Lamine Dieng

Driver drowsiness detection systems are often evaluated using data partitions that may retain observations from the same individuals across training and testing, making their ability to generalise to unseen drivers difficult to assess. This study investigates behavioural eye-based drowsiness detection under strict subject-independent evaluation using 65,929 facial image frames from four drivers in the NTHU Driver Drowsiness Detection dataset. Eye Aspect Ratio (EAR), Rolling Mean EAR, EAR Delta and an Eye Closure Indicator were represented over temporal windows of 10, 20 and 30 frames and evaluated using Logistic Regression, Random Forest, Long Short-Term Memory and Bidirectional Long Short-Term Memory classifiers under Leave-One-Driver-Out validation. Logistic Regression achieved the highest mean accuracy, whereas Random Forest at 30 frames produced the highest drowsy-class F1-score and Area under the Precision–Recall Curve. Controlled baseline analysis showed that temporal context contributed more substantially to performance than the additional engineered EAR features, while recurrent architectures did not consistently outperform the classical classifiers. Performance also varied across unseen drivers, and the modest absolute accuracy cautions against interpreting the results as evidence of deployment-ready detection. The study demonstrates the importance of separating driver identities during evaluation and provides a reproducible framework for examining temporal behavioural representations under unseen-driver conditions.

Welding processes generate considerable thermal gradients that lead to microstructural transformations and residual stresses, thereby affecting the integrity and fatigue performance of welded steel structures used in civil engineering and offshore applications. This study presents a coupled thermo-metallurgical-mechanical finite element model for TIG welding of S355 structural steel, aimed at quantifying the contribution of solid-state phase transformations to residual stresses. The model was validated experimentally in two stages. First, the predicted thermal cycles were compared with thermocouple measurements at six locations, showing a maximum error of less than 7% on the peak temperature and good agreement during cooling. Second, the predicted phase distribution was compared with scanning electron microscopy (SEM) observations, confirming bainite as the dominant phase in both the fusion zone and the heat-affected zone. Regarding residual stresses, transformation-induced plasticity is found to govern the relaxation of the longitudinal residual stresses, mainly through the strain associated with the phase fraction rate, whereas the volumetric strain alone has a limited effect. The evolution of the effective yield stress during cooling is also found to reduce the residual stresses, although to a lesser extent. The transverse residual stresses remain lower than the longitudinal ones, confirming that the highest stresses develop along the welding direction. These results provide practical insight for structural steel designers and fabricators seeking to predict and control residual stresses in welded assemblies.

Ultrasonic impact treatment (UIT) is widely employed as a post-weld treatment to enhance the fatigue performance of welded joints through the introduction of surface plastic deformation and compressive residual stresses. While its beneficial effect on fatigue life when applied at an earlier stage is well established, the influence of UIT on aged structures remains controversial in the literature. This study investigates the effect of UIT on the corrosion performance of S355 steel welded T-joints after accelerated corrosion-induced aging. As-welded (AW) and UIT-treated T-joints were subjected to salt spray exposure, followed by detailed microstructural and surface analyses to assess corrosion morphology and damage evolution. The results show that UIT induces significant surface plastic deformation and microstructural refinement in the weld toe region without promoting preferential corrosion aging or accelerated degradation. The aging behavior of UIT-treated joints, following accelerated environmental exposure, is comparable to that of the AW condition, with corrosion rates decreasing from 3.27 and 3.28 mm/year at 42 days to 1.32 and 1.26 mm/year at 126 days for AW and UIT specimens, respectively. These results indicate that the compressive residual stresses and surface modifications introduced by UIT do not adversely affect material durability. These findings clarify the role of UIT under such exposure conditions and demonstrate that UIT can be applied as a post-weld treatment to improve fatigue properties without compromising the long-term performance of structural steel welded joints.

This paper presents a numerical approach based on a simplified, scaled-down finite element (FE) model to predict the dynamic torsional behavior of a 24 kV dynamic power cable (DPC), since torsion-induced FE models are underrepresented in the current literature. This is the first paper to develop an FE model for this particular type of cable (i.e., without metallic armoring), and it is an innovative, new design presented by Nordiske Kabel og Traadfabriker (NKT). It produces unique torsion and traction data which are used to validate and calibrate the numerical model, respectively. The data can be used by cable designers or engineers to validate, compare, or produce their own FE models. The study primarily focuses on estimating the torsional stiffness of the cable, a key mechanical property during early design stages. The FE model is developed using ABAQUS 2025 and calibrated through experimental torsion and traction tests performed at Université Gustave Eiffel in Bouguenais, France. The numerical domain is reduced to a representative section of the cable, enabling efficient modeling while preserving the essential mechanical interactions. To validate the model, simulation results are compared with both experimental data and an analytical axial stress calculation. The results show strong agreement across all approaches, demonstrating the model’s reliability in capturing the linear torsional response of the DPC. This work confirms that a simplified FE model can serve as an effective predictive tool for torsional stiffness in the initial phases of offshore cable design.

Recent advances in vibration-based pavement assessment have enabled the low-cost monitoring of road conditions using inertial sensors and machine learning models. However, most studies focus on isolated tasks, such as roughness classification, without integrating statistical validation, anomaly detection, or maintenance prioritization. This study presents a unified framework for road roughness severity classification and predictive maintenance using multi-axis accelerometer data collected from urban road networks in Pretoria, South Africa. The proposed pipeline integrates ISO-referenced labeling, ensemble and deep classifiers (Random Forest, XGBoost, MLP, and 1D-CNN), McNemar’s test for model agreement validation, feature importance interpretation, and GIS-based anomaly mapping. Stratified cross-validation and hyperparameter tuning ensured robust generalization, with accuracies exceeding 99%. Statistical outlier detection enabled the early identification of deteriorated segments, supporting proactive maintenance planning. The results confirm that vertical acceleration (accel_z) is the most discriminative signal for roughness severity, validating the feasibility of lightweight single-axis sensing. The study concludes that combining supervised learning with statistical anomaly detection can provide an intelligent, scalable, and cost-effective foundation for municipal pavement management systems. The modular design further supports integration with Internet-of-Things (IoT) telematics platforms for near-real-time road condition monitoring and sustainable transport asset management.

Transportation networks are critical lifelines in national infrastructure but are increasingly exposed to risks arising from climate variability, cyber threats, aging assets, and limited resources. This paper presents a scoping review of 58 peer-reviewed studies published between 2015 and 2025 that examine the role of Artificial Intelligence (AI) in strengthening infrastructure resilience, with transportation systems adopted as the strategic case. The review classifies applications along five dimensions: technological approach, infrastructure sector, transportation linkage, resilience/security aspect, and key research gaps. Findings show that AI, machine learning (ML), and the Internet of Things (IoT) dominate current applications, particularly in predictive maintenance, intelligent monitoring, early-warning systems, and optimization. These applications extend beyond transport to energy, water, and agri-food systems that indirectly sustain transport resilience. Persistent challenges include affordability, data scarcity, infrastructural limitations, and limited real-world validation, especially in Sub-Saharan African contexts. The paper synthesizes cross-sector pathways through which AI enhances transport resilience and outlines practical implications for policymakers and practitioners. A targeted research agenda is also proposed to address methodological gaps, enhance deployment in resource-constrained settings, and promote hybrid and explainable AI for trust and scalability.

Driver fatigue continues to be a leading contributor to global road accidents, posing significant risks particularly in long-distance, night-time, and commercial driving contexts. Recent advances in artificial intelligence (AI), sensor fusion, and signal processing have led to the development of automated fatigue detection systems capable of providing early warnings to prevent collisions. However, the field remains fragmented in terms of data modalities, modeling techniques, evaluation standards, and deployment readiness. This systematic review synthesizes findings from 78 peer-reviewed studies published between 2014 and 2025, critically analyzing the use of behavioral, physiological, vehicular, and multimodal approaches for fatigue detection. The review identifies a clear methodological shift from handcrafted features and traditional machine learning toward deep learning models, particularly CNNs and hybrid CNN-LSTM frameworks, enhancing performance in visual and EEG-based fatigue detection. Despite high accuracies reported under controlled conditions (often exceeding 95%), only a minority of models were validated in real-world settings or evaluated for real-time, in-vehicle deployability. Explainability and personalization, essential for trust and usability in human-centered AI, were substantially underrepresented which featured in only 19.2% and 12.8% of the studies respectively. Moreover, the widespread use of small sample sizes, unimodal data, and non-standardized metrics impedes reproducibility and comparative analysis. Nonetheless, emerging trends such as federated learning, edge deployment, and privacy-preserving AI indicate a growing emphasis on ethical, scalable, and context-aware systems. This review concludes by offering a roadmap for future research: emphasizing multimodal integration, standardized evaluation, on-road validation, and transparent model architectures, to bridge the gap between experimental prototypes and real-world fatigue monitoring systems.

Unexpected machine failures in industrial environments lead to high maintenance costs, unplanned downtime, and safety risks. This study proposes a proactive predictive model using a hybrid of eXtreme Gradient Boosting (XGBoost) and Neural Networks (NN) to forecast machine failures. A synthetic dataset capturing recent breakdown history and time since last failure was used to simulate industrial scenarios. To address class imbalance, SMOTE and class weighting were applied, alongside a focal loss function to emphasize difficult-to-classify failures. The XGBoost model was tuned via GridSearchCV, while the NN model utilized ReLU-activated hidden layers with dropout. Evaluation using stratified 5-fold cross-validation showed that the NN achieved an F1-score of 0.7199 and a recall of 0.9545 for the minority class. XGBoost attained a higher PR AUC of 0.7126 and a more balanced precision–recall trade-off. Sample predictions demonstrated strong recall (100%) for failures, but also a high false positive rate, with most prediction probabilities clustered between 0.50–0.55. Additional benchmarking against Logistic Regression, Random Forest, and SVM further confirmed the superiority of the proposed hybrid model. Model interpretability was enhanced using SHAP and LIME, confirming that recent breakdowns and time since last failure were key predictors. While the model effectively detects failures, further improvements in feature engineering and threshold tuning are recommended to reduce false alarms and boost decision confidence.

Aluminum Conductors Steel-Reinforced (ACSR) conductors are typically used in overhead transmission lines. Corrosion is an important degradation mechanisms that might affect the lifetime of this essential electricity network component. Considering the complexity of conductors, it is difficult to predict the damage of these conductors in corrosive environments. The objective of this paper is to evaluate the effect of grease and conductor geometry on the mechanical properties of aluminum strand composing the envelope of ASCR conductors. Thus, ACSR wires and strands have been evaluated in corrosion by the mean of accelerated corrosion tests. Tensile, fatigue and torsion test results are presented to examine the effect of corrosion on aluminum strands. The influence of corrosion on mechanical characteristics is established by a decrease in ductility, maximum elongation and tensile strength for the longest exposition (336 days). This significant reduction in the internal layer of ungreased wires confirms the importance of the galvanic corrosion mechanism of aluminum wires. This evolution concerns only aluminum wires of non-greased conductors, confirming the crucial role of grease as protection against corrosion.