Infrastructures

Machine learning (ML) has emerged as a transformative tool in geotechnical engineering, offering data-driven alternatives to traditional empirical and analytical approaches. This paper presents a critical narrative review of recent advances in theory-guided machine learning (TGML) applications in geotechnical engineering, including soil-property prediction, constitutive modelling, pile capacity prediction, seismic soil–structure interaction, foundation performance, and slope stability. The review organizes the literature according to the physical mechanisms through which theoretical knowledge is incorporated into machine-learning methods, including physics-guided features, hybrid constitutive models, physics-informed neural networks, physics-constrained surrogates, probabilistic hybrid frameworks, and multiscale models. Emerging trends such as hybrid physics-informed models, explainable artificial intelligence (AI) frameworks, and integration with numerical simulations, real-time monitoring and data assimilation, and digital-twin-enabled infrastructure assessment are also discussed. The paper identifies key research gaps and proposes future directions for improving model robustness, reliability, and practical implementation in engineering design and infrastructure management. This review provides a structured roadmap for advancing AI-driven methodologies in geotechnical engineering, and their integration into smart and resilient infrastructure systems, while highlighting the critical issues that must be addressed for broader adoption in practice.

Worker responses to automated safety warning (ASW) systems have been widely examined across construction work environments to improve worker safety. However, the evidence of worker response is fragmented because individual studies only examine different types of ASW systems, hazards, work environments, response measures, and review scopes. Therefore, this review synthesizes evidence on worker responses to ASW systems to guide the evaluation and design of future ASW systems. To support this synthesis, five databases, including Scopus and Web of Science, were searched for articles published between 2015 and 2026. The screening process involved two domain experts and three large language models (LLMs) and resulted in 40 included studies. The review found shorter reaction times, hazard avoidance, and improved situation awareness with some ASW systems. Limited comparisons found that auditory-only warnings produced slower responses than vibrotactile warnings under controlled conditions. The review also found that perceiving or reacting to a warning does not always lead to protective action. Therefore, ASW effectiveness should be evaluated across the full worker response chain rather than by warning perception or reaction time alone. It is suggested that future research should examine protective action, repeated exposure, and ASW performance with real workers under realistic work conditions.

Models using machine learning (ML) algorithms have been increasingly adopted for predicting bearing capacity of shallow foundations, but rarely, their superiority to ordinary regression is evaluated using a controlled benchmark. This paper presents a controlled synthetic benchmark where linear, quadratic, and third-order polynomial regression, random forest (RF) model, and artificial neural network (ANN) are compared with the noise-free Meyerhof–Vesić analytical formula for the prediction of a shear-based allowable bearing pressure. Since the Meyerhof–Vesić formula with constant soil properties is an affine function of foundation depth D and width B in each USCS category, there is no curvature in the base generator. Two scenarios are created: (i) smooth scenario with 4% multiplicative Gaussian noise and (ii) complex scenario with fourth-order interaction. All models are analyzed under both soil-specific and pooled training approaches with equal stratified partitions; per-soil R², RMSE, MAE, and bias are presented with 95% bootstrap confidence intervals, and a blocked holdout with D ∈ [1.00, 2.50] tests extrapolation ability. There is no universally best model. Linear per-soil baseline gives the largest average per-soil R² under both approaches (0.9686 smooth; 0.9340 complex), equaling or surpassing analytical formula (0.9675; 0.9302), poly33 (0.9650; 0.9308) and ANN (0.9661; 0.9294). In a noiseless setting, all four baselines (analytical, linear, quadratic, cubic) perfectly reconstruct the target (R² = 1.0000), while RF gets only 0.8901 (pooled) and 0.5902 (per-soil), which proves that there is no point using a cubic polynomial over a linear one. ANN-pooled is the best model for the most nonlinear soil (CH in the complex scenario; R² = 0.9166 vs. 0.8734 for poly33), indicating that the ML approach brings benefits only when there is some curvature in the data. RF is consistently the worst model (R² = 0.8789 smooth; 0.8465 complex) and totally fails on a blocked holdout (R² = −0.059), proving inability to extrapolate. Globally pooled R² obscures soil-specific problems. Such observations apply to the specific synthetic generator studied; further empirical validation in the field using data from SPT/CPT and a load test is essential before making any suggestions.