To reliably calibrate suitable partial safety factors, useful for the specification of global condition describing structural safety level in considered design case, usually the evaluation of adequate failure probability is necessary. In accidental fire situation, not only probability of the collapse of load-bearing structure, but also another probability related to the people staying in a building at the moment of fire occurence should be assessed. Those values are different one from another in qualitative sense but they are coupled because they are determined by similar factors. The first one is the conditional probability with the condition that fire has already occured, whereas the second is the probability of failure in case of a potential fire, which can take place in the examined building compartment, but its ignition has not yet appeared. An engineering approach to estimate such both probabilities is presented and widely discussed in the article.
An advanced evaluation technique, helpful in the fire resistance assessment of a simple steel structure exposed to fire is presented and discussed in detail on the example of an unrestrained and uniformly heated steel beam. The proposed design methodology deals with the generalised probability-based approach in which the most probable failure point is formally identified. The random nature of all variables considered in the detailed analysis is taken into account. The critical temperature of the steel from which the considered beam is made of is accepted here as the authoritative safety measure. This temperature value is associated with the fire resistance limit state defined for the maximum acceptable value of failure probability. When forecasting the failure probability, not only the risk of a potential fire being initiated but also not being effectively extinguished is included in the calculation. Various levels of the target failure probability may be assumed in such the analysis, depending on the selected reliability class. They are specified in general by setting an appropriate value of the required reliability index β fire req. In the presented design algorithm no representative values of the considered random variables are specified. The critical temperature estimates obtained from these calculations are always less restrictive in comparison with the corresponding solutions computed after applying the conventional standard procedure.