Unsupervised learning methods for vibration-based damage detection

The basic premise of vibration-based damage detection is that damage will significantly alter the stiffness, mass, or energy dissipation properties of a system, which, in turn, alter the measured dynamic response of the system. Although the basis for vibration-based damage detection appears intuitive, its actual application poses many significant technical challenges. A fundamental challenge is that in many situations vibration-based damage detection must be performed in an unsupervised learning mode. Here, the term unsupervised learning implies that data from damaged systems are not available. These challenges are supplemented by many practical issues associated with making accurate and repeatable vibration measurements at a limited number of locations on complex structures often operating in adverse environments. This paper will discuss two statistical methods for approaching the unsupervised learning damage detection problem. The first method is density estimation and significance testing. The second method is statistical process control. Examples of these methods are applied to data from an undamaged and subsequently damaged concrete column.