In modern manufacturing environments, production system analysis is becoming more and more complex in consequence of an increasing use of integration and automation processes which call for far-reaching adjustments in the cost structures of firms: cuts on variable costs and corresponding increases in fixed costs with concomitant greater investment risks. Due to the sheer complexity of both their production systems and associated investment risks, firms tend to earmark sizable proportions of their investment resources for the system design and management phases. The most widely used analysis techniques are based on computeraided simulation models, i.e. tools which are able to simulate any, even the most complex, aspects of a production system and which guarantee highly accurate results close to real-case scenarios. Simulation techniques are used to obtain information on the behavior of a system by performing experiments on a representation of the system to be analyzed, called the model. These experiments are usually performed either on an existing system or on a system in the process of being designed. In the former case, simulation is preferred to other analysis methods in the following situations: when simulation offers costeffective solutions guaranteeing less complex analysis procedures, when, in the absence of other exact or approximate analysis techniques, it is the only efficient method or, conversely, in the event such techniques would be available, when the experimentation requirements are not applicable to the production system to be analyzed. With respect to the latter case, examples in point are either nonexistent systems, i.e. systems still to be designed, or system upgrades which are still at the planning stage and have thus not yet been completed. In these situations, simulations can be used to predict and/or improve the performance characteristics of the system concerned. In this connection, we also wish to mention that simulation techniques can only be used to evaluate a system, not reach an optimum solution. The relevant solutions must be worked out with the help of less complex techniques or by integrating the simulation model within an optimization routine capable of determining the optimal configuration of the system. Static allocation is a generative technique which was widely used when less efficient computing equipment was available and is still adopted today for preliminary analysis. It is the simplest method of modelling a client-server type process. On the one hand, the process is assumed to be unrelated to time; on the other, reasonable cross-resource interactions are not considered. When this technique is used, the performance characteristics of the manufacturing system are determined at a stage when only the process plans of the tasks to be carried out within a given timeframe are known. Consequently, major shortcomings of the static allocation method stem from the fact that the effects of possible interactions between different resources will not be considered and that satisfactory results will only be obtained when large volumes of semifinished products are involved. Today, as a result of the increasing adoption of just-in-time, lean production and other similar techniques, large volumes of work-in process are the exception and this cannot fail to impact production system design and analysis methods. Systems which require due attention to cross-resource interactions are generally analyzed based on queuing models, in which the individual input items of a production system are represented as queues which are then combined into a network. As these analytical models are based on simplified assumptions, their accuracy levels are not the highest, though the results are certainly both more accurate and more significant than those obtainable using the static allocation method. The models just mentioned (simulation, static allocation and queuing models) are based on the assumption that characteristic parameters such as interarrival times, service times and routing coefficients are either deterministic (static allocation) or probabilistic (queueing network and simulation). These hypotheses are certainly valid when historical data sets are available to describe the way these parameters are distributed, but they become less and less significant when it comes to developing a new system or when no or insufficient information is available on an existing system. In the latter case, experts on the system analyzed are asked to provide useful indications concerning the variables involved. Such information as is made available in this way is generally represented by linguistic phrases which can be properly represented using fuzzy numbers. The use of fuzzy numbers to estimate the performance characteristics of production systems is associated with a number of difficulties which stem in part from the fact that the methods used to execute mathematical operations with fuzzy numbers are as yet not generally accepted and in part from the explosion of the support of the fuzzy sets fired by the considerable number of computations to be performed. In this chapter we will analyse the problems associated with the description of vaguely known characteristic variables in a production system and the procedures to evaluate performance indices using queueing network models with fuzzy parameters. © 2005 Springer.
|Titolo:||Fuzzy performance evaluator of AMSs|
|Data di pubblicazione:||2005|
|Appare nelle tipologie:||2.1 Contributo in volume|