Research Of Hybrid Soft Computing Method For Structrural Damage Detection

In recent years, structural damage diagnosis developed rapidly in engineering.But some of them have some difficulties to large scale structure, and there are manyproblems to discuss deeply. As the civil engineering structure is complicated, thenumber of recognize methods are so many. It appears much new recognition in softcomputing method. The characteristic of these theories is not to need to build theaccurate mathematical model of the problem itself, but to get the result byprocessing the input data directly. Soft computing method includes fuzzy logic,rough set theory, artificial neural network and genetic algorithm. It is a new softcomputing method that the combination of rough set theory and artificial neuralnetwork. By using rough set theory to do data pretreatment, through the use of roughset attribute reduction and based-on the information entropy discretizationprocessing, it combines with artificial neural network method to identify thestructural damage detection in order to improve the artificial neural networkconvergence rate and to improve the efficiency of damage identification. Curvaturemode is a sensitive structural dynamic fingerprint to structural damage identification.Curvature modal and strain modal have the one-to-one correspondence relation.Curvature mode can be calculated by the measurement of acceleration. Bymeasurement of structural modal parameter, we can get the curvature modal bydifference. Based on the difference between non-damage structure and damagestructure, the damage location parameters can be certain. If there is a damage instructure, the stiffness will decrease. The curvature mode difference will change atthe location of damage. The main context of this passage is listed as follows.Firstly, finite element analysis model is established based on the fourteensimply supported steel truss. We analyse the structural modal parameter at damagesituation and non-damage situation. The structural dynamic fingerprints areextracted at any situation in order to process the damage database by reduct anddiscretization.Secondly, we measure the response of the structure under damage and non-damage situation in laboratory. Then we analyse the structural vibrationcharacteristic and get curvature mode by difference calculation. Through thecomparing with the data obtained from finite element simulation and the data frommeasurement, it is confirmed the accuracy of the finite element simulation.Then, rough set knowledge discovery method, and write the program ofinductive algorithm and discretization method by matlab. It is successful to doattribute reduction and discretization. It gets a good result that the conditionattributes decrease greatly and gets minimal attribute set at different damagesituation. And it is much easier to excavate rules.Finally, we use artificial neural network to train and recognize damage database.By using the trained network, we recognize the measurement of curvature mode.And we get a good result that we can make sure the damage span and location. Then,we analyse the effect of different noise size to identify result.