Preventive and condition-based maintenance of rail infrastructure is an important aspect in reducing interruptions and delays in train operations. In case of optimal implementation, it even helps to lower the overall maintenance costs by avoiding expensive instant repairs of sudden failures including possible incidental damages. For being effective in this context, asset managers need to estimate not only the current state of the rail infrastructure and its components but they also need to predict future conditions based on available data and measurements. Stochastic modelling has shown to be a promising way for tackling these tasks. However, uncertainties of the model results need to be evaluated then in order to make maintenance planning as solid as possible. Commonly, Monte Carlo (MC) simulations are used for analyzing the stochastic distributions of the model outputs whenever analytical solutions are not possible or difficult to obtain. In contrast to that, an interesting alternative for numerically deriving important statistical quantities related to the model results (such as mean or standard deviation) is given by the so-called Point Estimate Method (PEM). Depending on the details of the model under consideration, PEM can be shown to be even exact under certain constraints while often requiring just a small number of sample points to be evaluated. In contrast to that, the MC approach naturally yields approximate results only with a potential need for several hundred or thousand sample points in order to converge. The present contribution shortly reviews the mathematical background of PEM before demonstrating its performance based on three more or less academic examples from the wide field of railway asset management: i) track degradation, ii) reliability analysis of composite systems, iii) failure detection/identification using decision trees. Finally, advantages as well as limitations of the PEM approach in comparison to common MC simulations are discussed.
The different versions of the original document can be found in: