CNNによる橋梁異常診断における特徴周波数および時間成分の同定精度への影響

抄録

This study concerns an anomaly identification method that uses CNN classification of bridge vibration spectrogram images. A spectrogram displays both frequency and time characteristics in a single image. Damage to a bridge has different effects on vibration depending on the type. The effects of the occurrence of damage occur in frequency characteristics and damping characteristics. In this paper, the influence of the characteristic frequencies and time components of bridge vibrations shown in the spectrogram on the identification accuracy is considered, and a method for creating a spectrogram that effectively improves accuracy is clarified.