Interpolation issues in Sugeno-Takagi reasoning

In Sugeno-Takagi reasoning, rule premises describe fuzzy subspaces of inputs and rule consequents are linear input-output relations. Fuzzy models with this structure approximate global nonlinearities by a weighted average of local linear functions. It is shown in this paper that the interpolation between neighboring linear models depends on the relation between model parameters. In certain cases undesirable interpolation is obtained, or when model parameters are estimated from system input-output data, linear functions in rule conclusions may not represent local behavior of the system.