Virtual Air-Data Sensor Calibration Through a CANFIS Model
Manuela Battipede, Piero A. Gili, Marco Lando · 2002
This paper propose a neuro-fuzzy sensor calibration method, specifically applied on the sideslip angle β virtual sensor. The method is based on an ANFIS (Adaptive Neuro-Fuzzy Inference System) structure and attention is focused on how to preserve the linguistic interpretability of the Fuzzy Inference System (FIS) during the neural learning phase in order to improve mapping precision without deteriorate interpretability at the same time. The goal of the paper is to show how ANFIS model can be extended in a more fused and generalized neuro-fuzzy system, known as CANFIS (Co-Active Neuro-Fuzzy Inference System) with nonlinear or neural fuzzy rules. In particular, CANFIS model seems to be an effective approach to alleviating to so-called ”dilemma between interpretability and precision”, occuring during the learning phase. Data used for calibration are obtained through a series of simulations performed with a combat flight simulator augmented with a MIMO autopilot. Performance of different ANFIS/CANFIS configurations have been evaluated and it is shown how CANFIS architectures are very efficient as sensor calibration method.