UC San Diego
Ambient Excitation Based Model Updating for Structural Health Monitoring via Dynamic Strain Measurements
- Author(s): Martins, Benjamin
- Advisor(s): Kosmatka, John B.
- et al.
Structural health monitoring (SHM) technologies continue to be pursued for aerospace structures in the interests of increased safety and, when combined with prognosis, efficiency in life-cycle management. The current work is focused on developing and validating a method for in-situ health monitoring of aerospace structures. In particular, the current framework has been developed for use with response only vibration data using natural operating turbulence to provide the means of excitation. While the framework is general so as to work with a wide suite of sensor options, particular emphasis has been placed on fiber optic strain sensors as a lightweight, low cost, non-intrusive means of monitoring the vibration response.
At its core, the developed SHM system actively monitors a network of fiber optic strain sensors and utilizes the transient response data to calculate their associated power spectral densities (PSD). These PSD serve as the fundamental input to the developed SHM algorithm presented in the dissertation whereby comparisons between previously correlated model PSD and the current measured PSD are made. If anomalies between the correlated model and the measured data sets are detected, the developed SHM algorithm seeks to minimize the difference via updating of structural parameters underlying the structural model of interest (in the case of the presented work, a finite element model of the structure).
The SHM algorithm itself is an adaption of a statistical least-squares minimization based in concepts of non-linear parameter estimation and model correlation. The algorithm developed uses power spectra based residual error vectors derived from distributed vibration measurements to update a structural model through statistically weighted least-squares minimization. The output of the algorithm is a correlated finite element model which inherently produces estimates of the location, type, and severity of any detected damage as well as the uncertainty associated with these estimates. Throughout the dissertation the developed algorithm was shown, both analytically and experimentally, to successfully detect, locate, and quantify damage present in a structural system.