Retrofit-measure selection directly determines the energy, safety, and cost performance of existing-building renewal, creating an urgent demand for intelligent decision-support tools that can translate historical renovation experience into timely and feasible retrofit schemes. However, static weighting and explicit case matching cannot capture latent relationships between buildings and interventions or discount technologies whose suitability declines as standards, policies, and engineering practices evolve. This study combines dynamic value engineering, node2vec graph embedding, and CatBoost time-decay prediction in a hybrid recommender developed from 5000 Chinese retrofit cases and validated through EnergyPlus simulation and blinded expert assessment. Relative to random forest, HR@5 increased by 4.9 percentage points and simulated EUI reduction by 3.9 points, whereas the potentially obsolete measure rate decreased by 8.8 points and decision time by 0.15 h/case. The balanced fusion of attribute and graph similarity recovered both explicit and latent case relevance, while decay factors correlated with blinded expert ratings and downweighted time-sensitive measures as technologies evolved. This research delivers a dynamically adaptive, intelligent decision-making framework for the sustainable renovation of existing buildings, facilitating the transition of building stock renewal from an experience-based paradigm to a data-intelligence-driven process, thereby contributing substantively to the low-carbon evolution of the structural engineering and construction sector.
Fly ash-based geopolymer concrete (FA-GPC) has emerged as a promising low-carbon alternative to conventional cementitious materials; however, reliable prediction of its compressive strength remains challenging due to the coupled influence of compositional, chemical formulation, and curing-related parameters. This study develops a data-driven artificial neural network (ANN) framework for multi-parameter strength prediction and decision support in FA-GPC mixture design using a curated database comprising 563 mix designs collected from 55 published studies. Fourteen representative variables were systematically identified and organized into four physically meaningful attributes, including mixture proportions, molar formulation ratios, chemical composition of raw materials, and curing conditions, enabling integrated evaluation of their combined influence on strength development. A multilayer feed-forward artificial neural network was trained and validated using k-fold cross-validation to ensure robustness of the predictive framework. The developed model achieved R2 coefficients of 0.863 and 0.756 for training and validation datasets, respectively, with corresponding RMSE values of 7.07 MPa and 9.10 MPa. Comparative analysis with previously reported machine learning models demonstrated that the proposed framework maintains competitive accuracy while incorporating an expanded set of influential parameters. Variable importance evaluation indicated that H₂O/Al₂O₃ ratio, Na₂O/SiO₂ ratio, and water-to-solid ratio exert dominant control over strength development. The proposed framework further enables data-driven exploration of parameter combinations to support informed mixture design and performance-oriented material optimization. By linking predictive modeling with design-oriented insights, the study contributes toward the development of intelligent and resource-efficient construction materials, offering a computational tool to support sustainable decision-making in geopolymer concrete design.
This study introduces a computational approach for the free vibration analysis of graphene-platelet-reinforced functionally graded triply periodic minimal surface (GPLR-FG-TPMS) plates. The proposed approach seamlessly combines Chebyshev polynomials with moving Kriging (CMK) meshless method and the third-order shear deformation theory, resulting in a unified Chebyshev-based computational framework for high-fidelity vibration analysis of macroscale architected functionally graded structures. The third-order Chebyshev shear deformation theory (TCSDT) accurately captures transverse shear effects while naturally satisfying zero transverse shear stress conditions at the plate surfaces. Meanwhile, the CMK meshless method achieves high numerical accuracy using only scattered nodes, without requiring element connectivity. The complex TPMS architectures are homogenized using a relative-density two-phase model, while the graded GPL distribution is modeled based on the Halpin–Tsai micromechanical model under different reinforcement patterns. The proposed CMK–TCSDT formulation demonstrates excellent agreement with available benchmark solutions, with an average relative error of approximately 1.0%, thereby confirming its accuracy and robustness. In addition, new benchmark results for annular and heart-cutout GPLR-FG-TPMS plates are provided, demonstrating the effectiveness, robustness and computational efficiency of the proposed framework for future analyses and the design optimization of architected functionally graded composite plates.
Concrete filled steel tube (CFST) columns commonly endure coupled lateral impact and sustained axial load, which substantially reduces post-impact residual bearing capacity. Traditional theoretical analysis and numerical simulation involve complex modeling and high computational cost, while the existing machine learning research ignores the mechanical mechanism. The current research fails to provide explicit calculation formulas and systematic reliability evaluation. To address these deficiencies, this study first investigated the mechanical response and energy dissipation of CFST columns under coupled lateral impact and axial loading via numerical simulation. A novel Weighted Asymmetric Quantile (WAQ) loss function was proposed, which introduces damage-grading weighting and asymmetric penalty rules to achieve slightly conservative prediction. An improved entropy weight-grey relational analysis (IEW-GRA) was further developed for multi-index performance evaluation and hyperparameter optimization. Six hybrid models integrating back propagation neural network (BPNN) with metaheuristic algorithms were compared, and Sparrow Search Algorithm (SSA)-BPNN-WAQ was identified as the optimal prediction framework. On this basis, the explicit formulas of residual bearing capacity and damage probability were derived, which could rapidly quantify structural reliability without running computationally expensive numerical simulation. Engineering case analysis validated that enlarging section size, steel ratio and steel strength can effectively enhance anti-impact reliability. This work provides an integrated technical route from data-driven prediction to explicit formula derivation and reliability-based optimization. It offers a reliable and safety-oriented reference for anti-impact design of CFST members.
In Korea, the fire resistance of modular buildings is assessed per individual member, neglecting thermal interactions between adjacent members and modules, which leads to conservative and inefficient designs. To address this, this study proposes a coupled computational fluid dynamics–finite element method (CFD–FEM) analysis validated by a full-scale fire test following the British Building Research Establishment’s LPS1501-1 standard. The test measured structural temperatures to assess module-level structural fire performance, and corresponding CFD and FEM analyses were conducted to replicate the results. Their strong agreement verified the reliability of the proposed framework. The validated model was then used to simulate standard and natural fire scenarios to evaluate member temperatures and load ratios based on critical temperature methods according to ANSI/AISC 360-22 and the Korean design guidelines. The CFD–FEM approach offers a reliable and economical alternative to costly full-scale tests, providing insights for performance-based structural fire design in modular construction.
This research investigates the challenge of accurately predicting the crashworthiness of tubular nested (TNS) crash-box subjected to both axial and oblique loading conditions. Conventional theoretical approaches frequently exhibit limited generalizability under complex loading scenarios, while data-driven models often lack physical interpretability. This study proposes a comprehensive multi-method framework that synthesizes theoretical modeling, finite element (FE) analysis, and interpretable machine learning. A high-fidelity finite element model, validated against experimental data, is developed to characterize the crushing behavior of TNS and to identify the predominant geometric parameters influencing crashworthiness under various loading conditions. Drawing upon deformation modes identified through experimental observations and finite element analyses, an analytical model for axial crushing is initially formulated utilizing the simplified super-folding element (SSFE) theory. This model is subsequently expanded to accommodate oblique loading conditions. However, the theoretical model shows reduced accuracy under asymmetric deformation conditions. To address this limitation, machine learning models are introduced to improve prediction accuracy. An explicit formulation strategy is further developed to convert the learned relationships into interpretable analytical expressions, addressing the black-box nature of data-driven methods. Results demonstrate that the proposed framework significantly improves prediction accuracy under complex loading while maintaining physical interpretability. This approach provides a scalable solution for crashworthiness prediction and bridges the gap between physics-based and data-driven methods.
In topology optimization (TO) for additive manufacturing (AM), considering the anisotropy of formed materials induced by layer-by-layer AM processes is an emerging challenge. Based on this special process-related anisotropic constitutive relationship, a lightweight TO framework that considers both strength and stiffness requirements is proposed and explored in this work. Firstly, by introducing a print-off angle variable related to the anisotropy, the classical transversely isotropic model is extended to simulate the process-related anisotropic constitutive behavior. Then, based on the Hoffman failure criterion, a process-related anisotropic failure strength measurement is established. Furthermore, to achieve effective strength control, a global aggregation strategy based on the P-norm and error correction techniques is constructed. On this basis, the classical volume minimization is extended to include both anisotropic strength and stiffness constraints. Additionally, to address the convergence difficulties caused by the angle periodicity, an adaptive adjustment strategy for the angle variation is applied. The sensitivities related to the density and angle variables are derived in detail to adapt to gradient-based optimization algorithms. Typical numerical examples validate the effectiveness of the proposed method. The results reveal the inherent trade-off between lightweight, structural safety, and stiffness performance in the design. By effectively utilizing the process-induced anisotropy, the proposed algorithm reduces material usage while ensuring structural stiffness and strength requirements.
Conventional MLP-based physics-informed neural networks may suffer from spectral bias, gradient ill-conditioning, and limited local expressiveness in high-stiffness and high-frequency structural inverse problems. This study develops a physics-informed Kolmogorov-Arnold network incorporating finite-element semi-discrete dynamic equations into the loss function (FE-PIKAN) for identifying dynamic inputs and structural parameters in two-dimensional frames. Preassembled mass, damping, and stiffness matrices are used to construct the physics residual. Under a prescribed finite-element model, displacement responses are reconstructed from sparse observations, while base excitation and the equivalent flexural stiffness of a predefined candidate member are identified in separate tasks. A stiff single-degree-of-freedom benchmark first evaluates high-stiffness response prediction. A three-story finite-element frame then compares MLP, Fourier-feature MLP, fixed-grid KAN, and refined-grid KAN under the same residual framework. Under controlled numerical conditions, refined-grid FE-PIKAN achieves relative errors of 3.36% for base-acceleration inversion and 0.0522% for equivalent flexural stiffness inversion, with limited error variation under 1%, 2%, and 5% synthetic additive noise. Shake-table tests under harmonic and El Centro excitations show that FE-PIKAN recovers the main trends of external input and equivalent flexural stiffness, achieves relatively low stiffness errors in intact and time-invariant local-damage cases, and captures the approximate transition time and overall evolution trend in time-varying cases, although deviations increase near abrupt transitions. These results support the preliminary feasibility of FE-PIKAN for input identification and equivalent flexural stiffness inversion from sparse measured displacement data under controlled laboratory conditions.
Cold-formed steel–lightweight concrete (CFS-LWC) composite beams offer significant advantages in modular and lightweight construction, but their complex structural behaviour, governed by material nonlinearity, shear interaction, and local buckling of thin-walled sections, poses challenges for traditional design methods. While recent studies have validated numerical models for CFS-LWC beams, these approaches remain computationally intensive and may lack adaptability for rapid design iterations. This study develops a data-driven artificial neural network (ANN) surrogate model to predict the ultimate bending resistance Mu of built-up CFS–LWC composite beams. The study used a high-fidelity dataset comprising 960 finite-element simulations, which was calibrated using four experimental tests. This dataset spans beam lengths from 4310 to 8000 mm, steel section depths from 240 to 300 mm, lightweight concrete compressive strengths between 20 and 35 MPa, slab thicknesses representative of typical floor systems, and steel grades S280GD and S450GD. Inputs were z-score standardised, and model training used five-fold cross-validation with Latin Hypercube Sampling for the hyperparameter search. The final network (one hidden layer, 64 neurons, tanh activation function, and RMSprop optimiser) achieved R2≥0.98, with residuals centred near zero and limited spread, demonstrating strong generalisation within the studied domain. Model explainability via SHAP (Shapley additive explanations) indicates that steel yield strength and section geometry are the most influential features, consistent with structural mechanics. The proposed model shows good agreement with the FE database and can therefore provide more accurate predictions of the bending resistance of CFS–LWC composite beams than conventional analytical approaches. In addition, the surrogate model offers fast and reasonable capacity predictions, minimising the need for repeated FE analyses and making it well-suited for the preliminary design of CFS–LWC composite beams within the validated parameter range.
Industrial computer-aided engineering (CAE) analysis of moving contact assemblies often relies on finite element (FE) and component mode synthesis (CMS) methods because of their high accuracy in capturing local elastic deformation and contact behavior. However, their application to large-scale industrial simulations remains challenging due to the large number of degrees of freedom, expensive local deformation calculations, and repeated contact updates at moving interfaces. To address these challenges, this paper presents a parallel and memory-aware free-interface CMS framework for industrial-scale moving contact simulation. The framework integrates chunked block linear solves with factorization reuse, consistent interface load mapping under multi-point constraints, lock-free parallel post-processing for modal data construction, and parallel bounding volume hierarchy contact detection into a reusable CAE workflow. Three industrial examples, including a helical gear pair, a car differentia, and a gear-rack system, are used to assess accuracy and computational performance. The results show that the proposed method preserves FE-level accuracy while reducing peak memory of unchunked CMS method and achieving speedups of one order of magnitude in large drivetrain simulations.
Elastoplastic contact analysis presents a fundamental computational challenge: the asimultaneous treatment of material nonlinearity and boundary nonlinearity, when approached monolithically, leads to prohibitive computational costs and difficulties in algorithm convergence. A two-layer nested iteration strategy can effectively decouple these dual nonlinearities by reformulating the governing equations as a linear combination of elastic contact equations and elastoplastic equations. However, when implemented on a single-domain model, this strategy is still hindered by the need to reassemble the global stiffness matrix at every plastic iteration. To address this problem, a domain-decomposition B-differentiable Newton method (DD-BDNM) is proposed, integrating the nested iteration strategy with the FETI-BDNM. In this method, the structure is decomposed into non-overlapping subdomains. In the internal iteration layer, with the stiffness matrix of each subdomain held fixed, the contact equations are solved via the B-differentiable Newton method (BDNM) to derive the contact forces, interface forces, and rigid body motions. These quantities are then frozen in the external iteration layer to update the displacement and stress fields within each subdomain. Since the external iteration operates at the subdomain level, only those subdomains undergoing plastic evolution require stiffness matrix updates, while the remaining elastic subdomains remain unchanged. This selective stiffness update is the defining methodological contribution of the DD-BDNM, and is an architectural consequence of combining domain decomposition with the nested iteration strategy. Numerical examples demonstrate the accuracy, mesh convergence, computational efficiency, robustness, and engineering capability of the proposed algorithm.
Computer algebra systems are indispensable for symbolic linear algebra in engineering and scientific computing, but they often suffer from expression swell — a rapid growth in the symbolic expression size during computation. To mitigate this issue, we introduce two Maple packages: Lem (Large Expression Management) and Last (Linear Algebra Symbolic Toolbox). Lem reimplements Maple’s LargeExpressions module using an object-oriented approach that enables more effective control over expression growth. Last offers symbolic linear algebra routines with novel features, such as effective full-pivoting and sparsity preservation strategies. Together, these packages improve the performance of symbolic computations in Maple, offering engineers and scientists robust tools for symbolic matrix factorization and the solution of linear systems. Beyond immediate applications in engineering, science, and applied mathematics, they also provide a deeper understanding of the underlying principles of symbolic matrix factorization, paving the way for further advances and developments in other computer algebra systems.
To simulate the geotechnical mechanics problems across various scenarios in mining processes, such as granular transport in scraper conveyors, top-coal caving, rock fracture, and overlying strata collapse, this study proposes an numerical framework based on the Smoothed Particle Hydrodynamics (SPH) method. Integrating an elastoplastic constitutive model, the Drucker-Prager shear failure model, and the Grady-Kipp tensile failure model within a unified Lagrangian particle framework, the SPH model enables simulation from continuous media to discontinuous systems. The framework is validated through benchmark tests and engineering applications. Granular column collapse simulations demonstrate its capability in capturing free-surface flows and deposition morphology; rock beam bending fracture cases confirm the effectiveness of the G-K model in reproducing tensile crack initiation and propagation; block sliding along predefined weak interfaces validates the filler-particle strategy for simulating discontinuous behavior without explicit contact detection. The framework is then applied to typical mining scenarios: particle transport in a scraper conveyor, top-coal drawing, and coupled evolution of loose coal flow with overlying strata failure. For multi-material simulations involving substantial stiffness contrasts, adopting the global minimum critical time step is found to be essential for numerical stability, though at a notable computational cost. By filling pre-defined fractures with bonded-frictional filler material, the framework effectively simulates block sliding, collision, and collapse through kernel-based interactions. This study provides a versatile mesh-free simulation tool for the integrated numerical analysis of multi-process phenomena in mining geotechnics.
Three phase flows with phase change are central to predicting thermal hydraulic behavior in light water reactors during accident conditions. Gas-liquid-solid interactions govern key safety phenomena, including boiling and condensation during postulated loss of coolant accidents, debris bed cooling following core degradation, and melt fragmentation and solidification during fuel coolant interactions. Over the past decades, substantial progress has been made in computational modeling of these processes, spanning continuum multiphase formulations, Eulerian-Lagrangian approaches, and hybrid computational fluid dynamics and discrete element method (CFD-DEM) frameworks. This review combines recent advances in physical modeling, numerical methods, and multiphysics coupling strategies for three-phase flows with phase change relevant to nuclear thermal hydraulics. Eulerian-Eulerian and Eulerian-Lagrangian models for boiling, condensation, melting, and solidification are discussed, along with sharp-interface, diffuse-interface, and enthalpy-porosity methods for representing evolving phase boundaries. Key benchmark experiments including FARO, KROTOS, TROI, DEFOR, and POMECO are reviewed in relation to validation of fuel fragmentation, steam explosion energetics, debris bed coolability, and reflooding behavior. The treatment of phase change and interfacial transport in severe accident analysis codes and a selected computational frameworks are also compared. Outstanding challenges include transfer of pore scale information to component scale models, uncertainty quantification in interfacial closure relations, computational cost, and the limited availability of high-resolution experimental data under reactor representative conditions. Emerging developments in high performance computing, interface resolved simulation, and machine learning assisted closure development are discussed as supporting tools for improving predictive capability.
The Material Point Method is widely recognized for its robust hybrid Lagrangian–Eulerian formulation, which efficiently handles large deformations. However, its reliance on a global computational grid can introduce unphysical interactions between material points, an issue particularly evident when simulating granular soils such as sand or gravel. According to the principles of soil mechanics, these materials behave as a continuum until their void ratio exceeds a critical threshold; once this threshold is exceeded, individual grains lose contact with their neighbors. The Granular Material Point Method addresses this behavior by introducing disconnected, non-interacting clusters of material points based on the soil’s maximum void ratio. This is achieved algorithmically by dynamically adjusting multiple velocity fields and material indices in response to local changes in the void ratio. This manuscript presents the implementation of the Granular Material Point Method within the Uintah Computational Framework. It verifies its algorithm by simulating the impact collision between two elastic disks, demonstrating that its numerical accuracy matches that of the underlying grid-to-points interpolation method. Furthermore, three case studies, including dry sand settling, the silo problem, and the penetration of a rigid object into dry sand, are conducted to validate this proposed method and evaluate its computational performance. The results demonstrate that the Granular Material Point Method provides a physically consistent representation of granular soil dynamics during solid-to-gas phase transitions, effectively eliminates gaps arising from unphysical interactions between material points, and improves the determination of multi-material interfaces under frictional contact conditions.
This study addresses a multi-objective control problem for active suspension systems in in-wheel motor-driven electric vehicles (IWM-EVs), considering input delays and gain perturbations as practical implementation uncertainties. A full-vehicle active suspension model is employed to evaluate ride comfort in both the vertical and rotational directions during the design stage. Based on this model, a non-fragile static output-feedback controller is developed to achieve finite-frequency H∞/GH2 performance. Ride comfort is improved by minimizing the finite-frequency H∞ index, while the GH2 index is optimized to enforce safety-related constraints. A static output-feedback structure is adopted to enhance practical implementability, with both input delays and gain perturbations explicitly incorporated into the controller synthesis process. The resulting formulation leads to bilinear matrix inequality (BMI) constraints, which are inherently non-convex due to the static output-feedback design. To address these BMIs and balance conflicting performance objectives, a multi-objective particle swarm optimization algorithm is employed. Simulation results under uneven road conditions demonstrate that the proposed controller improves ride comfort while preserving safety-related performance, even in the presence of uncertainties, thereby providing a robust and practically viable solution for IWM-EV suspension systems.
Shrinkage porosity is a primary defect in investment castings, and numerical simulation plays an important role in shrinkage porosity distribution prediction. However, numerical simulation technology cannot obtain results in real-time conditions. Therefore, the research idea has changed from efficient numerical simulation to the construction of high-precision surrogate models. A digital twin (DT) model for investment casting, integrating Reduced Order Model (ROM) and machine learning (ML), is established to simulate the distribution of shrinkage porosity in castings. Based on offline data set established by ProCAST software, the temperature ROM which is constructed through Radial Basis Function (RBF) network interpolation and Proper Orthogonal Decomposition (POD) is developed to realize the rapid simulation of temperature field. Then, the supervised learning method is employed to extract temperature field features from thermal historical data to establish a ML model to predict the shrinkage porosity distribution. Based on ROM and ML, the position of the temperature measurement point is determined by the greedy algorithm. The POD coefficient is subsequently revised according to the measured data of thermocouple, so as to update the temperature field and determine the distribution of shrinkage porosity. Therefore, a shrinkage porosity prediction model based on real-time temperature measurement data is established to achieve rapid and accurate prediction of shrinkage porosity distribution. The results indicate that the error between the predicted temperature and the actual measured temperature is within 20 degrees C, affirming the proposed DT model enables effective and accurate prediction of the temperature field.
Efficient linear solvers are essential for large-scale reservoir simulation, yet the performance of conventional constrained pressure residual (CPR) preconditioners may deteriorate when well-reservoir coupling becomes strong. This study investigates the CPRW preconditioner as implemented in the open-source OPM Flow simulator (version 2024.10), an enhanced CPR variant that incorporates key well-related variables (e.g., bottom-hole pressure) into the coarse-level pressure correction. The paper provides a structural block-matrix formulation that clarifies the algebraic differences between classical CPR and CPRW. Rather than introducing a new preconditioning concept, the focus is on assessing the applicability, robustness, and performance of the existing CPRW formulation across a range of reservoir benchmark scenarios. Numerical experiments on nine models, including black-oil, highly heterogeneous, thermal, multisegment well, and real field-scale cases, show that CPRW improves linear convergence in strongly coupled settings. Key findings include: (1) CPRW reduces linear iteration counts relative to classical CPR in several cases with strong well-reservoir interaction and pronounced near-well nonlinearities; (2) when combined with algebraic multigrid and suitable fine-level smoothing strategies, it yields favorable runtime reductions and parallel performance in the tested cases; and (3) its modular structure leverages existing sparsity while preserving a coarse-level operator compatible with AMG, introducing only a small number of extra nonzero matrix entries associated with the well, enabling efficient integration into industrial simulators and supporting scalable parallel execution. These results position CPRW as a robust and scalable preconditioner for reservoir simulation workflows.
