Aggregate multiple radial basis function models for identifying promising process parameters in magnetic field assisted laser welding

摘要

The external magnetic field assisted laser welding (MFA-LW) exhibits great potential for reducing welding defects commonly encountered in laser welding (LW) process such as spatters, pores, and weld bead depressions. The quality of a welding joint of MFA-LW is closely related to the bead geometrical characteristics which are primarily determined by the process parameters. Since the relationships between the bead geometrical characteristics and process parameters cannot be expressed explicitly, it is impractical to identify promising process parameters intuitively. This paper proposes a method to aggregate multiple radial basis function models (MRBFs), including RBF models constructed with the Gaussian basis function, Thin-plate spline basis function, and Cubic basis function, for establishing relationships between the bead geometrical characteristics and process parameters. In the aggregating process, the weighting coefficients associated with each RBF model are determined according to their point-wise accumulative leave-one-out (LOO) errors. Accuracy comparison results show that the proposed MRBFs can take advantage of the prediction ability of each stand-alone RBF model and thus decrease the risk of adopting inappropriate RBF models. Furthermore, a genetic algorithm (GA) is adopted to identify promising process parameters, in which the fitness evaluations are replaced by the constructed MRBFs. Validation experiments at the obtained promising process parameters illustrate that the bead geometrical characteristics of experiments are in good agreement with the ones from the proposed approach. Therefore, the proposed approach can be used to identify promising process parameters for MFA-LW in practical production. (C) 2017 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.