9713730 O'Cinneide This grant provides funding for the development of computational methodologies for analyzing and predicting the performance of complex systems, such as manufacturing facilities and computer and telecommunication networks. The methodologies to be developed are based on theoretical models of the randomness in such systems and a theoretical understanding of the impact of randomness on performance. The principal methodologies to be considered are (a) methods based on Fourier or Laplace transform inversion; (b) heavy-traffic Brownian approximations; (c) Markovian methods (Neuts method of phases); (d) Markov Chain Monte Carlo methods; and (e) decomposition and other approximate methods. The research builds on a substantial body of literature in each of these areas. This research will enhance the ability of engineers and managers to assess the capabilities of award new systems and to predict the impact of modifications on existing systems. The computational methodologies to be developed are very different from the standard approach to these problems, namely, simulation, in which system performance is predicted based on a detailed and time-consuming process of mimicking the actual operations of the system on a computer. By bringing theoretical knowledge to bear on predicting performance, computational methodologies make possible very rapid computer analyses which may be sufficient for a given situation or which may be used as a preliminary step in preparation for a detailed simulation study. One specific goal will be to develop a methodology for the performance analysis of manufacturing systems that can take account of the key features already handled by existing software, such as variability in processing times and machine breakdowns, but with extensions to deal with some additional specific features: (a) Constant-Work-in-Process (CONWIP) conditions; (b) simple dispatching rules such as Easiest Due Date (EDD); and (c) multiple job types.
