A probabilistic approach for the detection of bolt loosening in periodically supported structures endowed with bolted flange joints. (1st August 2019)

• Record Type: Journal Article
• Authors: Yin, Tao; Wang, Xiang-Yu; Zhu, Hong-Ping
• Is Part Of: Mechanical systems and signal processing. Volume 128(2019)
• Page Start: 588
• Page End: 616
• Publication Date: 2019-08-01
• Subjects: Periodic structure -- Bolt loosening detection -- Spectral element method -- Transfer matrix method -- Model class selection -- Particle filter; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2019.04.010
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Bolted flange joint is to be modeled efficiently by spectral element method. Computational cost of proposed system model is irrespective of number of cells. A two-phased methodology is developed to identify posterior PDF efficiently. Abstract: From the literature, only very limited research activities have been carried out for fault diagnose of periodic structural system following model-based approaches. This paper focuses on the development of a practical methodology for modeling and detecting bolt loosening on periodically supported beam-type structure endowed with bolted flange joints, representing typical supported pipeline system in industry, through using measured modal parameters. Within the framework of periodic system, an efficient analytical model of the complete periodic system is first developed for dynamic analysis in frequency domain. The highly accurate spectral element method is employed to formulate the supercell-based dynamic stiffness matrix (DSM) of periodic cell containing bolted flange connection in the midspan, and the transfer matrix-based method is also developed for assembling the system DSM of entire periodic structural system through the obtained DSM of each individual cell, where the computational effort required in dynamic analysis of the complete periodic system with a large amount of repeated cells is almost comparable to a single cell. Then, in the proposed methodology, the statistical detection of bolt loosening is accomplished through two phases. The most plausible model class with appropriate parameterization complexity is first recognized by following the Bayesian model class selection strategy in the first phase. In the subsequent phase, the posterior probability density function of the stiffness scaling parameters is identified following the particle filter-based approach. To demonstrate and validate the proposed methodology, this paper reports not only the theoretical development but also a comprehensive series of numerical and experimental case studies, and corresponding results achieved are very encouraging.