Health state inversion of Jack-up structure based on feature learning of damage information. (1st May 2019)

• Record Type: Journal Article
• Authors: Lyu, Tao; Xu, Changhang; Chen, Guoming; Li, Qingyang; Zhao, Tantan; Zhao, Yipei
• Is Part Of: Engineering structures. Volume 186(2019)
• Page Start: 131
• Page End: 145
• Publication Date: 2019-05-01
• Subjects: Jack-up -- Structural damage -- Feature extraction -- Deep learning -- Health state inversion; Structural engineering -- Periodicals; Structural analysis (Engineering) -- Periodicals; Construction, Technique de la -- Périodiques; Génie parasismique -- Périodiques; Pression du vent -- Périodiques; Earthquake engineering; Structural engineering; Wind-pressure; Periodicals; 624.105
• DOI: 10.1016/j.engstruct.2019.02.004
• Languages: English
• ISSNs: 0141-0296

Abstract

Highlights: The health state inversion method for Jack-up structure is established. Deep Belief Network is investigated to extract the feature of the non-stationary output signal. Structural safety inversion test is carried out considering both punch-through condition and structural damage condition. The improved inversion method has higher identification accuracy in health state inversion. Abstract: How to detect structural damages in large offshore structures such as Jack-up platform, is always an essential engineering problem. In this study, a novel discrete inversion approach is proposed to identify health state in various conditions. In order to solve the ill-posed problems in inverse problem, Deep Belief Network is investigated to extract the feature of the non-stationary output signal. Furthermore, structural safety inversion experiment is designed to demonstrate the effectiveness of the proposed approach, and sensor layout optimization is formulated in terms of Modal Assurance Criterion and Fisher information matrix. The results show that the features of structural response can be learned in the Deep Belief Network, and the signal features which are beneficial to state recognition can be extracted adaptively. The improved inversion method has higher prediction accuracy in health state inversion.