Subspace‐based Mahalanobis damage detection robust to changes in excitation covariance

Abstract

Summary In the context of detecting changes in structural systems, several vibration‐based damage detection methods have been proposed and successfully applied to both mechanical and civil structures over the past years. These methods involve computing data‐based features, which are then evaluated in statistical tests to detect damages. While being sensitive to damages, the data‐based features are affected by changes in the ambient excitation properties that potentially lead to false alarms in the statistical tests, a characteristic that renders their use impractical for structural monitoring. In this paper, a damage detection method is presented that is robust to changes in the covariance of the ambient excitation. The proposed approach is based on the Mahalanobis distance of output covariance Hankel matrices, which are normalized with respect to possibly changing excitation properties. The statistical properties of the developed damage feature are reported and used for efficient hypothesis testing. Its robustness towards changes in the excitation covariance is illustrated on numerical simulations and successfully tested on a numerical offshore foundation model.

Sensor placement with optimal damage detectability for statistical damage detection

“…modal parameters [42,43] or features that are formed based on the subspace properties of output covariance Hankel matrices [36,37,44,45]. Some features from the literature [46][47][48] are even robust to changes in environmental conditions.…”

The detection matrix as a model-agnostic tool to estimate the number of degrees of freedom in mechanical systems and engineering structures

“…Unlike techniques based on subspace identification from output-only data 5,33 , this method does not rely on a complete identification of the system at hand -but only on counting the number of degrees of freedom and on monitoring changes to this quantity. In a way, our method is similar in spirit to others that are based on the monitoring of a damage-sensitive feature of measured data [34][35][36] , with the difference that our damage-sensitive feature, the number of degrees of freedom, has a clear physical meaning and is not affected by changes in the amplitude of the initial condition or in the environmental conditions. Its model-agnostic nature, however, makes it challenging if not impossible to detect where damage has occurred.…”

A data‐driven combined deterministic‐stochastic subspace identification method for condition assessment of roof structures subjected to strong winds

“…For a linear dynamical system, the data‐driven subspace identification method (SIM) is well suitable for the assessment of structure condition and damage detection 22–31 . In practice, the roof of the large‐scale structure in operation is hard to excite using artificial excitation forces and is almost impossible to exclude ambient excitation.…”

“…For a linear dynamical system, the data-driven subspace identification method (SIM) is well suitable for the assessment of structure condition and damage detection. [22][23][24][25][26][27][28][29][30][31] In practice, the roof of the large-scale structure in operation is hard to excite using artificial excitation forces and is almost impossible to exclude ambient excitation. Due to these restrictions, stochastic subspace identification (SSI) method has been developed, in which the unmeasured forces are modeled as white noise time series.…”