DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Acoustic-based process monitoring and closed-loop control in laser additive manufacturing
No full textLaser Additive Manufacturing (LAM) is a layer-by-layer production technique where
a laser beam melts and solidifies material within a powder bed or a substrate.
However, defects such as porosity, cracks, and balling can compromise the
mechanical properties and reliability of the produced parts. Therefore, to ensure
manufacturing quality in LAM, real-time monitoring is necessary. Traditionally,
high-speed cameras are used to monitor real-time process dynamics of melt pool,
detect defects, and conduct closed-loop feedback control. However, these optical
sensors are costly, difficult to install, and sensitive to variations in lighting conditions.
Acoustic sensors, on the other hand, serve as a cost-effective alternative as the
laser-material interaction generates process-specific acoustic emissions that
inherently contain information about the melt pool's stability and defect formation.
Therefore, this study explores an alternative monitoring approach using acoustic
signals. The proposed method consists of two interconnected tasks: inferring melt
pool visual characteristics using acoustic signals and implementing acoustic-based
closed-loop control. An unsupervised autoencoder model is trained to extract
latent-space representations of melt pool images. A convolutional neural network
(CNN) is then trained in a supervised manner to predict these latent-space
representations using acoustic signals. Experimental results demonstrate that CNN
achieves a high correlation coefficient of 94.91% in predicting the latent-space
features derived from the autoencoder, validating the feasibility of using acoustic
signals as an alternative for visual monitoring.
The study then explores real-time acoustic-based closed-loop control. By analyzing
process-induced acoustic signals, the system dynamically adjusts LAM parameters to
mitigate defect formation. By demonstrating the ability to infer visual defect
characteristics from sound and applying this knowledge for process control, this
research advances autonomous defect detection in LAM, improving process
reliability and reducing operational costs.Bachelor's degre
Data-driven framework for lane change prediction and assistance in intelligent transportation systems
No full textWith the development of autonomous driving technology, self-driving vehicles will
gradually become prevalent in human society. This advancement will significantly
enhance the safety, e!ciency, and intelligence of transportation systems, where
human-driven and autonomous vehicles share the roads harmoniously. However,
achieving this harmonious vision requires an understanding and prediction of human
drivers’ lane-changing behaviours, as inappropriate lane changes by human
drivers are a major cause of tra!c accidents. Therefore, one of the current major
challenges is how to predict human drivers’ lane-changing behaviours and assist
driving decision-makers in making reasonable, e!cient, and safe lane-changing decisions.
Additionally, it is crucial to reduce trajectory conflicts between vehicles
in micro-scenarios to make road tra!c safer, smoother, more e!cient, and more
environmentally friendly. To addressing these challenges, this thesis proposed a
data-driven lane change prediction and assistance framework for intelligent transportation
systems. Firstly, we developed a novel data-driven lane change trajectory
prediction method which enhances the transfer of knowledge from observed
to unknown scenarios using the few-shot learning (FSL) concept. Our proposed
method enables a pre-trained LSTM model to be quickly deployed in new scenarios
with only a few samples. Secondly, we developed a Lane Change Attention
Model (LCAM) based on the Transformer architecture to solve the lane change
manoeuvre prediction problem. Contrast experiments have been conducted on
several open-source tra!c datasets, which demonstrated that LCAM had made
some improvements compared to other baselines. Last but not least, we developed
a Lane-Change Responsibility-Sensitive Safety (LC-RSS) model to improve
the safety of lane-change decisions. The simulation results have shown that by
implementing LC-RSS, not only the safety of a single vehicle’s lane change trajectory
has been enhanced, but also the tra!c flow speed of the urban transportation
system has been increased.Doctor of Philosoph
Sustainable LC3 cement from calcined ferric sludge and MSW slag: enhancing reactivity with grinding aid
No full textThis project aims to explore and study the effects of Municipal Solid Waste (MSW)
slags and calcined ferric sludge as substitutes to calcined clay in limestone calcined
clay cement (LC3). The substitution aims to address environmental concerns such as
carbon emissions and solid waste generation from water treatment processes, as well
as landfill space consumption. The impact of grinding aid to reduce particle size will
also be studied on the powder samples. This project analyses the physical properties,
crystalline properties, and chemical properties of incorporating the slags and sludges
into concrete mixtures.
For samples that experience agglomeration when ball milling mechanical grinding
takes place, grinding aid was useful in preventing agglomeration. Sieve analysis was
possible due to prevention of agglomeration, resulting in higher yield of small particle
collection. However, for samples with no occurrence of agglomeration, grinding aid did
not have any effect.
Wet mode and dry mode particle analysis gave slightly different results. The usage of
modes depends on the sample used for testing. Especially for sludge, dry mode testing
was more accurate.
Smaller powdered samples resulted in stronger compressive strength of mortar mixes.
Some mixes (with grinding aid) produced mortar samples with ⅔ compressive strength
of conventional portland cement mixes.
In conclusion, this study advocates for the use of grinding aid for smaller particle size
collection for stronger concrete results. Incorporation of MSW slags and calcined ferric
sludge may be feasible for non-structural or light-load structural uses. More testing
should be done to possibly optimise the strength of this special LC3.Bachelor's degre
Multi-objective design optimization of cryo-polygeneration systems for urban microgrids: balancing cost-effectiveness and sustainability
No full textSmall- and medium-scale polygeneration systems provide multiple energy services to urban districts, including business parks, universities, and hospitals, offering significant energy, economic, and environmental benefits. These systems enhance energy efficiency, reduce cost, and lower emissions, especially in tropical urban areas with year-round cooling demand. This paper presents a multi-objective design methodology for polygeneration systems in tropical climates, integrating distributed energy technologies such as medium-scale gas turbines, solar photovoltaic, chillers, and energy storage. The proposed methodology adopts optimization approach to determine the optimal configuration and capacities of distributed energy systems to achieve various business goals, such as economic and sustainability objectives. The multi-objective design methodology employs a three-level optimization approach: simulation using the Transient System Simulation Tool, Pareto-based search conducted in Matrix Laboratory software, and an interface connecting the simulation tool with the optimization platform. This process generates a Pareto front of design solutions, balancing economic and environmental objectives. The proposed design methodology was applied to a case study of a polygeneration system at the Nanyang Technological University campus in Singapore, optimized across four superstructure configurations. Results show significant reductions in energy costs and CO2 emissions compared to the baseline, with a comparative analysis of various scenarios. The findings provide a comprehensive view of design options, allowing energy experts to balance economic and sustainability objectives for optimal system performance.National Research Foundation (NRF)Submitted/Accepted versionThis study is supported under the RIE2020 Industry Alignment Fund – Industry Collaboration Projects (IAF-ICP) Funding Initiative, as well as cash and in-kind contribution from Surbana Jurong Pte Ltd
Foodprints — following the spread of crops & languages in the world
No full textTaking an anthropological and linguistic approach, this thesis examines names given to 10 plants across 28 languages, including Chinese, Marathi, and Nahuatl. In doing so, it identifies recurring naming conventions, which will be discussed across two chapters.
The first chapter focuses on borrowing, analysing the motivations behind it — such as the desire to appear more prestigious or the need to fill lexical gaps — and the phonological adaptations loanwords undergo to fit the borrowing language.
The instinct to distinguish what is “ours” from what is “foreign” extends to plant names. The second chapter examines linguistic marking, a strategy used to differentiate foreign plants from local crops. This distinction is especially important in cooking, where mistaking a similar-looking plant may be a recipe for disaster. Linguistic marking can take various forms, including separate scripts (as in Japanese), specific lexical markers (for example, in Thai), metaphorical naming, or references to a plant’s roots (origins).
By investigating how different cultures receive, adapt to, and name new plants, this paper hopes to cultivate a deeper appreciation for the rich histories embedded in the names of these unassuming plants.Bachelor's degre
3D long-range object detection under rainy conditions
No full textAutonomous Vehicles (AVs) are at the forefront of modern transportation innovation, offering
the promise of safer and more efficient mobility. Traditionally, AV perception systems have
relied heavily on camera-based sensors. However, these vision-based systems are highly
sensitive to lighting variations and visibility constraints, such as glare, darkness, or fog. To
overcome these limitations, recent advances have integrated deep learning with LiDAR and
Radar sensors, which offer improved spatial awareness and range perception. Despite these
enhancements, object detection in AVs remains a significant challenge, particularly under
adverse weather conditions that can degrade sensor performance.
Among these sensors, LiDAR provides high-resolution 3D spatial data but suffers performance
drops in scenarios involving rain, fog, or snow. Conversely, Radar is more robust in harsh
weather but offers lower spatial resolution. Given their complementary characteristics, fusing
LiDAR and Radar data presents a promising approach to overcoming individual limitations and
enhancing perception robustness. This work aims to explore and implement a sensor fusion
strategy to improve object detection performance, particularly in challenging environmental
conditions.
The proposed fusion strategy involves preprocessing and aligning data from both LiDAR and
Radar sensors using temporal synchronization, followed by the application of a deep learning
detection model—specifically Faster R-CNN. This study leverages the RADIATE dataset, which
provides sensor data under diverse weather conditions. By comparing single-sensor performance
against fused data, we demonstrate how the integration of Radar and LiDAR enhances detection
accuracy and reliability, paving the way for more resilient AV perception systemsBachelor's degre
Design of CMOS bulk-driven operational amplifier
No full textFor low-voltage circuit design, the transistor threshold becomes one of the major obstacles that leads to the constrained voltage headroom under limited supply operation. To tackle the issue, bulk-driven CMOS design can permit lower supply operation but at the expense of finite device transconductance and so forth. The objective of the project is to investigate the circuit technique for low-voltage CMOS operational amplifier that permits bulk-driven circuit design to extend the performance in 40nm CMOS technology. This will result in useful analog signal-processing applications. The Cadence tools will be used for simulation and verification.Bachelor's degre
Vision-based autonomous unmanned aerial vehicle (UAV) precision landing on a moving platform
No full textThis report presents the development of a vision-based autonomous precision
landing framework of an unmanned aerial vehicle (UAV) targeting a moving
unmanned ground vehicle (UGV) platform. The proposed framework is
designed to operate entirely in GPS-denied environments by relying solely on
onboard sensors and computing for state estimation and flight control. A
downward-facing camera mounted on the UAV is used to detect an ArUco
marker placed on the UGV to enable real-time visual tracking and alignment.
Complementary off-the-shelf sensors including an optical flow sensor and a
Light Detection and Ranging (LiDAR) rangefinder are integrated into the UAV
system to provide velocity and altitude feedback for stable autonomous flight.
A custom-built A2-sized UGV features a cost-effective design with six
Dynamixel XL430-W250-T motors and mecanum wheels to enable
autonomous ground missions which serve as a mobile landing platform for the
UAV. Multiple flight experiments were conducted to evaluate the system’s
performance across varying altitudes and UGV speeds. The flight data results
demonstrate that the proposed framework achieves reliable marker detection
and precision landing under controlled conditions. This work contributes a
scalable and cost-effective solution for autonomous UAV precision landings in
dynamic environments without relying on any external GNSS or GPS
communication signals.Bachelor's degre
Enhancing fire disaster management in prediction, detection and evacuation applications using data-driven methods
No full textFire disasters continue to pose a considerable risk to both urban and rural settings, resulting in profound consequences for human life, infrastructure, and the environment. Fire disaster management is a complex field of study that focuses on reducing the adverse effects of fire-related emergencies on human life, property, and the environment. Effective management of fire disasters fundamentally depends on a cohesive framework that encompasses fire prediction, fire detection, and path optimization in evacuation. Although there are numerous studies that use advanced technologies combined with fire disaster management systems for higher efficiency and safety, there is still a lack of a highly efficient fire disaster management system that includes the management of different stages of fire disasters.
To promote the management of fire disasters and ensure the safety of both human lives and properties, three different hybrid approaches are proposed to solve the problems encountered, which start from the fire prediction stage, followed by the fire and smoke detection stage, and lastly the evacuation stage after fire for occupants. The key findings are summarized as follows:(1) Driving factors of fire are analyzed, and a deep neural network (DNN) method that can forecast the wildfire with high accuracy is proposed. (2) An improved deep learning algorithm based on YOLOv8n for smoke and flame detection is tested and shows better performance after improvements with the loss function and self-attention mechanism. (3) An IoT-based dynamic path optimization method for evacuees is proposed, which integrates Dynamic Graph Neural Networks (DGNN), Whale Optimization Algorithm (WOA), and Markov Decision Process (MDP) to enhance path efficiency and safety.Doctor of Philosoph
Non-reciprocal terahertz topological sensor on a silicon chip
No full textOn-chip non-reciprocal light-matter interaction improves sensor performance by leveraging direction-dependent differences in detection signals. Here, the experimental realization of a non-reciprocal terahertz topological sensor (NTTS) is reported through magneto-optical integration on a silicon valley photonic cavity chip, enabling dual-frequency non-reciprocal sensing. Through sensing the ultrathin polyimide membranes of varying thicknesses, non-reciprocal group delay sensitivities of 0.46 and 0.24 ns µm−1 are demonstrated for the detuned clockwise and counterclockwise cavity modes, respectively. NTTS also provides non-reciprocal sensitivity in the position sensing experiment, yielding group delay sensitivities of 2.63 and 3.21 ns mm−1. This non-reciprocal topological sensor enables a new sensing paradigm and can be integrated into compact topological photonic circuits for on-chip gyroscope, biochemical sensors, and environmental monitors.National Research Foundation (NRF)Published versionAll the authors acknowledge the research funding supported by the Na-tional Research Foundation Singapore, Grant No: NRF-MSG-2023-0002