Robust damage detection using Bayesian virtual sensors

Jyrki Kullaa

doi:10.1016/j.ymssp.2019.106384

Return

Robust damage detection using Bayesian virtual sensors

2020-01-01

PRE

Jyrki Kullaa *

DOI：10.1016/j.ymssp.2019.106384

Original

Original request for help

Save

Abstract

En 中文

Structural health monitoring based on output-only vibration measurements is studied. In order to get an early warning of structural failure, the signal-to-noise ratio of the measurement data should be high. This can be achieved by applying virtual sensing to a redundant sensor network data. A Bayesian estimate of the signal of each sensor in the network is derived and proved to be more accurate than the actual measurement. The virtual sensor data then replace the actual measurements in the subsequent steps of damage detection. Another issue is the environmental or operational influences, which can mask the effects of damage. They can be removed by acquiring training data under different conditions and utilizing the correlation between the virtual sensors. Each sensor is estimated using the remaining virtual sensors in the network. The measurement of the underlying variables is not necessary. The residual is the difference between the estimated and the actual data. Residual generation using the virtual sensors is discussed in detail. The first principal component of the residual is then used as a damage-sensitive feature. An extreme value statistics (EVS) control chart with appropriate control limits and subgroup size is plotted for damage detection. Numerical simulations were performed for a bridge structure subject to unknown random excitations and various environmental effects. Damage was modelled as an open crack in a steel girder. Damage detection was performed in the time domain using the virtual sensor data. Virtual sensors outperformed the raw measurements in damage detection, making an early warning more plausible. (C) 2019 Elsevier Ltd. All rights reserved.

Keywords:

Virtual sensing

Structural health monitoring

Damage detection

Noise reduction

Measurement error

Environmental or operational effects

Estimation

Unsupervised learning

Time-domain analysis

Sensor network

Vibration measurement

Statistical data analysis

AI Summary

Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.