Aircraft Damage Detection Using modeFRONTIER RSM

IAI structural technologies department produced a large amount of structural data concerning structural characteristics during landing. The data was gathered by two different means. The first was actual landing flight tests using strain gauges which were installed along the upper and lower aircraft skins. The second was finite-element model analyses which simulated both structural-fit and damaged-structure configurations under various landing conditions. The database which was gathered is then used to train several surrogate models which later predict the various sensors' response. The main models are Keras deep learning, modeFRONTIER stepwise regression, and modeFRONTIER anisotropic Kriging. For each model, a specific strain gauge was predicted, based on the various sensors readings (except the one which is predicted). Results showed good agreement using the trained models, which allows to detect skin damages in a clear manner and also detect the location of the damage. Using the Keras deep learning model the Honeycomb damage was not detected but the skin damage location was predicted accurately. On the other hand, the Stepwise Regression model predicts the honeycomb damage but not its location. The proposed models can be developed to post flight structural health monitoring tools and later, to real-time systems which will enable the detection of structure damage during flight.