Probabilistic design methodology of static systems using metamodels

The mechanistic model developed to model a physical system is, in many cases, too complex to be used in methods that aim to improve the performance of the system. In probabilistic design, where design variables are stochastic in nature, the most popular method to search for the best design is the Monte Carlo Method. However, as the complexity of the mechanistic model increases, so does the CPU time for the Monte Carlo method. Recent research shows how complex mechanistic models are being replaced by approximating functions, known as metamodels. This paper investigates the use of metamodels to replace mechanistic models in the probabilistic design of systems with static response. From reliability analysis theory, the First Order Reliability Method (FORM) is used to calculate the best design when provided with design specification. The speed and accuracy of three popular metamodels, the linear response surface model, the Radial Basis Function and the Kriging model are compared.