Percussion-based bolt looseness identification using vibration-guided sound reconstruction

Bolted connection functions to fasten and secure parts together for engineering structures. Newly developed percussion-based approaches have been proven as a fast and effective tool for bolt looseness identification; however, most of the existing studies use machine learning assisted approaches to classify percussion sounds and predict looseness conditions without investigating the relationship between percussion sounds and bolt vibrations. This paper presents a conceptual research to utilize bolt vibration signal to reconstruct percussion sounds, which significantly improves the validity and accuracy of bolt looseness identification. In the experimental study, a laser Doppler vibrometry was used to capture vibrational information of the test bolt; meanwhile, percussion sounds were collected by microphones. The relationship between sound and vibration signals was investigated using wavelet packet decomposition and correlation analysis. A new set of sounds were reconstructed by combination of the sound packets which showed the strongest correlation with the vibration signal. The reconstructed sound database was then transformed into spectrograms and trained by a two-dimensional convolution neural network to identify bolt looseness conditions.