INTELLIGENT CONTROL OF MATERIAL HANDLING SYSTEMS - Environmentally Conscious Materials Handling - page 65

Systems integration is the main issue for FMS to be successful, and flexible
hardware and manufacturing entities is only one part of the answer. The control
software to handle and integrate the flexible entities in the overall process is
equally important (Brennan and Norrie 2003).
In fact, the control software is often regarded as the critical part, as it
requires high expertise from developers. The complexity of the system and
time-consuming process for reconfiguration have often led to low understand-
ability of the system, which is an important problem to manufacturers, who are
not experts in manufacturing technologies.
FMSs are often composed of computer-aided or robotic assembly nodes, which
are connected by some form of material handling system. Each cell can auto-
matically handle either planned or unpredicted changes in the production flow.
The centralized control generally used in FMSs — which are based on princi-
ples and algorithms of classical control theories — would not scale very well for
such large systems as identified by Sandell et al. (1978). That was the main issue
leading to new approaches for manufacturing control. Bussmann (1998, p. 3) was
even more specific and clear in his conclusion:

Manufacturing systems on the basis of CIM (Computer Integrated Manufacturing)
are inflexible, fragile, and difficult to maintain. These deficits are mainly caused by
a centralized and hierarchical control that creates a rigid communication hierarchy
and an authoritarian top-down flow of commands.

3 DISTRIBUTED SYSTEMS

The experienced problems with complexity and maintenance led to new
approaches in the area of manufacturing control. Parunak (1995) states that
traditionally a centralizing scheduler is followed by control, which would
generate optimal solutions in a static environment, but no real manufacturing
system can reach this level of determinism. Even though scheduling of a
shop floor environment could be optimized centrally, the system would fail in
practice to generate optimal solutions due to the dynamic environment caused
by disturbances such as failures, varying processing time, missing materials, or
rush orders (Brennan and Norrie 2003).
In general, rescheduling and dissemination of new control commands are time
consuming and bring the centralized model to failure. Instead, Parunak (1996)
argued that manufacturing systems should be built from decentralized cooperative
autonomous entities, which — rather than following predetermined plans — have
emergent behavior spawned from agent interactions. He listed three fundamental
characteristics for a new generation of systems:
1. Decentralized rather than centralized
2. Emergent rather than planned
3. Concurrent rather than sequential