About

Dr. Kevin Han is an associate professor and Edward I. Weisiger Distinguished Scholar in the Department of Civil, Construction, and Environmental Engineering at North Carolina State University. He holds a Ph.D. in Civil Engineering with an emphasis on Construction Engineering and Management from the University of Illinois at Urbana-Champaign, a Master of Computer Science from the same institution, a Master of Science in Civil Engineering with a focus on Construction Engineering and Management, and a Bachelor of Arts in Architecture with a minor in Structural Engineering from the University of California, Berkeley. His research centers on developing and validating innovative computer vision and machine learning analytics that utilize visual data such as images, videos, and point cloud data, along with Building Information Modeling (BIM), to enhance construction project controls, improve site-to-office communication, and advance safety and hazard recognition and training. He is particularly interested in exploring new modalities of information and user interaction, including Virtual Reality, Augmented Reality, and Augmented Virtuality, as well as robotics for automating management of civil infrastructure systems, such as autonomous navigation for data collection and analytics. Dr. Han teaches courses related to BIM in Construction, Mechanical and Electrical Systems for Buildings, Visual Sensing for Civil Infrastructure Engineering and Management, and Design of a Robotic Computer Vision System for Autonomous Navigation. He is also the founder of the Construction Automation and Robotics Lab (CARL), where his research group focuses on automation and robotics in civil construction and infrastructure management.

Research topics

• Computer Science
• Computer Security
• Engineering
• Engineering management
• Construction engineering
• Human–computer interaction
• Artificial Intelligence
• Machine Learning
• Simulation
• Statistical physics
• Physics
• Real-time computing
• Business
• Mathematics
• Condensed matter physics
• Chemical physics
• Computer vision
• Economics
• Quantum mechanics

Selected publications

• TDFlow: Agentic Workflows for Test Driven Development

  2026-01-01

  articleOpen access1st authorCorresponding

  Kevin Han, Siddharth Maddikayala, Tim Knappe, Om Patel, Austen Liao, Amir Barati Farimani. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

  PublisherDOI

• Versatile Test Reactor Open Digital Engineering Ecosystem

  Insight · 2025-04-01

  articleOpen access

  ABSTRACT Modern design of nuclear facilities represents unique challenges: enabling the design of complex advanced concepts, supporting geographically dispersed teams, and supporting first‐of‐a‐kind system development. Errors made early in design can introduce silent errors. These errors can cascade causing unknown risk of complex engineering programs. The Versatile Test Reactor (VTR) Program uses digital‐engineering principles for design, procurement, construction, and operation to reduce risk and improve efficiencies. Digital engineering is an integrated, model‐based approach which connects proven digital tools such as building information management (BIM), project controls, and systems‐engineering software tools into a cohesive environment. The VTR team hypothesizes using these principals can lead to similar risk and cost reductions and schedule efficiencies observed in other engineering industries. This research investigates the use of a digital engineering ecosystem in the design of a 300‐MWt sodium‐cooled fast reactor. This ecosystem was deployed to over 200 engineers and used to deliver the conceptual design of the VTR. We conclude that initial results show significant reductions in user latency (1000x at peak use), the possibility of direct finite‐element‐analysis (FEA) integrations to computer‐aided design (CAD) tools, and nuclear reactor system design descriptions (SDDs) that we can fully link throughout design in data‐driven requirements‐management software. These early results led to the VTR maintaining milestone performance during the COVID‐19 pandemic.

  PublisherOA PDFDOI

• DistMLIP: A Distributed Inference Platform for Machine Learning Interatomic Potentials

  arXiv (Cornell University) · 2025-05-28 · 3 citations

  preprintOpen access1st authorCorresponding

  Large-scale atomistic simulations are essential to bridge computational materials and chemistry to realistic materials and drug discovery applications. In the past few years, rapid developments of machine learning interatomic potentials (MLIPs) have offered a solution to scale up quantum mechanical calculations. Parallelizing these interatomic potentials across multiple devices poses a challenging, but promising approach to further extending simulation scales to real-world applications. In this work, we present DistMLIP, an efficient distributed inference platform for MLIPs based on zero-redundancy, graph-level parallelization. In contrast to conventional spatial partitioning parallelization, DistMLIP enables efficient MLIP parallelization through graph partitioning, allowing multi-device inference on flexible MLIP model architectures like multi-layer graph neural networks. DistMLIP presents an easy-to-use, flexible, plug-in interface that enables distributed inference of pre-existing MLIPs. We demonstrate DistMLIP on four widely used and state-of-the-art MLIPs: CHGNet, MACE, TensorNet, and eSEN. We show that DistMLIP can simulate atomic systems 3.4x larger and up to 8x faster compared to previous multi-GPU methods. We show that existing foundation potentials can perform near-million-atom calculations at the scale of a few seconds on 8 GPUs with DistMLIP.

  PublisherOA PDFDOI

• A vision‐based weigh‐in‐motion approach for vehicle load tracking and identification

  Computer-Aided Civil and Infrastructure Engineering · 2025-03-16 · 10 citations

  articleOpen access

  With the rapid increase in the number of vehicles, accurately identifying vehicle loads is crucial for maintaining and operating transportation infrastructure systems. Existing load identification methods typically rely on collecting vehicle load data from weigh-in-motion (WIM) systems when vehicles pass over them. However, cumbersome installation, high costs, and regular maintenance are the main obstacles that prevent WIM from being widely used in practice. This study introduces the visual WIM (V-WIM) framework, a vision-based approach for tracking and identifying moving loads. The V-WIM framework consists of two main components, the vehicle weight estimation and the vehicle tracking and location estimation. Vehicle weight is estimated using tire deformation parameters extracted from tire images through object detection and optical character recognition techniques. A deep learning-based YOLOv8 algorithm is employed as a vehicle detector, combined with the ByteTrack algorithm for tracking vehicle location. The vehicle weight and its corresponding location are then integrated to enable simultaneous vehicle weight estimation and tracking. The performance of the proposed framework was evaluated through two component validation tests and one on-site validation test, demonstrating its capability to overcome the limitations of existing methods.

  PublisherOA PDFDOI

• Automated digital transformation for pedestrian suspension bridges using hybrid semantic structure from motion

  Automation in Construction · 2025-05-09 · 1 citations

  article
  PublisherDOI

• Analyzing the functioning of a graduate engineering class with multiple collaborating groups using Slack: An embedded mixed methods case study

  International Journal of STEM Education · 2025-10-01

  articleOpen accessSenior author

  Policy documents call for supporting STEM students in developing collaborative abilities for working in multidisciplinary teams. Courses with intra- and inter-group collaboration are therefore essential to prepare STEM students to participate in modern multidisciplinary professional environments. To analyze those courses, this paper develops a novel theoretical framework of the levels of functioning (individual, within team, across team, and whole group) that may occur. Using the novel framework as well as Communities of Practice theory and Social Interaction Theory, we analyzed a graduate engineering course that communicated through the Slack platform, using a case study design to examine students’ interaction. Social Network Analysis of 5969 Slack messages exchanged through the semester on channels for individual teams, sets of teams, and the entire class was complemented with qualitative analysis of interviews, class materials, observations, and the content of Slack messages. Findings reveal distinct patterns of intra- and inter-group participation. This study highlights how groups interacted through brokers, boundary objects, and tools. Moreover, subsets of teams displayed extensive interaction concerning related tasks, exemplifying “overlap” connections. Diverse patterns of brokerage were characterized. This paper concludes with a general approach for evaluating courses with multi-group collaboration. This approach can be used to diagnose complex multi-team classes, especially in hybrid or online courses where communication occurs through online platforms. This methodology holds promise for promoting effective collaboration and fostering teamwork skills among students in STEM fields.

  PublisherOA PDFDOI

• Motion-Based Communication as a Language: Formal Grammar Representation and Model-Free Decoding with Dense Optical Flow

  ASCE OPEN Multidisciplinary Journal of Civil Engineering · 2025-08-20

  articleOpen accessSenior author

  Transmitting information through movement or gestures has a long history and is especially valuable when other communication modalities are impaired or unavailable. This research explores motion-based communication (MBC), which enables collaboration among agents without relying on networking protocols or hardware. Unlike traditional robot-to-robot communication, which depends on networked data transfer and is vulnerable to failures or environmental interference (e.g., electromagnetic disturbances and crosstalk), MBC provides a robust alternative that remains unaffected by such disruptions. This work contributes to robotics by developing a motion-based pseudolanguage using formal grammar structured by an augmented transition network. Symbols in the language are encoded via directional shifts in motion rather than poses and are decoded by analyzing optical flow vectors without the need for deep-learning models. Evaluation in a real-world laboratory environment under nominal and dim lighting conditions shows baseline decoding accuracies of 94.0% and 87.6%, respectively, which improve to 95.8% and 88.2% with optimal tuning. Additionally, under nominal and dim lighting, frame rates of approximately 60 and 35 frames/s are achieved, respectively. Performance increases significantly when actively decoding messages, reaching 700–1,000 frames/s, after a region of interest is constructed during the process of requesting communication. These results validate the efficacy of using a model-free, language-based approach to robotic MBC.

  PublisherOA PDFDOI

• Dual‐reference approach for vision‐based structural displacement measurement using time‐varying Kalman filter

  Computer-Aided Civil and Infrastructure Engineering · 2025-07-06 · 7 citations

  articleOpen access

  Accurate displacement measurement is essential for structural health monitoring (SHM) to ensure infrastructure safety. Most previous vision-based displacement measurement methods either rely on static reference frames or lack dynamic error feedback, leading to performance degradation under real-world conditions. To address these challenges, this study proposes the dual-reference Kanade–Lucas–Tomasi (DR-KLT) method, which improves vision-based displacement measurements by dynamically integrating both the initial reference frame and the previous reference frame in the KLT tracker. The proposed method estimates the reliability of tracking by analyzing performance indicators such as corner tendency, bi-directional error, number of feature points, and optical flow magnitude. These estimates are incorporated into a time-varying Kalman filter for accurate displacement estimation. Validation through simulations, lab-scale, and on-site experiments demonstrate the method's robustness and superior accuracy compared to single-reference approaches. The results confirm that the DR-KLT approach effectively mitigates the limitations of conventional KLT-based tracking under unstable conditions such as occlusion or lighting variation, making it a reliable tool for real-world SHM applications.

  PublisherOA PDFDOI

• Analysis of two pedagogical approaches to foster discipline integrations in an educational data mining class using communities of practice

  International Journal of STEM Education · 2025-03-05 · 3 citations

  articleOpen accessSenior author

  This paper describes research into two pedagogical approaches to foster transdisciplinarity in a graduate engineering course that involves education and computer science. Leveraging the Communities of Practice framework, we examine how students majoring in computer science can integrate new knowledge from education and computer science to engage in an educational data mining project. The first course iteration sought to connect students from education and computer science disciplines through a blend of problem-based learning and traditional lectures. The second course iteration involved computer science students only, but included two instructors, one from computer science and the other from education. To evaluate these approaches, we conducted multiple student interviews and classroom observations. We found that pursuing interdisciplinary through student brokers had a localized student impact on discipline integration without creating an entire class transdisciplinary environment, proving particularly effective for students with backgrounds outside of computer science. However, it fell short of achieving an overarching integration of education knowledge across the entire class. In contrast, the co-teaching approach influenced class dynamics significantly as instructors honed their brokerage skills and introduced crucial components to the multidisciplinary toolkit. Students reinterpreted these elements within the context of their projects, leading to a deeper integration of education and computer science disciplines. However, while students did acquire more knowledge from both disciplines, they did not always achieve a comprehensive practical understanding of the class outcomes. Findings suggest that differences in instructional design can significantly impact how interdisciplinary integration forms within a class. Using CoP, we identified various models to foster disciplinary integration. The two pedagogical approaches used—student brokers and co-instructors—achieved some disciplinary integration, highlighting multidisciplinary, interdisciplinary, and transdisciplinary integration. Engaging in projects with multidisciplinary teams allows students to interact one-on-one while working on real projects, enabling them to negotiate their participation with peers and resulting in a deeper integration of the involved disciplines. This paper discusses the merits and the drawbacks of employing both approaches to build an interdisciplinary class.

  PublisherOA PDFDOI

• Digital engineering workflow for effective management of ITAAC using text analytics and 4D BIM concept for construction of nuclear power plants

  Nuclear Engineering and Design · 2025-06-04 · 1 citations

  article
  PublisherDOI

Recent grants

• IGE: Learning the Entire Pipeline: Analyzing and Improving Graduate Engineering Education through Communities of Practice

  NSF · $332k · 2021–2024

Frequent coauthors

• Mojtaba Noghabaei

  North Carolina State University

  32 shared

• Khashayar Asadi

  24 shared

• Alex Albert

  Florida Gulf Coast University

  17 shared

• Idris Jeelani

  University of Florida

  14 shared

• Mani Golparvar‐Fard

  13 shared

• Youngjib Ham

  12 shared

• Hyungchul Yoon

  11 shared

• Edgar Lobatón

  North Carolina State University

  11 shared

Education

• PhD, Civil and Environmental Engineering

  University of Illinois at Urbana-Champaign

  2016

• MCS, Computer Science

  University of Illinois at Urbana-Champaign

  2015

• MS, Civil and Environmental Engineering

  University of California Berkeley

  2012

• BA, Architecture

  University of California Berkeley

  2010

Awards & honors

• Edward I. Weisiger Distinguished Scholar