Lamb wave based ice sensing by neuronal network analysis of icing wind tunnel data

Abstract

Detecting an ice accretion in aviation is still a challenge today since heavy icing can cause catastrophic failure of aircraft. Ice sensors, which are able to detect icing conditions or the presence of ice on aircraft, provide the necessary information to either activate ice protection systems or to avoid the icing conditions. Within the last years, an ice sensor based on lamb waves has been developed where a lamb wave signal is sent through an icing prone structure. Ice accretion influences the lamb wave transmission, which is used to detect the presence of ice. This detection of the presence of ice has been successfully demonstrated in icing wind tunnel as well as in flight tests.

Since icing itself and especially the interaction of ice, the base structure and the ice accretion is a very complex phenomenon, the current understanding of the lamb wave signal does not allow to obtain a precise measure for the ice thickness on the structure. However, this is very desirable information from a pilot perspective, since it allows to calculate the ice accretion rate and the liquid water content of the atmosphere. This in the end is required to assess the severity of the icing encounter, which is the basis to decide the countermeasures that have to be taken to ensure the safety of flight.

In order to increase the precision of the ice thickness obtained by the lamb wave signal, an extensive icing wind tunnel test campaign was conducted to provide a solid set of data in different icing conditions. This dataset is used to train neuronal networks with the ice thickness as target function. The paper will provide an overview about the icing wind tunnel testing, the topology of the neuronal networks and the training. Finally some modeling and test results of the trained neuronal networks will be presented.