Open Access Open Access  Restricted Access Subscription or Fee Access

On the Use of Distance Measures in the Phase Portrait for Damage Localisation and Severity Assessments



Damage localisation and severity assessment are two essential levels in the field of structural health monitoring (SHM). One of the challenges in SHM today is to extract damage-sensitive features from the measured data and correlate them with the location and severity of the damage. Common types of structural damage, such as cracks, or even joint failures due to loosened bolts, usually result from a slow-time degradation process and show strong nonlinear behaviours. In this case, traditional features may not be able to distinguish between the undamaged and damaged structure and track the degradation process effectively and efficiently. In this paper, the damage features are described by a distance measure in the phase portraits to locate the damage and quantify its severity. The phase portraits are reconstructed from the measured time series of various conditions. Instead of focusing on the warp of the trajectory or the phase space topology, this paper introduces a simpler way, using a distance metric based on Mahalanobis distance (MD) to demonstrate the features in various damage scenarios and the undamaged one. Combined with a data-driven pattern recognition method, like Artificial Neural Networks (ANNs) and Gaussian Processes (GPs), this distance metric in the phase space can detect, locate and assess the severity of the damage in a standard classification approach. A simple laboratory structure is set up to verify the proposed feature. A comparison between ANNs and GPs for the damage severity problem is also done.


Mahalanobis distance, Phase space reconstruction, Damage localization, Severity assessment, Gaussian process.Text


Full Text: