Difficulties for performing life prediction analysis when corrosion is
- The seeds of some failures often exist before steady state operation
begins and result from lack of attention to precedent details.
- Conditions change with time, and nominally unexpected circumstances
develop for which there were no initial considerations.
- Multiple and explicitly different environments often occur on the same
subcomponent; further, multiple modes of corrosion may occur at each of
these multiple locations.
- Contaminants that produce corrosion often arise from other sources than
the component being considered. Such contaminants in steam power plants
result from: copper alloy condensers that produce copper ions in solution;
leaks in turbine seals that allow oxygen access; release of lead ions from
bearings; leaks in condenser tubes that allow chloride to enter.
- Predicting performance is an interdisciplinary project; often the
necessary disciplines are not involved.
- The schedule for producing the product often does not allow for a serious
assessment of performance; further, there is often insufficient money to
undertake both the analysis and necessary testing.
- Field experience is often not consulted and properly analyzed; on the
other hand the applicability of satisfactory field experience is sometimes
erroneously assumed to justify designs despite different, even slightly
different, conditions of exposure.
- Corrosion-related failures are inherently statistical as they involve
diverse inputs from the material and environment. For these reasons, even
under the best of circumstances, corrosion-related failure phenomena exhibit
"broad scatter" in their occurrences. Such scatter is often
erroneously regarded as just bad data rather than an inherent property of
- There are many who argue that responsible predictions are not possible
without an absolute understanding of fundamental atomic processes. This is
neither so nor generally possible. Ultimately, what many claim to be atomically
based predictions are nothing more than correlations with multiple adjustable
parameters. A more honest approach, and more reasonable, is simply to accept the
fact that good correlations are usually adequate if they are responsibly
Lifetime Prediction, Roger W. Staehle,
Adjunct Professor, Department of Chemical Engineering and Materials Science,
University of Minnesota, Staehle