Forecasting replacement of transit vehicle components using Markovian process modeling

Vehicle maintenance in the public transportation industry is a vital concern for the successful operation of a vehicle fleet. Forecasting transit vehicle component replacement is an important area for investigation. This paper utilizes a Markovian process model for forecasting the number of component replacements at particular mileage intervals. History indicates that Weibull and exponential distributions describe lives of electrical devices while Weibull and lognormal distributions describe lives of mechanical devices (Little). Probability distributions were determined for component replacement data and tested with chi-square goodness-of-fit procedures. Once a component with a hypothesized distribution did not reject the null hypothesis, the Markov application was performed. In the examination of twenty components, ten components were subjected to the Markov application. The other components did not satisfy chi-square goodness-of-fit testing for hypothesized probability distributions. There is good agreement between chosen probability distributins and observed data. That is, the differences between calculated mean distances to replacement and the data’s average replacement mileage are less than ten percent. The replacement forecasts were usually sufficient to satisfy data requirements. Areas of difference were seen in the first few intervals. Once the forecast stabilized, the forecast and the respective replacement data values were compared. The engine, the front axle, and the rear axle indicated good agreement between the forecast and the replacement data. The other components had a stabilized forecast in excess of the replacement data. Forecasts stabilized as the number of mileage intervals increased. As the vehicles reach greater mileages, component forecasts were more than ample in size. This result is evidence that component replacements are not completed due to the arrival of anticipated service life. This model is not difficult to apply and does provide insight into the number of component replacements needed for shelf-stock. However, after forecast stabilization, the model may provide a forecast that is not representative of actual needs. Based upon conversations with Greater Manchester Passenger Transport Executive, stock levels would be reduced by incorporating a Markovian process analysis.