DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Rethinking rehabilitation: new imaginations and narratives
No full textThe release of former Jemaah Islamiyah leader Para Wijayanto in May 2025 has reignited debate over radicalisation, reintegration, and the shifting nature of violent extremism in Indonesia. His transformation underscores not only a generational rift between old and new jihadists but also the underexplored role of masculinity in the radicalisation process. A gendered approach offers vital insights into both the pull of extremist ideologies and the possibilities for sustainable disengagement.Published versio
Cushion energy absorption analysis for a 3D printed material with lattice structure
No full textAdditive manufacturing, especially 3D printing, offers significant potential for
fabricating complex geometries, particularly lightweight lattice structures. This
dissertation focuses on polymer lattice structures. Based on the LS-DYNA simulation
platform, it systematically simulates the classic Split Hopkinson Pressure Bar (SHPB)
experimental method to investigate their dynamic response under high-speed impact
loading.
This dissertation first constructs an SHPB simulation model and analyzes the specimen
stress distribution, contact surface pressure distribution, and temporal strain evolution
patterns at measuring points on the incident and transmission bars. Based on
preliminary simulation results, the model was improved across three key parameters:
boundary conditions, bar material, and bar length. Second, simulations were conducted
on nine lattice structures with distinct geometries using the refined model, extracting
the following time-history curves through post-processing: incident waves, transmitted
waves, reflected waves, internal energy, and kinetic-to-internal energy ratios. Finally,
simulation experiments with varying striker bar velocities were conducted.
Simulations extracted key physical quantities including specimen von Mises stress and
contact surface pressure, and analyzed the temporal characteristics of strain rates at
measuring points on the incident and transmission bars. The results indicate notable
stress concentrations at structural unit of the lattice junctions and pore edge regions.
Furthermore, contact pressure undergoes significant spatial redistribution during
loading, which suggests high structural sensitivity to transient load paths. Simulations
with varying striker bar velocities further demonstrate that increasing the striker bar
velocity sharpens the stress wave steepness and raises the peak stress within the
specimens.
The simulation results provide a theoretical basis for optimizing polymer lattice
structures in impact protection and energy dissipation applications, and validate the
applicability of the adopted SHPB numerical simulation method for dynamic response
analysis of lightweight structures.Master's degre
Development of a step-counting system using pose estimation and ankle zero-crossing with end-to-end integration
No full textAccurate step tracking aids monitoring soldier movement during peacetime military route
marches. However, current step-counting methods rely on commercial devices, which may not
be optimal for military applications. This project aims to develop an open-source step-counting
framework that leverages pose-estimation and ankle zero-crossing to track steps.
The proposed system uses pose estimation to count steps when the ankle nodes cross each
other. A prototype was developed and tested on a Gait Database consisting of videos under
controlled conditions.
This project offers a low-cost, customizable, and scalable solution tailored for military use.
The system achieved a step-counting accuracy of over 99% and an Intraclass Correlation
Coefficient (ICC) of 0.93, demonstrating strong agreement with manual counting.
Future work includes improving step-counting performance and integration of Inertial
Measurement Unit (IMU) sensors, a virtual humanoid, and pose estimation within the step counting framework.Bachelor's degre
Strong mid-infrared optical harmonic processes mediated by excitons in layered indium selenide
No full textNonlinear optics in 2D materials has attracted significant attention due to their potential in various optoelectronic applications such as nonlinear optical detection, bioimaging, and nonlinear optical modulation. Optimising the nonlinear optical response in the infrared, especially the mid-infrared (MIR) region, is particularly desirable for various applications including tunable coherent sources, high-order harmonic generation, supercontinuum light generation, and frequency-comb-based spectroscopy. However, reports on MIR nonlinear optical responses in 2D materials are limited due to a lack of materials with suitable nonlinear optical properties. Van der Waals layered γ-phase indium selenide (γ-InSe) featuring broken inversion symmetry can serve as a promising candidate for MIR nonlinear devices. Beyond the noncentrosymmetric crystal structure, γ-InSe possesses unique out-of-plane dipoles, distinguishing it from other 2D materials such as black phosphorus (BP) and transition metal dichalcogenides (TMDs). Herein, multiphoton harmonic generation processes including second (SHG), third (THG), fourth (FHG) and fifth (FIHG) harmonic generation are investigated in γ-InSe in the MIR region. Notably, γ- InSe shows strong second and higher order nonlinearities. A metal-insulator-metal (MIM) plasmonic metasurface is further designed with large SHG and THG enhancement factors reaching 8400 and 32000 times, respectively. These results demonstrate the great potential of γ-InSe for next-generation MIR nonlinear photonic devices
Optimised processing of ridge waveguide semiconductor lasers
No full textQuantum Cascade Lasers have emerged as a powerful tool for spectroscopy
application, especially in the mid-infrared range, due to the high absorption lines of
the molecules. This allows for precise detection of gases, enabling applications like
environmental monitoring and industrial precision control to function. However, the
performance of the quantum cascade laser is extremely dependent on the fabricated
components, in particular ridge waveguides. Ridge waveguides play a critical role in
the performance of the quantum cascade laser as it is responsible for the light
propagation and minimising optical losses.
This project focuses on the optimised fabrication process of ridge waveguide quantum
cascade laser with the goal of improving their reliability and replicability.
Additionally, a comprehensive overview of quantum cascade lasers fundamentals as
well as the fabrication techniques like photolithography, etching, and deposition will
be covered. The optimised fabrication steps are then documented in detail.
By addressing these challenges, this project aims to contribute to the ongoing
advancement of quantum cascade lasers, allowing for the development of higher performance spectroscopic applicationsBachelor's degre
2D Ge-based crystals for enhanced thermoelectric performance
No full textThe global energy crisis and the need for waste heat recovery have driven research
into high-performance thermoelectric materials. Two-dimensional germanium-based
crystals (GeTe and GeSe) exhibit significant potential for thermoelectric property
optimization due to their unique electronic structure and low lattice thermal
conductivity. In this study, high-quality 2D GeTe and GeSe crystals were
successfully synthesized by flux-assisted chemical vapor deposition to systematically
investigate their microstructures, electrothermal transport properties and device
applications. The study further reveals the key role of flux-assisted method in
regulating the growth kinetics of 2D materials. Future work will focus on
high-pressure induced cubic phase GeSe stabilization to advance the application of
2D germanium-based materials in industrial-scale waste heat recovery and flexible
wearable devices. This research provides an experimental basis for developing how
to grow efficient thermoelectric materials.Master's degre
Hardware fault injection analysis for security evaluation
No full textAs embedded systems grow more essential to critical infrastructure and consumer gadgets, the significance of hardware security is escalating. This study investigates the efficacy of Clock Glitch Fault Injection (CGFI) in circumventing authentication in a secure AES-128 bootloader using the ChipWhisperer platform. More than 1,000 bug attempts were executed, resulting in around 100 successful bypasses (~10% success rate), enabling unauthorised AES decryption even if MACs were faulty. The resultant power traces, synchronised with known plaintexts, were examined by Correlation Power Analysis (CPA), effectively retrieving the whole 128-bit AES key from just 40–50 traces. The results show that CGFI-enabled side-channel attacks can work well and highlight the need for better protection against faults in embedded systems.Bachelor's degre
Meta-learning: enhancing model performance and adaptability across domains
No full textMeta-learning, also known as the process of acquiring the ability to learn new tasks efficiently, marks a key advancement in machine learning. It equips models with the capability to adjust to novel problems using very limited data. This stands in contrast to traditional deep learning methods, which often rely on vast quantities of labeled data, immense computing power, and still encounter challenges when applied to unfamiliar situations.
This research analyses three primary approaches to meta-learning: model-based, optimization-based, and metric-based approaches. Prototypical Networks, a key metric-based technique, were utilized due to their effectiveness in classifying data based on learned prototypes, making them ideal for few-shot learning scenarios.
Experiments were conducted on benchmark datasets (Omniglot and miniImageNet) using various optimizers (Adam, Adagrad, RMSProp, SGD with Nesterov), distance metrics (Euclidean, Mahalanobis, KL, and IS divergences), and image augmentation techniques (affine transformations, rotations, CutMix). The KL divergence combined with the Adagrad optimiser and affine augmentations emerged as the best-performing model, delivering superior accuracy and lower loss.
Learning rate schedulers such as StepLR, ReduceLROnPlateau, Cosine Annealing (Warm Restarts), and LinearLR were also incorporated in our experiments. Among these, Cosine Annealing (Warm Restarts) on KL/IS Divergences on the Adagrad optimiser provided notable improvements in training stability and accuracy. Ensemble (distance) learning techniques were explored as well but proved less effective, primarily due to the similarity of the probability-based distance metrics (KL and IS divergences), which potentially limited model diversity. Towards the end, ablation studies were conducted to systematically assess the impact of individual experimental components on overall model performance enhancement.
Our findings demonstrate that meta-learning significantly improves generalization and rapid adaptation, highlighting its potential for real-world applications. Future work includes exploring hybrid methods, reinforcement learning, and optimizing computational efficiency.Bachelor's degre
Hollywood may not be spared China's cultural ascent
No full textAs US-China trade tensions escalate, a potential Chinese ban on Hollywood films threatens to accelerate cultural and economic decoupling. With China now the second-largest film market, the move would deliver a major blow to American studios while reinforcing China’s push for cultural self-reliance. The potential fallout for Hollywood, including shifting market priorities, content censorship, and investor uncertainty may be devastating. China’s state-backed cinema seems set to build soft power and shape its own narrative. Can Hollywood adapt as the cultural influence it is familiar with is increasingly contested?Published versio