Improving scheduled maintenance by missing data reconstruction: A double-loop Monte Carlo approach

Abstract

This article describes a Monte Carlo–based approach for reconstructing missing information in a dataset used by General Electric for reliability analysis, which contains data coming from field observations at inspection of gas turbine components. The approach is based on a combination of maximum likelihood estimation technique to estimate the failure model parameters, Fisher information matrix to estimate the confidence intervals on the estimated parameters, and a double-loop Monte Carlo approach to estimate the missing equivalent starts (i.e. data of turbine state without the relative equivalent starts). The proposed methodology reduces the uncertainty in the estimation of the parameters of the turbine. The results of the application of the novel approach to a real industrial dataset are discussed along with a sensitivity analysis for the quantification of the robustness of the methodology to deal with different sizes of datasets.