DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Good grief, bad grief: exploring how interfaith families in Singapore navigate differences during the grieving experience
No full textSingapore prides itself with religious diversity and harmony, and interfaith relationships, marriages and family are part of the diverse social milieu. Indeed, religion remains a major social resource in today’s world, especially in matters relating to the otherworldly, such as death and the afterlife. Despite that, a critical gap in literature remains wherein little is explored about the interfaith family as a site of conflict and negotiation, especially in the grief experience in the Asian context. Taking into context Singapore’s religious diversity, this paper examines how interfaith families navigate differences, specifically in rituals and religious narratives, during periods of grieving, the challenges that they face, and the multitude of strategies and negotiations that these families adopt to make sense of life’s biggest mystery: death.Bachelor's degre
Numerical simulation of energy diaphragm walls in sand
No full textDue to climate change and the increasing focus on sustainable energy solutions, energy diaphragm walls have been introduced as a means of reducing greenhouse gas emissions. In winter, heat is extracted from the ground for heating, while in summer, heat is injected into the ground for cooling. However, there are growing concerns about the effects of thermal exchange on the stress–strain behaviour of geostructures, which could impact their long-term stability and performance.
The primary objective of this study is to investigate and evaluate the thermo-mechanical effects of heating on an energy diaphragm wall, with a particular focus on vertical strain development and earth pressure variation. A numerical model was developed using finite element analysis in ABAQUS, replicating a previously validated laboratory experiment to simulate the heating process.
The findings from the simulation reinforce the conclusion that thermal loading can significantly influence the mechanical response of diaphragm walls, potentially leading to additional internal forces and displacements. These results highlight the importance of accounting for thermal effects during the design and analysis of energy geostructures.Bachelor's degre
Developing an infrared rain gauge
No full textConventional rain gauges such as tipping bucket face challenges such as lack of real-time monitoring, mechanical wear and tears and maintenance requirements. This project aims to develop a rain gauge capable of real-time rainfall measurements using infrared sensing. The rain gauge utilises an infrared LED and photodiode to detect the amount of reflectivity of varying rainfall intensities based upon Arduino platform. Tests conducted under simulated rainfall conditions showed reasonably well response to heavy rainfall, however, sensitivity to light rainfall remained limited. The infrared rain gauge demonstrated potential as an alternative solution for real-time rainfall measurement, particularly in applications requiring quick response and low maintenance.Bachelor's degre
Context-aware techniques for real-time decision making in autonomous mobile robots (Sensor integration Part A)
No full textWith autonomous mobile robots (AMR) are being increasingly used in our daily lives, the navigation of AMR has become complicated. A suitable algorithm that allows the AMR to navigate reduces the need for constant supervision of the AMR. This project finds existing algorithm multimodal sensing to allow the AMR to navigate around environments that involves dynamic objects. Despite the many workable algorithms available, some of them were released more than five years ago and relies on older versions of software and libraries. Hence, this project facilitates the migration of the LVI-SAM algorithm from ROS Melodic to ROS Noetic by comparing versions of the libraries used and attempts to replicate the libraries used in LVI-SAM to run it on a ROS Noetic docker container.Bachelor's degre
Enhancing skin lesion diagnosis through meta-learning
No full textSkin lesion diagnosis is very crucial in the field of dermatology as early detection of
such skin lesions could allow patients to receive timely treatment. To assist doctors in
making fast and accurate diagnoses, deep learning and computer vision have emerged
as an approach to enhance the classification and analysis of skin lesions. However,
training deep learning models traditionally on small skin lesion datasets could cause
the models to overfit the data, fitting the noise rather than learning useful features.
Furthermore, skin lesion datasets exhibit a long-tailed distribution, characterized by
many images concentrated in a few common classes, while a small number of images
are spread across many tail classes. Training on such datasets would cause the model
to do well in classifying common skin lesions but fail to classify rare skin lesions due
to being exposed to only a small amount of such images. In this project, meta-learning
techniques such as MAML and Prototypical Network are explored to enhance skin
lesion classification performance, especially using small and unbalanced datasets. This
report demonstrates that meta-learning algorithms can effectively generalize across
skin lesion classes to a high degree of accuracy. Furthermore, this investigation also
explores how meta-learning techniques can improve cross-dataset skin lesion
classification. Our results show that both meta-learning techniques achieved notable
performance in both within and cross-dataset classification. However, Prototypical
Network generally outperforms MAML with the same amount of training in the metatrain phase and in cross-dataset skin lesion classification. These findings highlight the
viability of using meta-learning in improving skin lesion classification, allowing for
more reliable and accessible healthcare solutions.Bachelor's degre
Procedural generative design approach for multi-objective spatial layout planning with user interaction
No full textSpatial layout planning is a complex architectural problem with many constraints and a range of possible solutions. Finding solutions manually with traditional methods can be a tedious, time-consuming process. Hence, recent research is geared towards automated layout generation to aid space planners in creating more design alternatives, save time, and thereby ease the planning process. This thesis presents a procedural approach to spatial layout planning, incorporating user-defined inputs, and balancing automation with user control to generate adaptable layouts. Unlike traditional machine learning methods which tend to rely on large data sets and can introduce biases, the proposed procedural approach integrates concepts in cellular automata and graph theory to produce layout variations based on key user inputs of area, adjacency, and design boundaries. Three procedural algorithms have been developed here to generate layout variations deterministically from the input geometrical and topological requirements and additionally a user-specified seed that has virtually no upper limit. The model also mitigates limitations of data-driven techniques through user interaction, enabling designers to influence the generation process directly. A custom user interface is developed in procedural modelling software Houdini to allow the user to interact with the proposed generation process. Two real-world case studies are made to demonstrate the adaptability of the model to changing user input. In both cases, the proposed procedural model successfully generates layout variants that respond well to real-world design constraints. Preliminary work on developing an interface using game engine Unreal Engine further streamlines user input, highlighting the potential for more intuitive and user-friendly design experience through visual feedback and cues. Future improvements to the procedural model can incorporate more comprehensive algorithms and post-processing to address additional objectives of spatial layout planning. These advancements along with planned usability studies will enhance the practical potential of the proposed approach in fields of architecture and engineering.Doctor of Philosoph
An AI testbed for business sentiment analysis
No full textIn a data-driven business environment, the application of AI testing platforms has
become an important tool for optimizing business decisions and marketing strategies.
This research is dedicated to building an AI testing platform that analyzes and clusters
customer data using multiple machine learning algorithms to develop personalized
recommendation strategies. The platform analyzes customers’ basic information,
purchase history, browsing behavior and other data in-depth, and uses five classification
algorithms including Simple Bayes, SVM, LSTM, BERT and RoBERTa to analyze
customers’ sentiment and segment them into groups. Based on the analyzed data,
customized product recommendations and marketing strategies can be provided for
each type of group, which in turn improves customer engagement and conversion rates.
In addition, this study explores the methods of model optimization and validation
in AI testbeds, and compares the performance of several common AI models in
customer behavior prediction and sentiment analysis by analyzing their commercial data
applications. Meanwhile, drawing on the LSTM network structure in deep learning,
this study proposes an LSTM-based hybrid prediction model for long and short interval
data, which enhances the ability of LSTM in dealing with dynamic data correlations.Master's degre
Microstructure and mechanical properties of additively manufactured 410L/304L bimetal structure
No full textAdditive manufacturing has been a transformative approach to industrial production, with advantages of improved mechanical properties, complex geometries and simplified fabrication. 410L is a ferritic/martensitic stainless steel. 304L is an austenitic stainless steel. Both steels have a wide engineering applications. In this work, 410L was first deposited on an A36 substrate. Then, 304L was deposited on the 410L to fabricate a bimetal structure. The microstructures and mechanical properties of the additively manufactured 410L and 304L were studied. Metallurgical study was conducted to investigate the varied grain morphology in different regions. Microhardness test and uniaxial tensile test were conducted to investigate the mechanical performance of the two deposited stainless steel materials as well as bonding strength of the bimetal joint, with fractography study on the fracture modes as well. The study highlighted the impact of manufacturing processes on microstructure and mechanical performance, offering guidance for process optimization of the additive manufacturing of bimetal structures.Master's degre
Few-shot medical image classification: a comparative study
No full textFew-shot image Classification focuses on training models to classify images with
very few examples per category. Traditional models require large datasets with huge
amounts of labeled data, which is often really impractical to obtain. Few-shot
learning aims to overcome this limitation by enabling models to generalize from an
small number of examples. This project provides a foundation for understanding and
implementing few-shot learning techniques, with an emphasis on medical image
classification. By reproducing the experimental results from existing open-source
codebases, this project evaluates the performance of the latest state-of-the-art
methods. This project also includes a comprehensive review and comparison of
current approaches in few-shot medical image classification, highlighting their
advantages and limitations.Bachelor's degre
Wearable flexible circuit board for electrical impedance measurement
No full textThis report presents the design, development, and testing of a wearable flexible circuit board
employing Electrical Impedance Tomography (EIT) for detecting muscle contractions,
aiming for use in robotics and exoskeleton control applications. The study explores both
hardware and software components necessary for creating a reliable, portable, and
comfortable wearable system. A laboratory-based setup initially established the baseline
performance, which informed subsequent design of a compact printed circuit board (PCB)
system incorporating essential components such as Teensy 4.0 microcontroller, an enhanced
Howland circuit for stable current injection, multiplexers, and analog-to-digital converters.
Experimental evaluations involved assessing the effectiveness of various smoothing methods
to reduce signal noise in controlled mediums, including tap water containing chloride ions
and different gels, specifically UV-curable conductive hydrogel and standard electrode gel.
Among nine tested smoothing techniques, wavelet denoising consistently yielded superior
signal clarity, with the highest signal-to-noise ratio and the lowest root mean squared error.
The PCB-based system underwent preliminary validation through Arduino-based testing,
demonstrating functionality and confirming system integrity. The findings confirm the
feasibility of a wearable, flexible EIT device capable of accurately detecting muscle
engagement with optimized signal processing and reduced noise. This advancement lays a
foundation for further improvements in portable medical monitoring and rehabilitation
technologiesBachelor's degre