Hammering Inspection Using Deep Learning: Analyzing Wavelet-Transformed Images from Vibration Data

Bolt/nut loosening inspections are still predominantly performed by human experts using hammering methods. To enable non-experts to conduct these inspections, we propose a novel approach to identifying the degree of loosening in bolts and nuts by applying wavelet transforms to hammering vibration data and utilizing deep learning. In this method, vibration data near the bolt, collected via hammering inspections with an acceleration sensor, is converted into images using wavelet transforms. These wavelet images are then preprocessed to extract and highlight features related to loosening. The preprocessed images are subsequently analyzed using deep learning models to estimate the degree of loosening.