ABSTRACT
Deep learning model is a research hotspot in the field of artificial intelligence. As one kind of deep learning model, Stacked Auto-Encoder constitutes a deep neural network through stacking the auto-encoder. This paper proposed a method for damage detection of bridge structure based on the stacked auto-encoder. In this paper, a 5-layer stacked auto-encoder was constructed for damage detection, and nonlinear mapping relationship between modal shape and damage location as well as that between modal shape and damage degree were established through pre-training. A case study is presented focusing on the Åby bridge, a steel railway truss bridge with a span of 33.7 m in northern Sweden. In order to investigate the influence of white noise interference on the damage detection accuracy of the stacked auto-encoder, different Gauss white noise is introduced. The analysis results showed that the proposed damage detection method of bridge structure based on stacked auto-encoder has good accuracy and anti-noise performance.
