Statistical model for assessing the reliability of non-destructive testing systems by solving inverse problems

Abstract

Objectives. The wear monitoring of metal structural elements of power plants—in particular, pipelines of nuclear power plants—is an essential means of ensuring safety during their operation. Monitoring the state of the pipeline by direct inspection requires a considerable amount of labor, as well as, in some cases, the suspension of power plant operation. In order to reduce costs during monitoring measures, it is proposed to use mathematical modeling. This work aimes to develop a mathematical model of a diagnostic system for assessing the probability of detection of defects by solving inverse problems.Methods. A binomial model for assessing the reliability of monitoring, comprising the Berens-Hovey parametric model of the probability of detection of defects and a parametric model based on studying test samples, was analyzed. As an alternative to this binomial model, a computational method for assessing the reliability of non-destructive testing systems by solving an inverse problem was proposed. To determine the parameters of the defect detection probability curve, the model uses data obtained by various monitoring teams over a long period of power plant operation. To serve as initial data, the defect distribution density over one or more of the following characteristics can be used: depth, length, and/or cross-sectional area of the defect. Using the proposed mathematical model, a series of test calculations was performed based on nine combinations of initial data. The combinations differed in the confidence coefficient of the initial monitoring system, the parameters of the distribution of defects, and the sensitivity of the monitoring system.Results. The calculation data were used to construct curves of the probability density of detected defects as a function of the defect size, recover the values of the defect distribution parameters under various test conditions, and estimate the error of recovering the parameters. The degree of imperfection of the system was estimated using the curve of the detection probability of a defect by a certain monitoring system.Conclusions. Under constraints on the data sample size, the proposed methodology allows the metal monitoring results to be applied with greater confidence than currently used methods at the same time as evaluating the efficiency of monitoring carried out by individual test teams or laboratories. In future, this can be used to form the basis of a recommendation of the involvement of a particular team to perform diagnostic work.