Study On Testing Of Intelligent Structure By Optic Sensor Based On Neural Networks

In order to meet society developments, people are trying to research and develop intelligent structure or intelligent material system, and realize intelligent material and structure ultimately. The system can feel and echo inner and outer environment damage change, and realize functions, such as self-detection, self-diagnosis, self-control, self-adjustment, self-recondition and self-heal and so on. Feeling is the basic function for intelligent material and system.Factors of load and temperature, chemical and environment let material inner accrue microcosmic and macroscopical defects. When there are microcosmic and macroscopical defects, we believe the material is defected. Defections accrue enormous intimidate for structural safety, so the question is researched in academic and engineering broadly.Hurting to identify process is non-linear highly, complicated process of multi-variable, the signals distribute widely, and the signals are constituted by a flood of long-lost signals, handled with the general method, calculation quantity is great very, handling time is long, can't satisfy meet of hour monitors, because neural network has highly abreast, non-linear, permits the ability of mistake goodly and self-adapt, self-study ability ,suits to be used for the intelligent construction to detect hurt and valuation. This paper uses BP neural network suited to identify and valuate the gypsum piece. This paper proceeds primarily three researches contents:(l) identify of the gypsum intelligence piece hurts;(2) the harm position of the gypsum intelligence piece;(3) mark and evaluation of the gypsum intelligence for hurting degree.This thesis aimed at the gypsum intelligent structure by fiber optic sensor designs neural network identify model, adopts various training ways to train the network. The results show Levenberg- Marquardt way refrains more quickly than the traditional gradient descent way, but regularization way not only can increase trains speed , at the same time, but also can increase generalization ability of the network.