Asian Journal of Civil Engineering最新文献

This study investigates the synergistic use of Waste Glass Powder (WGP) as a partial replacement for fine aggregate and Recycled Coarse Aggregate (RCA) as a substitute for natural coarse aggregate in M30 grade concrete, with combined replacement levels ranging from 0 to 50% by weight. The objective is to enhance concrete sustainability by utilizing municipal glass waste and construction and demolition debris. A comprehensive experimental program evaluated fresh properties (slump, density), mechanical strengths (compressive, split tensile, flexural), and durability characteristics (chloride ion penetration, water absorption, shrinkage, and water permeability), alongside microstructural observations using Scanning Electron Microscopy (SEM). The results demonstrate that a 30% combined replacement yields optimal performance: compressive strength of 42.5 MPa, tensile strength of 5.5 MPa, and flexural strength of 7.35 MPa at 90 days—each surpassing or matching the control mix. Durability assessments confirmed acceptable chloride resistance, with chloride-ion penetration depths of 11.7 mm and RCPT values remaining low at approximately 30.1 Coulombs. SEM analysis revealed a densified matrix and improved interfacial transition zone (ITZ) bonding at this replacement level, driven by WGP’s pozzolanic reactivity. These findings establish that incorporating up to 30% WGP and RCA in concrete offers a technically viable and environmentally responsible solution for sustainable construction applications without compromising mechanical or durability performance.

Concrete with Ultra-High Performance (UHPC) is a next-generation cement-based material known for its ultra-high compressive strength, enhanced ductility, and extreme durability. By eliminating coarse aggregates and incorporating optimized fine particle packing, UHPC achieves strengths above 150 MPa. The synergy of low water-to-binder ratio, high-reactivity pozzolans like silica fume, and steel fiber reinforcement contributes to its exceptional mechanical and durability performance. UHPC offers self-consolidation, reduced permeability, and superior resistance to aggressive environments, making it ideal for critical infrastructure. Its ability to outperform traditional concrete in both structural and service life applications is transforming the future of construction. This study examines how well two sophisticated machine learning boosting models, Gradient Boosting algorithm (GB) and Extreme Gradient Boosting algorithm (XGBoost) performs while predicting the ultra-high performance concrete’s compressive strength. To evaluate this potential, a dataset consisting of 110 experimental results, compiled from existing literature, was employed to test and train the models. The GB model achieved a R² of 0.960 and Normalized Mean Square Error, NMSE of 0.041 on the train data and R² of 0.727 and NMSE of 0.452 on test data. The XGBoost model achieved a R² of 0.961 and NMSE of 0.039 on the train data and R² of 0.840 and NMSE of 0.160 on test data. These results demonstrate that XGBoost and GB, both have excellent predictive accuracy in modeling UHPC compressive strength and shown a significant improvement over the existing literature, Omar R. Abuodeh (2020). Overall, this research confirms that leveraging GB and XGBoost significantly enhances model performance and offers valuable insights into the compressive strength behavior of UHPC.

Intelligent prediction of the flexural strength of reinforced concrete beams remains a challenging task due to variation in accuracy and precision of machine learning algorithms. It also becomes harder for users to assess and select best model, since several machine learning models has their own limitation and advantages with different circumstances. To overcome this problem this study aims to evaluate and compare the performance of various machine learning algorithms in terms of their accuracy and efficiency. Linear Regression, Decision Tree, Support Vector Machine, K-Nearest Neighbors, Random Forest, Adaptive Boosting, Gradient Boosting, and Extreme Gradient Boosting models were evaluated. Hyperparameters of each model were optimized using Grid Search cross validation with Mean Squared Error used as the Performance Index. The predictive efficiency of each model was rigorously evaluated through four distinct statistical performance measures. The results of the analysis show that the Linear Regression model encountered issues of underfitting, while the Decision Tree model demonstrated signs of overfitting and constrained generalization capabilities. Additionally, the Adaptive Boosting model exhibited a minor overfitting concern. Moreover, the Support Vector Machine, Random Forest, and Adaptive Boosting models yielded comparable levels of accuracy. In contrast, the proposed Extreme Gradient Boosting model achieved superior performance characterized by exceptional generalization capabilities, as evidenced by its minimal mean absolute error of 2.08 kN-m, a root mean squared error of 3.09 kN-m, and the highest coefficient of determination of 98.50% on the test data.

This study uses full-scale Comprehensive experimental, analytical, statistical, and multiscale AI methods explicitly detailed to examine φ-factors for high-strength concrete (HSC) columns reinforced with externally bonded ternary hybrid composites. The research involves Seventy-five HSC columns precisely defined (120 × 120 × 600 mm) axial/eccentric loading explicitly, each reinforced with eight 8 mm longitudinal bars and eight 6 mm ties, tested under axial compression. The force–displacement curves showed different patterns depending on the specimen group: Specimens S1-S25 had higher strengths (up to 900 kN) with low displacement (up to 3 mm); Specimens S26-S50 had intermediate strengths (up to 850 kN) with moderate displacement (up to 7 mm); and Specimens S51-S75 had lower strengths (500–750 kN) with larger displacement (up to 9 mm). The mixes were created by systematically varying the amounts of cement, water, aggregates, fly ash, silica fume, steel fibers, and polymers to evaluate their effects on structural performance. Statistical analysis found significant relationships between material parameters and mechanical responses, with Fly ash and steel fibers explicitly confirmed strongest correlation (r = 0.3544), the strongest positive correlation to force capacity. A physics-informed neural network (PINN) was used to combine experimental data with physical mechanics, providing accurate predictions of φ-factors (R² = 0.9874). Combining first principles with data-driven learning via PINN outperformed traditional model-based methods. A reliability analysis showed that the optimal combination of steel fibers and polymers can significantly improve the strength and ductility of HSC-C, justifying a larger φ-factor than current codes. This has important implications for the performance-based design of advanced concrete systems. Additionally, the study offers a robust tool for determining the φ-factor in complex composite members, promoting more rational, economical, and reliable designs of high-strength concrete structures.

Pine cone trash is extremely combustible, increasing forest fire risk in pine-rich areas and requiring innovative, sustainable management. This study evaluates replacing Pine Cone Scale (PCS) waste in lightweight concrete with natural coarse aggregate. The study employs a hybrid framework that utilizes machine learning (ML) techniques such as ANN, Random Forest, XGBoost, and the Stacking ensemble in conjunction with Response Surface Methodology. This method is used to model and improve the compressive strength, slump, and hardened density of the concrete. Bootstrapped datasets help ML models overcome sample limitations. Use a central composite design (CCD) to make twenty experimental concrete mixes that are based on PCS. XGBoost and stacking models were more accurate, with R² values up to 0.9999 and mean relative errors less than 0.1%. RSM performed rather well, with R² values over 0.95. The optimized PCS-I and PCS-II concrete mixes meet ASTM C 330 and IS 456:2000 standards, respectively. Their decreased density and compressive strength of more than 17 MPa indicated that PCS may be a lightweight aggregate. This work shows the synergistic possibilities of RSM-ML integration for sustainable mix design and suggests a dual-benefit approach to turn waste biomass into an eco-efficient building material.

Predicting the compressive strength of concrete in built structures is crucial for assessing structural safety, addressing damage, adapting to regulatory changes, determining repair needs, and evaluating the strength of components for reuse in sustainability. Non-destructive testing (NDT) has been effectively used for decades, initially through empirical equations and, more recently, through machine learning (ML) models to predict the compressive strength of concrete in existing structures. Both traditional empirical equations and ML models have demonstrated promising results, but each has inherent limitations. This study introduces Empirical Physics-Informed Neural Networks (EMP-PINNs), which integrate empirical equations with artificial neural networks (ANNs) to combine the advantages of both methods, presenting an innovative fitting algorithm. First, a comprehensive dataset was generated using Generative Adversarial Networks (GANs) to enhance the training of machine learning models. This dataset includes NDT values such as rebound number and ultrasonic pulse velocity, with concrete strength as the target variable. Next, a regression equation was developed to serve as a physics-informed constraint within the EMP-PINN model. Finally, the empirical equation was combined with ANN to form the EMP-PINN model. The developed EMP-PINN model demonstrated strong convergence behavior, with a high correlation coefficient (R² = 0.92) and low prediction errors (MAE = 2.27, MSE = 9.29). Compared to traditional ANN, EMP-PINN showed superior generalization, particularly for extreme or unseen values. When applied to a real-world structure, the model achieved an average absolute prediction error of approximately 5.6%, validating its practical reliability. This Empirical Physics-Informed Neural Network paves the way for the broader application of physics-informed neural networks in engineering domains, where governing systems are often based on empirical equations rather than purely physical ordinary differential equations.

PPHCS are widely utilized around the world as essential components in horizontal structural systems, playing a key role in contemporary precast concrete construction practices. In this study, thirteen full-scale prestressed hollow-core slabs with varying shear span-to-depth (a/d) ratios were tested to failure to evaluate their ultimate load-carrying capacity. An artificial neural network (ANN) model was developed to predict this capacity, utilizing a comprehensive dataset that combines detailed experimental results from tested slabs with valuable data sourced from existing literature. A total of 431 PPHC slabs were used to train and validate the ANN model, utilizing key input variables including slab length, overall depth, effective prestressing force, area of prestressing steel, Prestressing load eccentricity, shear span-to-depth ratio, concrete compressive strength, and web width. The model’s predictive performance was thoroughly evaluated using multiple statistical metrics, delivering a strong coefficient of determination (R² = 0.948), along with a low root mean square error (RMSE = 43.433) and mean absolute error (MAE = 31.03). The experimental results were evaluated against the predictions from both the ACI code and the ANN model. The analysis indicated that the proposed ANN model provides a more accurate estimate of the ultimate load capacity of PPHCS compared to the ACI code.

The study explores the potential of incorporating shredded cigarette butt fibers into Stone Mastic Asphalt mixtures to improve performance. The study addresses the dual challenge of seeking durable road construction materials and managing cigarette butt waste. By assessing three CBF dosages in SMA mixes, the research introduces a sustainable solution. Key findings indicate that CBF inclusion enhances drain-down resistance, Marshall properties, and resistance to rutting and moisture-induced damage at high temperatures. Specifically, a 0.05% CBF dosage optimized elastic and dynamic modulus values, suggesting enhanced stiffness and resilience, while a 0.03% dosage improved fatigue performance. These results suggest that incorporating shredded CBFs can improve the structural integrity and durability of SMA mixtures. Furthermore, machine learning models, including XGBoost, Random Forest, and Linear Regression, were used to predict key mechanical performance parameters, with XGBoost and Random Forest demonstrating high accuracy (R² values up to 0.99, RMSE as low as 0.02, and minimal mean absolute error), thereby corroborating the experimental findings. This study offers an early experimental evaluation of shredded CBFs in SMA and uniquely uses machine learning to develop a predictive framework for performance evaluation. The approach supports the sustainable reuse of hazardous urban waste in road infrastructure, combining material innovation with environmental responsibility.

This article focuses on an environmentally friendly high-performance concrete (HPC) that efficiently uses cement and mineral additions. Using a three-factor experimental design, the impacts of substituting cement (PC) with marble dust or powder (MP) and granulated ground blast furnace slag (GGBFS) on the workability and hardened qualities of HPC are analyzed. 15 different mixes were synthesized. Data modeling was performed out with the help of the statistical program Design-Expert 13. Analysis proved successful, allowing the identification of mathematical models representative of the experimental findings. With R² values ranging from 0.76 to 0.95, models performed well in analysis of variance (ANOVA) for predicting all HPC characteristics examined. Findings show that GGBFS significantly enhances the workability. However, mixtures M6 (0.25PC + 0.75MP), M10 (0.5PC + 0.5MP) and M13 (0.75PC + 0.25MP) show maximum compressive strength (CS) after a week. After 28 days, mixes with the highest CS were M12 (0.5PC + 0.5GGBFS), M14 (0.75PC + 0.25GGBFS) and M15 (100%PC). The porosity (P) is also reduced in combinations M9 (0.25PC + 0.75GGBFS), M12, M14 and M15. The formulation M14 is ideal since it gives a better balance of attributes studied; M4 and reference M15 are almost identical in terms of characteristics, allowing for a reduced quantity of cement to be employed. Predicted values were validated experimentally and through optimization with an error margin of less than 2%, thereby proving that GGBFS and MP are feasible environmentally friendly building materials. Obtained findings provide important insights into the effective use of GGBFS and MP in cost-effective and eco-friendly HPC design.

This study examines cement production optimisation through Machine Learning models (ML). In this analysis, input variables are considered as the manipulating variables, such as feed and sepax power, while folaphone and elevator power are considered as the controlled variables. SHapley Additive exPlanations (SHAP) analysis and Local Interpretable Model-agnostic Explanations (LIME) for linear and tree-based models confirmed folaphone as the more significant outcome than elevator power. Models such as Regression Trees (RT), Ensemble Bagged Trees (EBT), Random Forest (RF), Coarse Tree (CT), Boosted Trees (BT), and Support Vector Regression (SVR) are trained on pre-processed data. Their performances are compared in terms of Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The BT and RT models yielded the highest accuracy for predicting folaphone complete data, with R² values of 0.9972 and 0.9963, respectively. Additionally, the RT model achieved the best accuracy in predicting elevator power with full data, yielding an R² value of 0.9877. This study identifies that folaphone is employed as a main controlled variable since fineness determination in the online context is challenging. The outcomes are a robust, data-driven optimisation framework that can be augmented with hybrid models for extension.