Neural Network Based System for Damage Identification and Location in Structural and Mechanical Systems

This is the final report of a three-year. Laboratory Directed Research and Development (LDRD) project at the Los Alamos National Laboratory (LANL). Recent advances in wireless. remotely monitored data acquisition systems coupled with the development of vibration-based damage detection algorithms make the possibility of selfor remotely-monitored structures and mechanical systems appear to be within the capabilities of current technology. However, before such a system can be relied upon to perform this monitoring, the variability of the vibration properties that are the basis for the damage detection algorithm must be understood and quantified. This understanding is necessary so that the artificial intelligence/expert system that is employed to discriminate when changes in modal properties are indicative of damage will not yield false indications of damage. To this end, this project has focused on developing statistical methods for quantifying variability in identified vibration properties of structural and mechanical systems Background and Research Objectives If accurate vibration-based damage detection is to be applied to in situ structures. sensitivity of vibration test results to environmental conditions and test procedures such as changes in temperature, traffic loading, wind, excitation method, etc. should be quantified to the extent possible. To date the vibration testing community has not developed methods to quantify the test-to-test variability in identified modal parameters such as resonant frequencies and mode shapes. Therefore, as a prerequisite to the development of sophisticated vibration-based damage detection algorithms, this project focused on developing methods to statistically quantify variability in modal parameters. These *Principal Investigator, e-mail: farrar @lanl.gov