Reliability Prediction Of Highly Safe Mechanical Systems Using Response Surface Assisted Tail Modeling

Abstract

Tail modeling method is a popular method for predicting reliability of highly safe mechanical systems. In classical tail modeling, only the limit-state values that belong to the tail part are used in tail model fitting, so majority of limit-state evaluations are wasted. This paper proposes response surface (RS) assisted tail modeling (RSATM) technique, where the limit-state evaluations are steered towards the tail part by using RS to reduce the amount of wasted data. The overall computational cost of RSATM is composed of two components: the cost related to RS model construction, and the cost related to tail modeling. To increase the accuracy of the RSATM, the computational budget should be allocated in a proper way. Two structural mechanics example problems are used to investigate the computational budget allocation in RSATM. It is found that the number of limit-state function evaluations allocated for RS construction can be limited between 1.5 and 2 times the number of coefficients in the RS model, and the rest of the computational budget can be allocated to tail modeling.

Keywords

• Reliability prediction
• Mechanical systems
• Tail modeling
• Response surface method