The application of design of experiment method in finite element model updating for estimating uncertainties of Laser Stitch Welded Structure

Abstract

Abstract The sensitivity based finite element model updating is a common method to improve the correlation between the initial finite element model and actual structure. The method is considered to be deterministic to correct the modelling parameters that lead to better predictions of the dynamic behaviour of a structure. However, for a complex jointed structure, there may contains variability and uncertainties that causing the structure become non-deterministic problems. Usually, variability in actual structure may arise from many sources such as geometrical tolerance and manufacturing process. Meanwhile, uncertainties in modelling usually results from the use of nominal material properties and imprecise joint stiffness. Therefore, this paper is intended to propose a novel approach to deal with modelling uncertainties and to perform the model updating of complex jointed structure by using statistical design of experiment (DOE) methods. Initially, the finite element model of the structure was constructed using CQUAD4 shell element and the area contact model (ACM) element connectors was used to represent laser stitch weld joints. Due to the lack of information on the structure and the welds, the input material properties used in the finite element model was based on nominal value of mild steel. Then, DOE methods such as Central composite design (CCD), D-optimal design and Latin hypercube (LHS) were used to estimate the new starting value for the material properties of the model. The results of the finite element models based on DOEs were compared with experimental data in terms of its natural frequencies. The correlations of the nominal FE model, CCD based FE model, D-optimal based FE model and LHS based finite element model with experimental counterpart were improved by using sensitivity based model updating. The results then were compared to identify the feasibility, accuracy and efficiency of the DOE methods in estimating reasonable starting material properties values and further overcome the uncertainties problems. The results show that DOE methods were efficient to be used in FE modelling and model updating since it capable to improve the accuracy of the prediction results.