Large industries such as mining, jute and sugarcane and steel plants need to use rail networks for transportation of material over wide geographically distant areas. Maintenance strategy of rail network is developed by understanding reliability of rails used in the rail track system. Reliability analysis of rails can be carried out by understanding the failure mechanism of rail through modelling and analysis of failure data. These failure data are time or usage dependent for certain conditions. In a probabilistic sense, rail failure is a function of its usage in terms of Million Gross Tonnes (MGT) for certain conditions. This paper is to analyse real life rail industry data, deal with the limitations of available data and develop predictive models for maintenance and replacement decisions. Parameters of the model are estimated using real world data with an application of non-homogeneous Poisson process.