A Simulation-Based Transfer Function Modeling Approach for Responsive Production Planning

In this thesis, a novel statistical metamodeling method is proposed to study the input-output dynamic relations of general queueing system, especially when the system going through transient state. This metamodeling approach incorporates discrete even simulation, statistical inference, analytical queueing analysis to estimate a set of transfer function models (TFMs), which fully describe the system dynamic in terms of outputs that are interested: first and second moment of work-in-process (WIP) and first moment of departure rate.;Empirical queueing examples with non-Markov property are studied, including single station-single server system, single-station multi-server with failure system and multi-station multi-server with reentrant flow system and failure, following the proposed approach. TFMs for such systems are obtained. Predictions of the system dynamics is estimated from these TFMs under given system. Cross validation of the predicted outputs with simulation results show that the proposed approach is very accurate in describing the system evolution under both transient and steady states of general queueing system. Also, the proposed TFMs has been applied to the general Jackson-network models, cross validation result shows that the estimation results are promising as well.;The above proposed TFMs is integrated into a production planning model for a 5 station in-tandem system with reentrant flow and machine failures. The demands are independent and random with mean and distribution known. The production planning model estimates both mean and variance of the expected total cost based on TFMs. Genetic algorithm (GA) is employed to find the Pareto Front of such a production planning model. Objectives from these Pareto Front points are compared with simulation results and proved the accuracy of the production planning model based on TFMs