A method for automated bolt-loosening monitoring and assessment using impedance technique and deep learning

Abstract

Traditional impedance-based bolt-looseness monitoring approaches using hand-crafted impedance features can lead to difficulties in quantitative damage severity estimation. In this study, a novel method for bolt-looseness assessment with automated impedance feature extraction is developed based on the integration of the impedance-based technique and deep learning algorithm. 1D CNN (1-dimensional convolutional neural network) – based bolt-looseness estimation models are designed to automatically extract and learn optimal features from raw impedance signals. The experimental verification shows that the proposed method can identify the location of loosened bolts and estimate loosening degree in a girder connection with high accuracy. The best training and testing errors of the 1D CNN-based looseness prediction are 0.063 and 0.081, respectively. The proposed method does not require pre-processing impedance signals and selecting proper frequency bands, so it has great potential for real-time bolt-looseness detection applications.