Abstract
This paper presents an analysis of promoted ignition test data based on formal statistical techniques. Logistic regression is identified as an appropriate method for modeling binary burn–no-burn test data. Logistic regression is applied to the data sets from promoted ignition testing of two Hastelloy® alloys, C-276 and G-3, previously published by Zawierucha et al., ASTM STP 1111, ASTM International, West Conshohocken, PA, 1991, pp. 270–287. The logistic regression model is used to predict the burn probability over a range of pressures in the nonflammable and flammable domains, and was shown to fit the raw data well. Confidence intervals for the model are also determined, allowing the uncertainty associated with the predicted burn probabilities to be quantified. The relationship between the amount of test data available and confidence levels is discussed, and recommendations are made to improve the current standard test methodology. In particular, more repeated tests are required at all pressure levels considered, and tests are also required at pressures in the nonflammable domain (in addition to the flammable domain). It is shown that there is a low level of statistical confidence associated with the definition of threshold pressure under the current methodology. It is concluded that the use of the logistic regression model is beneficial since it allows for the quantification of burn probabilities and confidence levels, which enables the definition of threshold pressure to be related to a critical burn probability threshold.
