Explosive Damage To Industrial Buildings: Assessment By Resampling Limited Experimental Data on Blast Loading

Assessment of damage to industrial buildings due to accidental explosions in air is considered. It is suggested to formulate the result of the damage assessment in the form of risk. The expression of risk embraces probabilities of foreseeable damage events (damage probabilities) and outcomes (consequences) of suffering these events. The situation is examined when blast loading imposed by an accidental explosion is predicted by a series of experiments. They yield a small‐size sample of blast loading characteristics. It is suggested to idealise the formation of explosive damage to industrial buildings by means of event trees diagrams. A quantitative analysis of these diagrams can be carried out by developing fragility functions for their branching points. Each branching point is used to represent a structural failure contributing to the final explosive damage. The fragility functions are applied to expressing the conditional probabilities of explosive damage. With these probabilities, a technique of frequentist (Fisherian) inference is applied to assessing the explosive damage. This technique is called statistical resampling (Efron's bootstrap) and applied as a practical, albeit not equivalent alternative to the Bayesian approaches. It is shown that statistical resampling is capable to yield confidence intervals of damage probabilities and can be applied almost automatically. It operates without using cumbersome methods of statistical inference developed in the classical statistics. The bootstrap confidence intervals do not contain any subjective information except the degree of confidence for which these intervals are computed. The degree of confidence must be chosen by the engineer. The bootstrap confidence intervals are applied to estimating damage probabilities on the basis of the small‐size sample of blast loading characteristics. An estimate of the risk of explosive damage is expressed as a set of bootstrap confidence intervals computed for damage probabilities and related outcomes of this damage.