Damage diagnosis of a building structure using support vector machine and modal frequency patterns

Abstract

A method using the support vector machine (SVM) to detect local damages in a building structure with the limited number of sensors is proposed. The SVM is a powerful pattern recognition tool applicable to complicated classification problems. The method is verified to have capability to identify not only the location of damage but also the magnitude of damage with satisfactory accuracy. In our proposed method, feature vectors derived from the modal frequency patterns are used after proper normalization. The feature vectors contain the information on the location and magnitude of damages. As the method does not require modal shapes, typically only two vibration sensors are enough for detecting input and output signals to obtain the modal frequencies. The support vector machines trained for single damage is also effective for detecting damage in multiple stories.