Study On BP Neural Network Of Structural Health Monitoring Based On Piezoelectric Impedence Technology

Piezoelectric impedance technology which is based on global vibration is a brand-new method for structure damage detection. It is one of the most popular research areas in recent years. In this paper, the applications of piezoelectric impedance technology and Artificial Neural Networks (ANNs) in domestic and abroad are respectively introduced, and the basic theories of them are analyzed as well; BP Networks based on the piezoelectric impedance technology are set up to diagnose the location of bolt loosening.The main contents are as following:1. The properties of electromechanical coupling are analyzed, and the relation between the mechanical impedance and electric impedance is illustrated as well as the coupling expressions are demonstrated. Besides, the fundament principle of piezoelectric impedance technology for structure health monitoring is analyzed.2. On the basis of piezoelectronic impedence technology, the experiment system is designed for detecting the piezoelectronic impedence which couples with the structural impedence which will be altered by the looseness bolt. The piezoelectronic impedence spectrums are analysised, and the results shows that the PZTs mounted on the structure can sensitively perceive the changes of structure; the nearer the changed place from PZT is, the greater change of this PZT's piezoelectronic impedence will be; for the same PZT, different bolts loosening lead to different spectrums appeared; similarly, for the same bolt looseness, different PZTs have different spectrums.3. The structure and algorithm of BP neural network are analyzed. BP network constructed by Matlab toolbox is used for exploring the application of BP network in diagnosing bolts loosening. The results indicate that these BP networks can show the area that the looseness bolt is staying, however can't tell which bolt is loosing in that area.4. The BP network constructed by C language of whose algorithm is altered partly can indicate the bolt loosening exactly, and it is easy to change the structure or algorithm for adapting the new situation of the number of the model increasing.