Damage Detection in a Benchmark Structure Using Long Short-term Memory Networks

Abstract

Early warning of structural damage is of great significance which can prevent the occurrence of structural accidents. Therefore, this paper proposes a damage detection framework for early warning of structural damage using Long Short-term Memory (LSTM) model. Firstly, raw sensor data are acquired from multipoint measurement device installed on the structure. Secondly, the raw sensor data are used to train the LSTM model after data preprocessing and feature extraction. Finally, the trained LSTM model outputs the results of damage detection when new sensor data are given as input. The simulation results based on a numerical benchmark structure with different damage patterns, imply that the proposed damage detection framework based on LSTM model is superior to all other compared models including BP Neural Network (BPNN), Random Forest (RF), Decision Tree (DT), Naive Bayes (NB) and Support Vector Machine (SVM).