Multifidelity co-kriging metamodeling based on multivariate data fusion for dynamic fit improvement of injection mechanism in squeeze casting

To improve the dynamic fit the injection mechanism in a squeeze casting machine, this paper proposes a novel multifidelity co-kriging (MFCK) metamodeling method, which fuses high-fidelity measured data with low-fidelity simulated data and considers data uncertainty and multivariate correlation influence to accurately predict response values when experimental sample data are insufficient. An MFCK model was established to predict the deformation and dynamic fit clearance, by selecting experimental and simulated values of deformation and temperature as the principal and covariates for correlation testing. The results indicate that the proposed MFCK model significantly improved the prediction accuracy by 34.18 %, 73.53 %, 41.57 % and 37.93 %, respectively, compared with the ordinary kriging model and finite element method. This method was applied to the multicycle injection process of a 2,500-kN squeeze casting machine, revealing the variation law of the fit clearance. The MFCK model improved the prediction accuracy of the fit clearance by 72.7 %, which is beneficial for process control. The accuracy and industrial applicability of the proposed MFCK method was thus verified. • Integrating simulation data into sparse experimental data to establish a multifidelity co-kriging metamodel. • Metamodeling for global prediction of an injection mechanism. • Considering multivariate correlation and data uncertainty to accurately predict the deformation and temperature. • Revealing the variation rule of the dynamic fit clearance in the injection mechanism of a squeeze casting machine.