Keywords: inverse method, load identification, health monitoring.

Advanced structures in aero-vehicles can suffer from abrupt external impacts or the extension of internal defects during service, potentially leading to catastrophic failure of aircraft without timely detection and repair. Motivated by aviation safety, on-line structural health monitoring (SHM) is a system with the capability of detecting and interpreting adverse 'changes' in a structure in real-time, so as to enhance reliability and reduce life-cycle costs [1]. A reliable on-line SHM system should possess automatic data acquisition and processing, structural condition assessment, and decision-making for corrective actions. In general, the development of successful health monitoring methods depends on two key factors: sensing technology and the associated signal analysis and interpretation algorithm. Instrumentation such as strain gauges and accelerometers are already used to determine the levels of overall flight loads, such as wing root bending moment and torque [2]. A better and continuous understanding of these load conditions is useful for more accurate structural health monitoring.

An inverse load identification problem in a simplified composite wing model is treated here by means of radial basis function (RBF) neural network and strain response data. It has been shown that this approach can be used to predict the load position and value. The network was tested to predict the load at locations different from the locations of the training data.

For a preliminary investigation, a simple model of an aircraft wing has been used rather than a real aircraft structural component. A composite wing model is connected to the cantilever aluminium beam and ten strain gauges have been installed on the top and bottom surfaces of the aluminium beam. The positions of gauges are arbitrary and independent of the solution routine. Experimental data collected by the above procedure are used to find the relation between load and strain. Each artificial neural network in the system with 10 input neurons in the input layer, 3 output neurons in the output layer were employed to model the applied load and the position of load. The proposed methodology was used to create RBF neural network models.

The results showed that RBF neural networks are capable of estimating the mechanical load and its position on a component in service inversely from the strain gauge data. The networks can model the relationship between the structural response and the loads. The results demonstrated an acceptable error. For a better representation of the minima, more training patterns must be introduced to the network.

RBF neural networks are fast in training and response. The selection of the training data has a key role in the capability of the network to accurately predict the load. It is important that the network is trained with the data that covers all ranges of the expected loading conditions. The successful network prediction on the testing data set illustrated the efficiency of this method and clearly showed that it is a reliable tool for prediction of the load. As shown, the error is low and well within average acceptable levels. It seems that this system can be embedded as a subsystem of total structural health monitoring system as it monitors the loads of in-service components.

1

Shkarayev S., Krashantisa R., Tessler A., An Inverse Interpolation Method Utilising In-Flight Strain Measurements for Determining Loads and Structural Response of Aerospace Vehicles, NASA Langley Research Center, 2002

2

Lin Ye, Ye Lu, Zhongqing Su, Guang Meng, "Functionalized Composite Structures for New Generation Airframes: a review", Composites Science and Technology, 65, pp 1436-1446, 2005. doi:10.1016/j.compscitech.2004.12.015

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £105 +P&P)