Scholarly Works, Industrial and Systems Engineering

Permanent URI for this collection

https://hdl.handle.net/10919/24292

Research articles, presentations, and other scholarship

Browse

Now showing 1 - 20 of 579

Deep Learning for Sequential Decision Making under Uncertainty: Foundations, Frameworks, and Frontiers
Büyüktahtakın, İ. Esra (2026-04)
Artificial intelligence (AI) is moving increasingly beyond prediction to support decisions in complex, uncertain, and dynamic environments. This shift creates a natural intersection with operations research and management sciences (OR/MS), which have long offered conceptual and methodological foundations for sequential decision-making under uncertainty. At the same time, recent advances in deep learning, including feedforward neural networks, LSTMs, transformers, and deep reinforcement learning, have expanded the scope of data-driven modeling and opened new possibilities for large-scale decision systems. This tutorial presents an OR/MS-centered perspective on deep learning for sequential decision-making under uncertainty. Its central premise is that deep learning is valuable not as a replacement for optimization, but as a complement to it. Deep learning brings adaptability and scalable approximation, whereas OR/MS provides the structural rigor needed to represent constraints, recourse, and uncertainty. The tutorial reviews key decision-making foundations, connects them to the major neural architectures in modern AI, and discusses leading approaches to integrating learning and optimization. It also highlights emerging impact in domains such as supply chains, healthcare and epidemic response, agriculture, energy, and autonomous operations. More broadly, it frames these developments as part of a wider transition from predictive AI toward decision-capable AI and highlights the role of OR/MS in shaping the next generation of integrated learning–optimization systems.
Adaptation to a Whole-Body Powered Exoskeleton: Human-Exoskeleton Coordination During Load-Handling Tasks
Park, Hanjun; Kim, Sunwook; Nussbaum, Maury A.; Srinivasan, Divya (Springer, 2026-03)
Whole-body powered exoskeletons can augment human performance and reduce physical strain in occupational settings, but little is known about how users adapt to these complex devices during practical work scenarios. We compared novice and experienced users during simulated, occupationally relevant load-handling tasks. Six novice users completed exoskeleton familiarization and stationary load-handling tasks in three sessions while five experienced users performed the tasks once. Task performance, biomechanical demands, and perceived workload were compared in each novice session vs. the experienced group. Novice performance improved substantially across sessions, with task completion time reduced by nearly 50% and movement jerk by 30%. However, performance gaps still persisted in session three, compared to the experienced users. Novices also used consistently lower angular velocities (up to 52% lower) and adopted greater hip flexion throughout the sessions. In contrast, differences in shoulder flexion, muscle activity, perceived exertion, and workload diminished more rapidly, with novices approaching experienced levels by session three. Novice users adapted to using a powered exoskeleton over multiple sessions, especially in movement patterns and muscle activation, but differences in task completion time, jerk index, and angular velocities indicated that novices did not attain the skilled coordination and efficiency of experienced users after three sessions. Our results highlight the likely need for extended familiarization and training for the current powered exoskeleton design and provide baseline data for the novice learning curve in occupational settings.
Not All Noises Are Equal: Investigating Auditory Distraction in Emergency Care Using the Tesseract Simulation Platform
Du, David; Lau, Nathan; Ojeifo, Olumide A.; Upthegrove, Tanner; Baber, Adam; Jones, Nathan A.; Parker, Sarah H. (SAGE Publications, 2025-09)
Auditory distractions in clinical environments can impair performance, yet their impact in emergency department (ED) waiting rooms remains understudied. This study investigates how distinct noise types (baby crying, conversations, and equipment alarms) and temporal patterns (continuous vs. intermittent) influence nursing triage performance. Thirty-two ED nurses completed standardized triage tasks within the Tesseract, an immersive audio-visual simulation platform replicating ED waiting room conditions. Preliminary results from 16 participants suggest that the effects of noise depend on both acoustic features and temporal structures. Subjective perceptions of distraction did not consistently align with measured outcomes. This work provides early evidence that auditory distractions influence clinical task execution in complex, task- and context-specific ways, underscoring the need for targeted mitigation strategies and soundscape-aware simulation training.
Sustainable Timber Supply Chain Optimization
Adams, Irma; Canuel, Austin; Lee, Minseo; Büyüktahtakın, İ. Esra (2026-02-23)
The U.S. timber supply chain faces mounting challenges related to capacity constraints, sustainability, and supply resilience at a time when federal policy calls for a rapid expansion of domestic timber production. Following the March 2025 executive order to reduce reliance on foreign timber imports, achieving near-term production targets requires a nationwide redesign of supply chain infrastructure under significant data and operational uncertainty. This study develops a data-driven optimization framework to support short-term, actionable planning for the U.S. timber supply chain. We propose a hybrid machine learning–mixed-integer linear programming (ML–MILP) model that captures the flow of timber from mills through distribution centers to demand points, with the objective of minimizing total transportation and facility-opening costs. U.S.–wide implementation is complicated by incomplete and fragmented data, particularly for mill counts, production levels, and facility locations. To address these gaps, we leverage machine learning models, including gradient boosting, ridge regression, and weighted K-Means clustering, to reconstruct a comprehensive national dataset and generate candidate distribution center locations informed by socioeconomic and environmental factors. The resulting MILP generates an infrastructure and flow plan and is evaluated through sensitivity and scenario-based analyses reflecting demand growth, transportation disruptions, and disaster impacts. Results highlight the dominant role of transportation costs, diminishing returns to capacity expansion, and heightened vulnerability in the South and West regions. Overall, the proposed framework provides policymakers and industry stakeholders with a scalable, sustainability-oriented decision-support tool for guiding domestic timber supply chain expansion under evolving policy objectives.
Cross-Attention Guided Data Sharing for Knowledge Transfer in Robotic AI Systems
Liu, Hui; Zeng, Yingyan; Qiao, Helen; Piliptchak, Pavel; Jin, Ran (2026)
Cross-robot transfer learning is crucial for building robotic AI systems that can generalize across diverse platforms and tasks by utilizing heterogeneous datasets. However, not all source samples are effective to improve the accuracy of the target AI task; while incompatible samples may lead to negative transfer and degrade model performance. This challenge is particularly in a connected robot fleet where robots differ in configurations, sensors, but are connected via Industrial Internet for sequential or parallel tasks. To address this, we propose Cross-Attention guided Proximal Policy Optimization (CAPPO), a reinforcement learning-based sample selection framework that adaptively identifies the most valuable source samples for a given target AI modeling task. Our method employs cross-attention mechanisms to capture fine-grained relevance between source and target samples, constructing informative state representations for a PPO-based selection policy. A task-driven reward function based on downstream performance improvement is created to enable the agent to learn efficient and adaptive selection strategies. Experimental results on a connected robotic fleet with different AI tasks show that our method consistently outperforms existing baselines under low-budget settings, demonstrating strong and robust knowledge transfer performance to train new robotic AI models.
Encrypted Fringe Projection Profilometry
Cheng, Yang; Gui, Zichen; Guan, Le; Jin, Ran; Li, Beiwen (2026-01-20)
We propose encrypted fringe projection profilometry (E-FPP), a co-keyed phasecode framework that embeds encryption directly into the measurement process of fringe projection profilometry (FPP). Two random phase fields and orthogonal temporal codes jointly encode the projected sequence, binding phase retrieval to legitimate key pairs. The method preserves standard three-step FPP operation while enabling key-dependent 3D reconstruction. Experiments verify accurate authorized reconstruction and strong resistance to single- and multi-frame inference, providing a privacy-preserving solution for optical metrology.
An Open-Source Framework to Design, Tune, and Fly Nonlinear Control Systems for Autonomous UAVs
Gramuglia, Mattia; Kumar, Giri Mugundan; Orlando, Giorgio A.; L’Afflitto, Andrea (American Institute of Aeronautics and Astronautics, 2025-01)
This paper introduces a freeware open-source software ecosystem to design, tune, and test control systems for autonomous vertical take-off and landing (VTOL) multi-rotor uncrewed aerial vehicles (UAVs) such as quadcopters and quad-biplanes. This environment comprises C++-coded flight stacks with a suite of control systems, a high-fidelity simulator that allows model-in-the-loop and hardware-in-the-loop tests, computer-aided design (CAD) models of UAVs, and a website for a broad overview of this project. This ecosystem aims to serve as a common platform for the aerospace control community and ease comparative analyses for control design techniques produced by multiple research groups.
Rigid and soft back-support exoskeletons affect biomechanical and perceptual demands, but in different ways, during simulated shingle installation
Choi, Jiwon; Kim, Sunwook; Usmani, Ahmad Raza; Barr, Alan; Harris-Adamson, Carisa; Nussbaum, Maury A. (Elsevier, 2026-02)
Passive back-support exoskeletons (BSEs) are promising but underexplored interventions to reduce the high physical demands of roofing shingling. Eighteen participants performed simulations of shingle installation tasks under 12 different conditions. These conditions included all combinations of three BSE levels (Rigid, Soft, and no BSE), two task orientations (peak-facing vs. side-facing), and two roof slopes (18° vs. 26°). Using the rigid BSE significantly reduced lumbar muscle activation (11–17%) compared to no BSE, without altering trunk flexion. In contrast, the soft BSE reduced trunk flexion (∼4%) without altering lumbar muscle activation. Both BSEs reduced perceived low back exertion (∼16%); however, the rigid BSE increased leg discomfort (∼26%), and the soft BSE increased shoulder exertion (∼19%). Our results suggest that using BSEs can be beneficial for shingle installation tasks but also highlight the importance of considering device-specific biomechanical benefits and associated trade-offs to ensure effective application.
Simulating U.S. Presidents for a Friendly Chat: Applying Generative AI to Study Political History
Nour, Sami; Ghaffarzadegan, Navid; Naugle, Asmeret; Godfrey, Joseph R. (Springer, 2026)
Advancements in generative artificial intelligence have created new opportunities to develop large language model (LLM)-based simulation models. By designing distinct personas and training them with relevant information, modelers can simulate a wide range of agents, representing diverse personalities, socio-economic backgrounds, and demographics. The potential of these simulation models, often referred to as generative agents, extends beyond creating average representations of groups; they can also be tailored to simulate specific individuals, predicting their responses or opinions under various scenarios. In this study, we take on the challenge of simulating 60 U.S. presidents to demonstrate how this approach can contribute to the study of political history. We simulate 60 generative agents using an LLM (GPT o1) primed on the inaugural addresses of presidents from 1789 to 2025. We then ask each simulated president the question, “what factors influence the economy?” We validate the simulated responses with other LLMs tasked with predicting which president is most likely to have given each response. We then use a causal loop diagram generation tool called SD Bot to extract variables and relationships from the text responses and depict mental models. Finally, we quantify and visualize presidents’ relative similarities to each other as a network.
Together or Apart: Designing Boundaries for Personal Intelligent Agents
Kang, Hyunmin; Lee, Seul Chan; Jeong, JiHyun; Kim, Hyochang; Cha, Minchul; Jeon, Myounghoon (ACM, 2025-11-10)
Personal intelligent agents (IAs) are increasingly embedded in everyday life, a trend accelerated by generative AI technologies. Despite their growing presence, these agents often remain fragmented across different life domains and environments. This workshop explores how to design integrated IA ecosystems emphasizing continuity, coordination, and human-centered values. Participants with varied perspectives will collaboratively develop frameworks, scenarios, and guidelines for cohesive personal agent systems that enrich user experiences holistically. By examining factors that shape users’ preferences for information integration or separation, we aim to inform the design of coherent, user-aligned multi-agent systems.
Fairness in machine learning-based hand load estimation: A case study on load carriage tasks
Rahman, Arafat; Lim, Sol; Chung, Seokhyun (Elsevier, 2026-01)
Predicting external hand load from sensor data is essential for ergonomic exposure assessments, as obtaining this information typically requires direct observation or supplementary data. While machine learning can estimate hand load from posture or force data, we found systematic bias tied to biological sex, with predictive disparities worsening in imbalanced training datasets. To address this, we developed a fair predictive model using a Variational Autoencoder with feature disentanglement, which separates sex-agnostic from sex-specific motion features. This enables predictions based only on sex-agnostic patterns. Our proposed algorithm outperformed conventional machine learning models, including k-Nearest Neighbors, Support Vector Machine, and Random Forest, achieving a mean absolute error of 3.42 and improving fairness metrics like statistical parity and positive and negative residual differences, even when trained on imbalanced sex datasets. These results underscore the importance of fairness-aware algorithms in avoiding health and safety disadvantages for specific worker groups in the workplace.
Toward Safer Diagnoses: A SEIPS-Based Narrative Review of Diagnostic Errors
Yen, Carol; Epling, John W.; Rockwell, Michelle; Vaughn-Cooke, Monifa (MDPI, 2026-01-21)
Diagnostic errors have been a critical concern in healthcare, leading to substantial financial burdens and serious threats to patient safety. The Improving Diagnosis in Health Care report by the National Academies of Sciences, Engineering, and Medicine (NASEM) defines diagnostic errors, focusing on accuracy, timeliness, and communication, which are influenced by clinical knowledge and the broader healthcare system. This review aims to integrate existing literature on diagnostic error from a systems-based perspective and examine the factors across various domains to present a comprehensive picture of the topic. A narrative literature review was structured upon the Systems Engineering Initiative for Patient Safety (SEIPS) model that focuses on six domains central to the diagnostic process: Diagnostic Team Members, Tasks, Technologies and Tools, Organization, Physical Environment, and External Environment. Studies on contributing factors for diagnostic error in these domains were identified and integrated. The findings reveal that the effectiveness of diagnostics is influenced by complex, interconnected factors spanning all six SEIPS domains. In particular, socio-behavioral factors, such as team communication, cognitive bias, and workload, and environmental pressures, stand out as significant but difficult-to-capture contributors in traditional and commonly used data resources like electronic health records (EHRs), which limits the scope of many studies on diagnostic errors. Factors associated with diagnostic errors are often interconnected across healthcare system stakeholders and organizations. Future research should address both technical and behavioral elements within the diagnostic ecosystem to reduce errors and enhance patient outcomes.
A Theoretical Foundation for IDEF0 Using Category Theory
Godfrey, Joseph (2025-09-26)
IDEF0 is a structured methodology in system engineering used to represent system functions. Category Theory is a branch of mathematics concerned with structures and relationships represented at a very high level of abstraction. Category Theory has recently attracted attention as a promising framework for developing a rigorous, formal theory for systems. As a contribution to this effort, we formulate IDEF0 using Category Theory. The formulation is largely one of translating between the specification of IDEF0 in natural language to the formal language of Category Theory.
Balance equations for physics-informed machine learning
Molnar, Sandor M.; Godfrey, Joseph; Song, Binyang (Elsevier, 2024-12)
Using traditional machine learning (ML) methods may produce results that are inconsistent with the laws of physics. In contrast, physics-based models of complex physical, biological, or engineering systems incorporate the laws of physics as constraints on ML methods by introducing loss terms, ensuring that the results are consistent with these laws. However, accurately deriving the nonlinear and high order differential equations to enforce various complex physical laws is non-trivial. There is a lack of comprehensive guidance on the formulation of residual loss terms. To address this challenge, this paper proposes a new framework based on the balance equations, which aims to advance the development of PIML across multiple domains by providing a systematic approach to constructing residual loss terms that maintain the physical integrity of PDE solutions. The proposed balance equation method offers a unified treatment of all the fundamental equations of classical physics used in models of mechanical, electrical, and chemical systems and guides the derivation of differential equations for embedding physical laws in ML models. We show that all of these equations can be derived from a single equation known as the generic balance equation, in conjunction with specific constitutive relations that bind the balance equation to a particular domain. We also provide a few simple worked examples how to use our balance equation method in practice for PIML. Our approach suggests that a single framework can be followed to incorporate physics into ML models. This level of generalization may provide the basis for more efficient methods of developing physics-based ML for complex systems.
Comparing in-home and bottled drinking water quality: regulated and emerging contaminants in rural Central Appalachia
Albi, Kate; Krometis, Leigh-Anne H.; Ling, Erin; Cohen, Alasdair; Xia, Kang; Gray, Austin D.; Dudzinski, Emerald; Ellis, Kimberly P. (IWA Publishing, 2025-09)
An increasing number of Americans rely on bottled water for household use, citing perceptions of poor in-home water quality and/or distrust of public water utilities. We analyzed in-home (n = 23), roadside spring (n = 4), and bottled drinking water (n = 36) in Central Appalachia. All samples were analyzed for regulated (bacteria, inorganic ions) and emerging (PFAS, microplastics) contaminants. Study survey results indicated the majority (83%) of participants viewed their in-home water quality as satisfactory or poor due to negative organoleptic perceptions. Coliform bacteria and sodium levels exceeding recommended levels were detected in 52% of home water samples, though detections varied by source, i.e., high sodium was more often observed in municipal water, while bacteria were more often observed in private system water. Bottled water samples did not exceed any regulations, though median microplastic concentrations were statistically higher (p = 0.001, Wilcoxon rank-sum test) than those recovered from in-home samples. PFAS compounds were detected in some in-home and bottled water samples at very low levels. While in general bottled water appears to be a safe drinking water source in these areas, the associated costs in time and money for lower-income households are considerable, and were estimated by participants as $68–400/month.
Muscle synergy analysis of short-term adaptation to arm-support exoskeletons during pseudo-static and dynamic overhead tasks
Park, Hanjun; Nussbaum, Maury A. (Elsevier, 2026-01)
Occupational arm-support exoskeletons (ASEs) can reduce shoulder muscle activity during overhead work, but their effects on muscle synergy structure and temporal activation remain limited. We examined the effects of using three different exoskeletons on muscle synergies during simulated overhead tasks. Muscle activity from 18 participants (gender-balanced) performing both pseudo-static and dynamic tasks across 24 conditions (three ASEs and a control condition) was analyzed using non-negative matrix factorization to extract synergy number, structure, and activation coefficients. Dynamic tasks recruited more muscle synergies (interquartile range: 2–5) than pseudo-static tasks (interquartile range: 1–3), with some task combinations showing modest increases with ASE use compared to the control condition. Synergy structure and temporal activation were generally similar across interventions (mean cosine similarity 0.74–0.92), but certain ASE-task combinations produced significant local changes in synergy structure. Using exoskeletons generally altered muscle weightings, shifting from primary arm-elevating and shoulder-stabilizing muscles toward modules involving neck and back muscles, suggesting compensatory strategies for device-imposed biomechanical demands. Activation time courses remained highly similar across most interventions during pseudo-static tasks, though dynamic tasks showed reduced peak magnitude with exoskeleton use. Our results indicate that while modular motor control is largely preserved with ASE use, device- and task-specific adaptations in synergy structure and temporal activation can occur. Future research should explore how ASE design features influence neuromuscular strategies and assess long-term adaptation of muscle synergies in occupational settings.
Workload Dynamics in Safety-Critical Monitoring Roles: Evidence from the Belgian Railway Network
Liu, Ning-Yuan; Triantis, Konstantinos P.; Roets, Bart (Taylor & Francis, 2025-05-24)
Increased mental workload leads to high stress levels or boredom during monitoring tasks, escalating the risk of human errors. Determining and quantifying an optimal mental workload level that maximizes operator performance presents a formidable challenge. Leveraging existing theoretical frameworks and literature on workload suboptimality, we constructed a quantitative System Dynamics model on workload and its impact on human error for safety-critical monitoring roles. This model is then rigorously tested and calibrated using real-world operational data from traffic controllers employed by the National Belgian Railway Infrastructure Company. The results point to (1) the support of operational data in the formulation of the feedback mechanism between an operator’s workload and human error; (2) the quantification of overload and underload thresholds; (3) the dynamics associated with both thresholds, where increased fatigue levels lead to a shrinkage of the ‘comfort zone’. The simulation model emerges as a potential approach for practitioners to assess the probability of human errors based on specific workload distributions. Beyond its immediate utility, the model also offers strategic insights for policy-making and schedule planning to enhance operator performance and ensure safety in socio-technical systems.
Testing for Heterogeneity in Data Envelopment Analysis
Triantis, Konstantinos P.; Mohsenirad, Saman (2025-01-07)
This paper introduces a comprehensive framework for detecting and conceptualizing heterogeneity in data envelopment analysis (DEA), aligning with the microeconomic production theory. Despite DEA’s significant advantage in evaluating DMUs based on their efficiency without assuming a specific functional form of technology, it critically relies on the comparability of these units. We address the persistent issue in DEA modeling that stems from the assumption of homogeneity among DMUs, which is often untested. We propose a novel methodological approach that serves as a testing framework for heterogeneity, predicated on minimal assumptions about data randomness. This framework provides a means to examine the biases introduced by technological disparities among DMUs and offers an approach for practitioners to ensure the validity of DEA modeling across diverse technological settings. This approach not only uncovers biases from technological differences but also serves as a preliminary step to enhance methods like clustering, aiding practitioners in verifying DEA’s applicability across varied technologies.
Evaluating the Accuracy of AI-Powered Ergonomic Assessments Using a Commercial Computer Vision System
Jamshid Nezhad Zahabi, Saman; Kim, Sunwook; Nussbaum, Maury A.; Lim, Sol (SAGE Publications, 2025-09)
Workers performing material handling tasks are at high risk of work-related musculoskeletal disorders (WMSDs). While AI-based computer vision tools claim to assess ergonomic risks with minimal input, their accuracy remains uncertain. This study evaluated a commercial AI system’s ability to estimate key parameters of the Revised NIOSH Lifting Equation (RNLE) by comparing its outputs to those from a marker-based motion capture system. Ten participants completed lifting tasks while being recorded by three cameras and motion capture sensors. The AI-analyzed video outputs were compared to ground truth data. Results showed significant inaccuracies in the AI’s estimates—especially for horizontal and vertical distances—leading to overestimated Recommended Weight Limits and underestimated Lifting Index values. Among the camera views, the side view produced the most accurate results, while the moving camera performed worst. These findings highlight the need for improvement in commercial AI tools before they can be reliably used in ergonomic risk assessments.
What drives the effective integration of lift assists in automotive assembly? Perspectives from operators, ergonomists, and manufacturers
Usmani, Ahmad Raza; Kim, Sunwook; Smets, Marty; Nussbaum, Maury A. (Elsevier, 2025-12)
Automotive assembly workers experience elevated risks of work-related musculoskeletal disorders due to frequent material handling. Lift assists (LAs) can reduce these risks by offsetting payload weights. However, integrating LAs into complex workflows can be challenging, and workers may choose not to use LAs to achieve other objectives. We interviewed 16 operators, nine ergonomists, and six LA manufacturers to capture diverse viewpoints. Content analysis revealed perspectives on LA usability, design, implementation, and operational concerns. Operators noted physical demands in initiating, turning, or stopping LAs, and emphasized lightweight designs, simplified controls, and structured training. Ergonomists reported retrofitting LAs into workflows not designed for LAs, creating integration challenges. LA manufacturers described balancing ergonomic goals with operational demands and evolving requirements, emphasizing the need for better design feedback. Our findings suggest that heavy equipment, complex controls, and limited training hinder successful LA implementation; we offer recommendations to improve future LA design and implementation.

Browse

Recent Submissions