DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
A data analytic approach for assessing XLPE cable insulation condition via resistance measurements
No full textThe insulation resistance (IR) test has been widely conducted by electricity companies to assess cable insulation health status due to its ease of applicability; however, there exist cases in which medium-voltage (MV) power cables pass the IR test but fail in service after reenergization. So far, historical IR data obtained via measurements have been recorded but have never been systematically studied to improve the accuracy of detecting unhealthy cables. This article proposes a data analytic approach using historical IR data to identify patterns for distinguishing between healthy and unhealthy cables and assess the cable insulation condition. The proposed approach first leverages a two-parameter Weibull analysis to link the failure probability of the cables to their ages. Such analysis sheds light on the classification of the cables with respect to their age and material. Next, a diminishing method (DM) is used to set the critical IR values and provide maximum detection of the unhealthy cables with minimum misclassification of the healthy cables as unhealthy. Finally, a self-organizing-map-based support vector machine (SOM-SVM) is used to classify the cables as healthy or unhealthy. The hybrid DM-SOM-SVM approach is applied to the historical IR data of 22 kV and 6.6 kV cross-linked polyethylene (XLPE) cables. Compared to current industrial IR criteria for insulation condition diagnosis, the proposed approach allows the detection of 18.5 and 1.8 times more unhealthy 22 kV and 6.6 kV XLPE distribution cables, respectively.National Research Foundation (NRF)Energy Market Authority (EMA)Submitted/Accepted versionThis work was supported in part by the SP Group, National Research Foundation, Singapore, through the Energy Market Authority from its Energy Programme (EP) under Award EMA-EP010-SNJL-005 and in part by Nanyang Technological University
Design of a high-fidelity audio amplifier
No full textThis project aims to design a noise resistance analogue audio amplifier with sound adjusting
capability to accommodate different music genres. Electronic music playback has been an
integral part of human civilisation since its invention. Current technology has evolved to
digital systems, and wireless configuration. However, they faced a key issue, where features
are less accessible by users due to the requirement of using an application and limited.
Traditional analogue amplifiers are less popular due to higher production cost, higher power
consumption, and more susceptible to noise. They are feature packed and is accessible to
the consumers to adjust to their preferences. This project aims to develop an analogue
amplifier with low noise characteristics and lower power consumption. This report has
shown that common mode choke and common mode rejection amplifier shows probable
noise reduction capability. The addition of automatic gain controller’s ability to reduce
distortion from excessive gain settings shows improvement on audio quality on higher
volumes. Reducing inrush current by implementing a soft starter showed promising result of
current limiting during the startup phase reducing probability of component failure with
addition of short circuit protection to prevent burning components. They have potential to
impact consumers through low cost solutions keeping signal integrityBachelor's degre
The technical architecture & design choices behind Hotpot City: Boiling Point
No full textHotpot City: Boiling Point is a narrative role-playing game (RPG) that combines branching dialogue, character stats, and player choices to shape the story. This project focuses on building a custom dialogue system using Ink and Unity, allowing in-game characters to respond dynamically based on the player’s skills and actions. Features include interactive quests, inventory management, and skill checks—common in classic RPGs. The result is a system that supports immersive storytelling and player agency in a richly designed game world.Bachelor's degre
Optimizing speech representation learning for enhanced noise robustness in downstream applications
No full textThe primary objective of this thesis is to enhance the effectiveness and efficiency of speech representations, specifically improving noise robustness for downstream applications. Current speech representation learning frameworks, despite their advanced foundational knowledge and powerful speech understanding capabilities, fall short in critical areas essential for real-world applications, such as adaptability to noise, expressiveness, and computational efficiency. For instance, their performance varies greatly across different levels of noise corruption, showing high vulnerability to distortion from external influences. Moreover, learning for domain adaptation and performing inference are computationally intensive due to the large scale and complex design of the model structures. Hence, this thesis introduces innovative solutions to bridge these gaps, offering substantial improvements over existing methods.
Adaptability to Noise: We address noise robustness by integrating Barlow Twins learning, an advanced regularization technique that strategically reduces channel redundancy and effectively disentangles speech features, as presented in chapter 3. This ensures that our models capture essential, noise-free features critical for accurate speech recognition, even in adverse acoustic environments, thereby ensuring more stable and robust performance to noise distortion.
Expressiveness: To enhance the expressivity of pre-trained models, we employ a parameter-efficient fine-tuning approach that incorporates the proposed deep filter tuning with Feature-wise Linear Modulation (FiLM)-inspired integration, as detailed in Chapter 4. This method refines the handling of speech nuances, allowing the models to more effectively express important information through targeted feature extraction from FiLM. This adaptation better accommodates diverse vocal attributes and acoustic variations.
Efficiency: Our proposed deep filter tuning strategy enables efficient adaptation of frozen pre-trained models with the aim of minimizing the number of trainable parameters. This approach optimizes computational resources and maximizes efficiency, particularly when faced with constraints in computational memory. In addition to the work, we also innovate in data compression through multi-band quantization, optimizing bit allocation by prioritizing perceptually significant features in chapter 5. This approach leverages psychoacoustic principles to enhance the efficiency of speech processing and is particularly beneficial for speech synthesis and voice conversion tasks, ensuring high fidelity in the outputs.
Our methods have been tested across tasks such as automatic speech recognition and speech reconstruction, demonstrating significant enhancements in accuracy, robustness, and processing efficiency. These improvements underscore the potential of our approaches to revolutionize speech representation learning, making it more adaptive, scalable, and context-aware.
By addressing the inherent limitations of current models, this research advances the field of speech representation learning, setting a new benchmark for future developments and applications in adaptive and efficient speech technology.Doctor of Philosoph
Redefining sport watching: the impact of process vs. outcome focus on prosocial behavior
No full textConsumer behavior studies on the process and outcome focus paradigm have primarily been conducted in the context of material goods, with implications largely confined to marketing strategies and individual-level consumer benefits, rather than broader societal outcomes. This study offers a novel extension of the process and outcome focus paradigm into sport watching, examining its effects on prosocial behavior. Using a 2 (process vs. outcome focus) × 2 (close loss vs. close win) between-subjects design, results did not provide evidence that process-focused sport watching leads to greater prosocial behavior than outcome focus. No evidence was found for the mediating role of hope and moderating role of game outcome in a first-stage moderated mediation analysis (Hayes Model 7). Nevertheless, our finding that process focus leads to greater hope compared to outcome focus serves as a stepping stone toward reorienting the sport-watching experience to emphasize the journey over the outcome—an insight which holds meaningful implications for sport marketing. Sport sponsors’ incorporation of process-focused sport watching into promotional campaigns may offer a refreshing alternative to the prevailing emphasis on game outcomes as the be-all and end-all—reshaping brand perceptions by positioning themselves as advocates of athletes’ journeys. By capitalising on emerging platforms like TikTok, where content creation thrives, sport media outlets can pioneer immersive process-focused sport watching experiences— boosting viewer engagement. Although these efforts require upfront investment, they present a valuable opportunity to drive growth in sport marketing while inspiring hope among sport consumers.
Keywords: prosocial behavior, process focus, outcome focus, hope, sport watchingBachelor's degre
ASEAN's trade networks as pillars of security and stability
No full textThe Southeast Asian grouping, ASEAN, will be better equipped to withstand geopolitical shocks and global challenges, especially when traditional security challenges become increasingly formidable, by treating its trade networks as shared security assets and building a resilient system inspired by stable physical structures.Published versio
Assembling artificial 2D Van der Waals heterostructure for optoelectronic devices
No full textTransition Metal Dichalcogenides (TMD) have been popular among two-dimensional
(2D) materials family for optoelectronic applications. Moreover, their oxides,
transition metal oxides (TMO) have gained recent attention due to the post-oxidation
benefits they offer in TMDs. Various oxidation techniques have been employed for
their seamless integration with TMDs for a wide variety of applications. This project
investigates the potential of van der Waals (vdW) heterostructures composed of
oxidized TMDs and their corresponding TMOs to enhance photoluminescence (PL)
unleashing new opportunities to integrate with optoelectronic devices. The study
focuses on fabricating 2D heterostructures using WSe₂ and MoS₂ which exhibit
promising luminescence characteristics at monolayer thicknesses. Mechanical
exfoliation and dry transfer techniques were employed to isolate and stack bilayer
flakes. A UV-free ozone oxidation process was optimized to selectively convert bilayer
regions into monolayers while avoiding photoluminescence quenching.
Photoluminescence and Raman spectroscopy confirmed successful monolayer
formation and enhanced excitonic emissions after oxidation. MoS₂ showed increased
PL intensity due to p-doping, while WSe₂ exhibited reduced PL under similar
conditions. Despite these doping effects, both heterostructures demonstrated strong PL
enhancement when stacked, affirming that structural configuration dominates optical
behavior. A device fabricated from the oxidized WSe₂/MoS₂ heterostructure exhibited
ohmic behavior with strong photocurrent response, validating its suitability for
optoelectronic applications.Bachelor's degre
Android apps development for image and video processing
No full textThis project presents a real-time emotion detection system for dogs and cats using computer
vision and deep learning techniques, implemented on an Android platform. Leveraging
MobileNetV2 for lightweight image classification, the system first detects the pet in the camera
frame, then classifies its emotional state into categories such as happy, calm, or angry. The model
was trained on a custom-labeled dataset and optimized for mobile performance using
TensorFlow Lite. Additionally, corresponding audio feedback is provided to enhance user
interaction. The application demonstrates an innovative approach to understanding pet emotions
in real time, enabling more intuitive human-animal communication.Bachelor's degre
Enhancing energy efficiency through predicitive maintenance modelling in HVAC system
No full textHeating, Ventilation, and Air Conditioning, or better known as HVAC, systems have grown to become a leading focus in the field of energy efficiency and sustainable development due to their high energy consumption in a building. The forefront strategy for improving HVAC system energy efficiency is predictive maintenance in Air Handling Units (AHUs), which targets the enhancement of energy performance in one of the highest energy consumption equipment by identifying potential faults before they occur. This study explores the application of Long Short-Term Memory (LSTM) networks on AHU operation data for fault predictions. Additionally, the study extends its investigation to the autoencoder network, utilising anomaly detection to eliminate misclassified predictions to enhance the model’s overall accuracy. This study also challenges the model’s transfer learning capabilities by training and validating on a fault-labelled time series and testing on a separate unlabelled time series dataset to generate fault predictions. Hyperparameter tuning is employed to optimise the models’ accuracy and robustness in predictions. Finally, the predictions are assessed through various analytical metrics and compared with real-world system behaviours to evaluate the model’s feasibility.Bachelor's degre
Development of new diagnostic assay for infectious diseases
No full textRapid molecular diagnostics are crucial for the management and timely intervention of infectious diseases, avoiding ineffective medication regimes and preventing disease progression. Mycoplasma genitalium, with its escalating resistance profile resulting in treatment failures and the Cytochrome P450 family subfamily C member 19 mutations,
impacting drug efficacy, represents two distinct diagnostic challenge in modern healthcare. However, their current diagnostic methods are time-consuming and complex. With this, RT-LAMP would be explored as an alternative diagnostic technique due to its simplicity. The RT-LAMP was adopted and optimised to focused on
enhancing the sensitivity, specificity and efficiency.Bachelor's degre