This study presents a new similarity measure method in order to detect and locate structural damage based on statistical pattern recognition paradigm. The similarity method is Bhattacharyya distance (BD) method that enables to calculate the distance or similarity between two random distributions with aid of their statistical properties. Feature extraction is an important step in the statistical pattern recognition process, which is conducted through fitting autoregressive (AR) models to the time series data in order to extract residuals of the models as the damage- sensitive feature. Based on the proposed methods, the similarity between pair sensors in baseline and damaged conditions is calculated in such a way that a sensor location associated with the largest distance value is identified as a damaged area. The performance and effectiveness of the proposed approaches are then verified by acceleration time histories from a three-story laboratory frame. Results show that the BD method based on using the residuals of AR model can detect and locate the damage precisely. Furthermore, this method is approximately able to estimate the damage severity.

Keywords