Bolt-looseness detection by a new percussion-based method using multifractal analysis and gradient boosting decision tree. (November 2020)

• Record Type: Journal Article
• Authors: Wang, Furui; Song, Gangbing
• Is Part Of: Structural health monitoring. Volume 19:Number 6(2020)
• Page Start: 2023
• Page End: 2032
• Publication Date: 2020-11
• Subjects: Bolt-looseness identification -- structural health monitoring -- percussion-based method -- multifractal analysis -- gradient boosting decision tree; Structural health monitoring -- Periodicals; Structural stability -- Periodicals; Strength of materials -- Periodicals; Nondestructive testing -- Periodicals; Constructions -- Stabilité -- Périodiques; Résistance des matériaux -- Périodiques; Contrôle non destructif -- Périodiques; Electronic journals; 624.17
• DOI: 10.1177/1475921720912780
• Languages: English
• ISSNs: 1475-9217

Abstract

Among various building blocks, bolted connections are the most widely used components, which can be employed to hold the integrity of entire structures. Looseness detection of bolted connections has been an attractive topic in multiple fields including aerospace and mechanical engineering, since loose-induced bolt failures may lead to costly disasters. Recently, several structural health monitoring methods have been applied to detect bolt looseness; however, they are often impeded in practical use due to the requirement for constant sensor–structure interaction. Thus, the potential of the percussion-based method in detecting bolt looseness has been noticed. In this article, considering the drawbacks existing in prior investigations (e.g. manual feature selection), a new percussion-based method was proposed to inspect bolt looseness. Based on the multifractal analysis and the joint mutual information maximization method, the feature sets of percussion-induced sound signals were selected automatically, which effectively avoided highly experienced personnel for manual feature selection. Subsequently, after feeding extracted feature sets into a gradient boosting decision tree model, we trained a classifier to achieve the identification of bolt looseness. Compared to current percussion-based methods for bolt-loosening detection, the method we proposed in this article has higher accuracy, which is proven by experimental results. Finally, as a rapid and non-invasive structural health monitoring approach, our method can be applied to detect damages in other structures and thus guides future investigations.