DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Bayesian dynamic programming approach for tracking time-varying properties in structural health monitoring
No full textStructural health monitoring (SHM) aims at monitoring the condition of structures such as buildings, bridges, and dams, using in-situ information extracted from measured data. Structural properties may change due to factors such as material aging and damage after extreme events (e.g., earthquakes and storms). Tracking changes under ambient excitations (e.g., wind, traffic, and tremor) is typical in SHM as it does not disturb normal operations. In principle, tracking results can aid in anomaly detection, supporting decisions such as early warning and maintenance strategies. How to do it properly and efficiently, however, is still challenging. The timing, frequency, and magnitude of changes are unknown and can occur in an apparently random manner. The properties extracted from SHM data under ambient excitations can have large estimation errors due to the lack of information arising from data (limited and noisy) and modeling (unknown excitation but modeled as a stochastic process). These can lead to spurious changes or even mask the actual changes in structural properties.
Motivated by the aforementioned issues, this thesis develops a Bayesian probabilistic theory with an efficient algorithm for tracking model properties from SHM data and applies the methodology to SHM based on operational modal analysis. In the proposed context, SHM data in the time domain is divided into non-overlapping segments of potentially varying sizes, over which model properties are assumed to be piecewise constant. Using a Bayesian model selection approach, statistically significant changes are then detected naturally from the change points. Exploiting the causality nature of the tracking problem, a dynamic programming approach is proposed for determining the best partitioning of data, which is otherwise computationally prohibitive by exhaustive search. A ‘pruning’ strategy is developed to further suppress the growth of computational effort with monitoring data size that keeps growing. Beyond conventional models where all parameters are either constant or varying in a particular time window for analysis, a ‘quasi time-invariant’ model is proposed that allows a mix of such variables, providing flexibility to achieve target monitoring objectives balancing statistical Type I (false alarm) and Type II (missed alarm) errors. The proposed methodology is investigated using synthetic and laboratory data, and applied to field data during a typhoon event and a full-scale ambient vibration test where practical challenges in modal property tracking are discussed.Doctor of Philosoph
Rethinking the message passing for graph-level classification tasks in a category-based view
No full textMessage-Passing Neural Networks (MPNNs) have emerged as a popular framework for graph representation in recent years. However, the graph readout function in MPNNs often leads to significant information loss, resulting in performance degradation and computational waste for graph-level classification tasks. Despite the common explanation of “local information loss,” the underlying essence of this phenomenon and the information that the MPNN framework can capture in graph-level tasks have not been thoroughly analyzed. In this paper, we present a novel analysis of the MPNN framework in graph-level classification tasks from a node category-based perspective. Our analysis reveals that the graph-level embeddings learned by MPNNs essentially correspond to category-based contribution measurements. Building upon this insight, we propose a groundbreaking Category-Based Non-Message-Passing (CANON) paradigm for graph-level representation learning. By leveraging a novel numerical encoding mechanism, CANON achieves superior performance even without incorporating structural information, surpassing state-of-the-art MPNN methods. CANON also offers substantial computational advantages, including a model size that is hundreds of times smaller and reduced time complexity, enabling faster inference and reduced costs for real-world applications, particularly in domains such as chemistry, biology, and computer vision. To further enhance the method's effectiveness, we introduce domain-specific structural incorporation. Our extensive experiments across multiple datasets demonstrate CANON's efficacy and its potential to serve as a highly efficient alternative to MPNNs, opening up new possibilities for graph representation learning and its downstream applications.Ministry of Education (MOE)National Research Foundation (NRF)This research/project is supported by the Ministry of Education, Singapore under its MOE Academic Research Fund Tier 2 (STEM RIE2025 Award MOE-T2EP20220-0006) and Tier 1 (RG16/24), and by the National Research Foundation, Singapore under its Industry Alignment Fund – Pre-positioning (IAF-PP) Funding Initiative
Highly rechargeable aqueous Sn-metal-based hybrid-ion batteries
No full textTin (Sn) metal, with its intrinsic resistance to the hydrogen evolution reaction (HER), holds great promise as an anode for safe and rechargeable aqueous Sn-metal batteries (ASBs). However, the major challenges for their practical deployment include uneven Sn deposition and low Sn2+/Sn4+ reaction reversibility. To mitigate these challenges, we design ASBs from both anode and electrolyte. First, a stannophilic silver-coated vertical graphene (Ag-VG) host improves the nucleation kinetics and uniform Sn deposition. Second, a biphasic H2O/ionic liquid (IL) electrolyte confines Sn2+ within the aqueous phase, suppressing the formation of Sn4+ at the cathode side and eliminating the usage of an ion exchange membrane. The biphasic electrolyte and Ag-VG host are coupled with various types of cathodes (herein, halogens, LiCoO2, and Li2MnO4) to fabricate full ASBs. Improved cycling stability and Coulombic efficiency are clearly observed. This work highlights a facile strategy for advancing ASBs.Ministry of Education (MOE)National Research Foundation (NRF)Submitted/Accepted versionThis research is supported by the National Research Foundation, Singapore, under its Singapore-China Joint Flagship Project (Clean Energy), and Ministry of Education, Singapore, by AcRF Tier 2 (MOE-T2EP50121-0006)
Image restoration and representation with deep generative models
No full textA fundamental challenge in the realm of computer vision lies in accurately modeling/characterizing image distributions. For example, in high-level vision tasks, better representations of images in the latent space can significantly enhance downstream processes such as image classification and segmentation. Similarly, in image restoration, a more accurate model for the distributions of clean images conditioned on degraded images can yield results with better perceptual quality. However, modeling these distributions is highly challenging due to the high dimensionality of both images and their latent codes.
In a relatively parallel research avenue, deep generative models have made remarkable strides, aiming to implicitly or explicitly model image distributions. These models offer potent tools for handling intricate distributions. However, directly applying these models is unfeasible and leads to suboptimal performances due to different domain priors and task formulations. The challenge of seamlessly and effectively integrating generative models into image restoration and representation tasks, along with incorporating task-specific priors, remains an open area for further exploration.
This thesis focuses on exploring the potential of deep generative models in image restoration and representation tasks, spanning from high-level vision tasks, i.e., image classification, to low-level vision tasks, i.e., image compression and restoration. Firstly, for the image classification task, we introduce a novel variational inference framework to obtain a latent representation with enhanced generalization ability. We implicitly model the posterior distributions of images given their latent codes using a generative adversarial network to disentangle domain-invariant features from the provided training data.
Subsequently, instead of implicitly modeling the posterior distributions, we explore the possibility of explicitly modeling them. Utilizing the power of Normalizing Flow, a generative model capable of obtaining the exact likelihood of a given sample, we apply it to the task of low-light image enhancement (LLIE), considering its inductive bias. While the flow-based model achieves promising results, the performance is constrained by the intrinsic limitations in the design of normalizing flow. Therefore, we delve into seamlessly integrating the prior knowledge of the LLIE task in raw image space into a diffusion framework to overcome the limitations in model designs.
Although our diffusion-based solution has promising results, it still has two main limitations: first, even with the unique advantages of enhancement in raw space, the significant storage overhead of raw images hampers its application. Consequently, in the subsequent two works, we explore joint compression by accurately modeling the latent distributions of images and leveraging the capabilities of autoregressive models to further enhance coding efficiency. Second, while the proposed work shortens the inference path from tens or hundreds to only three steps compared with commonly used diffusion models, it still requires iterative evaluations and causes inference overhead.
In the last work, we propose an acceleration strategy incorporating a distillation strategy with a novel self-consistency loss. We achieve SOTA performance in only one inference step in the super-resolution task, where the conditional distribution is relatively more complex.
In summary, the thesis makes three primary contributions. First, we showcase the effectiveness and unique advantages of generative-based image restoration and representation methods. Second, we propose diverse approaches to seamlessly integrate the capabilities of deep generative models with the domain knowledge specific to image restoration and representation tasks. To validate the effectiveness of these proposed methods, extensive experiments are conducted across multiple datasets. The experimental results unequivocally demonstrate that our methods outperform prior state-of-the-art models. The efforts and achievements presented in this thesis not only underscore the practical capabilities of image restoration and representation techniques but also provide fundamental support for future research and applications within the industry.Doctor of Philosoph
Identifying the knowledge and capacity gaps in Southeast Asian insect conservation
No full textInsects represent most of terrestrial animal biodiversity, and multiple reports suggest that their populations are declining globally due to anthropogenic impacts. Yet, a high proportion of insect species remain undescribed and limited data on their population dynamics hamper insect conservation efforts. This is particularly critical in tropical biodiversity hotspots such as Southeast Asia. To identify knowledge and capacity gaps in Southeast Asian insect conservation, we performed a quantitative review of insect occurrence records, studies for the region and global '#conservation' posts from Twitter. We found that occurrence records increased over time, and were dominated by butterflies. Overall, studies were largely focused on pest and vector groups, and insect conservation and ecology studies were lacking in many countries. Despite an increase in local authorships and funding sources over time, the majority of these were still located outside of Southeast Asia. In '#conservation' posts, insects were highly under-represented and insect-related content was biased towards popular groups such as bees and butterflies. We suggest potential solutions to address these gaps, such as integrative taxonomic approaches, and increasing regional collaborations and public engagements. Crucially, we stress the need for political will and funding to overcome the impediments towards insect conservation efforts in Southeast Asia.Ministry of Education (MOE)Nanyang Technological UniversitySubmitted/Accepted versionXRO was supported by the Nanyang President’s Graduate Scholarship from Nanyang Technological University. BT was supported by the Nanyang Scholarship from Nanyang Technological University. CHC acknowledges support from Pomona College and the Bezos Earth Fund. EMS and NP were supported by the Singapore Ministry of Education (MOE) Academic Research Fund (AcRF) Tier 2 grant (Grant number: MOE- T2EP30221–0020)
Development of keratin-based fibers fabricated by interfacial polyelectrolyte complexation for suture applications
No full textInterfacial Polyelectrolyte Complexation (IPC) is a convenient way to produce composite, micro-scale fibers. In this paper, we report the successful development of novel keratin-based IPC fibers and also demonstrate the feasibility of using these fibers as sutures through a proof-of-concept in vivo study. Two composite fibers were produced: chitosan-keratin (CK) and keratin-keratin (KK). These fibers were evaluated for their physico-chemical, mechanical and biochemical properties. In the dry state, the CK fiber had a greater Young’s modulus of about 2 GPa while the KK fiber registered a longer strain-at-break of about 100% due to the strain-stiffening effect. Notably, the keratins were found to assemble into amyloids within the composite fibers based on Congo red staining and Wide-Angle X-Ray Scattering. Functionally, both fibers were malleable could be weaved, braided and knotted. When used as sutures to close incisional wounds in mice over 21 days, these fibers were found to elicit minimal host tissue response and were partially degraded over the duration. Interestingly, the KK fiber evoked a lower extent of immune cell response and fibrous capsule encapsulation that was comparable to commercial, non-absorbable Dafilon® sutures. This work demonstrated the possibility of producing keratin-based IPC fibers which may find practicality as medical sutures.Ministry of Education (MOE)Nanyang Technological UniversitySubmitted/Accepted versionThis work was supported by the Ministry of Education, Singapore (AcRF Tier 1: RG7/22) and by Nanyang Technological University, Singapore Start-Up Grant through the School of Materials Science and Engineering
Coral skeletal isotopes (δ13C and δ11B) as indicators of seawater light attenuation and pH chemistry in the Singapore Strait
No full textThis study investigates the interaction between δ¹³C and δ¹¹B with terrigenous carbon dynamics in the Singapore Strait, a region characterized by distinct monsoon patterns and significant terrigenous input from surrounding peatlands.
We hypothesized that elevated levels of colored dissolved organic matter (CDOM) during the Southwest Monsoon would decrease light penetration, leading to more negative δ¹³C values in coral skeletons. Additionally, we expected that remineralization of terrigenous dissolved organic matter (tDOM) would acidify seawater, resulting in more negative δ¹¹B values in corals. Analysis
of Porites spp. corals from two plug cores (KUK and KUL) and seawater data from Kusu Island (2017-2020) revealed no significant correlation between CDOM and coral δ¹³C anomalies— deviations between coral skeletal δ¹³C values
and the δ¹³C values of dissolved inorganic carbon (DIC) in seawater— contradicting our hypothesis. Instead, variations in coral δ¹³C appear to be related to a reservoir effect associated with negative δ¹³C in seawater DIC, influenced by
tDOC remineralization. Although not statistically significant, the positive correlation pattern observed between δ¹¹B and seawater pH in the KUL core suggests that δ¹¹B might serve as a useful proxy for historical seawater pH and
acidification. This finding also supports the idea that Porites corals may regulate their internal pH in response to changes in seawater acidity, potentially influenced by tDOC remineralization. Inconsistencies in the KUK core could be attributed to data offsets from our age-depth model. Further research with extended sampling is needed to confirm δ¹¹B’s sensitivity to pH changes and understand its impact on coral physiology. This study highlights the complex
interplay between seasonal changes, carbon dynamics, and coral isotopic records.Master's degre
Microstructural characteristics, electrical conductivity and mechanical properties of Cu matrix composites reinforced with dual-phase borides
No full textThe effect of boride content on the microstructure, electrical conductivity, and mechanical properties of TiB2-MB2/Cu composites (M = Cr, Mo, and Zr) was systematically investigated. The composites were prepared via spark plasma sintering. The results indicated that the double borides were uniformly distributed in the Cu matrix without visible agglomeration, and excellent interfacial bonding was achieved between the borides and the Cu matrix. The electrical conductivity of the Cu composite with 5 wt% TiB2 and 15 wt% ZrB2 increased by 503 % compared to the Cu-20 wt% TiB2 composite. However, the hardness of the Cu composite with 5 wt% TiB2 and 15 wt% ZrB2 decreased by only 22 %. The wear resistance of the composites was significantly enhanced due to the synergistic effect of the dual-phase boride particles. The wear resistance of the Cu composite with 5 wt% TiB2 and 15 wt% ZrB2 was 4.12 times higher than that of the Cu-20 wt% TiB2 composite standard sample when tested under an applied load of 10 N. The worn surface of the composites exhibited minimal roughness. Therefore, the Cu matrix composites with dual-phase borides achieved desirable comprehensive performance in terms of electrical, mechanical, and tribological properties.Published versionThis work was funded by the National Natural Science Foundation of China (No. 52175184, 52305235, 52405213), the China Postdoctoral Science Foundation (No. 2023M742755), the Shaanxi Postdoctoral Science Foundation (No. 2023BSHEDZZ85), the Natural Science Basic Research Program of Shaanxi (No. 2024JC-YBQN-0502), the Zhejiang Provincial Natural Science Foundation of China (LQ24E020005)
Skin-inspired flexible and printed iontronic sensor enables bimodal sensing of robot skin for machine-learning-assisted object recognition
No full textIontronic devices manifest significant promise for versatile pressure-temperature sensing due to their substantial capacitance resulting from the electric double layer effect and the temperature-dependent ion mobility. Here, a skin-inspired printed iontronic bimodal sensor, capable of independently capturing pressure and temperature for robotic electronic skin is proposed. Through the elaborate design of the printable sensor electrodes, the decoupled resistance-capacitance dual measurement mode of the iontronic device is realized, which is used to measure temperature and pressure changes respectively. By leveraging the pressure-independent and temperature-dependent resistance changes of the iontronic film, the pressure-related capacitance value of the iontronic device is temperature compensated, with maximum pressure error of 5.1 % in 0–300 kPa pressure range. A bimodal pressure-temperature sensor array for robot electronic skin is further demonstrated, achieving high-precision temperature sensing and accurate pressure detection during various object manipulation as well as demonstrating a high object recognition accuracy rate of 98.3 %.Agency for Science, Technology and Research (A*STAR)Submitted/Accepted versionWe would like to acknowledge funding support from the Delta-NTU Corporate Lab through the A*STAR IAF-ICP (No. I2201E0013) and Delta Electronics Inc
A theoretical model for mass transition rate at liquid-vapor sharp interface
No full textPhase change phenomena like evaporation and boiling have traditionally relied on accommodation coefficients, determined experimentally over centuries. This paper presents a new formula for the mass transition rate at a sharp interface, derived from the diffuse interface model in a two-phase system. The mass transition rate is proportional to the difference in chemical potentials between liquid and vapor which is consistent with the classical thermodynamic phase transition criteria. By using the saturation state at the interfacial temperature as a reference, unmeasurable parts of the chemical potentials are eliminated. The proposed model consists of two terms: the first accounts for the relative pressure between the liquid's vapor pressure and the vapor pressure at the interface, while the second reflects the contributions of surface tension and curvature. An alternative formulation expresses relative pressure as the temperature difference between the interfacial temperature and the saturation temperature corresponding to the vapor's interfacial pressure. Unlike traditional models, the proposed formula does not require accommodation coefficients, which typically vary by case. Validation against literature data for both flat and curved surfaces as well as micro-scale test setup shows reasonable agreement, demonstrating the model's effectiveness.Agency for Science, Technology and Research (A*STAR)This work is supported by Agency for Science, Technology and Research (A*STAR), Singapore under the grant #C210415009