A Performance Comparison between Efficiency and Pheromone Approaches in Dynamic Manufacturing Scheduling

Introduction

These days competition is played in an environment characterized by high market shifting, rapid development as well as introduction of new technologies, global competition and customer needs focalization. Therefore, manufacturing environments are becoming more dynamic and turbulent than ever before. Traditional manufacturing facilities, however, are not able to cope with such environments, as no single facility can be flexible enough to cope with such a large magnitude of change in products and production requirements.

Szelke et al. (1999) identify that in the field of manufacturing, agility and reactivity can be achieved by operating both at system and control level. At the system level, the most common solution is the decomposition of the manufacturing system into smaller units, e.g. manufacturing cells, in order to achieve simplicity, specialization, scalability, fault tolerance, etc. At the control level, there are two ways to guide the complexity of operation management problems to simplicity, reactivity, scalability and fault tolerance: a) to enhance the reactivity and pro-activity of the scheduling and control systems by sophisticated new control techniques; b) to take advantages of distributed control.

Autonomous Agents Systems (AAS) are becoming very popular in several industries, such as manufacturing, telecommunications, medicines and so forth because of their ability to build very reactive, fast-learning and efficient distributed systems. In AAS, the focus is on the coordination and negotiation among intelligent autonomous agents. In manufacturing, such systems have demonstrated their ability to build up very agile and reactive systems from several viewpoints: enterprise integration and supply chain management (Swaminathan et al., 1996); dynamic system reconfiguration (Shen et al., 1998); learning in agent-based manufacturing systems (Monostori et al., 1996; Shen et al., 2000; Shen, 2002); distributed dynamic scheduling (Chiuc et al., 1995; Vancza et al., 2000); factory control architectures (Brennan et al., 1997); and implementation tools and standards.

In particular, the scheduling problem in real time is a difficult task in a dynamic environment. In order to operate in such an environment, a reactive scheduling or adaptive control needs to be developed.

The problem of scheduling in manufacturing systems concerns the allocation of resources to jobs over time. It is a decision-making process with the goal of optimizing one or more objectives (Pinedo, 2008). The objectives can be: minimization of the mean throughput time, tardiness of the jobs, minimization of the work in process, etc. Scheduling in Flexible Manufacturing Systems differs from other conventional job shop because each job can have alternative process plans and each operation can be performed on alternative machines. The scheduling problem is known to be NP-hard, i.e., the time required to solve the problem optimally increases exponentially with increasing problem size.

Most manufacturing systems operate in dynamic environments where, usually, inevitable and unpredictable real-time events may cause changes in scheduled plans. Examples of such events include machine breakdowns, demand changes in mix and volume, operational time of the manufacturing operations, etc.

MacCarthy and Liu (1993) addressed the nature of the gap between the scheduling theory and scheduling practice, and the failure of the classical scheduling theory to respond to the needs of practical environments. A decade later, Cowling and Johanson (2002) also addressed an important gap between scheduling theory and practice, and stated that scheduling models and algorithms are unable to make use of real-time information.