Detection Of Structural Damage By Inductive Learning Methods

Civil engineering structures serving in complicated environment often suffer different kinds of damages, accumulation of such damages easily results in catastrophes, so it is very important for safety of structures to detect the damages of structures.In recent years, machine learning methods are used to the detection of structural damage, which are interested by some researchers both in domestic and abroad, especially, the Artificaial Neural Network(ANN) methods have been already used by more and more scholars and obtained a quite good result. Inductive learning is a machine learning method with the basis of inductive inference, it is one of the most core and widespread applying branches in machine learning. Inductive learning is a method which can establish the mathematical model through the study of sample data and can resist the noise data, it can overcome the shortcoming of ANN, such as hard to explain and need long time to construct. Therefore, it has been already successfully applied to quite a lot domains. This thesis attempts to use some inductive learning methods to detect structural damage..This thesis detects the location and degree of a cantilever beam and a six-layer frame structure by ANSYS theory and inductive learning methods. Firstly, some commonly detect methods of structural damage have been introduced, and the present research situation related them has been issued;Secondly, the basic concepts, the pattern, the inference method and the classification of inductive learning have been discussed, which includes learning from examples & learning from observation and discovery. Three inductive learning methods(Divid-and-Conquer, Covering, Bagging) and RBFNN method are focused to analyse;Thirdly, the corresponding models of the cantilever beam and the six-layer frame structure are established by ANSYS, The different position and degree of structural damage are simulated, the first five rank or the first six rank natural frequency value of structural models are calculated both in the conditions of undamage and the damage;Finally, the above calculated natural frequency is preprocessed, and then the location and degree of a cantilever beam and asix-layer frame structure are detected by Matlab and RDS softwares. The effect of those inductive learning methods are analyzed through these test results.The thesis indicates that: (1) apply inductive learning methods to the structural damaging-detection research shows quite high research valuer not only on theoretically speaking but also through the experiment proof, especially the bagging;(2) inductive learning methods can be used in some simple structures, like cantilever beam, but it is not useful in the detection of the position of the construction structure;(3) the precision of the bagging, no matter in the cantilever beam structure or in the construction structure, is continuously stable, but the RBFNN is only useful to the structure which degree of noise is smaller than 50%.