Hybrid Simulation Environment

The developed hybrid simulation environment is basically a “virtual brewery”. Standard open source software is integrated with different programs to realize and combine diverse modelling approaches into a process image. Suggestions for a computer aided capacity planning and scheduling are generated – in this example for a scheduling problem – in the following way. The number of batches and beer varieties to produce in the next week is entered into the database. The simulation abstraction layer calculates all possible scheduling sequences and decides with the help of a search algorithm (genetic search) which are to send to the “virtual brewery” for simulation and evaluation. All simulation states are recorded into the database. Based on the recorded data the reporting generates scheduling suggestions for previously defined optimisation parameters.

As indicated above the core of the simulation and evaluation is the modelling of the “virtual brewery”. Processes can be classified in discrete-time (horizon of interest divided into a number of time intervals) and continuous-time processes (sequential, network and event based). In the production of beer both kinds of processes take place, e.g. computer controlled batch handling in the brew house and the continuous fermentation of the wort, creating the need for a modelling approach combining both.

Continuous- and discrete-time approaches can be integrated using Petri Nets i.e. Reference Nets. The Reference Nets represent the discrete timed part, modelling events, resources and constraints as well as process sequences. Needed parameters of continuous timed processes, like the fermentation time or wort ingredients are provided from the simulation abstraction layer using fuzzy systems, artificial neural networks and other modelling approaches. Vice versa key data needed for the continuous modelling, like raw materials, conditions and constraints is made available from the nets via the abstraction layer.

The application of Reference Nets allowed the creation of an object oriented net representation optimising the modular modelling capabilities of Petri Nets. Modules must not be combined in one great module but connect over predefined interfaces resulting in easy reusability as well as interchangeability of created modules. For example heat treatment of wort has a fixed set of interfaces enabling the simulation abstraction layer to exchange a traditional brew kettle with a stripping system in an investment analysis simulation run.

To top