Abstract: In lightweight optimization design of aluminum alloy frame, a surrogate model is usually constructed to replace its finite element model for reducing computation effort. For further enhance computation efficiency so as to establish a high-accuracy surrogate model with lesser sample points, an improved sequential sampling method named CVDA is proposed and compared with three existing sampling methods (OLHD, MDA and CVA) through a calculation sample. The results show that the prediction accuracy of surrogate model built with CVDA sampling is 11.1%, 11.1% and 30.3% higher than that with OLHD, MDA and CVA sampling methods respectively with a R2 value reaching 0.95. Then the CVDA method is applied to building the surrogate model of aluminum alloy frame in bending and torsion conditions, and with the same number of sample points, the prediction accuracy of surrogate model built with CVDA method is apparently higher than that with other sampling method. On this basis, a lightweight optimization is conducted on aluminum alloy frame, and as a result, the bending stiffness, torsional stiffness, 1st order bending modal frequency and 1st order torsional modal frequency increase by 2.54%, 3.38%, 0.68% and 2.75% respectively, and the mass reduces by 7.47 kg with a mass reduction rate of 7.20%, showing an obvious optimization result.

Key words: aluminum alloy frame, lightweight optimization design, sequential sampling, surrogate model