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
A cross-climate multi-stage approach to life-cycle carbon sensitivity analysis in BIPV-integrated building design
No full textAchieving carbon–neutral buildings requires design decisions that consistently align performance objectives across climate zones and design stages, yet most existing studies still focus on single climates or isolated parameters. This study develops a multi-stage framework for life-cycle carbon sensitivity analysis that integrates parametric modeling, Latin Hypercube Sampling, high-fidelity coupled BPS-LCA simulations and XGBoost surrogate models, combined with SHAP, Sobol, and ALE to systematically quantify both independent and interactive influences of climatic and design variables. A dataset of 6,000 design cases, generated and simulated for the Jockey Club Global Graduate Tower across six representative climates and calibrated with measured energy data, enables a surrogate model that achieves R2 ≈ 0.98 and NMSE ≈ 0.014 on five-fold cross-validation. The framework is further extended via transfer learning to four Hong Kong residential prototypes to examine its cross-prototype transferability. Results indicate a clear hierarchical sensitivity structure for life-cycle carbon. Climate zone is the primary driving factor, accounting for approximately 26–33 %, while the importance of BIPV configuration and early volume-window decisions is comparable to that of climate zone, and detailed construction and material choices contribute only about 9–10 %. The multi-objective optimization results further determined a series of climate-adaptive design thresholds and effective ranges (e.g., FFH ≈ 3.1–3.4 m, WWR ≈ 0.15–0.30, external wall/roof insulation ≈ 0.07–0.09 m) that can, across diverse climates, jointly ensure acceptable daylight and glare performance while achieving or closely approaching net-zero carbon targets. By emphasizing the coordinated configuration of climate–BIPV–massing–envelope–construction across climates and design stages, this study provides an interpretable quantitative basis for climate-aware, performance-driven carbon–neutral building design.</p
Layered bilateral feature fusion network for end-to-end defect segmentation on aging tiled building Façades
No full textMulti-modal information fusion of RGB and infrared (IR) images collected by unmanned aerial vehicles shows strong potential for automating large-scale building façade inspection. Existing methods predominantly rely on single feature fusion mechanisms, implicitly assuming that various defects exhibit congruent characteristics across the RGB and IR modalities. However, these approaches face challenges in real-world scenarios, where multiple defect types demonstrate distinct characteristics across modalities. To address it, this paper introduces the Layered Bilateral Feature Fusion Network (LBF2-Net), an end-to-end framework with a novel dual-branch architecture. Specifically, LBF2-Net consists of a complementary attention branch and a consistent attention branch. The complementary branch is designed to extract distinctive features unique to each modality, such as subsurface defects, while the consistent branch focuses on features that are manifested in both modalities, such as surface defects. To accurately delineate defect contours, LBF2-Net employs a bifurcated cascaded refining decoder that integrates local information from low-level features with defect context from high-level features. The effectiveness is demonstrated through field applications on tiled façades.</p
Multivalent ligands regulate dimensional engineering for inverted perovskite solar modules
No full textMultivalent, resonance-stabilized amidinium ligands enable stronger chemical coordination and reduced deprotonation compared with conventional monovalent ammonium ligands in low-dimensional perovskites. Here, we introduce a controllable one- to two-dimensional (1D-to-2D) structural transition strategy by systematically tuning ligand conformation, thereby modulating hydrogen bonding, π-π stacking, and basicity to elucidate the relationship between molecular structure, interfacial interactions, and resulting dimensionality. The 1D-amidinium perovskite structure, with its pronounced geometric anisotropy, impedes uniform surface coverage and defect passivation. In contrast, the 2D-amidinium perovskite forms a continuous, homogeneous interfacial layer, enabling more effective defect passivation and favorable energy-level alignment. With dimensionality control, inverted 3D/2D-amidinium perovskite solar cells deliver 25.4% power conversion efficiency (1.1 square centimeters, steady-state certified) and maintain >95% of their initial efficiency after 1100 hours of continuous 1-sun operation at 85°C.</p
Decoding the impact of low-saturation color background on working memory: insights from fNIRS, eye tracking, and subjective assessments
No full textThis study examines the impact of low-saturation background colors on working memory through a detailed multimodal analysis involving functional near-infrared spectroscopy (fNIRS), eye tracking, and subjective assessments. The aim is to identify colors that enhance or impede cognitive performance and uncover the physiological mechanisms involved. Thirty-eight participants completed a 2-back working memory task under nine different background color conditions, during which cognitive performance, brain activation patterns, and eye movement data were systematically recorded. The findings indicate that low-saturation cyan backgrounds significantly enhance cognitive performance by facilitating faster response times and higher accuracy, suggesting these colors help reduce cognitive strain. Conversely, red and purple backgrounds appear to increase cognitive load, slowing response times, and reducing accuracy. Increased activation in the dorsolateral prefrontal cortex was observed under cyan and blue backgrounds, correlating with improved cognitive metrics. While statistical analyses confirmed the impact of low-saturation background color on cognitive and physiological responses, the correlation analyses suggested a more nuanced relationship. Longer response times were associated with higher accuracy, especially in less optimal color conditions, indicating a possible compensatory mechanism to maintain performance. These findings highlight the potential of strategic color use to enhance cognitive function, with important implications for optimizing work and learning environments to more effectively support cognitive task performance. Further investigations should widen the scope of colors studied and consider different cognitive tasks to deepen the understanding of the role of color in cognitive environments, aiming to refine design strategies for optimal cognitive support.</p
UniStain: a unified and organ-aware virtual H&E staining framework for label-free autofluorescence images
No full textWhile hematoxylin and eosin (H&E) staining remains the gold standard for pathological diagnosis, its chemical-dependent workflow presents significant limitations, such as time-consuming protocols, hazardous reagent disposal and batch-to-batch variability in stain quality. We present UniStain, a breakthrough virtual staining framework that leverages label-free autofluorescence (AF) imaging and prompt-based deep learning to overcome these challenges. Unlike existing single-organ approaches that require multiple specialized models, our architecture enables versatile multi-tissue staining through a single model, significantly reducing computational overhead. The proposed crosspatch self-attention guidance (CPSG) mechanism addresses critical whole-slide image challenges by maintaining style consistency across adjacent patches and eliminating stitching artifacts. To support comprehensive evaluation, we curate and release the first multi-organ AF/H&E dataset with human tissue samples. Additionally, we introduce downstream clinical validation tasks including image retrieval and cancer subtyping analysis, thereby establishing a robust evaluation framework for virtual staining models. Quantitative assessments (image quality metrics, visual Turing tests) and downstream analyses demonstrate UniStain’s superior performance compared to existing image translation methods, achieving state-of-the-art results while eliminating chemical staining requirements. The dataset and code of UniStain can be found at https://github.com/TABLAB-HKUST/UniStain .</p
Enhanced dehumidification based on triply periodic minimal surface architectures
No full textTraditional fin-tube condensers (air-side) suffer from restricted flow disturbances and inefficient condensate drainage, constraining their performance in refrigerant dehumidifiers. This study introduces a novel condenser architecture based on triply periodic minimal surface (TPMS) structures to overcome these limitations. We designed and additively manufactured a full-scale condenser based on a typical TPMS Gyroid topology, optimized for uniform water flow distribution and minimal pressure loss. A series of experiments was conducted to illustrate the superiority, enhancement mechanisms, and parametric influences of the novel condenser. The thin-wall Gyroid condenser is 14 % lighter and achieves a 22 % reduction in overall size compared to a commercial wavy fin-tube condenser. Due to enhanced turbulent convection, low-temperature surface, and effective drainage capability, the Gyroid condenser shows improvements of 20 % to 90 % in condensation performance, accompanied by equivalent air pressure drops. Furthermore, in a controlled environment of 2 m3, the Gyroid condenser requires half the time to regulate the temperature and humidity to reach thermal comfort conditions compared to the wavy fin-tube condenser. Sensitivity analysis reveals that the condensation efficiency of the Gyroid condenser is primarily influenced by relative humidity, with negligible effects from air velocity or cooling water temperature. By examining both performance and enhancement mechanisms, this study pioneers the integration of TPMS architecture into dehumidification systems for thermal management and water collection applications.</p
Investigating urban wind vertical profiles during heatwaves with spatial synoptic classification using Doppler LiDAR: a case study in Hong Kong
No full textUrban ventilation plays a critical role in mitigating heat risks, yet its behavior under extreme heat remains poorly understood, especially for high-density cities. This study integrates Doppler wind Light Detection and Ranging (LiDAR) observations with the Spatial Synoptic Classification (SSC) system to examine vertical wind speed profiles during heatwave (HW) periods from 2020 to 2024 across four sites in Hong Kong representing diverse urban and topographic conditions, ranging from open coastal environments to compact high-rise districts, reclaimed port islands, and suburban downwind areas. Results show significant wind speed reductions during HW periods, especially under dry tropical (DT) conditions. Moist tropical (MT) conditions, characterized by high humidity and cloud cover, show stronger wind shear than DT, especially in the lower 500 m. Comparisons with wind profiles from physical and numerical modelling reveal overestimation of wind speeds under extreme heat, resulting from their assumptions of neutral stratification or use of seasonal mean forcings. The findings indicate the limitations of conventional profiles in representing wind conditions under extreme heat. By introducing SSC-based LiDAR profiles, this study provides insights into the improvement of boundary condition realism in microscale urban ventilation modelling and is helpful for the development of climate-adaptive planning strategies for high-density cities facing heat risks.</p
QSTAformer: a quantum-enhanced transformer for robust short-term voltage stability assessment against adversarial attacks
No full textShort-term voltage stability assessment (STVSA) is critical for secure power system operation. While classical machine learning-based methods have demonstrated strong performance, they still face challenges in robustness under adversarial conditions. This paper proposes QSTAformer—a tailored quantum-enhanced Transformer architecture that embeds parameterized quantum circuits (PQCs) into attention mechanisms—for robust and efficient STVSA. A dedicated adversarial training strategy is developed to defend against both white-box and gray-box attacks. Furthermore, diverse PQC architectures are benchmarked to explore trade-offs between expressiveness, convergence, and efficiency. To the best of our knowledge, this is the first work to systematically investigate the adversarial vulnerability of quantum machine learning-based STVSA. Case studies on the IEEE 39-bus system demonstrate that QSTAformer achieves competitive accuracy, reduced complexity, and stronger robustness, underscoring its potential for secure and scalable STVSA under adversarial conditions.</p
Roles of biochar in improving carbon mineralisation and sequestration in sustainable cement–based materials
No full textBiochar (BC) can enhance the carbon sequestration capacity of cementitious composites, offering a promising strategy for reducing carbon emissions. This study reviews the effects of BC on the carbon sequestration capacity of biochar–cement composites (BCC) from a microstructural perspective. Firstly, the effects of biomass type, pyrolysis conditions, and activation methods on the physicochemical properties of BC are discussed. Then, the effects of various types of BC on the hydration and carbonation reactions of cementitious matrices are summarised through micro-characterisation techniques. Moreover, BC significantly affects fresh properties, mechanical properties, and durability of cement composites. Additionally, the mechanisms of carbon sequestration in BCC are analysed. Both physical and chemical methods can significantly modify the microstructure of BC, thereby enhancing its CO2 adsorption capacity. With increasing CO2 concentration, both the hydration and carbonation processes in cement composites are further enhanced. Finally, the carbon sequestration abilities and economic benefits of BCC are quantitatively summarised through related case studies. The effects of the physicochemical properties of BC on low-carbon cementitious composites are systematically analysed, demonstrating that BC with different properties exhibits significant variations in the carbon sequestration capacity of cementitious composites. Therefore, optimizing the physicochemical properties of BC represents an effective strategy for reducing the carbon emissions of cementitious composites.</p
CoCoGesture: Towards coherent co-speech 3D gesture generation in the wild
No full textDeriving co-speech 3D gestures has seen tremendous progress in virtual avatar animation. However, due to the limited scale of 3D speech-gesture data, the existing methods often produce stiff and unreasonable gestures with unseen human speech inputs. To address this issue, we curate a large-scale co-speech gesture dataset covering diverse in-the-wild gesture types and propose a framework for generating plausible and diverse gestures from in-the-wild speech. Specifically, our curated dataset GES-X contains about 40M meshed postures across 4.3K speakers, which is 4× larger in word corpus and 15× more diverse in gesture motion distribution than the second-largest dataset, providing a solid foundation for diverse gesture generation. Considering the gesture sequence should be both natural and display generalization on in-the-wild speech audio, we propose CoCoGesture, a novel framework that is built upon a custom-designed pretrain-finetune training paradigm. At the pretraining stage, we aim to formulate a large generalizable gesture diffusion model by learning the abundant postures manifold provided by our GES-X dataset. Therefore, we scale up the large unconditional diffusion model to 1B parameters and pre-train it to be our gesture experts. At the finetune stage, we present the audio ControlNet that incorporates the human voice as condition prompts to guide the gesture generation. Considering the synthesized postures should be temporally coordinated with audio rhythmic while preserving the vividness and diversity, we design a novel Mixture-of-Gesture-Experts (MoGE) block. In particular, the MoGE block adaptively fuses the audio embedding from the human speech and the gesture features from the pre-trained gesture experts with a routing mechanism. Extensive experiments demonstrate that our proposed CoCoGesture outperforms the state-of-the-art methods on the zero-shot speech-to-gesture generation. Our dataset will be released on the project page: https://mattie-e.github.io/GES-X/.</p