22-08-2014, 02:53 PM
ARTIFICIAL NEURAL NETWORK APPROACH FOR FMS SCHEDULING
UNDER A RANDOM ENVIRONMENT
ARTIFICIAL.doc (Size: 65.5 KB / Downloads: 9)
ABSTRACT
A Neural Network based Production Control System for a Flexible Manufacturing Cell, operating in a highly random produce- to- order environment is presented. A multi-objective FMS Scheduler is designed to satisfy the desired values of multiple objectives set by the operator. For each production interval, a decision rule for each decision variable is chosen by the FMS scheduler. A unique feature of the FMS scheduler is that the competitive neural network generates the next decision rules based on the current decision rules, system status and performance measures. A commercial FMS is simulated to prove the effectiveness of the FMS scheduler. The result shows that the FMS scheduler can successfully satisfy the multiple objectives.
Introduction :
In the Worldwide competitive market, companies survival strongly depends on their ability to produce a wide range of low cost, high quality products, within very short lead times. Flexible manufacturing cells (FMCs) enable companies to achieve these essential standards. A FMC consists of several machines, equipped with facilities for automatic changing of parts and tools, inter-connected by handling and storage facilities, all computer controlled. A FMC is designed to process simultaneously a variety of part-types and to provide alternative processing routes for individual part types. A comprehensive survey of flexible manufacturing capabilities is given in Sethi and Sethi [1] and a classification scheme is given in MacCarthy and Liu [2]
An important capability of a FMC is routing flexibility [1]. Routing Flexibility enables operations to be processed by alternative machines with different efficiencies. This characteristic, when cleverly used, helps to resolve the occurrence of bottlenecks by balancing machine work loads.
A FMC may be operated under three basic situations. The simplest one is single-lot production where a known set of parts has to be produced with in a given period of time. The second situation is known as constant-mix, where parts are produced continuously in fixed and known proportions. The third and the most complex situation, is produce-to-order. In a produce-to-order situation a variety of orders arrive randomly. Each order has different processing requirements, individual arrival times and individual due dates. Thus, the production system is confronted with a continuously changing flow of jobs.
The Multi-objective FMS Scheduler
The Multi-objective FMS Scheduler is composed of two main modules : the competitive neural network and the search algorithm. The Figure depicts the execution sequence of the scheduler. The FMS operator gives the scheduler input date consisting of the desired relative objectives of evaluation criteria. Since the current and next decision rules do not affect the classification, both rules do not need to be given. Then the scheduler finds the matching class and presents the matching class to the search algorithm, which either finds the matching input vector and extract the next decision rules in the matching input vectors or selects the approximate next decision rule
Conclusion :
An FMS has the advantage of increasing the flexibility and productivity of discrete part manufacturing. The highly integrated system is becoming more complex to control and this causes a number of decision problems. A method is developed for the selection of associated decision rules on decision variables in order to obtain desired performance measures and system status at the end of each production interval. In the proposed method, a system control, strategy based on simulation technique, competitive neural network is suggested.
The results of this study strongly indicate that applying this methodology to obtain a control strategy in an effective material for coping with the complexity of FMS. Especially in a real time control system, it is useful to use pre-obtained control knowledge as a time-saving way to achieve prompt response in a dynamically changing environment.