A Bayesian block maxima over threshold approach applied to corrosion assessment in heat exchanger tubes

TL;DR

This paper introduces a Bayesian block maxima over threshold method for more accurate corrosion severity estimation in heat exchanger tubes, addressing the challenge of non-extreme data points in EVT analysis.

Contribution

It proposes a novel threshold-based EVT approach that improves inference robustness for corrosion assessment in heat exchangers.

Findings

Enhanced model robustness for corrosion data analysis

More accurate maximum pit depth predictions

Addresses non-extreme data in EVT context

Abstract

Corrosion poses a hurdle for numerous industrial processes, and though corrosion can be measured directly, statistical approaches are often required to either correct for measurement error or extrapolate estimates of corrosion severity where measurements are unavailable. This article considers corrosion in heat exchangers tubes, where corrosion is typically reported in terms of maximum pit depth per inspected tube, and only a small proportion of tubes are inspected, suggesting extreme value theory (EVT) as suitable methodology. However, in data analysis of heat exchanger data, shallow tube-maxima pits often cannot be considered as extreme; although previous EVT approaches assume all the data are extreme. We overcome this by introducing a threshold - suggesting a block maxima over threshold approach, which leads to more robust inference around model parameters and predicted maximum pit depth.