A Statistical Review of Experimental PV Burst Data Available In the Literature

Data on the consequences of fires and explosions can be expensive and difficult to obtain. When limited data are available, it is good practice to correlate the data with a 90 to 100 percentile curve (90 to 100% of the data falling on the conservative side of the correlating curve). Searching the literature for PV burst overpressure data, we have located over 400 measurements published between 1965 and 2007. These data span the overpressure range of interest for hazards assessment. Statistical analysis of this dataset shows that, for the subset of data on ground level burst of cylindrical vessels, the influence of parameters other than scaled distance can be simply considered as a lumped random error term. This error term follows a normal distribution, indicating that the data show no correlation with parameters other than the scaled distance. Given the quantity of data assembled, a correlation can be developed by doing a simple statistical curve fit of the data. The resulting curve can be scaled, based on the levels of conservativeness desired, to fall above 50 to 100% of the data points. Given the statistical properties of the dataset, we propose a 50 percentile correlation of the ground level cylindrical vessel data.