Integration of a production planning system and a manufacturing execution system based on multi-agent technology

For: Paul Verstraete

Manufacturing companies are often confronted to changes, caused either by
the client or the supplier. Examples are signicant changes in customer demand,
business opportunities for new types of production, mergers, lack or abundance
of personell, etc. The ability of a company to adapt to these changes determines
its chances of survival.
Crucial in this ability to change is the organisation and management of the
production processes. Three types of activities play a key role: tracking and
tracing, production control and optimisation.
Tracking and tracing (as done by Supervisory Control & Data Acquisition
(SCADA) systems or Manufacturing Execution Systems (MES) involves the
monitoring, logging, acces control and automation of the production processes.
These systems support decision taking by maintaining an up-to-date status re-
port of the shop floor. Furthermore, it maintains a history of the production
process.
Production optimisation provides optimisation of production for an abstrac-
tion of the reality. This abstraction is necessary because of the intractability of
manufacturing scheduling. The output of the production optimisation algorithm
is a production guideline, which can be rough. It does not necessarily contain
commands that are directly executable by the production floor. Because of the
intractability of manufacturing scheduling, the problem representation is the
key component to solve the problem. Unfortunately, the problem representa-
tion is dependent on a large number of factors. Furhtermore, these factors vary
over time. This makes case-by-case adaptations of the production optimisation
necessary.
Manufacturing control has to realise the optimisation goals during execution. Kanban implementations (or variations thereof) and dispatching rules are
the most popular industrial examples. In the current state-of-the art, most op-
timisation during execution is myopic. Futhermore, the optimisation performed
by manufacturing control systems does not take the sequence or allocation of
machines into account. Both Kanban implementations and dispatching rules
demand a large amount of tuning and require maintenance if the production
process changes.
Today, these three activities operate next to each other. As a consequence,
shopfloor supervisors and management lack up-to-date predictions of the near
future of the shopfloor. Furthermore, optimisation - before and during production - requires a lot of maintenance and happens on a case-by-case basis.
The K.U.Leuven PMA division (Production engineering, Machine design and
Automation) developed a manufacturing execution system that aims to reduce
the required maintenance to achieve optimisation. Furthermore, the execution
system provides an up-to-date prediction of the shop floor at every moment.
The manufacturing execution system always takes the latest shop floor information into account and has facilities for logging. The MES provides important
contributions to the state-of-the art at the level of production control. In fact,
the MES integrates and superseeds the functionality of a manufacturing control
system. In contrast to traditional manufacturing control systems, it predicts
future behavior and proactively takes measure to prevent production problems
from occuring.
The MES also provides important contributions at the level of adaptibility
to new production requirements. The MES has a decentralised design where
the production orders drive their own production. The MES has the following
characteristics:
 When searching for a solution, the order can consider multiple process
plans.
 Neither constraints (for instance precedence constrains) or decision vari-
ables (for instance operations to be executed) need to be known up front
to the MES.
 Decision variables and constraints may be added during execution. For
example new orders may become available during execution, as well as
extra machines, buers or transporters.
The thesis proposes and validates an integrated approach for tracking &
tracing, manufacturing control and production planning.
The approach allows to construct a single system that fullfills all three func-
tionalities, while allowing for improved support for changes form either client
or supplier side.
The system takes into account the output from a planning algorithm. It
compensates for possible simplications in the planning. As a consequence, a
planning can be rough and still provide good end results. The system uses the
capabilities of a holonic manufacturing execution system to execute the plan-
ning. The execution system provides local short term forecasts of the production
and is robust with regard to unexpected events (such as machine breakdowns).
The local short term forecast make the consequences of the optimisation con-
crete for the stakeholders involved in the production process. The robustness of
the manufacturing execution system diminishes the requirements on calculation
time for the planning system. Furthermore, the execution system provides sup-
port for using tracking & tracing information in control & optimisation. This
information is used by the intelligent beings in the environment to support the
agents in the execution system in performing their tasks.