The inverse determination of aerodynamic loading from structural response data using neural networks

Abstract

The prediction and monitoring of aircraft structural fatigue damage is vital for the safe operation of ageing aircraft. The ability to determine aerodynamic loading inversely using structural response data has the potential to significantly improve fatigue monitoring capabilities. This study examines how a Neural Network can be used to estimate and predict aerodynamic loading from structural response data. To simulate aerodynamic loading conditions F/A-18 Empennage fatigue test data which includes the application of both high frequency buffet and low frequency manoeuvre loading will be used. The neural network was trained using response data from several strain gauges as input and known applied loads as output. It was then tested with new data and compared to the known applied loading corresponding to the new data. The network was also tested in its ability to predict the aerodynamic loading across locations different to the locations of the training data.

Keywords:

• Neural networks
• Aerodynamic loads
• Aircraft fatigue

Keywords:

• AMS Subject Classification: 34A55
• 68T05
• 82C32
• 76G25

Acknowledgements

This project could not have been carried out without the support of Defence Science Technology Organisation (DSTO), Australia, which provided, along with financial assistance, access to data and technical resources required in order to carry out much of the work presented here and carried out by the DSTO/RMIT Centre of Expertise in Aerodynamic Loading (CoE-AL). In particular, the authors would like to thank Dr Douglas Sherman, Functional Head, Aircraft Loads from the Platform Sciences Laboratory for his vision, suggestions, support, initiative and thorough guidance of the work. The authors would also like to thank Dr Chris Blanksby for helpful discussions.