AI-Driven Structural Health Monitoring for Early Detection of Fatigue-Induced Failures in High-Rise Buildings

Abstract Structural Health Monitoring (SHM) plays an indispensable role in maintaining the safety and health of high-rise buildings, especially in early diagnoses of fatigue cracks, which are failure mechanisms that take time to emerge due to environmental and operational loads. The conventional SHM process involves monitoring, visual inspection, and conventional data acquisition using sensors, leading to failure mode maintenance based on failure detection instead of failure prevention. This paper proposes a new AI-based SHM approach using advanced machine learning classification for time series anomaly matrix analysis and convolutional neural networks to identify image defects. This paper uses transformer-based models for long-range structural response analysis using the wind turbine structural health monitoring dataset. The key reason for applying the approaches presented in the work is to extend the usage of the small pre-trained language models (LLMs). For image-based defect detection, advanced architectures, like YOLOv8 (You Only Look Once) and advanced CNN segmentation models, are used to identify the concrete cracks for the SDNET 2018 dataset. The best approach is then used to train and test the proposed system using an in-house optimised computational system for real-world applicability. The paper also presents a comparative analysis of the deep learning models, suggesting that up-to-date transformers work better than traditional SHM techniques for sensor anomaly detection. YOLO-based models are appropriate for the identification of defects in images. This brings major contributions to the research in utilising sensor data and image analysis combined with SHM and deep learning to achieve a systematic, predictive, and automated high-rise building monitoring technique.