Tools for Automated Structural Health Monitoring Using Deep Learning and Computer Vision Techniques

Structural Health Monitoring (SHM) is a critical task in the management of buildings, bridges, and civil infrastructures. With the evolution of Artificial Intelligence (AI), SHM through AI is gaining popularity from past few years. Therefore, this work aims in providing SHM solution for wall defects through computer vision techniques. The work presents the development of mobile application to fully automate SHM procedure for inspection of building, finding defects, and its analysis. For crack detection purpose, MobileNetV2 as an algorithm is incorporated to determine two anomaly condition of whether the wall has crack or not. Mobilenet showed good result with an accuracy of 95%. The defect detected image is then segmented with the U-Net algorithm that resulted a Dice Similarity of 0.88, to determine the area of the damaged portion. The work also comprises real-time SHM of the walls using You Only Look Once version 5 (YOLOv5) for various defects such as crack, spalling, and seepage. To integrate these algorithms into mobile application, the models are converted into TensorFlow Lite (TFLite) format with the aim to reduce the size of the model and computational power. The proposed software has potential to significantly improve efficiency and effectiveness of SHM, providing a user-friendly and accessible solution for defect detection and analysis on mobile devices.