Research On Feature Extraction And Diagnosis Techniques For Structure Damage

Damage will come into existence in the civil engineering structures during their lifetime with the effect of loads and other unknown factors. As a result, it sometimes will bring about significant economic losses and personnel casualties. So it is necessary to make efficient diagnosis, evaluation and prognosis for the health condition of serving structures. In this paper the damage feature extraction and diagnosis techniques for engineering structure based on signal analysis and nerve network are studied, main research work and conclusions are demonstrated as following:The necessity of civil engineering structural health monitoring (SHM) & damage detection is discussed firstly. Then the concept of SHM & damage detection and the architecture of SHM system are introduced. Moreover, damage detection techniques and their development are reviewed.In order to extract damage feature, the wavelet packet analysis methods of damage feature extraction are developed. The response signals of the ASCE benchmark structure are processed by using orthogonal wavelet packet transform, then wavelet package energy (WPE) on decomposition frequency bands are calculated to represent the structure condition. Researches show that for a signal the WPE distribution can describe the energy variation of its components, for a special damage the distribution of WPE is different at the different detection nodes and for different kinds of damage their WPE distributions are different each other. The wavelet package transform can give us a finer analysis approach for signal processing and feature extraction.To implement structure automatic diagnosis, a method of structure damage diagnosis is addressed based on wavelet packet analysis and BP neural network. Compared with other learning algorithm, the elasticity algorithm converges more quickly, and needs less learning time. It is suitable for pattern classification of structure condition. However, using signals from different detection nodes for the same damage,- the recognition correct rate and performances index of BP network are different.To aim at fixing the uncertainty caused by only using signals from single detection node in structure damage diagnosis, another diagnosis method is presented by means of multi-sensor feature fusion theory. Through fusing feature extracted from several different detection nodes, it can make different information complementary, and reduce the uncertainty of damage detection information. So precision and reliability of the diagnosis information is much more modified and the diagnosis accuracy was improved.The structure damage is a progressive process theoretically. In order to monitor the process efficiently, a feature extraction method of structure progressive damage is studied based on Hilbert-Huang transform (HHT). Acceleration vibration signals of a single-degree of freedom structure model and a multi-degree of freedom structure model are simulated by reducing the stiffness gradually. The signals are processed by using HHT to extract the instantaneous frequency. The extracted instantaneous frequency is obviously changed before and after damage coming into existence, which can be taken as a feature index to monitor the structure progressive damage.