A Data-Based Technique for Damage Detection Handling Environmental Variability During Online Structural Health Monitoring

Abstract

AbstractThe success of data-based online damage detection techniques depends upon the ability to detect the deviation from the previous measurements of the healthy system, changes in the material and/or geometric properties, boundary conditions, and system connectivity. Most of the data-based techniques extract features like frequencies, mode shapes, etc., for further processing. However, the random excitation, the varied environmental conditions, and the undesired measurement noise, often bring in the stochasticity in the features extracted from the vibration data. Also, the environmental variability due to temperature alters the material property of the structure, and creates an effect which is similar to that of the real damage. This fact emphasizes the need for techniques to differentiate the effects of environmental variability from damage, during diagnosis, using the output-only vibration data. In this paper, a data-based technique, which can effectively handle the environmental variability, as well as capable of locating the region of damage in the structure, using the acceleration time-history data, is presented. In this paper, Mahalanobis Squared Distance (MSD), which is popularly used in novelty detection, is used to handle the effect of environmental or operational variability (EOV) and simultaneously perform the damage detection, by treating the acceleration time-history of sensor nodes as feature vectors. With the confirmation of the presence of damage, subsequently, the spatial domain of damage is identified, by performing a sequential elimination approach, while forming the feature vectors for MSD evaluation. The results of the numerical studies using a synthetic data and benchmark data show that the proposed data-based technique using MSD is efficient in eliminating the variability and precisely locating the damage region, spatially on the structure.KeywordsStructural health monitoringDamage locationMahalanobis squared distanceEnvironmental and operational variabilityAcceleration responsesMeasurement noise