شبکه های عصبی مصنوعی برای شبیه سازی تولید کارگاهی

This paper explores the use of artificial neural networks (ANNs) as a valid alternative to the traditional job-shop simulation approach. Feed forward, multi-layered neural network metamodels were trained through the back-error-propagation (BEP) learning algorithm to provide a versatile job-shop scheduling analysis framework. The constructed neural network architectures were capable of satisfactorily estimating the manufacturing lead times (MLT) for orders simultaneously processed in a four-machine job shop. The MLTs produced by the developed ANN models turned out to be as valid as the data generated from three well-known simulation packages, i.e. Arena, SIMAN, and ProModel. The ANN outputs proved not to be substantially different from the results provided by other valid models such as SIMAN and ProModel when compared against the adopted baseline, Arena. The ANN-based simulations were able to fairly capture the underlying relationship between jobs' machine sequences and their resulting average flowtimes, which proves that ANNs are a viable tool for stochastic simulation metamodeling.

Scheduling has been defined as ‘the art of assigning resources to tasks in order to insure the termination of these tasks in a reasonable amount of time’ [3]. The general problem is to find a sequence in which the jobs (e.g. basic tasks) pass through the resources (e.g. machines), which constitutes a feasible and optimal schedule with respect to some specific performance criterion [7]. This represents the last stage of the planning activities before production takes place. Systems simulation has become a powerful decision-making instrument for job shop scheduling. It requires a few simplifying assumptions, captures many of the true characteristics of the real model, and provides good insights about the interactions and relationships between qualitative and quantitative variables. However, a major shortcoming of simulation is the need for expert assistance any time a change is required in a model [14] and [16]. One contribution to the simplification of the scheduling decision-making process might consist of the development of a system that can perform a rapid evaluation of different alternatives, without the necessity of computer simulation expertise. If a dynamic model of a system could be constructed and presented to the analyst as a black box, one of the major drawbacks of systems simulation would be overcome: the need for a human expert to carry out the simulation. Artificial intelligence (AI) is the generic name given to the field of computer science dedicated to the development of programs that attempt to replicate human intelligence. Artificial neural networks (ANNs) is one of the AI techniques that has gained an important role in solving problems with extreme difficult or unknown analytical solutions [12]. An ANN consists of an interconnected web of special units, called neurons, with associated connection weights that, after receiving a proper training, are capable of achieving a desired response to new inputs. Its ability of learning from examples makes ANN an extremely powerful programming tool when domain rules are not completely certain or when some amount of inaccuracy or conflicting data exist [13].

In this study, ANNs proved to be a viable tool for stochastic simulation metamodeling. The expected simulation output from the developed ANN models turned out to be as valid as the data generated from the conventional simulation packages, i.e. Arena, SIMAN, and ProModel. Their results proved not to be substantially different from the results provided by other valid models such as SIMAN and ProModel when compared against the adopted baseline, Arena. Despite the noise introduced by the stochastic nature of the interarrival and processing times, the ANN-based simulations were able to fairly capture the underlying relationship between jobs' machine sequences and their resulting average flowtimes. From the practical application standpoint, the developed models offer significant advantages regarding time consumption and simplicity to evaluate new job shop situations. The software operation is considerably fast when compared against conventional simulation software packages. This feature can become more and more relevant as the model is expanded and enlarged. In addition, ANN models, properly embedded in manufacturing decision support systems (DSS), may greatly contribute to the simplification of sequencing and scheduling decision-making processes. A neural network metamodel would be a tremendously useful tool for scheduling optimization packages where simulation of partial solutions is critical, as it is the case of genetic-algorithm (GA) based programs.