H.K. Tönshoff, M. Manns, and K. Spardel (Germany)
Modelling and Simulation Methods, Material-Flow Simulation, Neuro-Fuzzy, Production Planning
Flexible production lines are often planned in dynamic environment with very demanding time schedules. Since simulations require extensive modeling time, material-flow simulation is mainly used in later stages of the planning process. As are many artificial intelligence methods, neuro-fuzzy systems like CANFIS can be interpreted as fuzzy rules. Their use as a material-flow forecasting tool for early planning stages could support the planner with comprehensible estimates to improve the planning process. However, for a variety of hybrid neuro-fuzzy systems, including CANFIS, high input parameter dimensionality typical for material-flow simulations lead to unacceptable computing times. This paper proposes a method using CANFIS for learning fuzzy rules from simulation data. Results are given concerning if and how well the neuro-fuzzy algorithm CANFIS learns to transform simulation model parameters to performance measures.