Data-Driven Source Localization of Impact on Aircraft Control Surfaces
Li Ai, Vafa Soltangharaei, Rafal Anay, Michael J.L. van Tooren, Paul Ziehl
Abstract
Aircrafts are potentially subjected to damaging events during their service life. How to cope with impact events and impact related damage is a priority in the development of aircraft composite structures. The impact monitoring method presented in this paper utilizes acoustic emission (AE) based data to classify and thereby localize impact events. The method is implemented and tested on a full-scale aircraft elevator. This work builds on earlier research for the classification of AE signals acquired by a single sensor during impact events using a backpropagation neural network with two hidden layers [1]. The innovative aspect of the new method lies in the use of a deep learning algorithm to achieve the zonal localization of impact events. Compared to the backpropagation neural network method, the deep learning method can output localization results with improved accuracy without the need to extract signal features, such as time of arrival, signal strength and amplitude. For this paper, stacked autoencoder algorithms were applied. To train and test the performance of the new model, the same aircraft elevator impact test setup from prior work was used. A single sensor was attached to the spar of the elevator to collect the acoustic emission events. Impacting with steel spheres was conducted on the elevator skin at various distances from the impact source to the sensor. Results demonstrate the efficacy and potential of the deep learning-based approach for localization of impact events for aircraft elevators.