Predicting Buckling Load of Slender Hollow Rods Using Machine Learning: Model Comparison and Input Sensitivity Analysis

Application of artificial neural networks for the prediction of traction performance parameters

Machine Learning Framework for Reliability-Based Design of Screw Piles

Machine learning–assisted performance prediction of carbon fiber Mortise-Tenon connections in civil structures

This paper proposes a performance prediction and design method for carbon fiber mortise-tenon structures, integrating finite element simulation, experiment, and machine learning, to address the issues of experiment dependency and low design efficiency of such structures in composite materials. Through parametric finite element analysis, a dataset of 120 samples containing six key geometric parameters—tenon angle (a), tenon height (h), tenon root length (m), tenon neck width (n), specimen width (w), and thickness (T)—was established, and the prediction performance of nine machine learning regression models was systematically compared. The GradientBoostingRegressor model, after multi-stage hyperparameter optimization, achieved the best prediction performance, with a nested cross-validation R 2 of 0.76 ± 0.11 and MAPE of 4.10% ± 0.54%. To verify model reliability, seven groups of T300 carbon fiber mortise-tenon specimens with different geometric configurations were fabricated for uniaxial tensile testing, and full-field strain was obtained using Digital Image Correlation (DIC) technology. The results indicate that the mean absolute percentage error between machine learning predictions and experimental measurements is 9.99%, while the error between finite element simulation and experiments is 4.72%, demonstrating acceptable engineering accuracy. SHAP (SHapley Additive exPlanations) analysis reveals that tenon angle (a) is the dominant feature (mean |SHAP| = 1.296), followed by neck width (n, 0.539) and specimen width (w, 0.406), providing physically interpretable insights into the data-driven predictions. DIC strain contours confirm that high-strain bands concentrate at the tenon-neck/mortise-shoulder transition zone, consistent with the failure mechanisms captured by the model. This study confirms the effectiveness of machine learning in rapid performance prediction and key parameter identification for carbon fiber mortise-tenon structures, providing a new approach for the intelligent design of composite material connections.

Axial load and strain prediction of square FRP–concrete–steel double-skin tubular columns using machine learning and nonlinear finite element analysis

This study investigates the axial behavior of square fibre reinforced polymer–concrete–steel double-skin tubular columns (hybrid DSTCs) by integrating experimental data, nonlinear finite element analysis (NLFEA), and machine learning (ML). A hybrid dataset comprising 174 specimens was developed, including 24 experimental results from the literature and 150 additional specimens generated through validated NLFEA models. The models demonstrated strong agreement with experimental results, achieving R² values of 0.98 for axial load and 0.92 for axial strain. The expanded dataset facilitated parametric studies on cross-sectional dimension, outer FRP tube thickness, inner steel tube thickness, compressive strength of concrete, and FRP modulus of elasticity. Four ML models, including Multi-linear Regression (MLR), Decision Tree (DT), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), were employed to predict axial load (Pu) and axial strain (εcu). Model performance was evaluated using R², RMSE, MAE, and MAPE. The DT model achieved the highest accuracy for axial load (Pu) prediction (R² = 0.97, MAPE = 2.97 %), while XGBoost provided the best overall performance for both axial load (R² = 0.97, MAPE = 3.60 %) and axial strain (R² = 0.83, MAPE = 9.63 %). Feature importance analysis revealed that axial load (Pu) is primarily governed by FRP stiffness, steel yield strength, and concrete strength, whereas ultimate axial strain (εcu) is dominated by steel tube geometry and strength, confirming that axial load and strain are controlled by distinct governing mechanisms. The findings highlight the potential of FEM–ML integration to develop robust predictive tools for square hybrid DSTCs.

Explainable machine learning surrogates for lateral pile response prediction in soft clay

Predictive Modelling and Optimization of Slope Stability Using Numerical Simulations and Machine Learning Techniques

The current study combines numerical modelling and machine learning to identify the stability of applications in nail reinforced slope study. PLAXIS LE was used to develop different slopes having different soil properties including various values for cohesion (5, 10, 15 kPa), angle of internal friction (20°, 25°, 30°), unit weight (17, 18, 19 N/m³), and slope angle (30°, 35°, 40°, 45°, 50°, 60°, 70°). Safety Factors (FOS) prediction models such as Random Forest (RF), Linear Regression (LR), and K-Nearest Neighbors (KNN) have been developed using the parameters included in the study. The Random Forest model has shown a superior performance among the other models with the lowest Mean Absolute Error (MAE: 0.053) and Mean Squared Error (MSE: 0.006), taking into consideration the highest value of R² (0.957) and Adjusted R² (0.951) to indicate a better predictive accuracy. With R² values of 0.903 and 0.920, respectively, Linear Regression and KNN also showed considerable strength of results. The results mentioned above show the bright future of machine learning models with Random Forest in predicting slope stability and contribute to refining nail reinforcement strategies. It shall also provide an input for developing cost-effective and robust slope rehabilitation measures in a geotechnically unfriendly environment.