Mohamed Benazouz, Ariane Piel
CEA, LIST, Laboratory of Model Driven Engineering for Embedded Systems,
P.C. 174, Gif-sur-Yvette, 91191, France
{mohamed.benazouz, ariane.piel}@cea.fr
Keywords: Multi-Model Assembly Line, Malfunctions, Buffer Stock, Cyclo-Static DataFlow.
Introduction
As part of the SodaLite1 project, we consider an assembly line consisting of machines and inter-machine buffer stocks. Machines are subject to hazards and a line can be to produce different variants / versions of a product (i.e. a model assembly line mixed) that may have different processing times from one version to another. All this introduces an asynchronism between the different parts of the line and justifies the use of these zones buffers even for a perfectly balanced line (i.e. the time of treatment of a variant is the same for all machines). Buffer stocks then have the task of decoupling different machines in order to provide a smoothing effect on the treatment times and to prevent the total apparent failure time of the line is an accumulation of the duration of failures of all its machines.
The purpose of this work is to determine the minimum capacities of these buffer stocks as that the line reaches a target production rate. This sizing problem has been proven NP-difficult. In order to solve it, we had two ways to take into account the uncertainties of production and sequencing of the different variants.
The first is to do a study stochastic or worse-case, in which case the laws of probability of failures are to be integrated and the sequencing worst case is to be identified. This approach has been ruled out for reasons of increased complexity and duration reduced project. Moreover, the worse-case results that would have come out of it would not have been considered as acceptable by the participating company.
The other more pragmatic approach, which has been adopted, was to be in agreement with the technique used by companies for the validation of their lines, namely, the simulation of the line for a long production time considered as sufficient and representative of the different situations that may arise during the life of the line. This ensures for the resolution technique that we offer as much coverage of problematic cases than the validation approach itself.
