The purpose of this discussion is to describe an approach to predicting the earliest failures by corrosion and to compare this with conventional approaches that are largely based on evaluating the mean value of data or on using only engineering judgment. The essence of the approach proposed here involves using a statistical framework typified by the Weibull distribution with its statistical parameters quantified with physical quantities from the seven principal variables that control corrosion: e.g. pH, potential, species, metal composition, metal structure, temperature, and stress. The most important parameter affecting early failures is the shape parameter...