Regions within the seismic zone of the pacific volcanic quadrant are highly susceptible to frequent seismic activity, posing significant threats to both human life and infrastructure. This threat is particularly critical for the high-tech industries, which are holding a substantial global market share and highly vulnerable to earthquake-induced damages. The characteristics of earthquake ground motion play a crucial role in the effectiveness of structural control systems, especially for near-fault earthquakes, where low-frequency resonance can cause significant displacement in the isolation layer. To address these challenges, this study focuses on advancement of smart structural systems through the development of an intelligent semi-active isolation system using an innovative control strategy based on artificial intelligence. Specially, the study employs the deep deterministic policy gradient (DDPG) algorithm to enhance the control of seismic isolations systems. The control modules are tailored for distinct characteristics of both near-fault and far-field seismic events, ensuring effective mitigation of seismic impacts. The results demonstrate significant enhancements in structural performance using the DDPG-based framework, showing a reduction of approximately 68.85% in isolation layer displacement during near-fault earthquakes and a 57.65% reduction in superstructure acceleration during far-field earthquakes. These outcomes affirm the effectiveness of the DDPG-driven intelligent control strategies for advancing the resilience and service life of critical infrastructures in seismically active regions, contributing to the broader fields of structural control and prognosis.

The primary objective of this study is to predict the pullout capacity of belled piles by developing and evaluating various hybrid modeling techniques, including Evolution Strategy (ES), Moth Flame Optimizer (MFO), Grasshopper Optimization Algorithm (GOA), and League Championship Algorithm (LCA). Each hybrid model combines an artificial neural network (ANN) with an optimization algorithm, trained using a hybrid learning approach that incorporates back-propagation and least squares estimation, implemented in MATLAB. A total of 36 samples were used, with 25 designated for training and 11 for testing. The performance of each model was assessed using statistical metrics, namely the coefficient of determination (R²) and root mean square error (RMSE). Among the models, MFO-ANN demonstrated the highest predictive accuracy, followed by LCA-ANN, ES-ANN, and GOA-ANN, respectively. The results confirm the robustness and reliability of the ANN-based hybrid models in estimating pullout capacity.

Crack detection is vital for maintenance of civil structures. Recently, deep learning-based semantic segmentation models have shown promise in accurately identifying cracks. However, these methods require laborious manual data labeling. To address this, an unsupervised learning-based crack segmentation method was proposed, using a self-supervised Vision Transformer (ViT) as a backbone network to learn crack patterns from unlabeled images. A diverse crack image dataset with various crack types and backgrounds was used to train the network without time-consuming labeling, and to test the model after constructing ground-truths. The model was optimized with unsupervised contrastive loss function parameters, and probability thresholding was applied to enhance detectability by eliminating low confidence pixels, reducing false positives. On 1,399 test images, the unsupervised model achieved a mean F1-score of 75.02% and a mean Intersection over Union (mIoU) of 63.01%, with mIoU improving to 66.14% after thresholding, which shows great detection performance of unsupervised model. The model's application to high-resolution real crack images using a sliding window technique further demonstrated its suitability for field use, offering an efficient solution for real-time structural monitoring. These findings highlight the potential of unsupervised deep learning for crack detection, significantly reducing the need for manual labeling while delivering strong performance.

In recent years, artificial intelligence (AI) has been extensively deployed in different fields, especially in engineering. The ability of AI algorithms has been indicated in many research papers by providing accurate results compared to numerical and experimental approaches. Integrating Artificial Intelligence (AI) with Building Information Modeling (BIM) has opened new possibilities for enhancing structural design processes. BIM provides rich parametric data, while AI enables intelligent interpretation and prediction. This study develops a hybrid Artificial Neural Network-Genetic Algorithm (ANN-GA) model to predict the structural load performance of Reinforced Concrete (RC) buildings using parameters extracted from BIM models. Primary inputs include geometric properties, material strengths, reinforcement ratios, and layout configurations; outputs include structural indicators such as ultimate load capacity and deflection under standard loads. The model is trained and validated using empirical data from literature. The Artificial Neural Network (ANN) captures complex nonlinear input-output relationships, while the Genetic Algorithm optimizes network parameters like hidden layers, neurons, learning rate, and weights. Performance is evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R². This study offers a fast, simulation-free structural assessment method leveraging BIM and AI for early-stage decision-making. The ANN-GA performed strongly: RMSE 2.11 ± 0.36 kN and R² 0.9804 ± 0.0052. Compared to baselines, it outperformed linear regression (RMSE: 23.13 kN, R²: -1.13) and a fixed-architecture ANN (RMSE: 43.64 kN, R²: -8.01). Sensitivity analysis showed span length and reinforcement ratio were most influential (32% and 27%). In case-based validations, prediction errors stayed within ±5%, confirming practical applicability. The model was integrated into a BIM environment for real-time structural feedback using normalized inputs via Revit APIs or IFC-based workflows.