Condition Assessment of a Cantilevered I-Beam Using LSTM Deep Learning Algorithm

For maintaining and prolonging the service life of civil constructions, structural damage must be closely monitored. Monitoring the incidence, formation, and spread of damage is crucial to ensuring a structure’s ongoing performance. To give realistic means for early warning against structural deterioration, numerous monitoring and detecting approaches have been developed, such as vibration-based techniques, machine learning (ML) and especially deep learning (DL) algorithms. In this paper, the effectiveness of a deep learning technique known as long short-term memory (LSTM) for detecting damage in a steel cantilevered I-beam using a model-based approach has been investigated. For this purpose, a finite element model of an undamaged I-beam and several cases of I-beams with different damage combinations were prepared. The LSTM model was trained with the acceleration response sampled at different points on the top flange along each I-beam. Also, random normal noise was added to the acceleration signals for considering the effect of noise in the data extraction process. In this study, damage was defined as openings in the beam for simulating the stiffness reduction in the beam. It was noticed that by using the acceleration response of just one point on the top flange, the trained LSTM model can distinguish the undamaged beam from the damaged one with high accuracy.