Towards probabilistic data‐driven damage detection in SHM using sparse Bayesian learning scheme

Abstract

Despite continuous evolution and development of structural health monitoring (SHM) technology, interpreting a huge amount of sensed data from a sophisticated SHM system to extract useful information about structural health condition remains a challenge. Aiming to resolve this problem, a novel application of probabilistic data-driven damage detection method was proposed in the context of Sparse Bayesian Learning (SBL) scheme. The framework involves constructing a new structural damage index and establishing SBL regression model as reference base only using the data acquired in health state. The construction of the structural damage index is based on damage-sensitive frequency band, which is determined by NExT using vibration monitoring data. The structure will be classified to be damaged as the structural damage index based on new data deviates from the index predicted by SBL regression reference model, and further, the Bayes factor is adopted to quantify the damage degree. In addition, the relationship between the Bayes factors and the resonance frequency change rate is investigated in detail. The proposed methodology features the following merits: (i) It is probabilistic data-driven method exempting from physical model of the structure, excitation/loading information, and (ii) it belongs to the unsupervised model in need for structural damage detection, which can be formulated using only monitoring data from health state in the absence of monitoring data from damaged state. Damage detection and discrimination capabilities of the proposed methodology are verified using field monitoring data acquired from a cable-stayed bridge. Finally, a discussion of the SBL-based approach is made and further challenges pertaining to damage detection processes in the context of SHM are identified.

A Case Study of Enhancing the Data Science Capacity of an RCMI Program at a Historically Black Medical College

“…Data science as a new field has grown tremendously in the past decade due to its great potential to transform industries, the economy, health care, and scientific discovery. With data analyst skills becoming increasingly crucial for leveraging big data to gain a competitive edge, accelerate scientific discoveries, and advance public health [ 1 , 15 , 16 , 17 , 18 ], the inequitable representation of minority groups in the field, coupled with the lack of resources and infrastructure in minority-serving institutions, has become a major issue in the U.S.…”

Influences of Flood Conditions on Dynamic Characteristics of Novel 3D-Printed Porous Bridge Bearings

“…In other words, a FRF is a function employed to measure the response of a system to an excitation, divided by the excitation magnitude, in the frequency domain. Addtionally, the FRF based on monitoring data is basically influenced by the data noise [ 30 ].…”

“…In this paper, the parametric generation approach has been selected for creating the proposed model using CAD software due to its easy modification, instant design as well as better resolution. The model dimensions are based on the design of common bridge bearings used for Thailand highway bridges [ 30 ]. The model consists of two layers with a 50 mm unit cell size and a 2 mm thickness.…”

Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review

“…They also presented a strategy for updating the uncertainty with the acquisition of new measurements. Wang et al [ 71 ] also used Jeffreys–Bayes factor hypothesis testing to detect the structural damage in the Tianjin Yonghe cable-stayed bridge. They proposed a damage index obtained from the Natural Excitation Technique, which has both real and imaginary parts.…”