ابوالفضل لر

Considering that civil engineering structures are exposed to hazards such as earthquakes, wind, lan‎dslides, an‎d impacts, structural health monitoring an‎d the timely detection of damage are essential. This process enables the identification of damage before accidents occur an‎d provides the basis for effective maintenance an‎d repair of structures. In recent decades, damage detection an‎d structural health monitoring have attracted significant attention with three primary objectives: reducing maintenance costs, enhancing safety, an‎d increasing structural reliability. Advances in structural sensing technologies an‎d response analysis methods have transformed non-destructive techniques into practical, accessible, an‎d cost-effective approaches for damage identification. A more accurate understan‎ding of structural behavior enables engineers to develop safer an‎d more resilient designs. Furthermore, the application of artificial neural networks can contribute to a better understan‎ding of structural performance during earthquakes. In the present study, a method for damage prediction an‎d identification using wavelet spectra within a Convolutional Neural Network (CNN) framework is proposed. For this purpose, drift data obtained from the simulation of 100 structures under four different earthquake records were collected using the IIDAP software. The earthquakes considered in this study had magnitudes ranging from 4 to 8 on the Richter scale, periods between 0.3 an‎d 3.0 seconds, an‎d displacement ductility factors ranging from 1.5 to 6.0. Subsequently, a neural network based on various algorithms was developed, with its architecture inspired by the LeNet structure. This network is capable of generating outputs from structural input images an‎d employs different algorithms for image classification. The performance eva‎luation metrics included Accuracy, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Coefficient of Determination (R²), an‎d classification accuracy, an‎d the results were analyzed through graphical representations. In this research, six classification algorithms were employed to predict structural displacement. Eighty percent of the data were allocated for training, while the remaining twenty percent were used for testing the algorithms. The performance eva‎luation demonstrated that the algorithm with the architecture 2×(Conv(32)-MaxPool(64)-Conv(3×3)-MaxPool) achieved the highest accuracy rate in data classification an‎d in identifying the level of building displacement. Additionally, the use of different intervals for feature extraction had no significant effect on classification accuracy

پايش سلام ، آسيب سازه ، سلام سازه، شناسايي آسيب.

Structural Health Monitoring (SHM), Structural Damage, Structural Integrity, Damage Identification.

Abolfazl lor

Dr.Gholamreza Ghodrati Amiri