Health monitoring refers to the detection of structural damage and the assessment of the current state of engineered systems. Data interrogation for health monitoring encompasses many aspects of data analysis, signal processing, machine learning techniques, data management and feature extraction. Several examples of data interrogation algorithms are presented. First, a Holder exponent technique is developed to detect the presence of signal discontinuities caused by damage such as the rattling of loose components, and to identify when they occurred. Next, a statistical classifier is developed to quantify when changes in this feature are significant. The success of this method is shown both for data that contain an artificially introduced discontinuity and for an actual engineering system that has a rattle. Based on a statistical pattern recognition paradigm, damage classifiers are developed to assist a decision-making procedure related to the detection of damage. In particular, extreme value statistics are employed for the establishment of decision boundaries to minimize false positive and negative indications of damage.
Return to UQWG meetings