Development of a deterministic rail-wear prediction model

The management of the rail infrastructure in New Zealand has recently reverted back to public ownership and is now managed by the New Zealand Railways Corporation (NZRC). Current legislation requires the NZRC to produce long-term strategic plans for the management of the rail track infrastructure, in order to demonstrate proper management of the public asset. However, long-term strategic planning requires knowledge of not only the current condition of the asset, but also its future condition as determined from performance modelling, something previously not available for the rail infrastructure in New Zealand. This prompted a research study into the development of a deterioration model for New Zealand's rail track infrastructure. To date, the study has focused on rail wear; rail being one of the most expensive track renewal items in New Zealand. This paper describes the development of predictive models for 50 kg/m and 91&90 lb/yd rails using the "R" statistical software originally developed at the University of Auckland. A multiple regression analysis produced models with reasonably high coefficients of determination for both 'side wear on the high leg' and 'top wear on the low leg' on 50 kg/m curved tracks and for total wear on both high leg and low leg for 91&90 lb/yd rails. Both are 'absolute models', as the accuracy of the data collection techniques employed did not permit the development of an incremental model. Nonetheless, these will undoubtedly prove valuable to decision makers. However, the variability in the data suggests that a probabilistic model would be more appropriate than a deterministic model at the network level. In particular, rail wear would be better represented at the network level by a stochastic process such as the Markov model. Such a model is now in the process of being developed and will be reported in future publications.