Accurate soil classification is essential for pavement design; however, the traditional American Association of State Highway and Transportation Officials (AASHTO) classification system relies on extensive laboratory testing and subjective judgment. This study presents an artificial intelligence (AI) enhanced framework for AASHTO soil classification. A synthetic dataset of 349,015 samples was generated using parameter ranges for five AASHTO input variables to support model development. Four machine learning models were trained, analyzed, and compared where the random forest (RF) consistently achieved the highest accuracy of 100% among the four models in predicting AASHTO soil groups. Feature importance analysis indicates that percent passing the No. 200 sieve is the most influential factor, and under missing input scenarios. Additionally, the models remain reliable under partial input loss, though accuracy is most sensitive to the absence of percent passing the No. 200 sieve, dropping to 85.8%, while all other variables maintain accuracies of at least 93.1%. Prediction uncertainty using Monte Carlo simulations shows model performance within a 95% confidence interval. Overall, the proposed AI models can accurately and efficiently predict AASHTO soil groups using incomplete datasets for geotechnical engineering.

In data communication, limited communication resources often lead to measurement bias, which adversely affects subsequent system estimation if not effectively handled. This paper proposes a novel bias calibration algorithm under communication constraints to achieve accurate system states of the interested system. An output-based event-triggered scheme is first employed to alleviate transmission burden. Accounting for the limited-communication-induced measurement bias, a novel bias calibration algorithm following the Kalman filtering line is developed to restrain the effect of the measurement bias on system estimation, thereby achieving accurate system state estimates. Subsequently, the Field Programmable Gate Array (FPGA) implementation of the proposed algorithm is also realized with the hope of providing fast bias calibration in practical scenarios. A simulation about a numerical example and a practical example (for gyroscope’s angular velocity bias calibration) on MATLAB is provided to demonstrate the feasibility and effectiveness of the proposed algorithm.

Granular flow, such as hopper discharge and debris flows, involves complex multi-scale, multi-phase, and multi-physics coupling, posing significant challenges for numerical simulation. Over the past two decades, methods like the Discrete Element Method (DEM), Smoothed Particle Hydrodynamics (SPH), and Depth-Averaging Method (DAM), have been developed to address these problems. However, their applicability across different scales remains unclear due to differences in physical assumptions and numerical algorithms. Therefore, a comprehensive evaluation is critically needed. This study selects three typical methods (DEM, SPH, and DAM) to examine their convergence behavior, boundary condition implementation, and limitations in physical and numerical modeling. We numerically studied three extreme deformation flow cases with the three chosen methods. These cases include granular column collapse at the particle scale, flow-structure interaction at the laboratory scale, and reconstruction of the 2015 Shenzhen Guangming landslide at the field scale. By comparing the granular flow dynamics, deposition morphology, and structure interactions, and also the simulation accuracy and computational efficiency, we show the applicability of the three models across different scales. Further, we provide practical guidance for model selection in large-deformation flow problems in a granular system of different scales.

Accurate prediction of concrete compressive strength is fundamental for optimizing mix designs, improving material utilization, and ensuring structural safety in modern construction. Traditional empirical methods often fail to capture the non-linear relationships among concrete constituents, especially with the growing use of supplementary cementitious materials and recycled aggregates. This study presents an integrated machine learning framework for concrete strength prediction, combining advanced regression models—namely CatBoost—with metaheuristic optimization algorithms, with a particular focus on the Somersaulting Spider Optimizer (SSO). A comprehensive dataset encompassing diverse mix proportions and material types was used to evaluate baseline machine learning models, including CatBoost, XGBoost, ExtraTrees, and RandomForest. Among these, CatBoost demonstrated superior accuracy across multiple performance metrics. To further enhance predictive capability, several bio-inspired optimizers were employed for hyperparameter tuning. The SSO-CatBoost hybrid achieved the lowest mean squared error and highest correlation coefficients, outperforming other metaheuristic approaches such as Genetic Algorithm, Particle Swarm Optimization, and Grey Wolf Optimizer. Statistical significance was established through Analysis of Variance and Wilcoxon signed-rank testing, confirming the robustness of the optimized models. The proposed methodology not only delivers improved predictive performance but also offers a transparent framework for mix design optimization, supporting data-driven decision making in sustainable and resilient infrastructure development.

Reliable detection of traffic signs and lights (TSLs) at long range and under varying illumination is essential for improving the perception and safety of autonomous driving systems (ADS). Traditional object detection models often exhibit significant performance degradation in real-world environments characterized by high dynamic range and complex lighting conditions. To overcome these limitations, this research presents FED-YOLOv10s, an improved and lightweight object detection framework based on You Only look Once v10 (YOLOv10). The proposed model integrates a C2f-Faster block derived from FasterNet to reduce parameters and floating-point operations, an Efficient Multiscale Attention (EMA) mechanism to improve TSL-invariant feature extraction, and a deformable Convolution Networks v4 (DCNv4) module to enhance multiscale spatial adaptability. Experimental findings demonstrate that the proposed architecture achieves an optimal balance between computational efficiency and detection accuracy, attaining an F1-score of 91.8%, and mAP@0.5 of 95.1%, while reducing parameters to 8.13 million. Comparative analyses across multiple traffic sign detection benchmarks demonstrate that FED-YOLOv10s outperforms state-of-the-art models in precision, recall, and mAP. These results highlight FED-YOLOv10s as a robust, efficient, and deployable solution for intelligent traffic perception in ADS.

Multivariate anomaly detection plays a critical role in maintaining the stable operation of information systems. However, in existing research, multivariate data are often influenced by various factors during the data collection process, resulting in temporal misalignment or displacement. Due to these factors, the node representations carry substantial noise, which reduces the adaptability of the multivariate coupled network structure and subsequently degrades anomaly detection performance. Accordingly, this study proposes a novel multivariate anomaly detection model grounded in graph structure learning. Firstly, a recommendation strategy is employed to identify strongly coupled variable pairs, which are then used to construct a recommendation-driven multivariate coupling network. Secondly, a multi-channel graph encoding layer is used to dynamically optimize the structural properties of the multivariate coupling network, while a multi-head attention mechanism enhances the spatial characteristics of the multivariate data. Finally, unsupervised anomaly detection is conducted using a dynamic threshold selection algorithm. Experimental results demonstrate that effectively integrating the structural and spatial features of multivariate data significantly mitigates anomalies caused by temporal dependency misalignment.