Research On Damage Identification Structures Based On Response Surface Method

This paper is mainly identified the damage to the many degrees of freedomsystem and the continuum structure, and analyzed the noise and nonlinearity influenceon the result of the damage identification.1. A5DOF dynamic system is established and worked out corresponding to theoutput features of energy E and the log decrement feature LDaimed at the inputparameters of mass, stiffness, impulse magnitude and impulse location. Then theEand LDgenerated are used to analysis the sensitive to some of all the potential inputparameters by the general sensitivity analysis and the significant effects variablescreening. Because E and LDfeatures are highly correlated, the response surfacemodel about stiffness and location is established using natural frequencies as outputfeatures. At last, compared to the prediction of stiffness and location through thefraction factorial design and full factorial design, two designs are useful and the fullfactorial design could be better than the indentified the damage.2. The response surface models trained were used for damaged springsidentification to different noise, and proved that when the noise influence increases, theless accurate prediction of stiffness. At the same time, the stiffness of spring must benonlinear firstly because of nonlinearity. Then the nonlinear spring is supposed tolinear and indentified the damage. It can be seen that the prediction accuracy of thestiffness is lower than the location’s. Finally, the new response surface models are setup, and be shown that the damaged nonlinear spring may be located with thesatisfactory accuracy.3. For continuous structures, a finite element model is established that a beamwith ten potential locations at which small masses may be attached. And the responsesurface models developed for the beam-mass system using natural frequencies asoutput features are quite successful at predicting the number of masses on the beamand do a reasonable job locating masses given the limited amount of data they weretrained with. It is proved that the locations near the free end of beam are predictedbetter than those near the fixed end, and the potential of RSMs for the damageidentification problem need be developed further.