Machine Learning Framework for Reliability-Based Design of Screw Piles

Machine Learning–Aided Prediction of Shaft Resistance and Bearing Capacity Factors for Pile Foundation Design Using CPT Data

This study presents a data-driven framework for predicting the shaft resistance ( β ) and end bearing capacity ( N t ) factors of piles using cone penetration test (CPT) data. Traditional design methods often rely on empirical values that oversimplify soil–pile interaction by neglecting the influence of pile geometry. To address this, this research employs an evolutionary polynomial regression with a multiobjective genetic algorithm (EPR-MOGA) to develop design equations. A two-step methodology was used: first, robust models for shaft capacity ( Q s ) and end-bearing capacity ( Q t ) of drilled piles were developed from databases of 54 and 31 load tests, respectively. Validated using fivefold cross-validation, these base models demonstrated high predictive accuracy, achieving an average coefficient of determination of 0.92 for Q s and 0.97 for Q t . Explicit equations for the β and N t factors were then derived from these validated models. These equations inherently account for complex soil–pile interactions, showing increases with soil friction angle and reductions with pile slenderness ratio ( L / D ) at a diminishing rate. This eliminates the need for the additional caps on effective vertical stress required by conventional methods. Finally, the models were extended to driven piles by calibrating modification factors on an independent 36-test database, yielding values of 1.46 for Q s and 2.38 for Q t .

Probabilistic and Interpretable Machine Learning Framework for Predicting Pile Unit Base Resistance in Soft Soil

Accurate prediction of pile base resistance is essential for the safe and economical design of deep foundations, particularly in soft soils where load-transfer mechanisms are highly nonlinear and uncertain. This study develops a comparative, probabilistic, and interpretable machine learning framework for predicting pile unit base resistance using five input variables: applied load, settlement, effective pile length, axial stiffness, and SPT value. A Gaussian Process Regression model with an automatic relevance determination (ARD) Exponential kernel achieved the best performance, with RMSE = 262.11 kPa, R2 = 0.943 on an independent test set, and 95% prediction intervals with 96.46% coverage. Beyond record-level evaluation, a leave-one-pile-out validation (the first grouped validation applied to this database) showed harder generalization to entirely unseen piles, driven mainly by a per-pile level offset rather than shape mismatch (within-pile correlation = 0.975). A sequential next-stage scheme, calibrating this level from a pile’s early loading stages, then predicted its remaining segments with consistently strong agreement (Willmott’s d = 0.76–0.83), supporting practical extension of partial load tests. Interpretability was assessed using ARD, SHAP, permutation/ablation importance, and partial dependence/accumulated local effects analysis, identifying settlement as the dominant predictor. The framework combines accuracy, calibrated uncertainty, interpretability, and validated segment-level extrapolation for reliability-oriented pile assessment.

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

Bayesian-Optimized Ensemble Machine Learning for Predicting Settlement of Cohesionless Soil Under Loads

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

Predicting the critical buckling load of slender structural rods is essential for reliable and weight-efficient design of automotive steering and suspension linkages such as tie rods. This study evaluates the performance of four machine learning models such as artificial neural network (ANN), support vector regression (SVR), Gaussian process regression (GPR), and random forest (RF) in predicting the critical buckling load (Pcr) from geometric and material design parameters which are rod length (L), diameter (D), wall thickness (t), Young’s modulus (E), and initial geometric imperfection (δ₀). A dataset was generated using a parametric MATLAB code, and models were trained on an 80/20 train-test split with min-max normalized inputs. ANN and GPR achieved near-perfect predictive accuracy (R²=1.000), outperforming SVR and RF (R² = 0.92-0.96). Input sensitivity was assessed using permutation importance across all four models, complemented by Garson’s algorithm and the connection weight method. Rod length and diameter were consistently identified as the dominant parameters governing buckling resistance, jointly accounting for the majority of predictive importance; when length was held constant, diameter alone emerged as the leading parameter, followed by comparable contributions from wall thickness and Young’s modulus. Initial imperfection showed negligible influence according to all permutation-based methods, although the connection weight method disagreed sharply, illustrating a key limitation of weight-based sensitivity analysis relative to permutation-based approaches.

Machine Learning Prediction of the Ground Reaction Curve in Sand with the MATLAB GUI Platform

The proposed framework combined a curated database, neural network-based curve prediction, and hyperparameter optimization, providing a robust approach for evaluating the soil arching effect, providing a robust approach for evaluating the soil arching effect.