RELIABILITY ANALYSIS OF COMPARTMENT FIRE BY IMPROVED RESPONSE SURFACE METHOD

Lacking of explicit limit state function (LSF) is a problem in the reliability analysis of compartment fire, especially when the reliability calculation is coupled with fire dynamic analysis. A new response surface (RS) method is proposed from the view point of experiment design and multi-variable regression. Uniform design (UD) method is introduced to determine the simulated input parameter levels. Alternating Conditional Expectations (ACE) technique is applied to find the RS function. UD can allocate more parameters and more levels for each parameter without greatly increasing the calculation efforts. ACE can result a better regression for complex response surface. Coupling these two methods can achieve an efficient technique for constructing RS function. Based on RS, two reliability algorithms are proposed. A comparison with direct Monte Carlo simulation (DMCS) indicates that the proposed methods are efficient in reliability calculation.