Asian Journal of Civil Engineering最新文献

High-performance concrete (HPC) is widely used in infrastructure due to its enhanced strength and durability. However, under harsh environmental conditions—such as chloride exposure, freeze-thaw cycles, and variable temperatures—its long-term performance remains uncertain. Traditional models often fall short in forecasting multiple durability parameters due to their reliance on linear assumptions and isolated inputs. This study presents an integrated machine learning (ML) framework for multivariate prediction of key durability indicators: compressive strength and chloride ion penetration. Seven models—MLR, ANN, DTR, RFR, SVR, XGBoost, and LSTM—were developed and evaluated on a dataset incorporating material properties, mix design, curing conditions, and environmental factors. Performance was assessed using R², RMSE, MAE, MAPE, IOA, and a20 metrics. Results show that LSTM and XGBoost consistently outperform traditional models, achieving R² values of 0.965 and 0.942, respectively. Feature importance analysis revealed that the water/cement ratio, silica fume content, and exposure conditions were dominant predictors. The study highlights the potential of ML—particularly LSTM—for accurate, time-dependent durability forecasting in HPC. This framework can support predictive maintenance and service-life design of concrete structures exposed to aggressive environments.

Stiffened steel plate girders are very important for the safety and proper operation of modern structures and bridges. When people try to Figure out how much shear these girders can bear, they frequently make assumptions that don't always take into account how geometry, materials, and stiffeners are related in a complicated way. This research was indicated that machine learning (ML) prediction framework may make better guesses about how much shear strengthened steel plate girders can handle. We put together and changed published experimental results from 173 variables by using numerical methods from the field. As part of feature engineering, geometric and material ratios that are critical to aerodynamics were added. We trained Random Forest, XGBoost, CatBoost, AdaBoost, Decision Tree, and Multilayer Perceptron very carefully and changing their hyperparameters to see how well they worked. Both CatBoost and Random Forest did the best job of predicting, with R² scores over 0.97. SHAP analysis was used to find out that model works and what design features are most important for better shear strength. Engineers can utilize the framework to perform  designs that are both safer and cheaper with peer because traditional code-based methods. To help with real-time prediction for design purposes, a simple graphical interface was  built.

The current study presents a data-driven framework to predict the structural performance of CFRP-strengthened reinforced concrete beams with various web openings using deep learning. Experimental data from beams with circular and elliptical openings under different CFRP wrapping configurations were used to train and evaluate machine learning models. The key structural parameters cracking load, initial/post-cracking stiffness, strain, and energy absorption served as input features, while ultimate load was the target variable. Four deep learning architectures ANN, CNN, RNN, and LSTM were implemented using TensorFlow/Keras and optimized using early stopping, dropout regularization, and uniform hyperparameters. Model performance was assessed using multiple statistical metrics including R², RMSE, MAE, VAF, NSE, and LMI. RNN and LSTM outperformed others, achieving R² values above 0.96 on the test set, with minimal residuals and stable loss convergence. Visualization tools such as regression plots, REC curves, ROC curves, and Taylor diagrams further validated predictive accuracy. The model’s interpretability was enhanced through Sensitivity and SHAP-based analysis, which identified ultimate load and initial load as the most influential predictors in determining structural behavior. The proposed approach offers a robust alternative to traditional analytical modeling by capturing nonlinear interdependencies and feature interactions within structural systems. The study demonstrated that deep learning, particularly recurrent architectures, can provide accurate and interpretable predictions of RC beam behavior, supporting efficient retrofitting decisions and structural safety assessments in real-world civil engineering applications.

Construction project scheduling involves complex trade-offs between time, cost, and quality (TCQ), often under conditions of uncertainty. This paper presents a novel approach using a fuzzy multi-objective particle swarm optimization (Fuzzy-MOPSO) algorithm to address the TCQ optimization problem in uncertain environments. By integrating fuzzy set theory with MOPSO, the model accommodates imprecise data and stakeholder preferences, allowing for a more realistic representation of construction project dynamics. Three objective functions are considered: minimizing project completion time (PCT), minimizing project construction cost (PCC), and maximizing project quality index (PQI). A case study involving a real-world construction project is employed to demonstrate the effectiveness of the proposed methodology. Twenty-six Pareto-optimal solutions were obtained and analyzed through trade-off plots and correlation analysis. The performance of the Fuzzy-MOPSO algorithm is benchmarked against other popular multi-objective optimization techniques, including NSGA-III, MODE, and MOTLBO. Results show that the proposed algorithm outperforms existing methods in convergence, diversity, and solution quality, achieving a more balanced TCQ optimization. The findings suggest that Fuzzy-MOPSO is a robust and efficient tool for construction managers seeking optimal schedules under uncertainty, contributing to better decision-making and resource allocation in complex project environments.

Modern seismic design codes prioritize structural resilience by accommodating inelastic deformation, especially in steel and reinforced concrete (RC) buildings. While this approach allows for ductile behaviour under strong earthquakes, it often results in structural damage. To reduce such damage, passive energy dissipation systems are increasingly utilized. This study introduces an innovative base isolation technique for RC frame structures, employing a hybrid U-shaped damper that integrates steel, shape memory alloy (SMA), and a rubber core. The U-shaped components, made from steel and SMA, exhibit elastoplastic and super elastic behaviour, respectively, enhancing energy absorption and self-centring capabilities. The rubber core, modelled using the hyper elastic Ogden formulation, contributes flexible isolation and nonlinear damping. Positioned at the base, the hybrid system enhances seismic performance by combining the isolating properties of rubber with the damping and recentring benefits of steel-SMA elements. Nonlinear time history analyses using four earthquake records confirmed that the hybrid isolator significantly reduces structural responses compared to systems without it.

This research examines the seismic performance of imperfections in multi-story steel-framed buildings. The structural behavior of 4, 6, and 8-story edifices was analyzed under seismic stress utilizing ETABS software and data from the Halabjah, Northridge, and Kobe earthquakes. Stories, roof displacement, and strain responses were among the variables studied by nonlinear dynamic and pushover analyses. Steel bracing configurations (central core, external corner, and exterior side) significantly enhanced earthquake resistance by augmenting lateral stiffness, reducing displacement, and controlling drift. The optimal design identified was center-core bracing, which decreased displacement by 52.32% and drift by up to 59.99%. The reduced manufacture of plastic hinges was an additional advantage of bracing, enhancing energy dissipation and structural resilience. The study emphasizes the impact of building height, earthquake intensity, and irregularity on seismic performance. The benefits were more apparent in shorter edifices; nevertheless, optimum bracing methods were essential for taller structures due to heightened flexibility and lateral force requirements. These findings underscore the significance of steel bracing in seismic design and retrofitting approaches, emphasizing its need in reinforced concrete structures in earthquake-prone regions.

Modern seismic design codes and standards (e.g., ASCE 7–16, Eurocode 8, NEHRP, RPA, 2024) prescribe the use of the pseudo-acceleration (PSA) spectrum to determine seismic loads and forces. These codes provide spectral acceleration values for various structural periods to ensure buildings are designed to withstand earthquake forces. PSA is derived from the displacement response of a system, offering a theoretical approximation of seismic forces rather than a direct measurement of ground acceleration. Instead, it is computed based on the relationship between response displacement and natural frequency. In contrast, relative acceleration more accurately represents the actual motion of a structure during an earthquake, making it a more physically intuitive measure of seismic forces. However, PSA approximations introduce significant errors. This study analyzes acceleration and displacement response spectra for single-degree-of-freedom systems using a carefully curated dataset from the Pacific Earthquake Engineering Research (PEER) ground motion database across various site periods. Pseudo-response spectra were derived from horizontal displacement spectral ordinates, and differences between actual and pseudo-response spectra were examined. Based on this analysis, correction formulas were developed to improve PSA accuracy. The key findings of this study can be summarized as that the discrepancy between real and pseudo-acceleration spectra is significantly larger for near-fault motions than for far-fault records. Therefore, a conversion model that effectively adjusts PSA spectra, accounting for both distance and magnitude is proposed. Finally, the model’s validity is confirmed within the parameter limits of the used database.

This study explores the application of machine learning (ML) models to optimize the adsorption-based treatment of distillery wastewater using sugarcane bagasse fly ash as an adsorbent. Four ML models—random forest, extreme gradient boosting (XGBoost), artificial neural network, and k-nearest neighbors were developed to predict the removal rates of chemical oxygen demand, biochemical oxygen demand, total suspended solids, and color. The models were trained and validated using data from batch adsorption experiments conducted under varying conditions of temperature, contact time, agitation speed, adsorbent dosage, and particle size. Descriptive statistics indicated significant variation in both the parameters and treatment efficiency, reflecting the experimental conditions. The XGBoost model consistently outperformed other models, achieving the highest coefficient of determination values (0.9910–0.9991) and lowest root mean squared error and mean absolute error values across all target variables. Feature importance analysis and sensitivity analysis revealed temperature as the most significant factor influencing pollutant removal, followed by contact time and agitation speed. Validation of unseen data further confirmed the XGBoost model’s superior predictive accuracy. The study demonstrates the potential of ML, particularly the XGBoost algorithm, in optimizing adsorption-based processes for treating distillery wastewater. These models can predict treatment outcomes under various operational conditions, potentially leading to more efficient treatment strategies. The research contributes to the growing application of ML in environmental remediation and wastewater management, offering a promising approach to enhance the efficiency and sustainability of distillery wastewater treatment.

Any nation's progress is determined by the expansion and innovation of its infrastructure. In order to guarantee both structural efficiency and aesthetic appeal, the growing demand for high-rise structures in metropolitan settings calls for creative approaches to structural design. In high-rise building, voronoi, which are renowned for their capacity to produce organic, irregular patterns, provide a fresh framework for allocating loads and maximising material consumption. The performance of Voronoi-patterned steel frames under a range of load circumstances, including seismic and wind, is thoroughly examined using SAP2000 and parametric modelling in accordance with IS: 1893:2016 and IS: 875, respectively. Comparisons with conventional grid systems demonstrate how Voronoi patterns may result in lower material use and higher efficiency. According to the results, Voronoi grids are very advantageous for maximising the structural performance and visual attractiveness of tall steel buildings. This study lays the groundwork for future investigations into intricate geometric patterns in architectural design, indicating that Voronoi-based designs may eventually result in more inventive and sustainable skyscraper building.

Conventional modelling methods have been criticised to lack adequate applicability in characterising complex, nonlinear, uncertain associations between diffraction parameters and the phase composition of the modified concrete system. The purpose of this work is to conduct a study in which a framework based on machine learning can identify the phase in the bentonite calcite modified concrete based on X-ray diffraction (XRD) data. Concrete mixtures of different proportions of Ground Granulated Blast Furnace Slag (GGBS), bentonite and calcite were prepared, cured and tested to evaluate their compressive strength and the ideal mixture was chosen upon which further analysis was carried out. Concerning this mix, the 2θ-intensity profiles obtained by the X-ray diffraction data were taken to train and test two kernel-based regression models: Support Vector Regression (SVR) and Gaussian Process Regression (GPR), employing 80:20 ratio as the splitting line in the whole X-ray diffraction data. SVR showed good general capability because it successfully applied to high-dimensional data learning by optimizing on a margin. GPR conversely involved using a probabilistic kernel in order to establish latent nonlinearities and uncertainty on the data. The predictive performance was high in both models, where SVR had R² of 0.978 and 0.977 on training and testing respectively, and GPR had a R² of 0.982 and 0.979 on training and testing respectively. Also, the values of mean squared error confirmed the superiority of GPR. The results confirm the promising nature of machine learning, specifically SVR and GPR, as scalable and fully competent alternatives to established techniques in phase quantification, which are reliable and competent in capturing the microstructure, and there upon offers sensible guidelines in optimizing the microstructure of novel cementitious composites.