Bolt loosening detection based on audio classification

Abstract

As the most widely used coupling structure in electromechanical systems, bolt coupling is the important part in these systems. The reliability and strength of bolted joint are affected by pretension force, which is one of the most important factors to ensure the stability of bolt coupling. The inspection personnel hit the bolt with a hammer and judge the state of the bolt based on the sound. Although this method is very simple, the ability of the human ear to distinguish the knocking sound is poor, it can only distinguish the bolt with larger looseness. So a bolt loosening detection method based on audio classification is presented in this article. First, the hammering sound at different levels of bolt loosening was collected by smartphone. Then, the audio data were extracted to form a dataset. Finally, the support vector machine was used to train and test the dataset, and obtain the bolt loosening quantitative detection. A series of experiments were carried on to verify the accuracy and stability of this method. The results show that this method has high recognition accuracy and strong noise immunity. Therefore, this method can effectively reduce the occurrence of disasters.

Shear loading detection of through bolts in bridge structures using a percussion‐based one‐dimensional memory‐augmented convolutional neural network

“…The training and testing processes of the 1D‐MACNN are presented in Figure 8, and we can see that no overfitting exists. Moreover, to demonstrate the advantage of the proposed method () 1D‐MACNN, we compare its performance with other baseline methods including (1) MFCC + SVM (Y. Zhang, Zhao, et al, 2019), (2) PSD + DT (Kong et al, 2018), (3) IMSE + BPNN (Yuan et al, 2019), (4) 1D‐CNN, and (5) 1D‐CNN‐LSTM. For Method (1), the MFCC features (window length: 0.03 × sampling frequency; overlap length: 0.02 × sampling frequency) of each sound signal are extracted and flattened to construct a feature set (size: 494), and radial basis function is employed as kernel function in SVM (kernel band width parameter: 0.07; regularization parameter: 1).…”

“…Kong et al (2018) developed a percussion method based on the power spectrum density (PSD) of sound signals and the decision tree (DT) to identify the bolt looseness. Y. Zhang, Zhao et al (2019) proposed another percussion method that extracted the Mel‐frequency cepstral coefficients (MFCCs) of sound signals as features, and they trained a support vector machine (SVM) classifier to achieve bolt looseness detection. Based on the nonlinear properties, Yuan et al (2019) characterized the sound signals through the intrinsic multiscale entropy (IME) and distinguished different levels of bolt preload via a backpropagation neural network (BPNN).…”

“…Recently, another attractive nondestructive testing approach, namely, the percussion method, has been a favored direction in civil engineering by virtue of its low cost, its ease of implementation, and because there is no requirement for constant contact sensors. Many successful applications (Luk, Liu, & Tong, 2011) have confirmed the advantages of the percussion method, particularly in detecting bolt looseness (Kong, Zhu, Ho, & Song, 2018; F. Wang, Ho, & Song, 2019b; Yuan, Lv, Kong, & Song, 2019; Y. Zhang, Zhao, Sun, Su, & Xue, 2019). However, to develop the relationship between structural damages (e.g., bolt looseness) and percussion‐induced sound signals, the above percussion methods depend on machine learning (ML)‐based techniques, in which the features are manually crafted and then fed into various classifiers.…”

Loosening of Bolted Connections under Transverse Loading in Timber Structures

“…Xu [10] proposed a bolt loosening identification method based on a system nonlinear price reduction model and the experimental results showed that the model response solution could give a reliable and sensitive indication of bolt loosening. Zhang [11] collected quantitative audio of bolt loosening and used support vector machines to train and test the data to obtain quantitative detection of bolt loosening. Li [12] systematically studied the characteristics of the second-order output spectral transmittance along the physical structure of the sensor chain with bolts, and proposed a novel method based on the second-order output spectrum to detect multiple bolt loosening.…”

SHMnet: Condition assessment of bolted connection with beyond human-level performance

“…The local methods, or termed as non-destructive evaluation methods, include relative displacement sensors, 3 vision-based methods, [4][5][6] acoustoelastic effect-based methods, 7 piezoelectric impedance method 8 and piezoelectric active sensing method. 7,9 Most recently, percussion-based methods, 10,11 audiobased methods 12 and smart washer methods 13 have emerged. They are very sensitive to small damage, for example, partially loosened bolt, or loss of a part of preload.…”