An Enhanced Structural Damage Identification Approach Using Statistical Analysis and Optimization

Structural damage identification (SDI) is of paramount importance for ensuring the long-term safety and effectiveness of various structures. However, existing identification methods require careful selection of sensors, especially in terms of sensor combination optimization, which necessitates meticulous evaluation. Herein, we propose a novel approach for detecting structural damage from vibrational signals in planar steel frames, aiming to address and improve upon the current limitations. Our method first extracts relevant vibration signals from multiple joints and then replaces the vibration signal of the damaged joint with the signal difference between the damaged and undamaged joints. This innovative approach effectively highlights the features of structural damage for further analysis. Utilizing statistical analysis, we evaluate sensor performance based on mean, standard deviation, and scatter plots, obviating the need for meticulous sensor selection. Furthermore, by analyzing the robustness exhibited by sensor combinations and confirming their stability to a certain extent, we utilize a one-dimensional fusion convolutional neural network with a signal difference (1D-FCNND) composed of individual convolutional layers using one-dimensional traditional and separable convolution techniques to detect structural damage accurately and efficiently. We validate the practical application of our method using benchmark data from the Qatar University Grandstand Simulator (QUGS). The results demonstrate a significant reduction in sensor requirements, accounting for only 13.33% of the total required sensors, while achieving a notable enhancement 98.92% in accuracy using only five 1D-FCNND models. The simplicity and robustness of our method enhance identification performance and optimize sensor utilization, making it a practical and promising solution for structural health monitoring and damage identification in various engineering structures. The capability to simplify sensor selection and improve identification efficiency showcases the practical value of our method in advancing the field of structural health monitoring and damage identification.