Damage Detection For Pile Bent Based On The Support Vector Machine

Structural damage detection is a multi-discipline research covering several technical subjects. The global structure damage identification method based on vibration test is studied systemically, and two important damage identification indicators, Modal Strain Energy Change Rate (MSECR) and Residual Modal Force which are used widely in current, are discussed in the present thesis. Through comparing their ability on damage identification, it is appeared that the Residual Modal Force indicator has a better adaptability on the structure form, the extent of damage, the number of damage and the location of damage. Then Statistical method is introduced to structural damage identification in noisy environment. Monte Carlo method is utilized to simulate the environmental noise and this method is proved to be an effective way to locate structure damage in noisy environment. On the basis of determining the location of structure damage, the Support Vector Machine (SVM), which is a new kind of machine learning algorithm established on the small-sample statistics theory, is introduced. The damage extent in different positions is estimated using the fitting ability of SVM, which can overcome the disadvantage in traditional machine learning algorithms that requiring large training samples. Finally, the feasibility of the method is proved by being applied to a representative pile bent structure in port engineering.