The Study on the Modifying Structural Dynamic Models Based on Information Fusion

摘要

As we all known, there are two ways for the dynamical analysis and design of structures, the first one is to predict the dynamical property on the base of finite element model, The other is to discriminate, and then to confirm and modify the structural dynamics characteristics by structural experiments and inspections. FEM has been widely used for its short computation cycle and low expense, but practice proves that there is usually obvious error between FEM prediction and test result, so the model modification method is needed in order to decrease error and predict structural behavior correctly. The error between finite element model and actual test is lack of comparability for their error sources are completely different, so the method to modify the finite element model directly with test result is blind. Moreover, the finite element model is numerical form rather than function form, so it is hard to find the relationship between performance function and independent variable which conforms to the design requirement. In other words, even if the relationship can be achieved, the dissipation of the sparsity caused by the modification of the finite element model matrix will make the method complex to apply. In addition, the method of dynamic model modification is a gradually optimized progress, which needs a series of revisions and examination steps. This computation progress will call the finite element model uninterruptedly, which costs lots of time and workload. All the reasons above make the method of dynamic model modification hard to apply, so it is difficult to find a general method which is suitable for all the structures if the specific prior information has not been used well. This paper presents a new method of dynamic model dification which fuses the specific structural prior information, test information and finite element model numerical simulation information. It also presents a new fast working model based on the probabilistic techniques, which is named response surface model. This new model will replace the whole finite element model to complete the optimization progress. Moreover, regression analysis techniques can be used effectively in the optimization design room by making full use of prior information. Then the relationship between performance function and independent variable can be achieved. In addition, the computation time of optimization progress can be shortened to an acceptable level so as to make full use of the dynamic model modification method in the field of large complex structures. In conclusion, the dynamic model modification method presents in this paper is not only used for linear structures but for nonlinear structures.