Bayesian support vector machine for optimal reliability design of modular systems

Abstract

In a modular system, uncertainties will spread among coupled modules and cause system failure. To cope with this issue, the reliability-based design optimization (RBDO) of modular systems came into being. However, the solution of this design task is a nested triple-loop process, making the computational burden unaffordable for real-world systems. Thus, this paper endeavors to effectively mitigate this computational effort. The individual module feasible approach is first proposed to tackle the coupling effects of modules, whereby, the original optimization problem is converted into a conventional one. Then, the Bayesian-inference-based support vector machine is utilized to build the alternative model for the actual probabilistic constraint function, in the augmented reliability space. The alternative model is constructed using small number of model evaluations, which possesses enough precision everywhere in the augmented confidence region. Finally, the optimal decision scheme is obtained by solving the formulated conventional RBDO using the alternative model. The performance of the proposed method is investigated using several examples.