Kalman filter and neural network‐based icing identification applied to A340 aircraft dynamics Available to Purchase

The purpose of this paper is to maintain safe flight and to improve existing deicing (in‐flight removal of ice) and anti‐icing (prevention of ice accretion) systems under in‐flight icing conditions.Design/methodology/approach – A recent academic research on aircraft icing phenomenon is presented. Several wind tunnel tests of an experimental aircraft provided by NASA are used in the neural network training. Five ice‐affected parameters are chosen in the light of these experiments and researches. An offline artificial neural network is used as an identification technique. The Kalman filter is used to increase the state measurement's accuracy such that neural network training performance gets better. A linear A340 dynamic model is selected in cruise conditions. This linear model is simulated in time varying manner in terms of changing icing parameters in a system dynamic matrix. The obtained data are used in neural network training and testing.Findings – Airframe icing can grow in many ways and many points on aircraft. In this research, wing leading edge ice occurrence is only considered at the same level in both left and right wings. During ice growth other faults or anomalies are ignored.Originality/value – Existing icing sensors can only provide an indication about possible ice presence. They cannot give information of the exact level of ice. However, the efficiency of current control system of changed model decreases. The proposed technique offers a method to find out the model changes under icing conditions.