underlying the organ shortage crisis. Here, we report a novel technique for fabricating manufacturing conformal microfluidic devices for organ interface. 3D conformal printing was leveraged to engineer and fabricate novel organ-conforming microfluidic devices that endow the interface between microfluidic channels and the organ cortex. Large animal studies reveal microfluidic biopsy samples contain rich diagnostic information, including clinically relevant biomarkers of ischemic pathophysiology. Overall, these results suggest microfluidic biopsy via 3D printed organ-conforming microfluidic devices could shift the paradigm for whole organ preservation and assessment, thereby relieving the organ shortage crisis through increased availability and quality of donor organs.
  • Thumbnail Image
    Coverage, Secrecy, and Stability Analysis of Energy Harvesting Wireless Networks
    Kishk, Mustafa (Virginia Tech, 2018-08-03)
    Including energy harvesting capability in a wireless network is attractive for multiple reasons. First and foremost, powering base stations with renewable resources could significantly reduce their reliance on the traditional energy sources, thus helping in curtailing the carbon footprint. Second, including this capability in wireless devices may help in increasing their lifetime, which is especially critical for devices for which it may not be easy to charge or replace batteries. This will often be the case for a large fraction of sensors that will form the {em digital skin} of an Internet of Things (IoT) ecosystem. Motivated by these factors, this work studies fundamental performance limitations that appear due to the inherent unreliability of energy harvesting when it is used as a primary or secondary source of energy by different elements of the wireless network, such as mobile users, IoT sensors, and/or base stations. The first step taken towards this objective is studying the joint uplink and downlink coverage of radio-frequency (RF) powered cellular-based IoT. Modeling the locations of the IoT devices and the base stations (BSs) using two independent Poisson point processes (PPPs), the joint uplink/downlink coverage probability is derived. The resulting expressions characterize how different system parameters impact coverage performance. Both mathematical expressions and simulation results show how these system parameters should be tuned in order to achieve the performance of the regularly powered IoT (IoT devices are powered by regular batteries). The placement of RF-powered devices close to the RF sources, to harvest more energy, imposes some concerns on the security of the signals transmitted by these RF sources to their intended receivers. Studying this problem is the second step taken in this dissertation towards better understanding of energy harvesting wireless networks. While these secrecy concerns have been recently addressed for the point-to-point link, it received less attention for the more general networks with randomly located transmitters (RF sources) and RF-powered devices, which is the main contribution in the second part of this dissertation. In the last part of this dissertation, we study the stability of solar-powered cellular networks. We use tools from percolation theory to study percolation probability of energy-drained BSs. We study the effect of two system parameters on that metric, namely, the energy arrival rate and the user density. Our results show the existence of a critical value for the ratio of the energy arrival rate to the user density, above which the percolation probability is zero. The next step to further improve the accuracy of the stability analysis is to study the effect of correlation between the battery levels at neighboring BSs. We provide an initial study that captures this correlation. The main insight drawn from our analysis is the existence of an optimal overlapping coverage area for neighboring BSs to serve each other's users when they are energy-drained.
  • Thumbnail Image
    Data Analytics for Statistical Learning
    Komolafe, Tomilayo A. (Virginia Tech, 2019-02-05)
    The prevalence of big data has rapidly changed the usage and mechanisms of data analytics within organizations. Big data is a widely-used term without a clear definition. The difference between big data and traditional data can be characterized by four Vs: velocity (speed at which data is generated), volume (amount of data generated), variety (the data can take on different forms), and veracity (the data may be of poor/unknown quality). As many industries begin to recognize the value of big data, organizations try to capture it through means such as: side-channel data in a manufacturing operation, unstructured text-data reported by healthcare personnel, various demographic information of households from census surveys, and the range of communication data that define communities and social networks. Big data analytics generally follows this framework: first, a digitized process generates a stream of data, this raw data stream is pre-processed to convert the data into a usable format, the pre-processed data is analyzed using statistical tools. In this stage, called statistical learning of the data, analysts have two main objectives (1) develop a statistical model that captures the behavior of the process from a sample of the data (2) identify anomalies in the process. However, several open challenges still exist in this framework for big data analytics. Recently, data types such as free-text data are also being captured. Although many established processing techniques exist for other data types, free-text data comes from a wide range of individuals and is subject to syntax, grammar, language, and colloquialisms that require substantially different processing approaches. Once the data is processed, open challenges still exist in the statistical learning step of understanding the data. Statistical learning aims to satisfy two objectives, (1) develop a model that highlights general patterns in the data (2) create a signaling mechanism to identify if outliers are present in the data. Statistical modeling is widely utilized as researchers have created a variety of statistical models to explain everyday phenomena such as predicting energy usage behavior, traffic patterns, and stock market behaviors, among others. However, new applications of big data with increasingly varied designs present interesting challenges. Consider the example of free-text analysis posed above. There's a renewed interest in modeling free-text narratives from sources such as online reviews, customer complaints, or patient safety event reports, into intuitive themes or topics. As previously mentioned, documents describing the same phenomena can vary widely in their word usage and structure. Another recent interest area of statistical learning is using the environmental conditions that people live, work, and grow in, to infer their quality of life. It is well established that social factors play a role in overall health outcomes, however, clinical applications of these social determinants of health is a recent and an open problem. These examples are just a few of many examples wherein new applications of big data pose complex challenges requiring thoughtful and inventive approaches to processing, analyzing, and modeling data. Although a large body of research exists in the area of anomaly detection increasingly complicated data sources (such as side-channel related data or network-based data) present equally convoluted challenges. For effective anomaly-detection, analysts define parameters and rules, so that when large collections of raw data are aggregated, pieces of data that do not conform are easily noticed and flagged. In this work, I investigate the different steps of the data analytics framework and propose improvements for each step, paired with practical applications, to demonstrate the efficacy of my methods. This paper focuses on the healthcare, manufacturing and social-networking industries, but the materials are broad enough to have wide applications across data analytics generally. My main contributions can be summarized as follows: • In the big data analytics framework, raw data initially goes through a pre-processing step. Although many pre-processing techniques exist, there are several challenges in pre-processing text data and I develop a pre-processing tool for text data. • In the next step of the data analytics framework, there are challenges in both statistical modeling and anomaly detection o I address the research area of statistical modeling in two ways: - There are open challenges in defining models to characterize text data. I introduce a community extraction model that autonomously aggregates text documents into intuitive communities/groups - In health care, it is well established that social factors play a role in overall health outcomes however developing a statistical model that characterizes these relationships is an open research area. I developed statistical models for generalizing relationships between social determinants of health of a cohort and general medical risk factors o I address the research area of anomaly detection in two ways: - A variety of anomaly detection techniques exist already, however, some of these methods lack a rigorous statistical investigation thereby making them ineffective to a practitioner. I identify critical shortcomings to a proposed network based anomaly detection technique and introduce methodological improvements - Manufacturing enterprises which are now more connected than ever are vulnerably to anomalies in the form of cyber-physical attacks. I developed a sensor-based side-channel technique for anomaly detection in a manufacturing process
  • Thumbnail Image
    Data-driven Methods in Mechanical Model Calibration and Prediction for Mesostructured Materials
    Kim, Jee Yun (Virginia Tech, 2018-10-01)
    Mesoscale design involving control of material distribution pattern can create a statistically heterogeneous material system, which has shown increased adaptability to complex mechanical environments involving highly non-uniform stress fields. Advances in multi-material additive manufacturing can aid in this mesoscale design, providing voxel level control of material property. This vast freedom in design space also unlocks possibilities within optimization of the material distribution pattern. The optimization problem can be divided into a forward problem focusing on accurate predication and an inverse problem focusing on efficient search of the optimal design. In the forward problem, the physical behavior of the material can be modeled based on fundamental mechanics laws and simulated through finite element analysis (FEA). A major limitation in modeling is the unknown parameters in constitutive equations that describe the constituent materials; determining these parameters via conventional single material testing has been proven to be insufficient, which necessitates novel and effective approaches of calibration. A calibration framework based in Bayesian inference, which integrates data from simulations and physical experiments, has been applied to a study involving a mesostructured material fabricated by fused deposition modeling. Calibration results provide insights on what values these parameters converge to as well as which material parameters the model output has the largest dependence on while accounting for sources of uncertainty introduced during the modeling process. Additionally, this statistical formulation is able to provide quick predictions of the physical system by implementing a surrogate and discrepancy model. The surrogate model is meant to be a statistical representation of the simulation results, circumventing issues arising from computational load, while the discrepancy is aimed to account for the difference between the simulation output and physical experiments. In this thesis, this Bayesian calibration framework is applied to a material bending problem, where in-situ mechanical characterization data and FEA simulations based on constitutive modeling are combined to produce updated values of the unknown material parameters with uncertainty.
  • Thumbnail Image
    Deep Reinforcement Learning for Next Generation Wireless Networks with Echo State Networks
    Chang, Hao-Hsuan (Virginia Tech, 2021-08-26)
    This dissertation considers a deep reinforcement learning (DRL) setting under the practical challenges of real-world wireless communication systems. The non-stationary and partially observable wireless environments make the learning and the convergence of the DRL agent challenging. One way to facilitate learning in partially observable environments is to combine recurrent neural network (RNN) and DRL to capture temporal information inherent in the system, which is referred to as deep recurrent Q-network (DRQN). However, training DRQN is known to be challenging requiring a large amount of training data to achieve convergence. In many targeted wireless applications in the 5G and future 6G wireless networks, the available training data is very limited. Therefore, it is important to develop DRL strategies that are capable of capturing the temporal correlation of the dynamic environment that only requires limited training overhead. In this dissertation, we design efficient DRL frameworks by utilizing echo state network (ESN), which is a special type of RNNs where only the output weights are trained. To be specific, we first introduce the deep echo state Q-network (DEQN) by adopting ESN as the kernel of deep Q-networks. Next, we introduce federated ESN-based policy gradient (Fed-EPG) approach that enables multiple agents collaboratively learn a shared policy to achieve the system goal. We designed computationally efficient training algorithms by utilizing the special structure of ESNs, which have the advantage of learning a good policy in a short time with few training data. Theoretical analyses are conducted for DEQN and Fed-EPG approaches to show the convergence properties and to provide a guide to hyperparameter tuning. Furthermore, we evaluate the performance under the dynamic spectrum sharing (DSS) scenario, which is a key enabling technology that aims to utilize the precious spectrum resources more efficiently. Compared to a conventional spectrum management policy that usually grants a fixed spectrum band to a single system for exclusive access, DSS allows the secondary system to dynamically share the spectrum with the primary system. Our work sheds light on the real deployments of DRL techniques in next generation wireless systems.
  • Thumbnail Image
    Design and Fabrication of Piezoelectric Sensors and Actuators for Characterization of Soft Materials
    Cesewski, Ellen (Virginia Tech, 2020-08-27)
    The research presented in this dissertation supports the overall goal of creating piezoelectric measurement technology for the analysis and characterization of soft materials that serve as feedstocks (inputs) and products (outputs) of emerging biomanufacturing processes, including cell and additive biomanufacturing processes. The first objective was to define measurement challenges associated with real-time monitoring of material compositional profiles using biosensors in practical biomanufacturing and bioprocessing formats, as insight into a material's composition (i.e., concentration of a given biologic within a material or product) provides molecular-scale insight into processes and product quality. The second objective was to design, fabricate, and characterize continuous flow cell separation technology based on 3D printed self-exciting and -sensing millimeter-scale piezoelectric transducers and microfluidic networks for separation and characterization of expanded therapeutic cells. The third objective was to establish a sensor-based characterization approach for viscoelastic properties of hydrogels and gelation processes using high-order modes of piezoelectric-excited millimeter cantilever (PEMC) sensors and understand the influence of cantilever mode number on critical sensor characteristics, including sensitivity, dynamic range, and limit of detection. The first objective was addressed through a comprehensive review of recent progress in electrochemical and hybrid biosensors, which included discussions of measurement formats, sensor performance, and measurement challenges associated with use in practical bioprocessing environments. This critical review revealed that cost, disposability, form factor, complex measurement matrices, multiplexing, and sensor regeneration/reusability are among the most pressing challenges that require solutions through advancement of sensor design and manufacturing approaches before biosensors can facilitate high-confidence long-term continuous bioprocess monitoring. The second objective was addressed by creating a microextrusion-based additive manufacturing approach for fabrication of piezoelectric-based MEMS devices that enabled integration of 3D configurations of piezoelectric transducers and microfluidic networks in a one-pot manufacturing process. The devices contained orthogonal out-of-plane piezoelectric sensors and actuators and generated tunable bulk acoustic waves (BAWs) capable of size-selective manipulation, trapping, and separation of suspended particles in droplets and microchannels. This work suggests that additive manufacturing potentially provides new opportunities for the fabrication of sensor-integrated microfluidic platforms for cell culture analysis. The third objective was addressed through resonant frequency tracking of low- and high-order modes in dynamic-mode cantilevers to enable the real-time characterization of hydrogel viscoelastic properties and continuous monitoring of sol-gel phase transitions over a wide dynamic range using practically relevant hydrogel systems used commonly in additive biomanufacturing. This work suggests that high-order modes of PEMC sensors facilitate characterization of hydrogel viscoelastic properties and gelation processes with improved dynamic range and limit of detection that can complement the performance of low-order modes. Through this research, new approaches for sensor-based characterization of soft material composition and mechanical properties using millimeter-scale piezoelectric devices are presented as solutions for current challenges in biomanufacturing and biosensing to advance capability in real-time sensing of quality attributes among biomanufactured products.
  • Thumbnail Image
    Development and Assessment of Smart Textile Systems for Human Activity Classification
    Mokhlespour Esfahani, Mohammad Iman (Virginia Tech, 2018-09-13)
    Wearable sensors and systems have become increasingly popular for diverse applications. An emerging technology for physical activity assessment is Smart Textile Systems (STSs), comprised of sensitive/actuating fiber, yarn, or fabric that can sense an external stimulus. All required components of an STS (sensors, electronics, energy supply, etc.) can be conveniently embedded into a garment, providing a fully textile-based system. Thus, STSs have clear potential utility for measuring health-relevant aspects of human activity, and to do so passively and continuously in diverse environments. For these reasons, STSs have received increasing interest in recent studies. Despite this, however, limited evidence exists to support the implementation of STSs during diverse applications. Our long-term goal was to assess the feasibility and accuracy of using an STS to monitor human activities. Our immediate objective was to investigate the accuracy of an STS in three representative applications with respect to occupational scenarios, healthcare, and activities of daily living. A particular STS was examined, consisting of a smart socks (SSs), using textile pressure sensors, and smart undershirt (SUS), using textile strain sensors. We also explored the relative merits of these two approaches, separately and in combination. Thus, five studies were completed to design and evaluate the usability of the smart undershirt, and investigate the accuracy of implementing an STS in the noted applications. Input from the SUS led to planar angle estimations with errors on the order of 1.3 and 9.4 degrees for the low-back and shoulder, respectively. Overall, individuals preferred wearing a smart textile system over an IMU system and indicated the former as superior in several aspects of usability. In particular, the short-sleeved T-shirt was the most preferred garments for an STS. Results also indicated that the smart shirt and smart socks, both individually and in combination, could detect occupational tasks, abnormal and normal gaits, and activities of daily living with greater than 97% accuracy. Based on our findings, we hope to facilitate future work that more effectively quantifies sedentary periods that may be deleterious to human health, as well as detect activity types that may be help or hinder health and fitness. Such information may be of use to individuals and workers, healthcare providers, and ergonomists. More specifically, further analyses from this investigation could provide strategies for: (a) modifying a sedentary lifestyle or work scenario to a more active one, and (b) helping to more accurately identify occupational injury risk factors associated with human movement.
  • Thumbnail Image
    Development of Novel Attention-Aware Deep Learning Models and Their Applications in Computer Vision and Dynamical System Calibration
    Maftouni, Maede (Virginia Tech, 2023-07-12)
    In recent years, deep learning has revolutionized computer vision and natural language processing tasks, but the black-box nature of these models poses significant challenges for their interpretability and reliability, especially in critical applications such as healthcare. To address this, attention-based methods have been proposed to enhance the focus and interpretability of deep learning models. In this dissertation, we investigate the effectiveness of attention mechanisms in improving prediction and modeling tasks across different domains. We propose three essays that utilize task-specific designed trainable attention modules in manufacturing, healthcare, and system identification applications. In essay 1, we introduce a novel computer vision tool that tracks the melt pool in X-ray images of laser powder bed fusion using attention modules. In essay 2, we present a mask-guided attention (MGA) classifier for COVID-19 classification on lung CT scan images. The MGA classifier incorporates lesion masks to improve both the accuracy and interpretability of the model, outperforming state-of-the-art models with limited training data. Finally, in essay 3, we propose a Transformer-based model, utilizing self-attention mechanisms, for parameter estimation in system dynamics models that outpaces the conventional system calibration methods. Overall, our results demonstrate the effectiveness of attention-based methods in improving deep learning model performance and reliability in diverse applications.
  • Thumbnail Image
    Engineering-driven Machine Learning Methods for System Intelligence
    Wang, Yinan (Virginia Tech, 2022-05-19)
    Smart manufacturing is a revolutionary domain integrating advanced sensing technology, machine learning methods, and the industrial internet of things (IIoT). The development of sensing technology provides large amounts and various types of data (e.g., profile, image, point cloud, etc.) to describe each stage of a manufacturing process. The machine learning methods have the advantages of efficiently and effectively processing and fusing large-scale datasets and demonstrating outstanding performance in different tasks (e.g., diagnosis, monitoring, etc.). Despite the advantages of incorporating machine learning methods into smart manufacturing, there are some widely existing concerns in practice: (1) Most of the edge devices in the manufacturing system only have limited memory space and computational capacity; (2) Both the performance and interpretability of the data analytics method are desired; (3) The connection to the internet exposes the manufacturing system to cyberattacks, which decays the trustiness of data, models, and results. To address these limitations, this dissertation proposed systematic engineering-driven machine learning methods to improve the system intelligence for smart manufacturing. The contributions of this dissertation can be summarized in three aspects. First, tensor decomposition is incorporated to approximately compress the convolutional (Conv) layer in Deep Neural Network (DNN), and a novel layer is proposed accordingly. Compared with the Conv layer, the proposed layer significantly reduces the number of parameters and computational costs without decaying the performance. Second, a physics-informed stochastic surrogate model is proposed by incorporating the idea of building and solving differential equations into designing the stochastic process. The proposed method outperforms pure data-driven stochastic surrogates in recovering system patterns from noised data points and exploiting limited training samples to make accurate predictions and conduct uncertainty quantification. Third, a Wasserstein-based out-of-distribution detection (WOOD) framework is proposed to strengthen the DNN-based classifier with the ability to detect adversarial samples. The properties of the proposed framework have been thoroughly discussed. The statistical learning bound of the proposed loss function is theoretically investigated. The proposed framework is generally applicable to DNN-based classifiers and outperforms state-of-the-art benchmarks in identifying out-of-distribution samples.