Our paper with Frederico Cruz (see here for his homepage) is accepted for publication in the Journal of Applied Mathematics. This paper has been lying around for quite some time, but now we managed to finish it up and get it published.
This paper provides an overview of the queueing modeling issues and the related performance evaluation and optimisation approaches for manufacturing networks. Such networks are represented as queueing networks. The performance of the queueing networks is evaluated using an advanced queueing network analyzer: the generalized expansion method. Secondly, different model approaches are described and optimized with regards to the key parameters in the network (e.g. buffer and server sizes, service rates, and so on.).
The numerical results for the buffer allocation problem, the server allocation problem, and the joint buffer- server allocation problem show that significant gains can be achieved in manufacturing systems. Specifically, set- ting the buffers and servers in an appropriate way greatly affects the throughput for these manufacturing systems. This is important as these systems need to be as highly utilized as possible, given the high investments. Our models and optimizations show that the optimal configurations are not always straightforward and thus advanced models and solution methods are needed. We have followed a queueing network approach with finite buffers, as this re- sembles reality the closest. This modeling approach is of course much harder than e.g. infinite queueing networks.
We see based on the various experiments that our solution methodology is powerful and suitable for the different types of models handled in this paper. This offers managers and manufacturing systems designers a powerful tool to work with. We also noted that while the addition of the first extra server gives a certain amount of increase in the throughput the addition of the first buffer space generally will give a lower increase. In other words in order to achieve the same increase in throughput by only using buffers, we need more extra buffer spaces rather than only a few server space.