Fleet sizing in the chemical industry under uncertain demand, transit and unloading times

A large number of chemical railcar fleet managers currently use static spreadsheet models to help them size their fleets. The nature of the ordering and operating environment make this task suitable for solving by simulation modeling techniques. This dissertation examines the usefulness of a hierarchy of queueing models in predicting fleet sizes under conditions of uncertain demand, transit and unloading times for rail cars. Three queueing models, which differ in their level of data aggregation, are tested along with the simple deterministic model. In addition, a full simulation model is developed for each fleet.; The output of the full simulation model is used as a reference fleet size for evaluating the fleet sizing accuracy of the queueing models. Four fleets are used to test the models under varying network size and order frequency. The models are used to predict fleet sizes under three different service level policies: 10%, 5%, and 2% delayed orders. An order is considered delayed if the shipment was not made on the day the shipment was ordered. This research found that with the data set used the queueing models studied do not consistently predict fleets similar in size to those generated by the full simulation model. The models do predict larger fleets, at all performance levels studied, than the simple static models currently in use. One queueing model, under conditions of small delays between orders, appears to predict fleet sizes consistent with the full simulation model for all fleet sizes. Additional study is needed to further define the ordering frequency conditions under which this model can accurately predict fleet sizes.