A cumulative damage approach to modeling atmospheric corrosion of steel

Date of Award

2014

Degree Name

Ph.D. in Materials Engineering

Department

Department of Chemical and Materials Engineering

Advisor/Chair

Advisor: Douglas. C. Hansen

Abstract

Past attempts to develop models that predict atmospheric corrosion rates used statistical regression, power-law", or other approaches that result in linear or simple nonlinear corrosion rate predictions. Such models were calibrated by statistically comparing corrosion test results to predictions based upon long-term (e.g., annual) deposition measurements of chloride aerosols and/or SO2. Relative humidity, if explicitly considered, was only used to define the amount of time during the year when conditions were thought to be favorable for corrosion. Most models ignored temperature effects but those that do only consider annual averages. A new approach was constructed to predict corrosion rates using the concept of cumulative damage. This new model is analogous to some types of fatigue models and is based upon the Eyring equation, which was originally developed to predict the dependence of chemical reaction rates on levels of the presumed acceleration factors. The model makes hourly weight loss predictions, which when added together makes longer-term "cumulative" predictions. Principal advantage of using hourly predictions is that the effects of diurnal and seasonal temperature cycles and related changes to relative humidity are explicitly considered. The stochastic nature of atmospheric contaminants is considered as well. An inverse modeling approach using Monte Carlo simulations was used to fit various candidate models to proxy environmental characterization data representing conditions at corrosion test sites. Proxy data (measured elsewhere and for other purposes) was used to infer the stochastic environmental severity at sites where corrosion tests were conducted. Such data included hourly SO2 and ozone data obtained from the Environmental Protection Agency's Air Quality System database, longer-term chloride deposition data from the National Atmospheric Deposition Program's on-line database, and hourly weather data from the U.S. Air Force's 14th Weather Squadron. Proxy data was used so that if this current research proved successful, follow-on work could lead to a practical methodology that design engineers could employ to make realistic predictions without having to explicitly characterize the environment at a location of interest. Cumulative predictions made using such data were statistically compared to quarterly corrosion test results that came from a DoD Strategic Environmental Research and Development Program (SERDP) funded effort and other related programs that used the same testing protocols. Billions of simulations were conducted whereby coefficients employed by the candidate model were randomly varied and the individual predictions statistically compared to test measurements in order to identify the most accurate model. Each candidate model was calibrated by considering hourly data for an entire year at multiple locations in order to quantify interactions between acceleration factors. The degree of fit between the model results and test measurements at the calibration sites was very high (R2=̃ 0.99). When the optimum model was applied to locations where corrosion tests were conducted but not used for calibration, the fit was not quite as good, but was still quite high (R2=̃0.86). Analyses were conducted to identify ways to further improve accuracy, thus laying the framework for future efforts."

Keywords

Steel Corrosion Simulation methods, Materials Science, Aerospace Materials, Automotive Materials, atmospheric corrosion, cumulative damage, mass loss, modeling, predictions, environmental severity, climate zones, surf zones, urban pollution, Monte Carlo simulations

Rights Statement

Copyright © 2014, author

Share

COinS