MPS-based LS-SVR Metamodeling Technique for Sheet Forming Optimization

Several metamodeling techniques have been developed in the past two decades to reduce the computational cost of desian evaluations, such as finite element (FE) simulation code. With the increase of complexity and scale of practical engineering problems, popular metamodeling techniques involving response surface methodology (RSM), Kriging (KG), radial basis functions (RBF), and multivariate adaptive regression splines (MARS) confront difficulties in solving nonlinear optimization problems, such as sheet forming optimization. In this paper, a mode pursuing sampling (MPS)-based least square support vector regression (LS-SVR) method is applied for sheet forming problems. The advantage of MPS is that the samples are concentrated near the current local minima of the design space and yet still statistically cover the entire design space. Therefore, the MPS is used to obtain local optimum samples, and corresponding metamodel is constructed by the LS-SVR based on these samples. Finally, the drawbead design is successfully optimized by the proposed metamodeling technique. The optimization results demonstrate that the MPS-based SVR is feasible for real engineering problems.