DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
GPU acceleration for full-field optical coherence tomography
No full textOphthalmic imaging is an essential and non-invasive tool used by clinicians to better understand, diagnose and manage retinal diseases. Optical coherence tomography (OCT) has increasingly been regarded as the gold standard for retinal imaging owing to its micrometer-scale resolution and rapid image acquisition. Full-field swept-source OCT (FF-SS-OCT) is swiftly gaining attention for its exceptional high volumetric scanning rate realized through a high-speed camera that can capture the back-scattered light from the entire sample at once. As such, the temporal resolution of the OCT system will be enhanced and motion artifacts during imaging can be minimized. However, due to the sheer amount of raw data recorded by the high-speed camera and the use of computationally intensive numerical post-processing techniques to correct motion artifacts and noise, the post-processing time to reconstruct the OCT volumes will be prolonged and undesirable in a clinical setting.
In this project, GPU acceleration was used to accelerate the post-processing pipeline and reconstruction of OCT images. With the use a GPU library known as ArrayFire, a GPU pipeline was created and implemented on 3 different sample models – synthetic phantom, mirror and ocular model. We achieved significant time reduction the post-processing processes for synthetic phantom, mirror and ocular model respectively without affecting image quality. In particular, the time consumption for the intra-volume motion correction in ocular model decreased from 43.03 s to 13.30 s. Additionally, signal averaging was found to increase signal-to-noise ratio by decreasing the background noise floor in OCT images. Our developed GPU pipeline could facilitate the future adoption of FFOCT and its functional extensions, such as optoretinography, in clinical settings.Bachelor's degre
Designing an EMG signal-based prosthetic hand
No full textSingapore's rapid development relies heavily on foreign workers, who often face significant occupational hazards, particularly the risk of severe hand injuries from machinery. The loss of a hand can significantly impact daily life, affecting basic tasks, diminishing quality of life and even causing these foreign workers to lose their only way of income.
My final year project addresses this challenge through the development of an Electromyography (EMG)-controlled prosthetic hand that prioritizes affordability and accessibility. Inspired by the struggles faced by individuals, especially the foreign workers in Singapore who suffer occupational hand injuries, this project aims to create an affordable alternative to existing prosthetics.
EMG technology measures the electrical activity generated by muscle contractions. Instead of traditional, invasive needle electrodes, this project employs surface EMG sensors, which are non-invasive and more comfortable for long-term use. These sensors detect muscle activity and transmit the data to an Arduino Uno microcontroller, which interprets the signals and activates servo motors to move the prosthetic hand in real-time. The hand itself is designed using 3D printing, which enables lightweight and customizable construction while keeping costs low.Bachelor's degre
NLOS identification using UWB-IR measured channel impulse responses
No full textWith the widespread application of indoor positioning technologies in fields such as
navigation and the Internet of Things (IoT), Ultra-Wideband Impulse Radio (UWBIR) has become a research hotspot due to its high penetration ability and resistance to
interference. However, Non-Line-of-Sight (NLOS) signal propagation can
significantly degrade positioning accuracy, making NLOS identification a key factor
in enhancing the reliability of UWB-IR ranging systems. This study systematically
investigates the signal propagation characteristics under Line-of-Sight (LOS) and
NLOS conditions, based on UWB-IR measured Channel Impulse Response (CIR).
Through a combination of experimental and theoretical analysis, various LOS and
NLOS scenarios were designed, and feature parameters such as Root Mean Square
Delay Spread, Energy Rise, and Skewness were extracted. Their distribution patterns
and classification capabilities were analyzed. The results show that delay-related
features, such as Mean Excess Delay and Root Mean Square Delay Spread, perform
optimally in NLOS recognition. Combining these features with the Support Vector
Machine (SVM) algorithm achieves 100% classification accuracy, reducing the
ranging error by 91.23% to 0.18 meters.
However, this study still faces limitations, such as the unverified generalization of the
model in complex multi-obstacle scenarios, providing directions for future research
improvements.Bachelor's degre
Design and development of an immersive game (A)
No full textThe global video game market has experienced significant growth throughout the
decade, with the PC segment contributing notably to this expansion. With the
increasing popularity of video games as a form of entertainment and escapism, the
demand for innovative and immersive game experiences continues to rise.
This project therefore demonstrates the design and development of a horror-themed
video game using the Unity Engine, created by a team of three EEE students, with its
primary goal to represent the school in submissions to the Games Development
World Championship (GDWC) or Indie Games Festival (IGF). As EEE students, this
project provides a unique experience by exploring the technical and creative
challenges involved in developing an immersive experience.
This report delves into the creative design and conceptualization of the initial idea as
well as the technical development of 3D environments, interactive elements, artificial
intelligence for enemy behavior, and sound design by leveraging the use of Unity
Game Engine, Blender and Audacity.
This report would also include playtesting feedback from volunteers, challenges
faced during the creation of the game and the different iterations created due to the
user feedback obtained, offering a valuable insight into the game’s overall reception.Bachelor's degre
Imaging with flat optics sensors
No full textFlat optics, particularly metalenses, represent a significant advancement in
optical design, offering compact and efficient alternatives to traditional lenses. This
project explores the imaging performance of flat-optic devices through simulation
and experimental implementation. A high-resolution optical spectrometer was
designed using a transmission grating and a Petzval-like lens system to analyze the
frequency response of a metalens. Simulations were conducted in MATLAB and
Zemax Optics Studio to optimize wavefront propagation, spectral dispersion and
optical design of the spectrometer. A temporary spectrometer setup was built, and its
performance was analyzed through spectral measurements and their comparison with
results of simulations. While the spectrometer demonstrated effective spectral
dispersion, further mechanical refinements are required to improve system stability
and minimize alignment errors. Futural work includes developing a more robust
spectrometer setup and do the spectrum analysis on the frequency responds in
different metalens set ups. It is expected to extend the analysis to a broader
wavelength range, including infrared and ultraviolet spectra, to further understand
the potential of flat optics in advanced imaging applications.Bachelor's degre
Low temperature chemical upcycling of Polyurethane via aprotic polar solvent assisted aminolysis
No full textAminolysis presents itself as a promising method of chemically depolymerizing
Polyurethane (PU) foams under mild conditions to obtain functional
monomers/oligomers to be used in other polymerization reactions. While various
aminolysis methods have been proposed in established studies, many of them still take
place at elevated temperatures and with excessive addition of amine containing reagents.
Furthermore, additional energy and chemical waste is generated during the purification
of the recyclate products obtained in these studies.
This final year project demonstrates the ability to perform aminolysis at lower
temperature ranges (80°C-110°C) than previously reported and with lower amount of
amine reagents needed by utilizing Aprotic Polar Solvents to increase the nucleophilicity
of amine reagents. The obtained amino-functionalized urea chains and hydroxyl
functionalized polyols were then characterized using Gel Permeation Chromatography
(GPC) and Proton Nuclear Magnetic Resonance (
1
H NMR).
In addition, the obtained recyclate products are utilized in their entirety, without
requiring further purification methods, to obtain Poly (urethane urea) (PUU) foams with
relatively high recyclate content through the addition of diisocyanates. These PUU
foams demonstrated arguably satisfactory compressive strength and modulus,
characterized through Instron compression test, and thermal stability, characterized
through differential scanning calorimetry (DSC), Thermogravimetric analysis (TGA)
and dynamic mechanical analysis (DMA). These PUU products also underwent Fouriertransform infrared spectroscopy (FTIR) characterization to observe the formation of key
functional groups against their PU precursors. However, further work is required to be
done to verify the exhibited mechanical and thermal characteristics of these PUU
samples.Bachelor's degre
Post-fire behaviour of wire arc additively manufactured aluminum alloy 5356
No full textThis experimental study tends to investigate the residual material properties of wire arc additively
manufactured aluminum alloy after exposure to elevated temperatures. The aluminum coupons
were additively manufactured using AA5356 feedstock wires through wire arc additive
manufacturing (WAAM) using cold metal transfer plus pause technology. A total of 52 WAAM
aluminum alloy dog-bone tensile coupons were created, considering three test orientations with
respect to the printing direction ( = 0, 45 and 90), different fire exposure temperatures ( =
100C, 200C, 300C, 350C, 450C, 500C and 550C) and two different cooling methods (aircooling and water-cooling). The full range of stress-strain curves and the key mechanical
properties of WAAM aluminum alloy coupons after exposure to elevated temperatures were
obtained by uniaxial tensile testing. The anisotropic material behavior related to the test
orientation was investigated, and the effect of the cooling method on residual material properties
was discussed. In addition, the relationship between the geometric undulations inherent to the
WAAM process and mechanical properties was investigated.
The results showed that Young’s modulus and yield strength were stable up to the temperature of
450°C, with slight reductions beyond this threshold. Ultimate tensile strength and ductility were
significantly influenced by the exposure temperature as well as the extraction orientation, with a
notable reduction found at the intermediate temperature range of 200-400C particularly at the
45 and 90 orientations. Cooling methods had minimal impacts on Young’s modulus and yield
strength (0.2% proof stress); however, they significantly affected ultimate strength and ductility
at intermediate temperature, where water-cooled coupons exhibited lower performance. Surface
undulations or geometric irregularities showed little correlation with residual mechanical
properties. The results of this study provide valuable insights into the structural performance and
post-fire assessment strategies of WAAM aluminum alloy components, thus making the use of
WAAM technology in structural engineering safer and more effective.Bachelor's degre
PS-CAD: local geometry guidance via prompting and selection for CAD reconstruction
No full textReverse engineering CAD models from raw geometry is a classic but challenging research problem. In particular, reconstructing the CAD modeling sequence from point clouds provides great interpretability and convenience for editing. Analyzing previous work, we observed that a CAD modeling sequence represented by tokens and processed by a generative model does not have an immediate geometric interpretation. To improve upon this problem, we introduce geometric guidance into the reconstruction network. Our proposed model, PS-CAD, reconstructs the CAD modeling sequence one step at a time as illustrated in Figure 1. At each step, we provide three forms of geometric guidance. First, we provide the geometry of surfaces where the current reconstruction differs from the complete model as a point cloud. This helps the framework to focus on regions that still need work. Second, we use geometric analysis to extract a set of planar prompts, that correspond to candidate surfaces where a CAD extrusion step could be started. Third, we present a step-wise sampling to generate multiple complete candidate CAD modeling steps instead of single-tokens without direct geometric interpretation. Our framework has three major components. Geometric guidance computation extracts the first two types of geometric guidance. Single-step reconstruction computes a single candidate CAD modeling step for each provided prompt. Single-step selection selects among the candidate CAD modeling steps. The process continues until the reconstruction is completed. Our quantitative results show a significant improvement across all metrics. For example, on the dataset DeepCAD, PS-CAD improves upon the best published SOTA method by reducing the geometry errors (CD and HD) by 10%, and the structural error (ECD metric) by about 13%.Published versionThis work was partially supported by the National Natural Science Foundation of China (62271467, 62476262, 62206263, 62306297, 62306296, 62202076), the National Key R & D Program of China (2024YFC3308000), Beijing Nova Program, Beijing Natural Science Foundation (4242053, L242096), and China Postdoctoral Science Foundation (2022T150639)
Machine learning enabled dynamic scheduling of a job shop
No full textJob shop production is a manufacturing process that caters to small, often custom or unique orders, prioritizing flexibility and adaptability in meeting demand. However, due to the complexity of the production system, scheduling has become a challenge. In addition, dynamic events, such as job insertions and machine failures, can increase the lateness of the jobs. Several approaches to job shop scheduling problems have been researched, such as Heuristics, Meta-Heuristics, and Machine Learning, with Deep Reinforcement Learning (DRL) being the most recent. However, most studies of DRL focus on job insertions, with limited attention to machine failures, a critical real-world issue. This study develops a multi-agent DRL approach that integrates machine failures into scheduling decisions. By assigning dispatching and sequencing rules across machines under varying conditions, the model adapts dynamically to disruptions. Comparative analysis demonstrates that the proposed approach outperforms conventional dispatching rules in managing machine failures. There is an exception when it comes to total cumulative tardiness, where it is less effective than the Earliest Completion Time (ECT) when paired with advanced sequencing rules. However, distribution comparison results in the model have more consistent and predictable results. Currently, the model operates on a small-scale artificial neural network (ANN) with limited features. Future improvements could involve a more sophisticated architecture tailored to specific processes, enhancing adaptability in dynamic environments.Bachelor's degre
Attitudes of Singaporeans toward the Hokkien dialect
No full textSingapore is currently facing an issue where dialectal usage has been on a steep decline. There are multiple factors that have contributed to this and many of which are intertwined. As the Hokkien language continues to decline in usage, the idea of language maintenance comes up. This paper aims to explain the factors involved in the decline and to explore possible solutions. The method used in this study would be through qualitative interviews, where the focus is on the participants and their stories told. This paper provides an alternative perspective to quantitative interviews and gives deeper insights to the participants' thoughts. Key findings for this paper seem to align with most previous studies, even though they were done in the past. The key takeaway is that there are numerous factors affecting this language shift and more disciplines need to work together to alleviate it. Hokkien usage is still on a decline, there are less and less speakers of it. Future researchers can work on the idea of emotions in languages and the cultural, practicality side of it. This opens doors for new ideas to be brought in by the Government, or it could even be an initiative from the ground-up.Bachelor's degre