Asian Journal of Civil Engineering最新文献

Structured design methodologies are widely discussed in academic research, yet their use in professional practice, especially at the early design stage, remains limited. This is largely because such methods often fail to capture tacit knowledge, intuition, and contextual reasoning that guide real design work. To address this gap, we propose a machine learning (ML) framework that analyzes past design cases and predicts the likelihood of concept adoption. A dataset of 32,154 instances was used to train and compare four models: Logistic Regression, Random Forest, XGBoost, and Artificial Neural Networks. Among these, XGBoost showed the highest accuracy and interpretability. Features such as feedback score, decision time, and designer experience proved to be the most influential predictors of adoption. Rather than replacing intuition, the framework is intended to complement it, providing interpretable, data-driven insights that improve the usability and acceptance of design methods. The findings suggest that ML can strengthen the bridge between academic methodologies and practice by creating adaptive, human-centered tools for decision support.

The construction industry plays a pivotal role in socio-economic development but remains a major contributor to environmental degradation due to emissions, noise, and excessive resource consumption. Traditional scheduling frameworks primarily focus on minimizing project duration and cost, often overlooking environmental sustainability. This study proposes a novel hybrid multi-objective optimization model the Latin Hypercube Sampling–Non-dominated Sorting Genetic Algorithm III (LHS-NSGA-III), which integrates Latin hypercube sampling for improved population diversity with NSGA-III for robust many-objective optimization. The developed resource-constrained time-cost-environmental trade-off (RCTCET) model simultaneously minimizes project completion time (PCT), project completion cost (PCC), and project environmental impact (PEI), enabling informed and sustainable decision-making. A comprehensive case study involving 25 interdependent construction activities, each with multiple execution modes and diverse environmental footprints, is used to validate the model’s applicability. The optimization process generates a diverse set of Pareto-optimal solutions, which are further analyzed using clustering, trade-off plots, and correlation analysis. Comparative evaluation with established metaheuristics demonstrates the superiority of the proposed approach in terms of solution diversity, convergence, and hypervolume metrics. This research establishes the feasibility and effectiveness of incorporating environmental objectives into construction scheduling and provides a scalable framework for sustainable project delivery in alignment with global environmental performance targets.

The Modified Adaptive Weight Approach (MAWA) is a widely used and relatively straightforward method for addressing time–cost optimization problems, which are typically formulated as multi-objective optimization tasks. Metaheuristic algorithms are particularly effective for these problems since they iteratively refine a randomly generated population of candidate solutions. However, a noted drawback of the standard MAWA is its assignment of uniform weight factors to all solutions, without considering their individual fitness or distribution in the search space. To overcome this limitation, this study introduces a novel multi-objective framework that integrates the Rao-2 algorithm with MAWA, yielding a set of Pareto-optimal solutions. The performance of this hybrid MAWA–Rao-2 model was evaluated using benchmark construction project case studies from the literature, each consisting of 146 activities. The obtained results were compared with Hybrid heuristic meta-heuristic (HHMH), non-dominated sorting TLBO, non-dominated sorting Aquila optimizer, non-dominated sorting AOA reported in the literature. Findings demonstrate that the MAWA–Rao-2 algorithm serves as a robust and efficient approach for solving time–cost trade-off problems (TCTP) in construction engineering and management.

This study uses statistical modelling and Response Surface Methodology (RSM) to evaluate the design, optimisation, and performance evaluation of Fly Ash-GGBS-based Self-Compacting Geopolymer Concrete (SCGPC), both with and without steel fibres. The self-compatibility requirements were considered when creating mixtures in compliance with EFNARC guidelines. Three types of concrete were made: conventional self-compacting concrete (SCC), self-compacting geopolymer fibre-reinforced concrete (SCGPFRC) with varying percentages of steel fibre, and self-compacting concrete using fly ash and GGBS as binders. Both fresh and hardened properties were evaluated, and the material’s durability was determined through abrasion resistance testing. RSM was used in conjunction with a quadratic model to explore the effect of input variables on compressive, flexural, and split tensile strengths. Although the model’s prediction dependability was restricted, it had appropriate precision. When compared to the SCC and SCGPFRC processes, the SCGPC demonstrated significantly improved flowability and passing ability. The addition of steel fibres resulted in an increase in flexural and split tensile strengths of 16.23% and 41.90%, respectively, at an optimal fibre content of 1.65% (SCGPFRC2), despite the fact that compressive strength decreased somewhat in SCGPC compared to SCC. Despite the fact that the statistical model’s prediction accuracy varies slightly, the experimental results show that SCGPFRC has excellent mechanical performance and durability, establishing it as an environmentally friendly and high-performing material for structural applications.

Vibration control in tall buildings is a critical aspect of modern structural design, directly influencing safety, occupant comfort, and long-term durability. Due to their slender and flexible geometry, high-rise structures are especially susceptible to dynamic forces such as wind and seismic activity, which can induce resonance and lead to structural damage or collapse. One effective strategy for mitigating such vibrations involves the use of end shear walls specialized shear wall elements that connect the extremities of reinforced concrete core walls across all floors. These walls enhance diaphragm stiffness and reduce stress concentrations at shear wall ends, contributing to improved dynamic stability. This research examines how end shear walls affect the seismic behavior of two 30-story structures, utilizing far-field earthquake records and nonlinear dynamic analysis. The model incorporating end shear walls demonstrated a 22% reduction in peak acceleration, a 76% drop in variance, and a 51% decrease in standard deviation in the X-direction, as confirmed through SPSS analysis. Q-Q plots further revealed up to 38% vibration reduction, underscoring the effectiveness of end shear walls in enhancing seismic resilience and improving overall dynamic performance in high-rise structures.

The complex, nonlinear, and random interactions between diffraction parameters and the phase composition of altered concrete systems are often not adequately represented by standard modelling methods. The objective of this research is to employ data from X-Ray Diffraction (XRD) to test the applicability of advanced machine learning methods to phase identification and classification in concrete modified with magnesium chloride and ground granulated blast furnace slag (GGBS). After preparing, curing, and testing various mixes of concrete using different proportions of GGBS and magnesium chloride to measure their compressive strength, the most suitable combination was selected for further microstructural analysis. The two models trained were Vision Transformer (ViT) and Extreme Gradient Boosting (XGBoost). Applied and tested on the 2θ intensity profiles obtained in XRD data. In XGBoost, the original XRD dataset was used as input for the model. As an additional improvement to the dataset, a synthetic dataset is available. It was synthesised with the help of Generative Adversarial Networks (GANs). Both datasets were similar in the measurement of the Mean square error MSE and R² values. Additionally, the performance of the vision transformer is also superior to that of the Convolutional Neural Network (CNN). The declassification of XRD images of conventional concrete and magnesium chloride-modified concrete was then verified using a confusion matrix. The findings reveal that XGBoost and vision can provide comparable results. A transformer can be a helpful method for accurately interpreting XRD data, providing new insights. It allows identifying and defining phases more precisely in cement-based materials, which is also in terms of classification.

This study presents a comprehensive multi-objective optimization framework for retrofitting projects by integrating seven critical performance dimensions: time, cost, quality, energy consumption, safety, environmental impact, and client satisfaction. A novel opposition-based non-dominated sorting genetic algorithm III (OBNSGA-III) is proposed to address the high dimensionality and complex trade-offs inherent in retrofitting decisions. Key innovations include the dual application of opposition-based learning during population initialization and offspring generation, the use of a bivariate normal distribution to model quality as a function of time and cost, and the application of fuzzy logic for safety risk evaluation. The proposed framework is validated using a real-world case study involving 11 retrofitting aspects and 33 intervention options. The OBNSGA-III algorithm successfully generated 18 Pareto-optimal solutions. Among them, the best-performing solution achieved a project duration of 30 days, a quality index of 0.913, and a client satisfaction score of 4.7, outperforming benchmark algorithms such as NSGA-III, MOPSO, and OB-MODE across 13 standard performance indicators, including hypervolume (0.92) and generational distance (1.35). These results underscore the model’s ability to deliver diverse, high-quality trade-off solutions under real-world constraints. The TCQESEC framework provides a robust decision-support tool for project managers and policymakers, enabling sustainable, efficient, and client-centric retrofitting strategies in complex urban infrastructure environments.

This research investigates the application of hybrid artificial neural network (ANN) models that are optimized using Grey Wolf Optimization (GWO), Particle Swarm Optimization (PSO), and Artificial Bee Colony (ABC) algorithms to predict the punching shear capacity (PSC) of flat slabs made of recycled aggregate concrete (RAC). To ensure robust model development, a database of 101 experimental specimens was compiled, processed, and divided into training (70%) and testing (30%) datasets. Regression error characteristic (REC) curves, Taylor diagrams, and statistical indices (R2, RMSE, MAE, WMAPE, WI, and LMI) were used to extensively assess the model’s accuracy, external validation, and uncertainty analysis. ANN-ABC has the best predictive value, as it delivered the lowest values of errors and R² of 0.9593 (training) and 0.9527 (testing). The robustness of its analysis was also supported by narrow uncertainty bounds, favourable rankings across all evaluation criteria, and analysis of the REC curve (AUC = 0.755 during training and 0.563 during testing). On the contrary, ANN-PSO worked with moderate accuracy, whereas ANN-GWO worked with the lowest accuracy. The sensitivity analysis showed that effective depth, reinforcement area, and water-to-cement ratio were the most sensitive parameters that control PSC behavior. Unlike previous PSC prediction studies that relied mainly on tree-based or kernel-based ML methods, this work is the first to benchmark swarm-intelligence-optimized ANNs (ANN-ABC, ANN-PSO, ANN-GWO). This hybridization improves predictive accuracy while ensuring robustness and interpretability for RAC structural applications.

Paper presents a validated method of metaheuristic-trained adaptive activation neural networks that are supposed to integrate mechanics with imperfections and reliability calibration into a unified process. The method starts with a physics-regularized column state encoder that carries equilibrium and energy consistency while compactly summarizing geometry, residual stresses, imperfections, and boundary conditions. A state latent informs a stability-energy guided adaptive activation network, where metaheuristic tuning of nonlinear activations is supposed to adjust predictions toward mechanical principles and increase fidelity. The third stage, imperfection manifold synthesizer, produces statistically and physically realistic imperfection fields conditioned on the latent states and stability sensitivities thereby enlarging sparse experimental catalogs. Building on this, a reliability-preserving resistance calibrator that solves the inverse reliability problem and stretches strength reduction factors and dispersion surfaces smoothed out across shapes and load conditions completes the process. Finally, a code-integrable decision map constructor compress calibrates surface into interpretable rule tables and charts while verifying reliability against adversarial imperfections in process. Across pooled datasets of hot-rolled and welded sections, mean resistance errors attained by the method is almost 3% with calibrated uncertainty coverage and reliability factors well positioned with respect to code target requirements. Beyond accuracy, the framework adds a transferable latent state, a defensible reliability link, and compact design maps on the path toward safer and more implementable design standards.

The present study intends to study the mechanical properties of fiber-reinforced geopolymer concrete (FRGC) experimentally and validate the results using state-of-the-art machine learning. At 28 days, compressive strength (CS) of 72.6 MPa, and split tensile strength (STS) of 9.4 MPa were recorded. The CS deteriorated with a further increase in the NF content due to the cluster formation, while the STS increased with increasing fiber dosage. The hybrid ML models, including Random Forest (RF) and XGBoost (XGB), were fine-tuned using Grid Search (GS) and Giant Armadillo (GA) algorithm based on 5-fold cross-validation. The GA-XGB model had maximum accuracy to predict CS (R² = 0.988, RMSE = 0.032 MPa) and STS (R² = 0.985, RMSE = 0.029 MPa) in testing sets. SHAP analysis supported molarity and SS/SH as the influencing factor for CS, and meanwhile, NF for STS. Shapley additive explanations (SHAP), Partial dependence (PDP) and Individual Conditional Expectation (ICE) plots confirmed these trends by showing the nonlinear effects of each independent variable on strength predictions. A graphical user interface (GUI) was also designed to aid practical use, which the user can use to insert long short mix parameters and receive instant predictions for CS and STS. The close match between experimental measurements and ML predictions, along with the importance of explainability and GUI integration, proves that the developed FRGPC design framework is robust, transparent and usable in real applications.