Abstract:
Rail renewal is one of the most expensive track management expenditures in New Zealand. This research developed both project level deterministic and network level stochastic rail wear models using local data collected over a period of 10 years. These models will facilitate efficient long-term planning of rail renewal actiyities and have the potential to produce significant economic benefits. Deterministic rail wear models were developed using multivariate regression analysis for High Leg Side wear, and Low Leg Top wear for 50kg/m rails, and High Leg Total wear and Low Leg Total wear for 91&90lb/yd rails for curved tracks on the Class-A Principal Lines network. The models express rail wear as a function of rail age, annual gross tonnage and rail/wheel angle of attack. Through an 'expert opinion survey', it was verified that these models capture only a subset of factors that affect the rate of rail wear. A comparison of model predictions with actual data was not undertaken. However, they have R2 values ranging from 0.6 to 0.8, and therefore they can provide very useful results for track asset management. The deterministic models produced are 'simple and ready to use', and enable a whole-of-life cost approach to be taken in rail track renewals at project level. The main feature of this research is the introduction of a Markov theory based stochastic model for use in network level rail wear predictions. It was established that rail wear is affected by a number of factors that are inherently probabilistic; therefore rail wear is likely to be better represented by a stochastic process, especially at network level. A prototype stochastic model was developed which forecasts network level rail wear distributions based on probabilities. Four homogeneous subnetworks were included in the model development, all belonging to the Class-A Principal Lines network in New Zealand. Stochastic model predictions were compared with actual network condition distributions for the period 2000 to 2005. For each of the five rail wear bands, the predicted rail wear distributions are generally within 5% of the actual observations. Hence the model has been validated as a prospective tool to be used in rail renewal planning. The stochastic model can be used to forecast long term funding profiles required to maintain the network within a given level of service, as well as simulate the effects of changes to the level of funding on the network condition distribution. These features of the prototype model have been demonstrated on the SOkg/m rail curves network in Class-A Principal lines. Further research is required before this prototype stochastic model can be readily used at the industry level. To assist in 'total' network rail renewal planning in New Zealand, rail sections other than SOkg/m, 91 lb/yd and 90lb/yd on curves, must be incorporated into the models. Consideration must also be given to forecasting rail renewals due to defects other than rail wear. These have not been incorporated in the developed stochastic model. Stochastic models also have the potential to be developed to forecast the occurrence of rail fatigue defects as well as degradation of sleepers, using similar procedures introduced in this thesis.