| Driver turnover is a significant problem for full truckload (TL) carriers that operate using point-to-point (PtP) dispatching. The low quality of life of drivers due to the long periods of time they spend away from home is usually identified as one of the main reasons for the high turnover. In contrast, driver turnover is not as significant for less-than-truckload (LTL) carriers that use hub-and-spoke transportation networks which allow drivers to return home more frequently. Based on the differences between TL and LTL, the use of a relay network (RN) has been proposed as an alternative dispatching method for TL transportation in order to improve driver retention. In a RN, a truckload visits one or more relay points (RPs) where drivers and trailers are exchanged while the truckload continues its movement to the final destination.;In this research, we propose a new composite variable model (CVM) to address the strategic TL relay network design (TLRND) problem. With this approach, we capture operational considerations implicitly within the variable definition instead of adding them as constraints in our model. Our composites represent feasible routes for the truckloads through the RN that satisfy limitations on circuity, number of RPs visited, and distances between RPs and between a RP and origin-destination nodes. Given a strict limitation on the number of RPs allowed to be visited, we developed a methodology to generate feasible routes using predefined templates. This methodology was preferred over an exact feasible path enumeration algorithm that was also developed to generate valid routes. The proposed approach was successfully used to obtain high quality solutions to largely-sized problem instances of TLRND.;Furthermore extending the original CVM formulation, we incorporate mixed fleet dispatching decisions into the design of the RN. This alternative system allows routing some truckloads through the RN while the remaining truckloads are dispatched PtP.;We analyze the performance of our models and the solutions obtained for TLRND problems through extensive computational testing. Finally, we conclude with a description of directions for future research. |
