Accurate and efficient simulation tools are essential in robotics, enabling the visualization of system dynamics and the validation of control laws before committing resources to physical experimentation. Developing physically accurate simulation tools is particularly challenging in soft robotics, largely due to the prevalence of geometrically nonlinear deformation. A variety of robot simulators tackle this challenge by using simplified modeling techniques-such as lumped mass models-which lead to physical inaccuracies in real-world applications. In contrast, high-fidelity simulation methods for soft structures, such as finite element analysis, offer increased accuracy but lead to higher computational costs. In light of this, we present a discrete differential geometry-based simulator that provides a balance between physical accuracy and computational efficiency. Building on an extensive body of research on rod and shell-based representations of soft robots, our tool provides a pathway to accurately model soft robots in a computationally tractable manner. Our open-source MATLAB-based framework is capable of simulating the deformations of rods, shells, and their combinations, primarily utilizing implicit integration techniques. The software design is modular for the user to customize the code-for example, add new external forces and impose boundary conditions-to suit their requirements. The implementations for prevalent forces encountered in robotics, including gravity, contact, kinetic and viscous damping, and hydrodynamic and aerodynamic drag, have been provided. We provide several illustrative examples that showcase the capabilities and validate the physical accuracy of the simulator. The open-source code is available at https://github.com/StructuresComp/dismech-matlab. We anticipate that the proposed simulator can serve as an effective digital twin tool, enhancing the Sim2Real pathway in soft robotics research.
The structures of protein complexes allow us to understand and modulate the biological functions of the proteins. Integrative docking is a computational method to obtain the structures of a protein complex, given the atomic structures of the constituent proteins along with other experimental data on the complex, such as chemical cross-links or SAXS profiles. Here, we develop a new discrete geometry-based method, wall-EASAL, for integrative rigid docking of protein pairs given the structures of the constituent proteins and chemical cross-links. The method is an adaptation of efficient atlasing and search of assembly landscapes (EASAL), a state-of-the-art discrete geometry method for efficient and exhaustive sampling of macromolecular configurations under pairwise intermolecular distance constraints. We provide a mathematical proof that the method finds a structure satisfying the cross-link constraints under a natural condition satisfied by energy landscapes. We compare wall-EASAL with integrative modeling platform (IMP), a commonly used integrative modeling method, on a benchmark, varying the numbers, types, and sources of input cross-links, and sources of monomer structures. The wall-EASAL method performs similarly to IMP in terms of the average satisfaction of the configurations to the input cross-links and the average similarity of the configurations to their corresponding native structures. But wall-EASAL is more efficient than IMP and more robust against false positive cross-links in the context of binary integrative rigid docking. Although the current study uses cross-links, the method is general and any source of distance constraints can be used for integrative docking with wall-EASAL. However, the current implementation only supports binary rigid protein docking, i.e., assumes that the monomer structures are known and remain rigid. Additionally, the current implementation is deterministic, i.e., it does not account for some uncertainties in the cross-linking data, such as noise in the cross-link distances. Neither of these appears to be a theoretical or algorithmic limitation of the EASAL methodology. Structures from wall-EASAL can be incorporated in methods for modeling large macromolecular assemblies, for example by suggesting rigid bodies or restraints for use in these methods. This will facilitate the characterization of assemblies and cellular neighborhoods at increased efficiency, accuracy, and precision. The wall-EASAL method is available at https://bitbucket.org/geoplexity/easal-dev/src/Crosslink and the benchmark is available at https://github.com/isblab/Integrative_docking_benchmark.
This study explores the effect of a scaffold geometry on cell distribution within a cell culture system with specific environment conditions, employing computational modeling and discrete phase method (DPM) to analyze cell motion and attachment. Using a cubic polycaprolactone (PCL) scaffold designed with controlled porosity and architecture, simulations were performed using DPM to examine the interaction between scaffold design and fluid flow dynamics and how the cells motion is affected. The findings highlight that the role of the scaffold architecture and fluid conditions are essential in the optimization of cell seeding efficiency. The results provide a foundation for developing tailored scaffold designs that enhance cellular interaction, adhesion, and proliferation, thereby supporting successful tissue regeneration applications.Clinical Relevance- Large bone defects often require invasive surgical interventions that come with significant risks and variable success rates. Bone tissue engineering (BTE), combined with computational fluid dynamics (CFD) analysis, provides a powerful approach for designing scaffolds that enhance bone regeneration. By understanding how fluid dynamics influence cell viability and growth within porous scaffolds, this study helps optimize scaffold design to improve osteogenic performance. These insights can lead to more effective and reliable bone graft alternatives, reducing the need for complex surgeries and improving patient outcomes in orthopedic and reconstructive medicine.
The present study introduces a web-based framework that provides real-time generation, visualization and meshing of venous valve geometry using analytically defined constructional parameters. The proposed framework provides a parametric and physiologically inspired representation of venous valve geometry to support investigations of deep vein thrombosis (DVT), where thrombus formation is known to occur in valve sinus regions. The interface allows manipulation of physiological parameters, including vein radius, sinus bulge, sinus length, leaflet spacing, leaflet thickness, and leaflet extension. This enables the modeling of both healthy and pathological valve morphologies and ensures reproducibility. Based on the parameter values, the framework generates a two-dimensional (2D) outline of the geometry and a three-dimensional (3D) lab-on-a-chip-style geometry. It supports the direct export of stereolithography (STL) files, numerical simulation, or microchannel fabrication. The framework also generates high-quality meshes suitable for computational fluid dynamics (CFD) simulations using a hexagonal lattice-based discretization strategy, combined with Delaunay triangulation and masking. Parameter sweeps reveal the influence of sinus geometry on flow space, the volume of the recirculation zone, and regions prone to thrombus formation due to low shear. The web-based framework can assist the experimental thrombogenesis studies by enabling interactive parametric exploration and rapid prototyping. The developed framework can serve as a scalable tool for thrombogenesis research, bridging computational design, simulation and experimental validation. The modular, rapid geometry-generation pipeline is designed for integration with real-time Digital Twin platforms to support patient-specific venous flow modeling and simulation.
Current methods for evaluating scleral staphyloma morphology fail to provide curvature data of global scleral deformation. This study aimed to develop a quantitative method based on discrete differential geometry for analyzing scleral deformation caused by staphyloma. This study retrospectively analyzed 128 eyes of 73 patients with pathological myopia. All patients underwent orbital magnetic resonance imaging and three-dimensional (3D) reconstruction. The discrete Gaussian curvature measures (DGCMs) and discrete mean curvature measures (DMCMs) of all vertices on the ocular 3D model were calculated. We further established a computational method for the degree of scleral staphyloma expansion (E/U ratio). The posterior scleral DGCM (pDGCM) and DMCM (pDMCM) standard deviations (SDs) were significantly greater in the staphyloma group than in the non-staphyloma group (0.069 ± 0.026 vs. 0.025 ± 0.005 and 0.335 ± 0.096 vs. 0.154 ± 0.027, respectively; both P < 0.0001). The E/U ratio was strongly linearly correlated with both the pDGCM and pDMCM SDs (both P < 0.01). For staphyloma diagnosis, the areas under the receiver operating characteristic (ROC) curves for pDGCM and pDMCM SDs were 0.994 (95% confidence interval [CI], 0.984-1.000) and 0.993 (95% CI, 0.982-1.000), respectively (both P < 0.001). The variation in the curvature of the posterior sclera is significantly greater in eyes with staphyloma than in those without. This variation is highly specific and sensitive for staphyloma diagnosis. This method, based on discrete differential geometry, enables the direct quantification of scleral deformation, potentially providing a quantitative basis for the diagnosis and evaluation of staphyloma.
Protein loop modeling remains a fundamental challenge in computational biology due to the inherent flexibility of loops and their critical role in biological functions. In this work, we employ a discrete distance geometry formulation, efficiently solved using the Branch-and-Prune algorithm, with a key innovation being the incorporation of hydrogen atoms into the model. Hydrogen atoms bonded to N and C α in the protein backbone introduce additional geometric constraints, and their inclusion is particularly justified in the context of nuclear magnetic resonance (NMR) experiments, where short-range hydrogen-hydrogen distances can be detected and provide valuable structural information. By integrating these experimentally accessible constraints into the modeling process, we refine the representation of protein conformations. Computational experiments demonstrate that incorporating hydrogen atoms reduces the conformational space, leading to a more constrained and biologically realistic model. Comparisons with hydrogen-free formulations confirm that our approach improves agreement with known protein structures, further highlighting the relevance of distance geometry methods in structural refinement.
Acoustic metamaterial development is a rapidly expanding field, which has shown great success in many applied scenarios. The development of acoustic metamaterials is enabled by modelling techniques such as the Finite Elements Method (FEM), which allow for the determination of acoustic parameters of structures with different shapes and sizes. However, complicated geometries within narrow regions of such structures require the consideration of thermoviscous phenomena in modelling, which significantly increases the computational load, at times rendering structural optimization infeasible. This study presents a method for determining acoustic sound absorption parameters of an arbitrary-shape acoustic metamaterial cell in a discretized representation space. The proposed method utilizes a reduction in model dimensionality, using a 2D numerical model to represent a 3D structure, which allows for reduced computational complexity while retaining model accuracy. The method was validated with a simple geometry and compared to methods commonly used in the field including equivalent fluid modelling and the Transfer Matrix Method. The method was validated using impedance tube measurements of generated arbitrary geometry metamaterial cells. A numerical mesh sensitivity study was conducted utilizing 20 generated numerical models. Recommendations regarding the mesh density within a 2D thermoviscous acoustics domain were presented, with acceptable modelling error levels (under 0.5%) achieved by models with strongly reduced mesh density, with a maximum model complexity reduction of 72.4%.
In this study, a novel algorithm for computing red blood cell (RBC) geometry was developed as the first step of a quantitative model for RBC-ATP release. This model relied on the developing coordinate-invariant computational framework of discrete exterior calculus (DEC). The algorithm for the first time in literature was formulated in an implicit manner, utilized a Lie-derivative based vertex drift contribution to ensure the mesh was well-behaved throughout deformation, and was able to obtain RBC equilibrium geometries in an efficient manner. This algorithm was shown to be highly stable, quantified through tracking the RBC membrane energy. Equilibrium geometries were shown to agree with literature in in vivo observations, and qualitatively reproduced phenomena seen in in vivo experiments where RBCs are subjected to solutions of varying osmolarity. This DEC algorithm will be applied in future work to fluid-structure interactions of RBCs, and has application to a multitude of open cell biology problems.
Efficient targeting of the olfactory cleft remains a key barrier to olfactory-targeted intranasal therapy and emerging nose-to-brain (N2B) delivery strategies. However, the upstream aerodynamic mechanisms governing aerosol access to the olfactory cleft during natural inhalation remain insufficiently characterized. A standardized representative sinonasal model reconstructed from high-resolution CT data of 32 healthy adults was used to evaluate the effects of administration plane, aerosol particle size, and administration angle on olfactory cleft deposition. Airflow and particle transport were simulated using a lattice Boltzmann-discrete particle method (LBM-DPM) framework across 108 parameterized conditions under natural inhalation. A geometry-consistent 3D-printed nasal model combined with radiotracer-based SPECT/CT imaging was used to experimentally validate deposition trends across administration planes. Under nebulized delivery during natural inhalation, administration plane and aerosol particle size were the primary determinants of olfactory deposition efficiency, whereas administration angle exerted minimal influence. Shallow insertion facilitated upstream aerosol transport toward the olfactory cleft, with particles of approximately 7 μm achieving the highest and stable deposition efficiency across conditions. A modest interaction between insertion depth and particle size was observed without altering the optimal delivery configuration. In vitro radiotracer experiments demonstrated consistent deposition trends across administration planes compared with numerical simulations, supporting the model predictions. Under physiological inhalation, shallow nozzle positioning combined with intermediate-sized aerosol particles represents an optimal parameter configuration for olfactory-targeted intranasal aerosol delivery. These findings provide quantitative guidance for optimizing intranasal administration parameters and may support the development of nebulized delivery systems for olfactory-targeted therapies.
Effective inhaled drug delivery depends on formulation, device, and a clear understanding of aerosol transport in the respiratory tract. This study explores how lower airway anatomical complexity affects airflow and particle deposition in the main respiratory airways. Seven 3D bronchial models with varying airway generational depths were developed. Numerical simulations were conducted using the discrete phase model (DPM) to simulate aerosol transport and deposition under transient inspiratory flows at inhalation peak flow rates of 60 and 120 L/min. Our results showed that simplified models with three bronchial generations could underestimate the deposition efficiency in the main respiratory airways by up to 65.6% compared to the seven-generation lower airway model. Additionally, disparities in deposition outcomes and flow characteristics diminished with increasing lower airway complexity, with minimal changes in flow dynamics and deposition observed beyond the model with six bronchial generation. This study suggests that airway models incorporating at least up to the 6th generation of bronchial branching may be required to sufficiently capture the implications for inhalation therapy design and the development of reliable in silico testing frameworks. Notably, insufficient lower airway detail not only compromises lung deposition estimates but also affects airflow dynamics throughout the entire respiratory tract, including the upper airway. This study highlights the importance of retaining detailed geometry in the design of inhalation therapies and the development of in silico testing frameworks.
We present a comprehensive computational model to simulate the coupled dynamics of aqueous humor flow and heat transfer in the human eye. To manage the complexity of the model, we make significant efforts in meshing and efficient solution of the discrete problem using high-performance resources. The model accurately describes the dynamics of the aqueous humor in the anterior and posterior chambers and accounts for convective effects due to temperature variations. Results for fluid velocity, pressure, and temperature distribution are in good agreement with existing numerical results in the literature. Furthermore, the effects of postural changes and wall shear stress behavior are analyzed, providing new insights into the mechanical forces acting on ocular tissues. Overall, the present contribution provides a detailed three-dimensional simulation that enhances the understanding of ocular physiology and may contribute to further progress in clinical research and treatment optimization in ophthalmology.
In magnetron sputtering-based gas aggregation sources, nanoparticle formation and yield are strongly influenced by the flow-regulated transport and residence time of particles within the condensation chamber. However, the effect of internal geometric parameters on flow structure and nanoparticle growth is not well understood. In this study, computational fluid dynamics (CFD) coupled with a discrete phase model (DPM) is employed to investigate how magnetron radius affects flow characteristics, particle transport, and their implications for nanoparticle formation. The results show that increasing the magnetron radius significantly enhances axial flow uniformity and suppresses vortex structures near the inlet. This shift from radial diffusion-dominated to primarily axial transport effectively reduces particle trapping and wall deposition. Furthermore, the regulation of flow structure modifies particle residence time distributions, which is considered a key factor associated with nanoparticle growth potential and size evolution in gas-phase synthesis. Larger magnetron radii promote more stable transport pathways and improve particle transmission efficiency, thereby improving particle transmission efficiency and providing more favorable conditions for nanoparticle formation. These findings indicate that geometric optimization can simultaneously enhance transport efficiency and influence the conditions potentially favorable for particle growth, providing valuable guidance for the design of high-yield nanoparticle synthesis systems. Overall, this work provides insight into how flow field characteristics influence nanoparticle transport and potential growth behavior, offering a foundation for optimizing magnetron sputtering-based nanoparticle synthesis.
Biological memory networks are thought to store information by experience-dependent changes in the synaptic connectivity between assemblies of neurons. Recent models suggest that these assemblies contain both excitatory and inhibitory neurons (E/I assemblies), resulting in co-tuning and precise balance of excitation and inhibition. To understand computational consequences of E/I assemblies under biologically realistic constraints we built a spiking network model based on experimental data from telencephalic area Dp of adult zebrafish, a precisely balanced recurrent network homologous to piriform cortex. We found that E/I assemblies stabilized firing rate distributions compared to networks with excitatory assemblies and global inhibition. Unlike classical memory models, networks with E/I assemblies did not show discrete attractor dynamics. Rather, responses to learned inputs were locally constrained onto manifolds that 'focused' activity into neuronal subspaces. The covariance structure of these manifolds supported pattern classification when information was retrieved from selected neuronal subsets. Networks with E/I assemblies therefore transformed the geometry of neuronal coding space, resulting in continuous representations that reflected both relatedness of inputs and an individual's experience. Such continuous representations enable fast pattern classification, can support continual learning, and may provide a basis for higher-order learning and cognitive computations.
This work presents a novel fully automated computational framework for optimizing profile extrusion dies, aiming to achieve balanced flow at the die flow channel outlet while minimizing total pressure drop. The framework integrates non-isothermal, non-Newtonian flow modeling in OpenFOAM with a geometry parameterization routine in FreeCAD and a Bayesian optimization algorithm from Scikit-Optimize. A custom solver was developed to account for temperature-dependent viscosity using the Bird-Carreau-Arrhenius model, incorporating viscous dissipation and a novel boundary condition to replicate the thermal regulation used in the experimental process. For optimization, the die flow channel outlet cross-section is discretized into elemental sections, enabling localized flow analysis and establishing a convergence criterion based on the total objective function value. A case study on a tire tread die demonstrates the framework's ability to iteratively refine internal geometry by adjusting key design parameters, resulting in significant improvements in outlet velocity uniformity and reduced pressure drop. Within the searching space, the results showed an optimal objective function of 0.2001 for the best configuration, compared to 0.7333 for the worst configuration, representing an enhancement of 72.7%. The results validate the effectiveness of the proposed framework in navigating complex design spaces with minimal manual input, offering a robust and generalizable approach to extrusion die optimization. This methodology enhances process efficiency, reduces development time, and improves final product quality, particularly for complex and asymmetric die geometries commonly found in the automotive and tire manufacturing industries.
Fluid flow through rock fracture networks was experimentally and numerically studied based on an enhanced discrete fracture network (DFN) model that explicitly characterizes 3D void geometry within rough-walled fracture. Fluid flow tests under different hydraulic gradients J were conducted on a series of DFN samples created by a 3D printer. Meanwhile, numerical simulations were performed based on the enhanced DFN model solving the NS equations and conventional DFN model solving the Reynolds equation, respectively. The validity of the simulations was verified by comparison with flow tests. Then numerical investigations were extended to amend the permeability estimated by the Reynolds equation to seek for an acceptable approximation to the calculation of the Navier-Stokes (NS) equations. The results indicate that the enhanced DFN model can better capture the nonlinear flow caused by surface roughness and aperture heterogeneity, providing more realistic fracture representation and more accurate results. The conventional DFN model overestimates permeability by up to 82 % compared to the flow test result, while the enhanced DFN model give more accurate permeability with a fewer error of 5.3 %. As the fracture number or surface roughness increases, the critical hydraulic gradient Jc that defines the onset of the nonlinear flow decreases. For the linear flow regime under J < Jc, a model that can directly compare the equivalent permeability estimated by NS equations and Reynolds equation was proposed. This is important for assessment of permeability of fracture media where the conventional DFN model solving Reynolds equation is primarily utilized to reduce computational burden.
In this paper, the radiation characteristics of Anabaena cells and their structural influences are systematically analyzed by using the discrete dipole approximation and experimental validation. While previous studies have often focused on the impact of either external morphology or internal composition in isolation, a systematic and quantitative analysis of the influence weights of both internal and external microstructures on all three key radiative properties (absorption, scattering, and phase function) is still lacking. Theoretical calculations show that the cytoplasm and the number of cell bodies are the main microstructures affecting the absorption cross-section of columnar Anabaena cells, and the influence of these two microstructures needs to be considered when constructing the model. The effects of cell wall, geometry, lipid nucleus, nuclear zone, and starch nucleus are more limited, and these microstructures can be simplified when constructing the model to improve the computational efficiency. For the scattering cross section, the microstructures that play a major role in influencing the scattering cross section are, in order, the number of cell bodies, cytoplasm, nuclear zone, and cell wall, which need to be taken into account when constructing the model. In addition, the effect of lipid nuclei needs to be additionally considered when calculating the exact numerical size of the scattering cross-section. In contrast, the effects of the geometry and starch nucleus of Anabaena are relatively minor and can be simplified for the modeling. For the scattering phase function, the exact microstructural parameters that need to be ensured when modeling are the geometry and the cytoplasm. In addition, the effect of the number of cell bodies needs to be taken into account when studying the scattering phase function near the scattering angle of 0°, while the rest of the microstructure can be simplified when modeling. These findings provide an important basis for the establishment of an accurate radiation transport model for microalgae, and it is recommended to focus on the key structural parameters during model construction, while the secondary structures can be appropriately simplified to improve computational efficiency. This study is an important theoretical guide for optimizing the design of photobioreactors.
The complexity of the processes occurring in both natural and artificial joints necessitates carrying out the analysis on a 3D model based on already existing mathematical models. All the presented numerical calculations define qualitative conclusions about the influence of certain parameters of endoprostheses on the values of stresses and strains arising in polyethylene parts of hip and knee endoprostheses. The obtained results make it possible to reveal "weak points" in the studied models and thus counteract the later effects resulting from premature wear of the endoprosthesis components. The study included a numerical analysis of the stress and strain distribution of polyethylene components of hip and knee endoprostheses working with the most commonly used material associations in this type of solution. The most common are metal alloys and ceramics. The analyses were carried out using ADINA and Autodesk Simulation Mechanical software. Geometric models were designed based on current solutions used by leading endoprosthesis manufacturers. The load models adopted are based on models commonly used in musculoskeletal biomechanics. Particular attention was paid to modeling the resistance due to friction at the hip endoprosthesis node. To build the hip endoprosthesis model, eight-node 3D solid elements were used. Due to the axisymmetric geometry of the model, the resulting discrete model consisted of 10,000 cubic elements described by 10,292 nodes. In the case of the knee endoprosthesis, a finite element mesh was adopted for the calculations, which was built with 3600 3D solid cubic elements and 4312 nodes. The accuracy of the adopted numerical model did not differ from the generally used solutions in this field.
This dataset reports the characterization and data processing methodology of 45 individual AISI 316L single melt tracks, fabricated by powder blown laser beam directed energy deposition (DED-LB) metal additive manufacturing. The melt tracks were deposited across a parametric combination of process parameters: powder size distributions, carrier gas flow rates, and laser spot diameter-laser power sets. The measured melt track properties include the average melt track width, height, cross-sectional area, and the powder catchment efficiency. Optical profilometry was used to extract the melt track dimensions and to calculate the powder catchment efficiency. In addition, the corresponding particle stream spatial distributions and particle velocity distributions were measured across the deposition flow parameters by processing high-speed image data. The median particle Stokes number for each flow condition was reported for comparability with other discrete coaxial nozzle systems with particle-laden flows. This dataset can aid in the validation of computational simulations of particle-laden flows from three-jet nozzle systems and the validation of DED-LB models which predict the melt track properties from known process parameters.
Structure-based drug design (SBDD) has advanced with deep generative models, but bridging the gap between continuous atomic coordinates and discrete atom types remains a challenge. Current approaches, such as diffusion and flow matching models, often fail to unify these heterogeneous modalities, relying on separate strategies or ill-fitting Euclidean metrics for discrete variables. This lack of a consistent framework limits generative models' ability to capture the geometric and chemical structure of protein-ligand complexes. We present MolPIF, a parameter interpolation flow mechanism designed to unify the generation of continuous and discrete molecular variables. Unlike traditional flow models that operate in sample space, MolPIF interpolates between distributions in the parameter space, theoretically recovering Wasserstein-2 optimal transport for continuous coordinates and establishing Fisher-Rao geodesics for discrete atom types. We further incorporate a geometry-enhanced learning strategy to improve the capture of atomic contexts. Extensive evaluations on the CrossDocked2020 dataset demonstrate that MolPIF outperforms baselines in binding affinity, chemical validity, geometric fidelity, and chemical space coverage. Additionally, MolPIF exhibits versatility in lead optimization and offers flexible prior distribution selection (such as Laplace), establishing a robust paradigm for SBDD. Source code is freely available at https://github.com/BLEACH366/MolPIF.
The accurate and rapid prediction of hazard gas dispersion fields in urban environments is essential for guiding emergency sensor deployment and enabling real-time risk assessment. However, the computational cost associated with Computational Fluid Dynamics (CFD) simulations hinders their use as real-time forward models, while simplified Gaussian plume models lack the fidelity to resolve building obstruction effects. This study proposes a morphology-guided conditional Generative Adversarial Network (cGAN) framework designed to achieve real-time gas dispersion field modeling in urban environments with complex building configurations. The urban area is discretized into 50 × 50 m grid cells, each characterized by six morphological parameters describing building geometry. K-means clustering categorizes these cells into distinct morphological types. High-fidelity dispersion datasets are then generated for each type using Lattice Boltzmann Method (LBM) simulations. Each sample encodes building geometry, release location, wind speed, and time as multi-channel input images, with the corresponding gas dispersion concentration field is recorded as the output. Two cGAN architectures, Image-to-Image Translation (Pix2Pix) and its high-resolution variant (Pix2PixHD), are employed to learn the mapping from input features to dispersion fields. Model performance is evaluated using four complementary metrics: Fraction within a Factor of Two (FAC2) for prediction accuracy, Normalized Root Mean Square Error (NRMSE) for precision, Fractional Bias (FB) for systematic error, and Structural Similarity Index (SSIM) for spatial pattern fidelity. A case study is conducted across a 1176 km2 urban district in China. The results demonstrate that under varying wind speeds (0.5-1.5 m/s) and temporal scales (5-60 s), and across five morphological categories, the Pix2PixHD-based model achieves 92.5% prediction accuracy and reproduces 97.6% of the spatial patterns. The proposed framework accelerates computation by approximately 18,000 times compared to traditional CFD, reducing inference time to under 0.1 s per scenario. This sub-second capability enables real-time concentration field estimation for emergency management, and provides a physically informed, computationally feasible forward model that can potentially support sensor-based gas source localization and detection network planning in complex urban environments.