DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Superstructure design, data-driven mixed-integer optimization, and guidelines for techno-economic-environmental enhancement of blended amine-based CO2 capture in natural gas combined cycle power plants
No full textBlended amine-based post-combustion CO2 capture (PCC) for natural gas combined cycle (NGCC) power plants lack efficient superstructure design, viable mixed-integer optimization approach, and practical guidelines to reduce capture cost, heating energy consumption, and amine emissions. This study introduces an efficient superstructure design comprising 12 feasible configurations, a viable data-driven mixed-integer optimization approach, and design and operation guidelines to enhance the techno-economic-environmental performance of decarbonizing a 400 MW NGCC power plant using Monothanolamine/Piperazine (MEA/PZ). First, we develop a high-fidelity process simulator to decarbonize the NGCC power plant using MEA/PZ. The validated simulator is then employed to create the superstructure design consisting of the 12 feasible configurations to reduce the heating energy consumption, capture cost, and amine emissions. Due to the cyclic nature of the closed-loop PCC process, each steady-state simulation takes 20 min to complete. To expedite the process, we develop a one-hot vector deep neural networks (OHV-DNN) model based on 4220 synthetic data cases, enabling accurate prediction of key performance indicators for the superstructure design in just one second. This DNN model is subsequently used to formulate economic mixed-integer optimization problems, which can be solved within one minute. Under the optimal conditions for capturing 90% of CO2 and emitting less than 1 ppm of MEA and PZ concentrations, the MEA/PZ-based optimal design significantly reduces the capture cost by 14.34% and 24.39% and cuts the heating energy consumption by 17.21% and 30.82% compared to the conventional design using MEA/PZ and MEA. Additionally, the analysis reveals the importance of design selection and optimal operating zone, as well as the impact of decision variables on the techno-economic-environmental performance of the MEA/PZ-based optimal design. The findings and the proposed approach are highly beneficial for decarbonizing NGCC power plants and can be extended to other concentrated CO2 sources.Ministry of Education (MOE)National Research Foundation (NRF)Public Utilities Board (PUB)Submitted/Accepted versionThis research is supported by the National Research Foundation, Singapore, and PUB, Singapore’s National Water Agency under its RIE2025 Urban Solutions and Sustainability (USS) (Water) Centre of Excellence (CoE) Programme, awarded to Nanyang Environment & Water Research Institute (NEWRI), Nanyang Technological University, Singapore (NTU). This research is also supported by the Ministry of Education, Singapore, under its Academic Research Fund Tier 1 (RG63/22). Additionally, this research receives support through Schmidt Sciences, LCC. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of the National Research Foundation, Singapore and PUB, Singapore’s National Water Agency
Ionic electroactive polymer actuators using quaternary ammonium iodide (QAI) containing hydrogels operational at low voltages
No full textQuaternary ammonium iodide (QAI)-containing hydrogels with different alkyl side chain lengths were synthesized and used to develop ionic electroactive strip actuators. The strip bent at relatively low voltages, exhibiting a 5.0 mm bending displacement over the 30 mm strip length at 0.6 V, for example. The bending displacement largely depended on the applied voltages and alkyl side chain lengths. The anticipated bending mechanism is the movement of the iodide anions toward the anode, thereby driving volume contrast in the strip near the anode (volume expansion) and cathode (volume contraction). This mechanism was experimentally confirmed by monitoring the location of the iodide anions in the strip by using scanning electron microscopy with energy-dispersive X-ray spectroscopy (SEM-EDX) and confocal microscopy. The obtained QAI-containing actuators are operational at relatively low voltages and in water-containing environments and also have antimicrobial properties, which would be useful for biological environment applications.Ministry of Education (MOE)Submitted/Accepted versionThis work was supported by Academic Research Fund (AcRF) Tier 2 from Ministry of Education in Singapore (MOE-MOET2EP10121-0005)
Analyzing and comparing LM architectures for named entity recognition
No full textThis project aims to compare and evaluate the performances of Encoder-only LLM,
Decoder-only LLM and Encoder-Decoder LLM for Named Entity Recognition (NER).
Despite Encoder-only models being known to be better than generative models such
as Decoder-only models for NER tasks, due to the rapid growth of generative artificial
intelligence development, there have been numerous research done to find ways to
improve capabilities of these generative models in various fields, with NER being one
of such fields.
As such, this project aims to compare the difference in performance between the
different model architectures as well as to seek a way to improve the performance
of Encoder-Decoder/Decoder-only models to match that of Encoder-only models to
potentially find a way to make NER tasks more accessible and cost-effective in practical
applications.Bachelor's degre
Can LLM agents recognize demographic heterogeneity in economic games?
No full textHuman societies are diverse and the recognition of such diversity is central to research in social science. As large language models (LLMs) like GPT-4 are increasingly used to simulate human behavior, a key question arises: Can LLMs capture the diversity at individual level and, more importantly, recognize the structure of human society by treating each group of individuals as an integral part of the society? We address this question with two experiments using GPT-4 agents across five canonical economic games. In Experiment 1, the GPT-4 agent is assigned a specific demographic identity (e.g., age, gender, income) and makes decisions with explanations. In Experiment 2, the GPT-4 agent remains neutral while its opponent is assigned demographic attributes. We find that GPT-4’s decisions and rationales do shift in response to demographic prompts, echoing some human-like patterns (e.g., greater generosity toward disadvantaged groups). However, these shifts remain superficial. Rather than capturing realistic differences between subgroups, GPT-4 agents often adjust all their responses uniformly relative to a no-demographics baseline, fail to recognize each demographic group as a part of the population. This limitation indicates that current LLMs do not truly internalize the structure of human social diversity. If an LLM agent cannot distinguish how different groups contribute to outcomes, it may misrepresent heterogeneity and produce biased conclusions in research or policy applications. Our findings underscore the imperative for cautious implementation of LLMs in individual-level simulations and broader practical applications.Submitted/Accepted versio
Biophysical approach for precision membrane engineering of lipid nanoparticle systems
No full textMultivalent ligand-receptor interactions involving soft-matter, membrane-enveloped biological and biomimetic nanoparticles (e.g., virus particles, exosomes, vesicles, lipid nanoparticles) at cellular membrane interfaces are critical to a wide range of biological functions and pathologies. This class of multivalent interactions has been widely explored by biological and biophysical measurement approaches but it has proven analytically challenging to track corresponding multivalency-related nanoparticle shape deformation processes due to limitations in conventional measurement approaches. Such deformation processes are biologically important and influenced by a balance between the multivalent binding interaction energy and membrane bending energy of soft-matter nanoparticles. From an engineering perspective, developing measurement approaches to characterize multivalency-induced nanoparticle shape deformation at lipid membrane interfaces can help to provide a biophysical understanding about how various design parameters affect the membrane nanomechanical properties of nanoparticles. As a model experimental system to characterize soft-matter nanoparticle-membrane interactions, this thesis presents a localized surface plasmon resonance (LSPR)-based measurement platform to characterize the binding interaction of ligand-modified lipid vesicles with a receptor-functionalized supported lipid bilayer (SLB) platform and to obtain nanomechanical insights into multivalency-induced vesicle shape deformation processes based on a combination of experiment and theory. Within this scope, analytical models of the LSPR-related physics and multivalent ligand-receptor complex dynamics were developed in order to quantify structural and energetic aspects of vesicle deformation. By utilizing this measurement approach and corresponding analytical models, it was possible to elucidate how key parameters like receptor and ligand densities, vesicle size, cholesterol fraction in vesicles, and solution pH affect multivalency-induced vesicle attachment and shape deformation processes. This study discusses the results in terms of the analytical merits of the LSPR technique to track nanoparticle shape deformation at lipid membrane interfaces compared to other measurement options as well as insights into how multivalent interactions affect the nanomechanical properties of lipid vesicles. The measurement approach and analytical framework developed in this thesis can be broadly extended to evaluate the multivalent interactions of different types of membrane-enveloped biological and biomimetic nanoparticles with engineered properties and also to demonstrate the utility of membrane biophysics approaches to study interfacial phenomena related to nanoparticle-membrane interactions in general.Doctor of Philosoph
Attacking and defending multi-robot systems
No full textMulti-robot systems have offered a solution to completing tasks efficiently. However, with such
large systems also comes a large attack surface, presenting itself as a possible security risk in
the cyber world. This project aims to explore Byzantine attacks on the navigation and
exploration workloads in ROS-based multi-robot systems. By analysing the architecture of
these workloads and their implementations, several attacks were identified with the aim of
disrupting each workload execution, either through delay of the workload completion or
causing the workload to fail to accomplish its goals. Through simulating the attacks on the
workloads with a ROS-based simulator, the report evaluates the severity of each attack to show
the vulnerabilities in the existing implementations of these ROS-based systems. Additionally,
several defensive controls were proposed to improve the robustness of these workload
implementations.Bachelor's degre
BlockTicket - an ethereum-based event ticketing system
No full textThis study investigates how blockchain technology, particularly in Ethereum, can
address the fundamental challenges of ticketing systems, namely fraud, scalping,
and the lack of a transactional record. The proposed decentralized ticketing
system consists of a full-stack implementation including smart contracts, a
backend server, a frontend website, and a mobile application. It introduces a two-
layer contract architecture separating event management from ticket operations,
utilizes IPFS for metadata storage, and implements a dynamic verification
mechanism. The system aims to enhance ticket security, transparency, and control
while balancing safety and user experience. Experimental results validate the
effectiveness of the solution in enhancing ticket traceability and regulating the
secondary market.Bachelor's degre
GaussianAnything: interactive point cloud latent diffusion for 3D generation
No full textWhile 3D content generation has advanced significantly, existing methods
still face challenges with input formats, latent space design, and output
representations. This paper introduces a novel 3D generation framework that
addresses these challenges, offering scalable, high-quality 3D generation with
an interactive Point Cloud-structured Latent space. Our framework employs a
Variational Autoencoder (VAE) with multi-view posed RGB-D(epth)-N(ormal)
renderings as input, using a unique latent space design that preserves 3D shape
information, and incorporates a cascaded latent diffusion model for improved
shape-texture disentanglement. The proposed method, GaussianAnything, supports
multi-modal conditional 3D generation, allowing for point cloud, caption, and
single/multi-view image inputs. Notably, the newly proposed latent space
naturally enables geometry-texture disentanglement, thus allowing 3D-aware
editing. Experimental results demonstrate the effectiveness of our approach on
multiple datasets, outperforming existing methods in both text- and
image-conditioned 3D generation.Submitted/Accepted versio
VR-AI lab safety training: a virtual reality game with AI-powered guidance for hazardous scenario response
No full textThis report presents the design and implementation of a single-player Virtual Reality
(VR) training system enhanced with Artificial Intelligence (AI) for immersive and
adaptive laboratory safety education. Traditional methods of lab safety instruction often
lack interactivity and fail to simulate real-world urgency, limiting student engagement
and retention. To address these gaps, we developed a Unity-based VR application
that simulates hazardous lab scenarios—such as chemical fires—allowing users to
interact with virtual equipment and perform safety protocols in real time. An AI
assistant powered by GPT-4 is integrated via LangChain, offering context-sensitive,
natural language guidance during training. Semantic caching with Redis optimizes
response latency and reduces redundant API calls by storing and retrieving AI responses
based on game state and query embeddings. A Rust-based backend handles session
management and scoring persistence through PostgreSQL, ensuring scalability and
clean data separation.
The system emphasizes modularity through dedicated Unity managers for transitions,
game logic, and API interaction. Scoring is dynamically calculated based on hazard
resolution time and completion of weighted safety subtasks. Initial validation was
conducted through manual testing of gameplay mechanics, interaction reliability, and AI
response behavior. This work contributes a modular, scalable framework for AI-guided
VR safety education and demonstrates how experiential learning models, supported
by intelligent tutoring systems, can enhance training in safety-critical environments.
Future directions include expanding to multiplayer, introducing voice-based AI queries,
and increasing the fidelity of lab equipment simulations.Bachelor's degre
Direct 4D printing of hydrogels driven by structural topology
No full textFour-dimensional (4D) printing combines shape-morphing materials and three-dimensional (3D) printing technology, enabling efficient fabrication of complex shape-changing structures. However, 4D printing of hydrogels into structures with complex shapes suffers from poor printability, which limit their practical applications. Here, we present an efficient strategy for direct 4D printing of hydrogels, leveraging intricate structural designs and highly viscous hydrogels. This strategy facilitates programmable shape-morphing through precise control of filament spacing and orientation, resulting in gradient swelling behaviours when the structures are immersed in a Ca2+-ion solution. Our study also reveals the critical role of printability in improving shape-morphing performance. On this basis, we propose a practical solution to enhance the shape-morphing capability of hydrogels with limited inherent performance by improving their printability through the addition of viscous additives such as MC or PVA. Overall, this strategy expands the list of hydrogels suitable for 4D printing, demonstrating compatibility with both synthetic and natural hydrogels, including Alginate/Methylcellulose (ALG/MC), gelatin methacryloyl (GELMA), and ALG/polyvinyl alcohol (PVA). Various sophisticated plant-inspired shape-morphing behaviours can be achieved in 4D-printed hydrogels through precise control of structural topology. The combined strategy of employing highly viscous hydrogel with intricate structural design demonstrates vast potential for applications in biomimetic soft robotics.Nanyang Technological UniversityNational Research Foundation (NRF)Published versionThe authors acknowledge the financial support from the Singa-pore Centre for 3D Printing (SC3DP) and the National Research Foundation, Prime Minister’s Office, Singapore under its Medium-Sized Centre funding scheme