DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Device-free human sensing in a deep-in-building environment
No full textHuman motion localization, tracking, and recognition technologies have gained increasing attention and significance across multiple domains in recent years. In particular, in applications such as disease monitoring and rehabilitation, elderly care, as well as emergency scenarios like fire hazards and counter-terrorism, achieving accurate and rapid indoor human motion tracking and recognition can significantly reduce costs and even save lives.
Traditional sensing and recognition systems have primarily relied on vision based methods or other existing sensor technologies. However, such approaches often entail high costs, limited flexibility, and may be unavailable in critical emergency situations.
To address this issue, this project integrates micro-Doppler features with ma chine learning techniques to develop an active RF-based wireless sensing system. The proposed system can achieve near real-time, accurate indoor human motion recognition and classification without relying on existing sensors, achieving an accuracy of over 90%.
Furthermore, we explore methods such as changing the model architecture and applying data augmentation techniques to enhance the system’s robustness, accuracy, and generalization capability.Master's degre
Large-area chemical vapor deposition growth and characterisation of 3D boron nitride foams
No full textThis project mainly focuses on optimising the chemical vapor deposition parameters for the large-area growth of 3D boron nitride foams on the nickel foam templates. The primary goal is to achieve uniformity, overall BN coverage on 3D nickel foam structure and achieve free-standing foams by investigating the effects of various parameters such as argon and hydrogen gases flow rates. Parametric studies were conducted to obtain optimal parameters that produced uniform and with great structural integrity BN foams. These foams are characterized using scanning electron microscopy, where detailed analysis takes place to identify how the change in flow rates affect the quality of the BN foams and how it can be further improved on. This study demonstrates how the CVD process is crucial for achieving and scaling up 3D BN foam productions which is suitable for application such as thermal interface devices and thermal fillers.Bachelor's degre
Strategies for implementing green logistics in Singapore part B: packaging
No full textSingapore has made significant strides in promoting sustainability, yet challenges persist in achieving
the targets outlined in the Singapore Green Plan 2030. Various stakeholders, including the government,
have taken measures to encourage businesses to adopt green practices and improve sustainability.
Despite these efforts, full achievement of the set targets remains an ongoing challenge. Several of these
companies, especially Small and Medium Enterprises (SMEs) struggles with implementing
environmental conscious business practices.
This study uses quantitative and qualitative data to better comprehend the key drivers and barriers
influencing the adoption of green logistics practiced by SMEs in Singapore. It examines the current
state of green logistics adoption within Singapore’s logistics industry and explores packaging waste
management strategies through case studies from Germany and South Korea. These international
examples offer potential approaches that Singapore can consider strengthening its sustainability agenda.
Findings suggest that the initiative currently proposed in Singapore are perceived as ineffective, and
that companies, particularly SMEs, remains reluctant to adopt sustainable packaging practices.
The report explores the potential barriers hindering businesses from adopting sustainable packaging
practices and identifies key motivating factors that encourage their transition. Through the findings, it
has been concluded that issues such as lack of collaborations between various stakeholders, cost
concern, ineffective government incentives, and lack of awareness and expertise are contributing
contributors to the slow adoption of sustainable packaging practices.
Recommendations put forward are based on the current challenges encountered and aim to enhance the
adoption rates of sustainable packaging practices in Singapore. Furthermore, this study evaluates the
effectiveness of government initiatives in promoting sustainable packaging among businesses.
Additionally, the report highlights the limitations of these recommendations and offers suggestions for
future research and improvements.
Ultimately, achieving the ambitions of the Singapore Green Plan 2030, multi-stakeholder collaboration
involving the government, businesses, and customers is essential. A unified approach among these
diverse stakeholders will be pivotal to the successful implementation of green logistics practices in
Singapore.Bachelor's degre
Foundation model-based control and simulation for robotic arms
No full textThe use of robotic arms has transformed industries like manufacturing, healthcare,
and logistics, enabling complex operations usually performed by human hands to be
easily automated. While current control methods are still pre-programmed and based
on pre-defined trajectories, they are less flexible and adaptive than needed in more
dynamic real-world environments. This current project evaluates incorporating the
latest artificial intelligence models, mostly CLIP (Contrastive Language–Image
Pretraining) and LLaVA (Large Language and Vision Assistant), into a more
interactive and adjustable robotic arm controller system. This project aims to use
these base models to produce a simulation setup in which robotic arms can be
manipulated through vision and text input, minimizing hardcoding and real-time
adjustment allowance.
This project utilizes CoppeliaSim(V-REP), a powerful robotic simulation software
that creates a platform to design various robotic arms and movements. The initial
phase of this project involved with intensive literature review and research on
reinforcement learning algorithms and their integration using machine learning
models CLIP and LLaVA. The second phase focused on configuring the codes
previously created that need to be enhanced and ways to connect the external Python
environment, Visual Studio Code to CoppeliaSim. The third phase involves the
incorporation of a machine learning model for image processing and enhancing deep
reinforcement learning models, which allows the robotic arm to imitate learning.
Looking ahead, additional work will focus on refining the system's multi-agent
capability, optimizing the interface between external coding platforms and
CoppeliaSim, and real-time testing to analyze how the system would operate under
real conditions. The study can significantly enhance the operation of robotic arms,
cost-cutting in programming, increase efficiency in task implementation, and make
more intelligent and responsive robotic systems in various industries.Bachelor's degre
Cost optimization algorithms for distributed data storage and access
No full textWith the rapid growth of distributed storage systems and the increasing demand for efficient data management, optimizing operational costs has become a critical challenge for service providers. In these systems, minimizing the combined costs of data storage and network transfer is essential, particularly as data is replicated across multiple geographically dispersed servers to ensure availability and meet user demands. This thesis focuses on addressing this cost optimization problem by developing and analyzing online algorithms that manage data placement and migration in both homogeneous and heterogeneous distributed systems, where storage cost rates of servers are uniform and distinct, respectively.
In the context of homogeneous systems, we first explore the cost optimization problem within a learning-augmented setting, leveraging simple binary predictions about inter-request times. We propose an online algorithm that achieves a consistency ratio of (5+α)/3 under perfect predictions and a robustness ratio of 1+1/α under arbitrary predictions, where α ∈ (0, 1] indicates the level of distrust in the predictions. We analyze the impact of prediction errors on performance, and establish a lower bound of 3/2 on the consistency of any deterministic learning-augmented algorithm, thus showing that achieving a consistency near 1 is unattainable. Furthermore, we validate our approach with real data access traces, confirming that improved prediction accuracy enhances algorithm performance.
To further enhance algorithm performance, we propose a randomized online algorithm employing variable storage durations for data copies. By modeling and solving an optimization problem, we derive an optimal probability density function for storage periods that allows our algorithm to achieve a competitive ratio of 1 +√2/2 ≈ 1.7. An illustrative example showcases the tightness of our competitive analysis, and experiments demonstrate that our randomized algorithm outperforms existing state-of-the-art deterministic algorithms.
We also investigate the cost optimization problem with a fault tolerance requirement, where a predefined minimum number of copies must be maintained to mitigate the impact of disruptions or failures. Our research derives lower bounds on the competitive ratio, expressed as min{n/k , 2}, where k denotes the minimum number of copies that must be maintained across n servers. We develop optimal online algorithms with competitive ratios that match these bounds, thereby enhancing the reliability of data management in the face of potential disruptions.
In the context of heterogeneous systems, we first contest a previous claim regarding the competitive ratio of an existing algorithm. Specifically, we disprove the assertion that this existing online algorithm is 2-competitive, demonstrating that its competitive ratio is, in fact, at least 3. Subsequently, we introduce our online algorithm tailored for heterogeneous systems, achieving a competitive ratio of max{2, min{μ_max, 5/2}}, where μ_max denotes the maximum-to-minimum ratio of storage cost rates among servers. We corroborate our theoretical analysis through simulations and comparative experiments against the existing algorithm.
In summary, this thesis presents a series of online algorithms designed to address cost optimization challenges in both homogeneous and heterogeneous distributed storage systems. The findings contribute to the understanding of competitive analysis in these contexts, and while they provide insights into practical solutions, further research can be carried out to explore additional avenues for improvement.Doctor of Philosoph
Delhi-Bangkok partnership: a new strategic calculus?
No full textChina's assertiveness elevates India's partnership with Thailand to strategic significance. Indian Prime Minister Narendra Modi has sought to leverage BIMSTEC to convert a friendly relationship into a potential regional counterweight against China. Is this partnership meaningful or merely symbolic? While challenging India's "Act East"
policy capabilities, it raises the question: Can it effectively counter Chinese hegemony, or will economic realities prevail?Published versio
Analysing the impact of climate change on subdaily rainfall frequency and design
No full textThis study investigates the impact of climate change on sub-daily rainfall frequency and design rainfall intensity in Singapore. The weather generator LARS-WG 7.0 is used to simulate future daily rainfall projections under various global climate models. These daily values are then disaggregated into sub-daily rainfall using the Huff rainfall distribution method. From the sub-daily data, annual maximum rainfall intensities are derived and Intensity-Duration-Frequency (IDF) curves are generated to observe trends over time.
The results indicate an overall increase in rainfall frequency, along with shifts in the period of wet and dry spells. These findings are critical for Singapore's climate resilience, as understanding potential extreme rainfall events allows for proactive infrastructure and drainage system design. Ensuring adequate capacity to handle future extreme rainfall events will help prevent flooding, maintain structural stability, and minimize disruption to daily life, which in turn contributes to the continued well-being and prosperity of the nation.Bachelor's degre
Novel view synthesis based depth perception for robotic arm control
No full textVisual servoing is a popular method used for controlling a robot’s motion using
visual feedback. In Image Based Visual Servoing or IBVS, the control process
relies on minimizing the error in 2D image space, without relying on a 3D
model of the environment. However, access to depth information can improve
the system’s performance by helping in the computation of the Image Jacobian
matrix. Traditionally, this depth information is obtained via separate sensors
such as LiDARs or using multiple cameras to achieve stereoscopic perception
of the environment. Although effective, these methods tend to be expensive as
compared to using single traditional cameras.
In recent years with the onset of improved Computer Vision models, especially
in the domain of Image Generation, the topic of Novel View Synthesis (NVS)
has gained traction. The method involves generating a completely new image of
the target object, viewed from a different angle. This dissertation demonstrates
the use of NVS to simulate a stereoscopic perception of the scene, to obtain
depth information of the object. Results from conducted experiments demonstrate
an effective system for controlling a UR5E Robotic Arm, that reduces
reliance on external depth sensors in dynamic and unstructured environments.
This approach bridges the gap between sensor-free depth estimation and effective
robotic control, providing more accessible servoing systems.Master's degre
Transnational crime in Cambodia and Indonesia: strengthening regional and national responses
No full textTransnational crime networks in Cambodia and Indonesia have been responsible for heinous cybercrime, including illegal online gambling and scams. These networks target vulnerable populations and undermine human security. Addressing these challenges requires stronger law enforcement cooperation, regulatory reforms, and coordinated regional and national strategies to dismantle the criminal networks.Published versio
Developing towards an organic colourful passive radiative cooling paint that achieves sub-ambient temperature in tropical daytime
No full textPassive radiative cooling presents a promising solution for mitigating the effects of global warming and reducing energy consumption, particularly in tropical regions where high temperatures and humidity exacerbate the Urban Heat Island (UHI) effect. This project aims to develop an innovative, water-based, colorful passive radiative cooling paint capable of achieving sub-ambient temperatures under tropical daytime conditions. The study explores the formulation and optimization of paint systems, focusing on achieving high solar reflectance (300–2500 nm) without compromising aesthetic versatility. By utilizing advanced materials such as nano-silicone resins, optimized barium sulfate (BaSO₄) particle sizes, and protective additives, the developed paint successfully combines thermal performance with visual appeal. Experimental results demonstrate that the best-performing formulation achieved a total solar reflectance (TSR) exceeding 94%, outperforming commercial paints in both cooling efficiency and durability. This research contributes to sustainable cooling technologies by addressing the dual challenges of high performance and integration into urban environments.Bachelor's degre