Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Dimension-reduced Chapman-Kolmogorov equation for high-dimensional stochastic dynamical systems
No full textRandom vibration analysis of high-dimensional dynamical systems is a fundamental problem in science and engineering, yet it remains challenging due to the curse of dimensionality. While dimension-reduced formulations have been developed for differential-type equations governing time-variant probability density, such as the Fokker-Planck equation, no equivalent formulation has been established for the integral-type Chapman-Kolmogorov (CK) equation, despite its theoretical importance and computational advantages. In this paper, a novel dimension-reduced Chapman-Kolmogorov (DRCK) equation is established governing the transient probability density function (PDF) of any quantity of interest in high-dimensional Markov systems. The derivation is conducted based on the projection of the full Chapman-Kolmogorov equation onto the dimension-reduced space. It is established that the intrinsic transition probability density (TPD) of the DRCK equation is the conditional expectation of the original TPD. Further, the short-time approximate intrinsic TPDs under both Gaussian and Poisson white noise excitations are derived analytically, enabling practical numerical implementation. The proposed DRCK equation provides a mathematically rigorous and computationally efficient framework for high-dimensional stochastic systems. Numerical examples are developed to demonstrate its accuracy and effectiveness. The DRCK equation thus provides a new tool for reliability assessment and uncertainty quantification in complex engineering applications.</p
3D SA-LoIFM: A three-dimensional framework for segmentation-assisted learning of informative feature matching in liver CT-3D US rigid registration
No full textThe registration of tomography (CT) and ultrasound (US)images can improve lesion detection and intervention success. However, it faces the challenge of global registration due to the random scanning pose and limited field of view of US. Most existing liver CT-US registration methods rely on manual initialization or tracking systems, which are time-consuming and increase operation complexity. The most commonly used automatic vessel geometric information-based methods are only applicable to images under specific conditions. In this study, we propose a 3D segmentation-assisted learning of informative local feature mapping framework for global rigid registration of CT-3D US, namely 3D SA-LoIFM. Multi-label segmentation masks are integrated with the intensity image as the input for deep feature extraction to provide corresponding semantic information. Informative points with large intensity variances are sampled to save memory consumption of the 3D framework and avoid imbalanced learning of large, less informative or noisy regions. During testing, an outlier rejection procedure based on geometric consistency and smoothness criteria is introduced to refine the matched points. Various experiments have been carried out to assess the importance of the proposed improvements and the performance of the whole framework. A five-fold cross-testing registration error of 6.16 ± 1.88 mm is achieved over 42 paired liver CT & 3D US training datasets. The testing registration error is 6.26 ± 2.06 mm, with a rotation error of around five degrees on 74 testing datasets, which cannot be successfully registered by the vessel-based registration methods. Our code is available at https://github.com/hbcfx/3D-SA-LoIFM</p
Zonal flows driven by libration in rotating spherical shells: the case of periodic characteristic paths
No full textThis work investigates the weakly nonlinear dynamics of internal shear layers and the mean zonal flow induced by the longitudinal libration of an inner core within a spherical shell. Building on the work of He et al. (2022 J. Fluid Mech., vol. 939, p. A3), which focused on linear dynamics, we adopt a similar set-up to explore the nonlinear regime using both asymptotic theory and numerical computations, with Ekman numbers as low as E = 10−10. A specific forcing frequency of ω = √2Ω, where Ω denotes the rotation rate, is introduced to generate a closed rectangular path of characteristics for the inertial wave beam generated at the critical latitude. Our approach extends previous results by Le Dizès (2020 J. Fluid Mech., vol. 899, p. A21) and reveals that nonlinear interactions are predominantly localised around regions where the wave beam reflects on the boundary. We derive specific scaling laws governing the nonlinear interactions: the width of the interaction region scales as E1/3 and the amplitude of the resulting mean zonal flow scales as E1/6 in general. However, near the rotation axis, where the singularity of the selfsimilar solution becomes more pronounced, the amplitude exhibits a scaling of E−1/2. In addition, our study also examines the nonlinear interactions of beams that are governed by different scaling laws. Through comparison with numerical results, we validate the theoretical predictions of the asymptotic framework, observing good agreement as the Ekman number decreases.</p
Occlusion-Aware Contingency Safety-Critical Planning for Autonomous Driving
No full textEnsuring safe driving while maintaining travel efficiency for autonomous vehicles (AVs) in dynamic and occluded environments is a critical challenge. This article proposes an occlusion-aware contingency safety-critical planning approach for real-time autonomous driving. Leveraging reachability analysis for risk assessment, forward reachable sets (FRSs) of phantom vehicles (PVs) are used to derive risk-aware dynamic velocity boundaries. These velocity boundaries are incorporated into a biconvex nonlinear programming (NLP) formulation that formally enforces safety using spatiotemporal barrier constraints, while simultaneously optimizing exploration and fallback trajectories within a receding horizon planning framework. To enable real-time computation and coordination between trajectories, we employ the consensus alternating direction method of multipliers (ADMMs) to decompose the biconvex NLP problem into low-dimensional convex subproblems. The effectiveness of the proposed approach is validated through simulations and real-world experiments in occluded intersections. Experimental results demonstrate enhanced safety and improved travel efficiency, enabling real-time safe trajectory generation in dynamic occluded intersections under varying obstacle conditions. The project page is available at: https://zack4417.github.io/oacp-website/.</p
3-Dimensional Pixel-based Multi-port Antennas
No full textWe extend the conventional 2-dimensional (2D) versatile pixel antenna concept to 3-dimensional (3D) structures. The idea is to conform a pixel surface onto a 3D object and develop appropriate feeding mechanisms to form a 3D pixel-based multi-port antenna. We focus on a pixel-based antenna cube with four antenna ports for investigation and experimental verification. Different from 2D antennas, the antenna current on the cube surface can flow from one oriented plane (xz-, yz- or xy-plane) to another when the cube is excited. This allows more flexibility for pixel connections to achieve spatial and polarization diversity, as well as further miniaturization. As the cubic geometry has rotational symmetry, the analysis of mutual coupling among multiple ports can also be simplified so it is more convenient to deal with. Furthermore, decoupling structures can be designed and printed on internal faces of the cube to further enhance multi-port antenna performance. To provide experimental verification an example design with 4-ports operating at 1.77 GHz in a pixel cube with a side length of 24.4 mm (equivalent to only 0.14 λ0, where λ0 is the wavelength in free space) is provided.</p
TrajLens: Visual Analysis for Constructing Cell Developmental Trajectories in Cross-Sample Exploration
No full textConstructing cell developmental trajectories is a critical task in single-cell RNA sequencing (scRNA-seq) analysis, enabling the inference of potential cellular progression paths. However, current automated methods are limited to establishing cell developmental trajectories within individual samples, necessitating biologists to manually link cells across samples to construct complete cross-sample evolutionary trajectories that consider cellular spatial dynamics. This process demands substantial human effort due to the complex spatial correspondence between each pair of samples. To address this challenge, we first proposed a GNN-based model to predict cross-sample cell developmental trajectories. We then developed TrajLens, a visual analytics system that supports biologists in exploring and refining the cell developmental trajectories based on predicted links. Specifically, we designed the visualization that integrates features on cell distribution and developmental direction across multiple samples, providing an overview of the spatial evolutionary patterns of cell populations along trajectories. Additionally, we included contour maps superimposed on the original cell distribution data, enabling biologists to explore them intuitively. To demonstrate our system's performance, we conducted quantitative evaluations of our model with two case studies and expert interviews to validate its usefulness and effectiveness.</p
Novel Insights into Solid-State Batteries Through Phase Modulations: Dielectric Phase and Liquid Crystal Phase
No full textAll-solid-state batteries (ASSBs) are widely regarded as one of the most promising candidates for next-generation energy storage technologies. Among the various components of ASSBs, solid polymer electrolytes (SPEs) have attracted significant attention due to their excellent mechanical toughness, low densities, ease of processing, and good interfacial contact with electrodes. In recent years, liquid crystal polymer (LCP) electrolytes have emerged as a research hotspot. Unlike traditional classifications of dielectric and non-dielectric phases, the unique ordered self-assembled structures of LCP electrolytes can provide highly efficient ion transport pathways. This perspective presents a systematic perspective on regulating the performance of lithium-ion batteries (LIBs) (especially ASSBs) through the synergistic combination of dielectric and liquid crystal (LC) phases. The aim of this work is to offer detailed and timely insight into the advantages and disadvantages of polymers and their composite electrolytes from the perspectives of dielectric and ferroelectric phases, while also evaluating the potential of LCPs from the viewpoints of LC and non-LC phases. By combining the advantages of dielectric and LC phases, this work envisions a future for SPEs where ferroelectric LCPs and their composites emerge as a new class of SPEs.</p
Global increase in rain rate of tropical cyclones prior to landfall
No full textMost studies on tropical cyclone (TC) rain rate focus on long-term variability, yet the short-term (days or shorter) variations across the TC lifecycle, with a particular focus on the period before landfall, are most critical because they strongly influence flood risk. Using satellite data, we show that, globally, the mean rain rate of TCs increases by over 20% from 60 hours before landfall to the time of landfall. This increase occurs across hemispheres, ocean basins, intensity categories, and latitudes, although the magnitude varies. As a TC approaches the coast, land-sea thermal contrasts raise low-level humidity over land, while frictional differences enhance convergence, upward motion, and instability on the offshore side of the circulation. These conditions collectively promote increased convection and precipitation of TCs as they near landfall. Our findings critically strengthen the current understanding of TC precipitation dynamics and support more effective flood management.</p
Spatiotemporal graph structure similarity evaluation for pure zero-shot anomaly detection tasks
No full textIn the field of equipment health management, the scarcity of fault samples poses a notable challenge for anomaly detection. Although current few-shot learning approaches have shown potential, they remain inadequate for addressing pure zero-shot scenarios where neither training nor validation sets contain fault samples, and detection must proceed without reliance on any labeled sample. This challenge is further aggravated by the concealment of the distributional divergence between normal and abnormal samples due to environmental noise interference, thereby exacerbating detection complexity. To overcome these limitations, a pure zero-shot anomaly detection framework driven by spatiotemporal graph structural similarity is developed. Specifically, a local structure-embedded graph (LSEG) is constructed to describe the feature distribution in a non-Euclidean space, where the subgraph isomorphism technique is employed to learn and differentiate the structural characteristics between normal and abnormal samples. Furthermore, to enhance the graph feature extraction ability, an inductive variational graph autoencoder (IVGAE) is designed, where the final output is derived by calculating reconstruction errors and using an inner product decoder. Finally, the model output is compared with a detection threshold determined using the dynamic interquartile range (IQR) method to identify anomalies in the test set. Experimental validation on the self-owned axial flow pump and two public datasets resulted in F1-Scores of 0.9613, 0.8928 and 0.7317, and Recall rates of 0.9603, 0.8144 and 0.6297, respectively, thereby demonstrating the model's generalizability and effectiveness of the model.</p
Metal-based biomaterials for treating bone diseases
No full textBone-related diseases resulting from accidents, illnesses, and injuries have become increasingly common in recent years. Treating these conditions poses significant challenges, including prolonged recovery times, high costs, and unpredictable outcomes, which can lead to complications such as infections and reduced muscle strength. Although autologous bone transplantation is regarded as the "gold standard" for addressing bone diseases, its application is often limited by complications at the donor site and the risk of infection. This underscores the urgent need to explore alternatives to autogenous bone transplantation. In response, a range of biomaterials for bone repair have been developed, with metal-based biomaterials emerging as effective adjuncts that enhance and optimize the repair and regeneration of bone tissue. These materials can actively influence the bone repair process through mechanisms such as inductive osteogenesis, immunomodulation, and pro-angiogenesis. This review begins by highlighting the biological effects of metal-based biomaterials, followed by a comprehensive overview of their macro- and micro-scale classifications and applications for treating various bone diseases. Finally, the review addresses future directions and challenges associated with the use of metal-based biomaterials in bone repair, aiming to propose promising strategies for the treatment of bone-related diseases.</p