Reliability-based design optimization of hinge sleeve using adaptive E-SVM

To achieve efficient reliability analysis and effective weight reduction of hinge sleeve, a reliability-based design optimization (RBDO) method based on the Bayesian optimization algorithm and adaptive ensemble of support vector machine with different kernel functions (E-SVM) is proposed in this paper. Firstly, the Bayesian optimization algorithm is used to search the optimal ensemble weights of the polynomial and Gaussian kernel functions in SVM. Based on this, the two kernel functions are linearly weighted, and the optimal hyperparameters of the SVM model are determined using the K-fold cross-validation to ensure the robustness of surrogate modeling. Secondly, an adaptive reliability analysis method for the E-SVM model is used, which quantifies the uncertainty of the empirical estimation of failure probability by calculating the bounded failure probability range. In each iteration, information parameter point is selected and added to the training sample set to achieve adaptive update of the SVM model. The effectiveness of the proposed method is verified by numerical examples and the application of the automobile front axle. Finally, the proposed method is applied to the lightweight design of hinge sleeve. Under the reliability requirements of stress, deformation, and lifespan, the weight of the hinge sleeve after optimization is reduced by 202.6 kg, achieving a weight reduction ratio of 3.3%.