Crimping is widely used in production of large diameter submerged-arc welding pipes. Traditionally, the designers obtain the technical parameters for crimping from experience or trial-errors by experiments. However, it is difficult to obtain the ideal crimping technical parameters with this method immediately at present. To tackle this problem, a new method coupled with response surface methodology and finite element method is proposed to design crimping technical parameters and save the design time of crimping. In this paper, the crimping forming process is simulated by finite element (FE) code ABAQUS. Taking the crimping of X80 steel Φ1219mm×22mm×12000mm welding pipe for instance, the simulation data from the arrangement of simulation which is constituted by the optimal latin hyper-cube sampling approach is treated as sample point. Four types of response surface methodology which included four-order polynomial function, orthogonal polynomial function, kriging and radial basis function is discussed, where the response surface model based on radial basis function is proved more efficient than other types of response surface methodology to construct surrogate model. The results showed a good agreement by a comparison with simulation results and remarkably predicted the crimping quality. Thus, the presented method of this research provides an effective path to design crimping parameters.
Comparison of Radial Basis Function Neural Network and Response Surface Methodology for Predicting Performance of Biofilter Treating Toluene
