Research on damage diagnosis of steel truss bridge structures using a hybrid 1DCNN-LSTM Network

Abstract

This paper presents a novel approach for structural damage diagnosis by applying a deep learning (DL) model that combines the feature extraction capabilities of a One-Dimensional Convolutional Neural Network (1DCNN) with the time-series data processing strength of a Long Short-Term Memory (LSTM) network. While 1DCNN excels at extracting features from data, it struggles with capturing long-term dependencies in time-series sequences. On the other hand, LSTM is effective at learning and analyzing long-term relationships but faces challenges in balancing computational weights and has slower processing speeds. To validate the effectiveness of the proposed method, the study utilizes time-series data collected from an accelerometer sensor system on the Chuong Duong steel truss bridge. The results demonstrate that the proposed hybrid model outperforms the standalone deep learning models-1DCNN and LSTMachieving accuracies of 91.6%, 84.5% and 81.4%, respectively, on the test dataset.

Keywords: Damage diagnosis, one-dimensional convolutional neural network, long short-term memory network, time-series data, steel truss bridge.