Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 93
Edited by:
Paper 60

A Neural Approach to Crack Identification in Shafts using Wave Propagation Signals

L. Rubio and B. Muñoz-Abella

Department of Mechanical Engineering, University Carlos III of Madrid, Leganés, Spain

Full Bibliographic Reference for this paper
, "A Neural Approach to Crack Identification in Shafts using Wave Propagation Signals", in , (Editors), "Proceedings of the Tenth International Conference on Computational Structures Technology", Civil-Comp Press, Stirlingshire, UK, Paper 60, 2010. doi:10.4203/ccp.93.60
Keywords: cracked shafts, elliptical cracks, non-destructive dynamic test, neural network, crack identification, wave propagation, inverse problems, health monitoring.

The presence of cracks in rotating mechanical elements, such as shafts, is quite dangerous due to the fatigue effect generated by their work in bending and torsion while rotating. Experience shows that fatigue cracks in rotating shafts propagate with an elliptical shape front [1]. Different methods to detect and identify cracks in mechanical and structural elements have been developed for decades [2]. In this work, the wave propagation in a cracked shaft has been numerically analyzed as the first step to establish a non-destructive method for the detection and identification of cracks using a simple dynamic test, and in which not only position and depth, but also the shape of the front has been taken into account. Cracks with increasingly depths and shape fronts have been considered and studied. The three-dimensional finite element model has been used with an initial velocity to simulate an impact load at the free end of the shaft. The differences observed in the registered stress signals due to the different cracks, offer a procedure to solve the inverse problem of identifying a crack in a shaft [2]. It has been found a strong relation between some properties of the signal (amplitude or area under the curve) and the properties of the crack (depth and shape of the front) [3]. The processed results have been used to identify the defect using a neural network approach. Two types of neural networks have been analyzed: back propagation neural networks (BPNN) and radial basis function networks (RBFN). The input data of the neural network selected for the input layer are the amplitude and area below the curve. The output layer is made of two variables: the depth of the crack and the shape of its front, which are the needed parameters for the identification. Both artificial neural networks enable a very good estimation of the depth of the crack to be obtained. The main parameters of performance efficiency analyzed for the estimation of the crack depth give very promising values. However, the estimation of the shape of the front using any of the neural networks is not as accurate. In this work we propose a new method for crack identification based on the analysis of the wave propagation in a cracked shaft and on the use of a very extended and reliable technique which are the artificial neural networks.

L. Rubio, B. Muñoz-Abella, "Flexibility coefficients of a shaft with an elliptical front crack", Proceedings of the 9th International Conference on Vibrations in Rotating Machinery, 657-667, 2008.
P. Cawley, R.D.J. Adams, "The location of defects in structures from measurements of natural frequencies", Journal of Strain Analysis for Engineering Design, 14(2), 49-57, 1979. doi:10.1243/03093247V142049
L. Rubio, B. Muñoz-Abella, G. Loaiza, "Numerical simulation of wave propagation in cracked shafts", Dymat Conference, 2009. doi:10.1051/dymat/2009245

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)