Artificial neural network model to extract modal participation factors for damage localization in beam-like structures

Abstract

This paper proposes a novel method for localizing structural damage in a beamlike structure based on changes in modal participation factors that are extracted using an artificial neural network model. The forced vibration response of a structure is expressed as the superposition of its natural modes after the transient response decays out. For the same excitation conditions, the modal participation factors of the natural modes change when the structure is damaged. Two convolutional neural network models were developed to calculate the modal participation factors from a vibration image and predict the damage location. The two models were trained separately using appropriate datasets, which were extracted for nominal and damaged cantilever beams, and were combined into a single artificial neural network model. This model successfully localized damage in the beam for the test dataset, and its visual explanations were demonstrated and discussed. The model was also tested with experimental data and successfully detected damage in a cantilever beam. The simulation and experimental results strongly support the potential of the proposed damage localization method.