|
|
|
|
|
All corrosion phenomena are inherently variable in their initiation and propagation, even under the best of circumstances such as in a single heat of homogeneous material and tested under well controlled conditions. There is nothing about corrosion that is deterministic. While this does not imply that the mean behavior cannot be reasonably modeled, it does mean that there is inherent variability in the initiation and propagation of all corrosion phenomena.
In addition to corrosion phenomena being inherently variable, even the same material purchased with the same specifications is variable according to inevitable variations in processing during initial preparation of the material and in its fabrication. This is the heat-to-heat or lot-to-lot variability.
The third level of variability is associated with environments in general and of local surface environments in particular.
It is necessary, then, to accept the reality that predicting performance must account for these three levels of variability.
When it is suggested that the statistical nature of corrosion needs to be quantified, visions of an unending number of tests and specimens seem to appear. This does not follow. An approach to developing predictions based on practical applications of Weibull statistics is described by Abernethy [11].
Without a statistical approach to testing and analysis, corrosion phenomena seem disconnected and scattered. With a statistical approach, what might have been considered "scatter" becomes coherent and can be expressed as a part of the Weibull slope (in the Weibull framework); this slope is also known as the "dispersion." Conventionally, results from corrosion testing are analyzed to obtain an average value, and a range of results is often expressed. This range is, in fact, the data that can be used to evaluate the statistical dispersion.
An important result of statistically based testing, and using the Weibull based analysis as described by Abernethy [11], is the capacity to identify the early failures. While the mean value gives a central characteristic of the data, the dispersion indicates when the early failures can occur. For example, with a relatively small number of specimens in test, it is possible to determine a Weibull slope that can be legitimately extrapolated to small fractions of elements. Thus, it is possible to estimate when 0.01% of elements might fail, and this would guide the approach to monitoring and inspections.
In general, accelerated testing using more intense stressors, e.g. stress, temperature, and environmental concentrations, produce less dispersion of data than found in engineering service. This means that while acceleration may be obtained while comparing mean values between the accelerated test and the application, at the fraction failed of 0.01%, the amount of acceleration could be negligible. This is a consequence of the application, with its lower stressors, having greater dispersion of data than the accelerated test having higher stressors and a more narrow dispersion.
Lifetime Prediction, Roger W. Staehle, Adjunct Professor, Department of Chemical Engineering and Materials Science, University of Minnesota, Staehle Consulting Co.