Damage Identification in High-Rise Buildings Using Deep Learning Techniques

Abstract

AbstractIn the last few years, structural damage identification has been considered a strong area of research by the structure engineers community. Vibration-based damage identification using machine learning algorithms has shown a tremendous advantage over other damage identification procedures. Although existing damage detection methods have been adapting machine learning principles, the majority of machine learning-based procedures extract fixed features. Depending on the framework under investigation, their output differs significantly across different data patterns. During the training process, deep learning technique convolution neural networks (CNNs) will fuse and adjust two key sets of an assessment assignment (feature extraction and classification) into a sole learning block that is manually identified in advance. This capability not only improves classification accuracy, but it also improves computational performance. In the proposed work, a damage identification study has been carried out using deep learning CNN algorithms. The performance of two CNN models one-dimensional (1D) CNN and two-dimensional (2D) CNN is discussed. This approach is verified using an analytical model of G+20 storey building modelled in FEM-based software. Probability of damage (POD) is used as the damage indicator in the analysis. It has been witnessed that the presentation of 1D CNN is superior to 2D CNN for damage diagnosis.KeywordsStructural health monitoring (SHM)Damage identification (DI)Convolution neural network (CNN)Machine learning algorithm (MI)Deep learning algorithms (DL)