GIST Scholar
Not a member yet
30271 research outputs found
Sort by
Enhancing collaborative signing songwriting experience of the d/Deaf individuals
No full textSongwriting can be an important means of developing the personal and social skills of d/Deaf individuals, but there is a lack of research on understanding and supporting their songwriting. We aimed to understand the d/Deaf people's songwriting experience for the song signing genre, which visually represents music with sign language and body movement. Through two workshops in which mixed-hearing individuals collaborated in songwriting activities, we identified the potentials and challenges of the songwriting experience and developed a music-sensory substitution system that multimodally presents music in sound as well as visual, and vibrotactile feedback. The proposed system enables mixed-hearing partners to have better collaborative interaction and signing songwriting experience. Consequently, we found that the process of signing songwriting is valued by d/Deaf individuals as a means of musical self-expression and social connecting, and our system has increased their musical engagement while encouraging them to express themselves more through music and sign language. © 2024 Elsevier LtdFALSEsciescopu
Methane Enrichment from Hydrogen-Natural Gas Blends via Tryptophan Assisted Semi-Unstirred Hydrate Formation
No full textROS2 based Distributed Task Assignment for Heterogeneous Unmannded Vehicles: Evaluation through SITL and HITL Hwan-Yong Park School of Mechanical and Robotics Engineering Gwangju Institute of Science and Technology
No full textThis paper proposes a ROS 2 based distributed dynamic task assignment framework for heterogeneous unmanned vehicles (UVs) engaged in large scale, time critical missions and validates it through Software and Hardware in the Loop (SITL/HITL) experiments. Re- current motion patterns—wide area search, multi view observation, and manoeuvring by altitude—are first formalised as operational modes and automatically transformed into real time ROS 2 tasks by parameterised path generation software. Reference implementations of the Asynchronous Consensus Based Bundle Algorithm (ACBBA) and a Distributed Auc- tion are then encapsulated as modular ROS 2 nodes with DDS interfaces, enabling direct comparison with a centralised Hungarian benchmark. A multi device test bed that couples Jetson Orin Nano companion computers, Pixhawk 6C mini flight controllers, and Gazebo Classic/Garden simulators reproduces realistic sensing and control loops, allowing quanti- tative evaluation of algorithm performance. Across scenarios containing 3 – 12 tasks and 3 – 6 agents, the Distributed Auction matches the centralised optimum in wide area search, whereas ACBBA outperforms all other methods in multi view observation and altitude strat- ified manoeuvre. These findings indicate that, with appropriately designed reward functions and message protocols, ROS 2 middleware can support real time, near optimal coordination of mixed UAV/UGV teams without a single point of failure. Furthermore, the source code and datasets produced in this study are expected to facilitate subsequent research aimed at enhancing the robustness and scalability of autonomous swarm mission planning. ©2025 Hwan-Yong Park ALL RIGHTS RESERVED|본연구는대규모 ·시간제약임무를수행하는이기종무인이동체(UV)를대상으로, ROS 2기반분산동적작업할당프레임워크를제안하고 SITL및 HITL환경에서타당성 을입증한다.먼저광역탐색,다시점관찰,고도별기동의반복경로패턴을운용모드로 정식화한후,파라미터화된경로생성소프트웨어를통해이를실시간 ROS 2과업으로자 동변환하였다.이어서Asynchronous Consensus Based Bundle Algorithm(ACBBA)과Dis- tributed Auction을 DDS 인터페이스를 갖춘 모듈형 ROS 2 노드로 구현하여, 중앙집중형 Hungarian 알고리즘과 직접 비교 가능한 구조를 마련하였다. Jetson Orin Nano · Pixhawk 6C mini, Gazebo Classic/Garden을연동한다중장치 SITL · HITL테스트베드를구축함으 로써실제센서 ·제어루프를모사한조건에서알고리즘성능을정량평가하였다. 3 – 12 개과업및 3 – 6개에이전트시나리오에서 Distributed Auction은광역탐색모드에서중앙 집중해와거의동일한결과를보였고, ACBBA는다시점관찰및고도분할기동모드에서 최우수 성능을 기록하였다. 이 같은 결과는 보상 함수와 메시지 프로토콜만 적절히 설계 하면 ROS 2 미들웨어 상에서도 UAV-UGV 혼합 팀의 실시간 ·준최적 협조가 가능함을 시사한다.아울러,본연구를통해개발한소스코드와데이터셋은자율군집임무계획의 탄력성 ·확장성향상을위한후속연구에기여할것으로기대된다. ©2025 박환용 ALL RIGHTS RESERVEDMaster1 Introduction 1
1.1 Motivation 1
1.2 Literature review 2
1.3 Contributions and Outline 4
2 Preliminaries and Background 6
2.1 Multi Robot Task Assignment Problem 6
2.2 Hungarian Algorithm 7
2.3 Distributed Auction Algorithm 9
2.4 Asynchronous Consensus Based Bundle Algorithm (ACBBA) 13
3 Problem Design 22
3.1 Multi-waypoint task modeling 22
3.2 Reward Function Design 22
3.3 Modes of Operation 25
3.4 Scenarios 27
4 Implementation 30
4.1 Software Platform and Physics Engine Simulator 30
4.2 Software Development 33
4.3 Hardware for Experiments 42
4.4 SITL in Multi Devices 44
4.5 HITL in Multi Device 56
5 Conclusion 69
References 7
Dual Benefits of Manganese Recovery and Carbon Mineralization via pH Swing-Assisted Carbonation Process in Iron and Steelmaking By-products
No full textEx-situ carbon mineralization utilizing iron and steelmaking by-products are a promising approach for the circular economy by enabling by-product recycling, resource recovery, substantial CO2 sequestration, and the production of value-added solid carbonates (e.g., CaCO3, MgCO3). However, the variability in the physicochemical properties of these by-products poses challenges in predicting leaching, precipitation, and carbonation behavior, which is generally influenced by their generation processes. In this context, this study focuses on the recovery of manganese (Mn) and CO2 sequestration utilizing blast furnace slags (air-cooled, IS-A; water-cooled, IS-W), and electric arc furnace slag (EAF) and its dust (EAF-D) via a pH swing-assisted carbonation process. Leaching behavior for magnesium (Mg), aluminum (Al), silicon (Si), calcium (Ca), iron (Fe), and manganese (Mn) in IS-A and IS-W was identical, though differences emerged in the precipitation behavior of major elements due to co-precipitation. The IS-A and IS-W included higher Ca content (IS-A, 21.4 wt%; IS-W, 24.3 wt%) compared to EAF and EAF-D, resulting in impressive CO₂ storage capacities (IS-A, 93 kg CO2/ton slag; IS-W, 128 kg CO2/ton slag). In contrast, EAF and EAF-D had higher Mn content (EAF, 19.4 wt%; EAF-D, 34.7 wt%), with concentrations achieving up to 27.4 wt% for EAF and 46.7 wt% for EAF-D, alongside notable recovery efficiencies (EAF, 29 %; EAF-D, 68 %). The carbonation process produced high purity CaCO3 (92–98 %). These findings underscore the potential of iron and steelmaking by-products as feedstock for carbon mineralization and resource recovery, thereby contributing to a more sustainable society. © 2025 Elsevier LtdFALSEsciescopu
Auxiliary Decoder-Based Training for Sound Event Detection with a Pretrained Model
No full textSound Event Detection (SED) identifies the types and timestamps of events within audio clips and has applications in diverse areas such as audio captioning, wildlife tracking, and equipment monitoring. These applications are critical for extracting and analyzing meaningful information from audio recordings. This thesis introduces an innovative approach to enhance SED by incorporating an auxiliary decoder into the output of the final layer of the convolutional block. This integration significantly im- proves SED capabilities, enabling more accurate recognition and classification of sound events. Furthermore, this thesis proposes a one-stage training method to enhance SED model performance, replacing the conventional two-stage approach, which involves a CRNN training process followed by fine-tuning. By employing integration and corre- lation branches, this network combines embeddings from multiple sub-networks. The method unifies the objectives of both stages by utilizing an auxiliary decoder, allowing joint training of the pre-trained transformer encoder and convolutional layers. Applied to the DCASE 2024 Challenge Task 4 dataset, the model’s performance was evaluated using the Polyphonic Sound Detection Score (PSDS) and the Mean Pairwise Area Un- der the Curve (MPAUC). These evaluations assessed the impact of the auxiliary module and the new training method, comparing the results with state-of-the-art SED models. The experimental results on the DCASE 2024 Challenge Task 4 evaluation dataset con- firmed that the proposed method outperformed state-of-the-art SED models leveraging pre-trained architectures, achieving a performance improvement of 0.029 in the sum of the two evaluation metrics. The one-stage training approach demonstrated its abil- ity to utilize resources more efficiently compared to the traditional two-stage training method, while enabling domain-specific sound event boundary detection.|사운드 이벤트 검출(Sound Event Detection, SED)은 오디오 클립 내에서 이벤트의 유형과 발생 시간을 식별하며, 오디오 캡셔닝, 야생동물 추적, 장비 모니터링과 같은 다양한 분야에 응용한다. 이러한 응용은 오디오 기록에서 의미 있는 정보를 추출하고 분석하는 데 필수적이다. 이 논문에서는 SED를 향상시키기 위해 합성곱 블록의 마지막 계층 출력에 보조 디코더(auxiliary decoder)를 통합하는 혁신적인 접근 방식을 제안한 다.이러한통합은 SED의성능을크게향상시켜사운드이벤트의인식및분류정확도를 높인다. 또한, 이 논문에서는 기존의 합성곱신경망 및 재귀신경망(CRNN) 구조에 대한 훈련 과정과 후속 미세 조정 과정을 포함한 2단계 접근 방식을 대체하기 위해 SED 모델 성능을 향상시키는 1단계 훈련 방법을 제안한다. 이 네트워크는 통합 및 상관 관계 브랜 치를 활용하여 여러 서브 네트워크의 임베딩을 결합한다. 이러한 방법은 보조 디코더를 사용하여 두 단계의 목표를 통합하며, 사전 학습된 트랜스포머 인코더와 합성곱 계층을 함께 학습할 수 있도록 한다. DCASE 2024 Challenge Task 4 데이터셋에 적용하여, 모 델의 성능은 복합 음원 점수(PSDS)와 평균 쌍별 면적 아래 곡선(MPAUC)을 사용하여 평가한다. 이러한 평가는 보조 모듈과 새로운 훈련 방법의 영향을 측정하며, 최신 SED 모델과의 성능을 비교한다. DCASE 2024 챌린지 Task 4 검증 데이터셋에서 실험 결과 – iii – 는 제안된 방법이 사전 학습된 아키텍처를 활용한 최신 SED 모델과 성능 비교 시, 두 성능지표의 합에서 0.029 더 높은 성능을 보여줌을 확인했다. 1단계 훈련 방법은 기존 2 단계 기반 훈련 방식보다 자원을 효율적으로 사용할 수 있으면서, 특정 분야에 특화된 음원 사건 구간 탐지를 가능하게 만들 수 있음을 확인했다.MasterAbstract (English)
Abstract (Korean)
List of Contents
List of Tables
List of Figures
1 Introduction
1.1 Overview of Sound Event Detection
1.2 Problem definition
1.3 Aim of Thesis
2 Baseline Models
2.1 Overview
2.2 Feature Extractor
2.3 Embedding Extractor Using Pre-trained Model
3 Proposed Method
3.1 Overview
4 Experiments
4.1 Dataset
4.2 Evaluation Metric
4.3 Experiment Setup
4.4 Results
5 Conclusion
5.1 Conclusion
5.2 Further Study
References
Acknowledgement
Self-Assembled Peptide-Gold Nanoparticle 1D Nanohybrids Functionalized with GHK Tripeptide for Enhanced Wound-Healing and Photothermal Therapy
No full textGlycyl-l-histidyl-l-lysine (GHK) tripeptides are known for their remarkable therapeutic potential, including wound-healing, anti-inflammatory activity, and cellular regeneration. However, their clinical application has been significantly hindered by poor biological stability and limited efficacy in a physiological medium. In this study, we introduce a sophisticated approach to overcome these limitations by developing supramolecular peptide nanofiber-gold (Au) nanoparticle (NP) hybrids functionalized with GHK tripeptides. By strategically manipulating peptide self-assembly and NP integration, we demonstrated a useful platform that enhances both therapeutic efficacy and material stability. Our methodology involves the precise engineering of 9-fluorenylmethoxycarbonyl-diphenylalanine scaffolds with GHK and KHG tripeptides, enabling robust nanofibril formation through π-π stacking and hydrogen bonding. Critically, we discovered that the specific amino acid sequence significantly influences the surface exposure of lysine, directly impacting the nanohybrid’s wound-healing capabilities. The resultant nanohybrids exhibit exceptional characteristics: Au NPs are spatially confined within the peptide nanofibers, achieving a remarkably uniform size distribution of approximately 3 nm. These nanohybrids demonstrate superior near-infrared (NIR) light absorption and photothermal conversion efficiency, enabling effective eradication of cancer cells and organoids killing under NIR irradiation. This dual-functional nanohybrid integrates biocompatible and enzymatically degradable peptide scaffolds to achieve synergistic wound-healing and cancer-killing effects. By mitigating the cytotoxicity and biodegradability issues associated with conventional photothermal agents, our system provides a promising strategy to improve postoperative cancer therapy and promote tissue regeneration. This work highlights the potential of peptide-inorganic nanohybrids in advancing multifunctional therapeutic platforms for cancer treatment and tissue repair. © 2025 American Chemical Society.FALSEsciescopu
Visualizing Ice-Binding Specificity of Designed Peptides via Fluorescence-Based Ice Plane Affinity (FIPA) for Cryopreservation Applications
No full textCryopreservation requires inhibition of ice growth and recrystallization, where antifreeze proteins (AFPs) and peptides play crucial roles. To investigate how these molecules interact with ice, we employed Fluorescence-based Ice Plane Affinity (FIPA), enabling real-time visualization of peptide binding on crystallographic ice planes. We designed Fmoc-WW-based amphiphilic peptides with varying stereochemistry to modulate ice-binding specificity. FIPA analysis using single crystal ice aligned along a- and c-axes revealed that peptide chirality affects preferential binding to basal or prism planes. Peptide self-assembly morphology was characterized via TEM and Cryo-TEM to support structural understanding. These results suggest that chirality and sequence-controlled peptides can directionally regulate ice growth, providing a strategy to develop efficient, low-toxicity cryoprotectants that mimic or outperform natural AFPs
Aerodynamic flow analysis using conditional convolutional autoencoder in various flow conditions and application to CFD-based design optimization
No full textThis study investigates the accuracy and efficiency of a convolutional autoencoder in predicting flow solutions of diverse characteristics, including strong local nonlinearity and unsteady wake vortices. Modifications to the standard U-net method were made suitable for non-Cartesian CFD mesh topology, enhancing solution accuracy. Additionally, conditions for predicting flows in unseen environments are integrated into a bottleneck layer between the encoder and decoder structures, guiding flow interpolation or extrapolation and parameter types. For direct comparison, this study uses a proper orthogonal decomposition (POD)-based ROM with linear reconstruction using dominant basis vectors from the flow solution space. Interpolation and extrapolation of generalized coordinates are performed using Gaussian process regression (GPR) and Long Short-Term Memory (LSTM) networks, respectively. The Conditional Unet (CUnet)’s accuracy is demonstrated through inviscid transonic airfoil flows, capturing shock waves effectively. Additionally, it can also be used for predicting the flow field of the three-dimensional shape of the Onera M6 wing. Vortex shedding flows around an Eppler airfoil at a 16-degree angle of attack in turbulent conditions were well-resolved, with root mean squared errors under 1% compared to full-order CFD results. Remarkably, the CUnet’s computational efficiency is highlighted as the wall clock CPU time for these 2D flows was less than one second. Finally, the ROM’s effectiveness is further validated through successful multi-point shape optimization, minimizing wave drag of RAE 2822 airfoils across subsonic to transonic conditions. This resulted in a maximum drag reduction of 37.38% at Mach 0.74 without performance degradation at off-design conditions. © 2025 Elsevier B.V., All rights reserved.TRUEscopu
Enhanced Catalytic Activity via Rapid Two-Electron Transfer in Low-Spin Fe(II) Complex and Spin-State Dependent Proton Reduction Pathways
No full textThe growing interest in green hydrogen gas production has brought significant attention to the development of efficient proton reduction catalysts. A comprehensive understanding of proton and electron transfer processes within catalyst complexes is crucial for developing efficient catalysts. While the proton transfer process is influenced by the Brønsted acid used, electron transfer is an intrinsic property determined by the molecular orbitals and spin states of complexes. Complexes that rapidly transfer electrons are associated with high catalytic performance. In this study, we present a first example of low-spin FeII complex that utilizes the π* orbital of ligand for rapid two-electron transfer, resulting in exceptional catalytic performance for hydrogen gas evolution. The consecutive two-electron transfer rate was measured at 33.24 s-1, and in combination with proton transfer, the catalyst achieved an extraordinarily high turnover frequency (TOF) of 224,643 s-1 for hydrogen gas production. Conversely, a high-spin Fe(II) complex produced hydrogen gas at a relatively low TOF of 8848 s-1. These comparative experiments confirmed that the observed high catalytic efficiency is unique to the low-spin FeII complex, attributed to its distinct electron transfer mechanism. © 2025 American Chemical Society.FALSEsciescopu
Magneto: Enabling Multimodal Haptic Feedback on Paper through Magnetic Fields
No full textPaper, with its inherent versatility and adaptability, has long served as an accessible medium for interaction across different use cases. However, its analog nature constrains it to relatively fixed, static uses once created and limits its ability to provide richer sensory feedback or respond to changing contexts in the way digital interfaces can. To address this limitation, we introduce Magneto, a multimodal haptic system that augments paper interfaces with force, vibration, and thermal feedback through magnetic fields. The system combines neodymium magnets and steel plates in patch modules, using electromagnets for force and vibration and Zero Voltage Switching (ZVS) induction modules for thermal feedback. The technical evaluation confirms Magneto’s operation within human perceptual thresholds across the suggested feedback modalities. Applications of Magneto are explored in educational, gaming, and prototyping contexts, illustrating its potential to transform static paper interfaces into dynamic, interactive systems. © 2025 Copyright held by the owner/author(s)