.png){kind=icon}
.png){kind=icon}
Return
Semi-supervised structural damage assessment via autoregressive models and evolutionary optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.istruc.2023.105762.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Strategies based on Structural Health Monitoring (SHM) allow identifying the occurrence of damage in sensitive structures, such as bridges, tall buildings, dams, stadiums, among others. However, few strategies allow determining the exact damage locations and intensities through the structure's dynamic responses, i.e., accelerations, displacements, strains, etc. Many structures around the world are reaching their service lives and yet are still being used, as their simple replacement is economically impractical. Therefore, having an advanced SHM technique may present itself as a favorable solution for evaluating the structure's condition under changes in their vibration properties. In this paper, the four main autoregressive models (AR, ARMA, and their exogenous input counterparts ARX and ARMAX) are combined with Differential Evolution (DE) algorithm and employed to deal with multiple time series patterns of vibration signals. The procedure of model order optimization is performed both with a proposed semi-supervised objective function and compared with the well-known Akaike criterion (AIC). The semi-supervised design consists in exploring the behavior of progressive deterioration to localize the focus of it. Damage levels are calculated for each sensor based on the Mahalanobis distance (MD) of the resultant autoregressive coefficients. This enhanced framework is applied to a three-story frame and a real -world bridge (Z24 bridge). The suggested strategy reaches similar or better damage localization in all cases, while significantly outperforming the compared literature in terms of computational cost and time, making it more promising for real-time SHM.
Keywords:
Structural Health Monitoring
Metaheuristics
Autoregressive Models
Differential Evolution
Damage localization