Experiments, science and probability theory in building new models for long-term pitting corrosion in marine environments

This paper describes some aspects ofthe processes involved in the development of high-quality probabilistic models for corrosion loss and for maximum pit depth for steels exposed to marine environments. Such models increasingly are required for advanced infrastructure life-cycle management to predict likely future levels of deterioration and for the setting of rational inspection regimes. Most models currently available are empirical. They have high levels of uncertainty. The models described herein are based on appropriate levels of representation of underlying corrosion science fundamentals, including the recognition of the important role of microbiological activity and model calibration using a probabilistic approach to data interpretation. This requires careful interpretation of existing literature data and has required obtaining new data to elucidate particular aspects. A brief overview of this work is given with emphasis on the modeling of maximum pit depth using extreme value distributions. Reinterpretation of existing pit depth data was found to be necessary for consistency with new observations of pit depth development in actual in-situ tests (as distinct from laboratory observations). As a result, the new model provides maximum pit depth as a function of time based on understanding of the underlying pitting process with time. It has shown also that long-term pit depth is most appropriately represented by the Frechet extreme value distribution since this is consistent with the underlying corrosion and bacterial processes. This new interpretation of data provides estimates of pit depth uncertainty as a function of exposure time. It is shown that this can be applied even for existing data sources that were not designed to elucidate probabilistic estimates of uncertainty. The conventional application of the Gumbel extreme value distribution can lead to serious errors in long-term extreme pit depth predictions.