2017 年 55 巻 6 号 p. 289-296
Collection of big data for corrosion under insulation (CUI) of pipes has been a major concern in maintenance management. It is expected that new inforamtion is extracted from the collected data. To extract the statistical information from the data, statistical treatment should be done carefully. This paper investigates statistical treatment of metal loss growth rate of CUI (hereafter referred as CUI rate) for the collected data under the project of Ministry of Economy, Trade and Industry (METI) . The variability of CUI rate is normally modelled as Gumbel distribution. However, it is often reported that the variability of CUI rate depends on the inspection period. Especially, the variability in CUI rate in shorter inspection period prones to increase. It is supposed that the CUI rate data in longer inspection period has been used for the evaluation. However, it is clear that the statistical treatment is important for the accurate estimation of CUI rate. This paper shows that the variability of the wall thickness is not negligible for the statistical evaluation of CUI rate and proposes a new statistical model including the variation of thickness. When the nominal thickness value is used as the initial thickness, the evaluated CUI rate includes the uncertainty of the thickness. Thus, the evaluated CUI rate is not the true value, but only apparent value. The validity of the model is confirmed for the data in database, and by Monte Carlo Simulation. Finally, it will be shown that the apparent CUI rate is expressed mathematically using moment generating function.