New damage-sensitive feature for structures with bolted joints

Abstract

Abstract.Compared with riveted and welded joints, bolted joints have advantages of easily dismantled, high load carrying and load-transferring capacity. However, bolted joints are also weaker components of assembled structures. Structural damage detection (SDD) on bolted joints is much required in the field of structural health monitoring (SHM). A new SDD method is proposed for damage identification of structures with bolted joints based on residual error of AR model in time series analysis. Firstly, a new data standardization process is defined to maintain the information of damage location. Then, a new structural damage feature sensitive to structural damage is developed based on the standard deviation of AR model residual errors. To verify the proposed method, a bolted joint structure is designed and fabricated in laboratory, connection damages of structures are simulated by loosening the bolted joints. The acceleration responses of structures with bolted joints under healthy and damage cases are acquired. Finally, the SDD is performed by traditional DSF and the new DSF. The illustrated results show that the proposed method is a hybrid tool for the bolted joint damage detection with the new damage-sensitive feature. In addition, some related issues will be discussed as well. IntroductionThe vibration-based SDD technique has become an effective way in SDD [1,2]. Normally, SDD can be achieved by comparing the structural characteristics extracted from structural reference state and damage state respectively. Zhou et al [3] used the hierarchical clustering analysis and similarity measure to distinguish structural damage state from health state. Most of these methods can be divided into two categories: model based and feature based. The model based methods need to construct the structural dynamic model. Khatir et al [4] used the co-ordinate modal assurance criterion for SDD. Khatir et al [5] used the modal scale factor and natural frequencies for structural damage detection and localization. To the feature based methods, especially for those based on time series analysis, are found to be less complicated and more sensitive to local damage [6].Most of time series analysis based methods need to construct a time series model that fit for vibration response data and achieves SDD via comparing the extracted features from structural reference (healthy) and damage states. Chen and Yu

Probabilistic machine learning for detection of tightening torque in bolted joints

“…The results’ analysis indicates that it was possible to recognize a reduction in the torque values from the 80–60 cNm conditions to 55–5 cNm based on the proposed DI . Compared to methods in the literature using modal parameters, just a considerable loss of tightening torque is detected with statistical confidence, as performed by Luo and Yu, 23 where only changes for tightening were observed from 20 Nm in the undamaged condition to 2.5 Nm in the damaged one. Furthermore, it is also important to stress that the structure tested by these authors consisted of a single bolted joint, which is a setup quite sensitive to variations in the applied preload.…”

“…22 To the best of our knowledge, the best safe detection of lack of tightening torque currently present in the literature is in the order of approximately 10 times less than a torque condition considered healthy. 23 An effective way to treat the inherent nonlinearities seems to be by applying a higher-order spectrum, as the Volterra series. 24 Luo and Yu 23 proposed a new method for damage identification of structures with bolted joints based on a residual error of the Auto-Regressive model in time series analysis.…”

“…23 An effective way to treat the inherent nonlinearities seems to be by applying a higher-order spectrum, as the Volterra series. 24 Luo and Yu 23 proposed a new method for damage identification of structures with bolted joints based on a residual error of the Auto-Regressive model in time series analysis. In this particular work, the authors could only observe torque changes when the undamaged condition's level reduced from 20 Nm to 2.5 Nm in the damaged one.…”

“…22 To the best of our knowledge, the best safe detection of lack of tightening torque currently present in the literature is in the order of approximately 10 times less than a torque condition considered healthy. 23 An effective way to treat the inherent nonlinearities seems to be by applying a higher-order spectrum, as the Volterra series. 24…”

A comprehensive review on health monitoring of joints in steel structures

“…FD signal processing techniques, such as Discrete Fourier Transform (DFT), FRF, Strain Frequency Response Function, Frequency Domain Decomposition (FDD), and Multiple Signal Classification (MUSIC), are widely employed for the health monitoring of diverse infrastructures [197]. Subsequently, AR models were deployed to estimate residual errors, which served as indicators of structural damage [198]. Xu et al presented a method for detecting nonlinearity within a system based on Empirical Mode Decomposition (EMD).…”