| This research concerns the performance improvement of queueing networks with synchronization stations. Queueing networks are models found in the manufacturing, transportation, communication and computer areas. A classical approach to evaluate their performance uses discrete-event simulation. While simulation provides accurate performance estimates, it becomes computationally expensive when we seek to study the sensitivity of performance metrics with respect to design parameters. To alleviate this computational cost, we develop analytical models of queueing networks via parametric decomposition. Advantageously, analytical models are fast to solve and still provide accurate performance estimates.; The first part of our research concerns the performance improvement of airfield operations, which represent the steps aircraft need to undergo at an airfield in air logistics systems. We use synchronization stations to model a mechanism for air traffic control, as well as a mechanism for designing operations concurrently instead of sequentially.; The second part of our research concerns kanban-controlled serial production lines. We assume there is a finite supply of raw material and a finite supply of demand in finished goods. The line can be organized into cells, and we model cells with synchronization stations. We then propose two algorithms to solve the optimal card allocation problem. The first algorithm is a heuristic based on simulation. It provides an optimal answer when tested against exhaustive simulation, but it is computationally expensive. The second algorithm is a heuristic built in two steps. In the first step, the optimal card allocation problem is formulated as a mixed integer nonlinear programming problem (MINLP) with the analytical model obtained via parametric decomposition. In the second step, a simulation-based search is performed in the neighborhood of the MINLP solution.; In the third part of our research, we characterize exactly a synchronization station fed by three independent Poisson processes with shut-down levels via a Markov renewal approach. We compare this system with an alternative system comprising two synchronization stations, each with two buffers, organized in a cascading fashion. We show that those two systems are not equivalent. We provide guidelines for arranging the three input processes so that the cascading system represents most accurately the original system. |
