Keywords: sandwich structures, finite element analysis, vibration-based features, artificial neural networks, damage identification.

This study investigates the effectiveness of the combination of global and local vibration characteristics [1,2] of a glass fibre-reinforced plastic sandwich beam-like structure when introduced to artificial neural networks (ANN) for the severity and location prediction of various damage with different extent at different locations. A finite element modelling and analysis tool is used to obtain the dynamic characteristics of both intact and damaged cantilever sandwich beam-like structures. To create various damage scenarios, the elements located at the upper surface of the beam are removed in the finite element models by simulating delamination type damage. In finite element models, five different damage severities are simulated at 26 different locations along the span of the beams. On these 130 individual models, normal mode dynamic analyses are performed in order to find the first three undamped natural frequencies and corresponding displacement mode shapes. Then changes in natural frequency arising from damage are calculated for each out-of plane bending mode and curvature mode shapes are obtained from the normalised displacement mode shapes by using a central difference approximation. All global (changes in natural frequency) and local (curvature mode shape) vibration analysis data of intact and damaged beams are then assembled. Having gathered these vibration data sets, the sensitivity analyses, aimed at finding the necessary parameters, are performed for the damage quantification and localisation. Following this, supervised feed-forward backpropagation artificial neural networks with one hidden layer are designed and different combinations of input (reduction in natural frequencies, maximum absolute difference in curvature mode shape and its spatial location along the span of the beam) and output (damage severity and damage location) pairs are then introduced to these networks for the training and the validation runs. Having completed the training and validation processes, the designed ANNs are also tested for new damage cases and checks are made for the severity and location predictions. The results obtained from new test cases show that the selection of vibration-based analysis features which are considered as input data for ANNs is very crucial from the accuracy point of view of the identification of the damage. The ANN test results also indicate that the predictions are close to the expected target values with acceptable deviations.

1

M. Sahin, R.A. Shenoi, "Vibration-based Damage Identification in Beam-like Composite Laminates", Proceedings of the IMechE, Part C, Journal of Mechanical Engineering Science, 217(6), 661-676, 2006. doi:10.1243/095440603321919581

2

M. Sahin, R.A. Shenoi, "Quantification and Localisation of Damage in Beam-like Structures by using Artificial Neural Networks with Experimental Validation", Engineering Structures, 25(14), 1785-1802, 2003. doi:10.1016/j.engstruct.2003.08.001

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 £145 +P&P)