Condition monitoring of natural gas distribution networks is a fundamental prerequisite for evaluating safety of the operation during the lifetime of the system. Due to the high level of uncertainty in the observed data, predicting the operational reliability of the networks is complicated. Moreover, there is a fluctuation in most of the monitoring data in different time scales, as most of the derived data tend to be of non-stationary nature and are complex to model or forecast. Therefore, a more realistic data driven approach for developing a reliability framework needs to be considered. This paper aims at proposing a probabilistic model to predict the complexity of the non-stationary behaviour in monitoring data. It also aims at developing a novel framework for the time dependent reliability assessment of a natural gas distribution system using condition-monitoring data. To this end a methodology by integrating Empirical Mode Decomposition (EMD) and Hierarchical Bayesian Model (HBM) is developed. The advantages of the methodology are demonstrated through a case study of a Natural Gas Regulating and Metering Station operating in Italy. Based on pressure data acquired from the case study, the model is able to predict overpressure thus directly avoiding unnecessary maintenance and safety consequences. (C) 2019 Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.