Review on recent advances in structural health monitoring paradigm for looseness detection in bolted assemblies

Abstract

The integrity of bolted joints is still a challenging problem owing to the gross or localized slip at the interfacial surfaces of the joints when subjected to external disturbances such as vibrations. This slip escalates the interfacial movement and, thus, leads to a decrease in preload levels, that is, looseness of the bolted assemblies. In the last decade, modal analysis, wave propagation, and percussion methods were traditionally used to detect looseness in bolted connections. With an increase in computational power, machine learning algorithms such as neural networks, random forests, decision trees, and support vector machines complemented the traditional methods in accurate looseness estimation. Subsequently, this integration paved the path for real-time health monitoring of bolted joints. This paper summarizes recent investigations on looseness detection in bolted assemblies based on traditional methods and machine learning algorithms. The working principle, advantages, challenges, and applications of the aforementioned methods are also detailed in this paper. Apart from these investigations, the latest studies on the Internet of Things-based health monitoring of structures are also reviewed to explore their adaptability in remote monitoring of bolted connections for damage detection in the future.

Bolt Looseness Detection in Brake Disc of High-Speed Rail Using Wavelet Packet Decomposition and 1D Convolutional Neural Network

“…The hammer tapping method is one of the most widely used methods for bolt looseness detection in flange connection and steel pillar connection. 3,4 A hammer is used to tap the head of the bolt to generate an excitation signal. Different types of response signals at different locations of the bolts or the connected components can be acquired using different transducers, including the acoustic signal through acoustic sensors, [5][6][7][8][9] the stress wave signal through piezoelectric transducer [10][11][12][13][14] and vibration signal through accelerometers.…”

Addressing data scarcity using audio signal augmentation and deep learning for bolt looseness prediction

“…In the automobile industry, 23% of the service issues are related to the looseness of bolted connections. Moreover, 12% of all new cars encounter adopted to recognize looseness in bolted connections [7]. In all these approaches, sensors are attached to the test structures to collect the response signals.…”

“…The collected audio signals are investigated to evaluate looseness in bolted joints. In the past five years, the integration of the percussion approach with machine learning (ML) models for looseness identification in bolted joints has witnessed rapid growth [7]. For instance, Kong et al [26] adopted a decision tree model to classify bolt looseness conditions in a multi-bolt flange using hand-crafted frequency features extracted from percussion-induced audio signals.…”

“…As ML models work on hand-crafted/manual features, finding an optimal feature/classifier combination for every damage case is laborious [29]. Subsequently, DL models, such as convolutional neural networks (CNNs), which fuse automatic feature extraction and classification strategies in their framework, are employed for improved bolt looseness classification [7,30]. These models are trained using spectrograms [31], cloud images [32], mel-frequency cepstral coefficients [33], etc.…”

An In-Depth Study of Vibration Sensors for Condition Monitoring

“…Looseness: Vibration signals might show non-synchronous peaks at different frequencies when there is looseness, which denotes loose connections [84]. There may be inconsistencies in the amplitude, particularly when loads or operating circumstances change.…”