STP1137: A New Approach to Power-Model Regression of Corrosion Penetration Data

    McCuen, RH
    Professor, University of Maryland, College Park, MD

    Albrecht, P
    Professor, University of Maryland, College Park, MD

    Cheng, J
    Professor, Nanjing Institute of Chemical Technology, Nanjing,

    Pages: 31    Published: Jan 1992


    Corrosion penetration data have traditionally been fit with the power model using a log transformation of both the exposure time and the penetration. The transformation has several disadvantages, including (1) the resulting model gives biased estimates of the penetration, (2) the sum of the squares of the errors in penetration are not minimized even though the sum of the squares of the logarithmic errors are minimized, and (3) the logarithmic transformation results in greater emphasis being placed on the penetration for the shorter exposure times.

    An alternative method of fitting power models is discussed. The numerical calibration method is shown to produce better fitting and more rational power models than is provided with a logarithmic transformation. Unbiased power models can also be fitted. The fitting procedure is more complex than that for the log transform method.

    The two fitting methods were compared using 32 sets of corrosion penetration data. The average standard error ratio for the numerical procedure was 58 percent of that for the logarithmic analyses, which suggests that the numerical procedure consistently produced better penetration estimates for the largest measured exposure time. Thus, more accurate 75-year projections of penetration should be expected with the numerical model. Analyses of the records suggest that record lengths of at least 10 years are necessary to produce accurate model coefficients.


    Atmospheric corrosion, penetration, structures, weathering steel, power model, numerical optimization

    Paper ID: STP19754S

    Committee/Subcommittee: G01.04

    DOI: 10.1520/STP19754S

    CrossRef ASTM International is a member of CrossRef.