Uncertainty Assessment of Component Degradation Model Based on Support Vector Regression

Degradation modeling and condition assessment of critical components are important issues in the maintenance of nuclear power plant, but modeling uncertainties must be taken into account seriously by considering the stochastic nature of degradation and observation process. Based on support vector regression algorithm, this article proposes a wall thinning model for carbon steel pipes in a nuclear power plant using in-service inspections data and further performs the uncertainty quantitive assessment for the proposed model. In the beginning, Latin hypercube sampling method is used to create new sample sets from the original observation with certain distribution of the mean values which are assumed from the observed data. Furthermore, part of the reconstructed sample sets are chosen as training sets to develop a wall thinning model and the remaining samples are used as test sets to verify the model. By comparing model predicted wall thickness values of the test sets and the observed values, a quantitative assessment of the degradation model uncertainty is obtained. The obtained results demonstrate that the deviations between observed thickness values and average model predicted values fluctuate around 1%, while model predicting variances are much smaller than the observed variances. This report concludes that the proposed support vector regression model for component degradation can provide accurate condition assessments with rather small variance.