Analysis and modeling of diffuse ultrasonic signals for structural health monitoring
MetadataShow full item record
Structural Health Monitoring (SHM) refers to the process of nondestructive autonomous in situ monitoring of the integrity of critical engineering structures such as airplanes, bridges and buildings. Ultrasonic wave propagation is an ideal interrogation method for SHM because ultrasound is the elastic vibration of the material itself and is thus directly affected by any structural damage occurring in the paths of the propagating waves. The objective of this thesis is to provide a comprehensive damage detection strategy for SHM using diffuse ultrasonic waves. This strategy includes a systematic temperature compensation method, differential feature extraction methods optimized for discriminating benign surface condition changes from damage, and data fusion methods to determine the structural status. The temperature compensation method is based upon a set of pre-recorded baselines. Using the methods of baseline selection and baseline correction, a baseline that best matches a monitored signal in temperature is provided. For the differential feature extraction, three types of features are proposed. The first type includes basic differential features such as mean squared error. The second type is derived from a matching pursuit based signal decomposition. An ultrasonic signal is decomposed into a sum of characteristic wavelets, and differential features are extracted based upon changes in the decomposition between a baseline signal and a monitored signal. The third type is a phase space feature extraction method, where an ultrasonic signal is embedded into phase space and features are extracted based on changes of the phase portrait. The structural status is determined based on a data fusion strategy consisting of a threshold selection method, fusion at the feature level, and fusion at the sensor level. The proposed damage detection strategy is applied to experiments on aluminum specimens with artificial defects subjected to a variety of environmental variations. Results as measured by the probability of detection, the false alarm rate, and the size of damage detected demonstrate the viability of the proposed techniques.