Several Ensemble Modeling Approaches Based On Model Selection Strategy For Reliability-based Design Optimization Under Model Uncertainty

The model uncertainty is widely existed in the process modeling of the quality characteristic for the sophisticated mechanical product,since internal and external factors in manufacturing process affect the precision of experimental data.At present,the ensemble modeling approach has been widely used for their capability of high computational efficiency and low computational cost.However,existing studies only combine all candidate surrogate models via linear weighted sum method without considering the model uncertainty,which lead to inferior candidate models in the ensemble modeling process.As a result,they may lead to an inaccurate quality estimation,which undermine the accuracy of ensemble models and affects the reliability assessment and the optimal result.Aiming at addressing the drawback of exising ensemble approaches,the model slection strategy is adopted where redundant models are eliminated before constructing an ensemble model.With the purpose of constructing an accurate and robust ensemble model for the reliability-based design optimization,a new ensemble modeling approach is proposed to select potential superior models for the construction.By adopting the model selection strategy,redundant models are eliminated from a given set of candidate models,and then remaining candidate models are constructed via the heuristic scheme.Several major research tasks and corresponding conclusions from this dissertation are summarized as follows:(1)The existing RBF ensemble model for process modeling of the sophisticated mechanical product combines all of candidate RBF models as a final model.The existing RBF ensemble model does not account for any uncertainty in the form of the kernel function.To overcome the issue,this dissertation proposes a robust RBF ensemble modeling approach that considers the uncertainty in the model form of the kernel function.The Stepwise strategy is adopted to the RBF model selection where redundant RBF models are eliminated before constructing a robust RBF ensemble model.The proposed RBF ensemble modeling approach is applied to a bridge-type amplification mechanism to illustrate its effectiveness.The results reveal that proposed RBF ensemble modeling approach not only improve predictive accuracy,but also enhance the predictive robustness.(2)The existing Kriging ensemble model for process modeling of the sophisticated mechanical product combines all of candidate Kriging models as a final model.The existing Kriging ensemble model does not account for any uncertainty in the form of the correlation function.To conquer the issue,this dissertation proposes a robust Kriging ensemble modeling approach that considers the uncertainty in the model form of the correlation function.0-1 programming is adopted to the Kriging model selection where redundant Kriging models are eliminated before constructing a robust Kriging ensemble model.The proposed Kriging ensemble modeling approach is applied to a circular parallelogram mechanism to illustrate its effectiveness.The results reveal that proposed Kriging ensemble modeling approach not only improve predictive accuracy,but also enhance the predictive robustness.(3)The existing ensemble model for process modeling of the sophisticated mechanical product combines all of candidate models as a final model.The existing ensemble model does not account for any uncertainty in the form of the model.To conquer the issue,this dissertation proposes a robust ensemble modeling approach that considers the uncertainty in the model form of the model.Bayesian model averaging is adopted to the model selection where redundant models are eliminated before constructing a robust ensemble model.The proposed ensemble modeling approach is applied to a corner-filleted parallelogram mechanism to illustrate its effectiveness.The results demonstrate that proposed ensemble modeling approach is robust and efficient in dealing with different kinds of black-box problems in terms of accuracy,efficiency,robustness.(4)The existing surrogate models for the limit state function of the sophisticated mechanical product could generate inaccurate quality estimation.As a consequence,it may result in a failed design during the subset simulation of a reliability model and the reliability-based design optimization(RBDO).Based on the proposed ensemble modeling approaches,this dissertation presents a new RBDO scheme for the the sophisticated mechanical product by integrating the desirability function approach to deal with multiple responses optimization.The proposed ensemble modeling approach is applied to a corner-filleted parallelogram mechanism to illustrate its effectiveness.The results reveal that the ensemble modeling approach for the subset simulation of a reliability model and the RBDO can provide a deasible and efficient way in actual engineering application for the sophisticated mechanical product.