Application of statistical pattern recognition methods for structural damage detection under various ambient conditions

Structural health monitoring is an economical and reliable strategy for infrastructure condition assessment. In recent years, researchers have tried to propose algorithms based on statistical pattern recognition techniques. Studies show these algorithms can be successfully used to detect structural damage. Variability of operational and ambient conditions during data acquisition should be considered as an important factor in applying statistical pattern recognition methods in practical applications. This paper studies the efficiency of statistical pattern recognition methods on the damage detection of structures under various operational and ambient conditions. The data is obtained from an experimental study on an eight degrees of freedom mass spring system. Ambient vibration is applied to the mass spring system using random excitation. In order to simulate various ambient conditions, the amplitude level of the input force has been varied. By applying the statistical pattern recognition methods, the ability of these methods to damage detection under various ambient conditions is discussed. Two common approaches of statistical pattern recognition are considered. These approaches are autoregressive model accompanied with using control chart and Mahalanobis distance for outlier analysis. Results show the importance of considering the statistical pattern recognition methods for structural damage detection under various operational and ambient conditions.