Unsupervised Structural Damage Diagnosis Based on Change of Response Surface Using Statistical Tool

(Application to Damage Detection of Composite Structure)

抄録

Most structural health monitoring systems adopt parametric methods based on modeling or non-parametric methods such as artificial neural networks. The former methods require modeling of each structure, and the latter methods require a large number of data for training. These methods demand high costs, and it is impossible to obtain training data of the damaged state of an in-service structure. By the present method, damage is detected by judging the statistical difference between data of the intact state and the current state. The method requires data of the undamaged state, but does not require complicated modeling or data for training. As an example, the present study deals with the detection of delamination of a composite beam. Damage is detected from the change of strain data using statistical tools such as the response surface and F-statistics. As a result, the new method successfully diagnoses the damage without the need to use modeling or data of the damaged state.