Online Adaptation of Takagi-Sugeno Fuzzy Inference Systems

Adaptive algorithms for data-driven models are often of fundamental importance in order to identify real-time processes, which possess a time-variant behaviour that would make a time-invariant model too inaccurate. Beyond that, an insufficiency of amount, distribution and/or quality of actual recorded measurement data can occur, such that the model cannot meet the expectations at a particular time. In this case, the incorporation of new recorded data into previously generated models can improve the model's accuracy and reduce the bias or model error captured due to original noisy data. Moreover, models based on vague analytical or even linguistic expert knowledge can be refined and detailed with adaptive modelling methods. In this paper algorithms and strategies for adapting a special kind of data-driven models, namely so-called Takagi-Sugeno fuzzy inference systems, are demonstrated. Validation results from simulation studies with respect to updating Takagi-Sugeno fuzzy models based on real-life measurement data obtained from engine tests are included at the end of the paper.