CRACK DETECTION IN BEAM-LIKE STRUCTURES USING A POWER SERIES TECHNIQUE AND ARTIFICIAL NEURAL NETWORKS

MARTA.B. ROSALES; CARLOS P. FILIPICH; FERNANDO S. BUEZAS

Vienna, Austria

Congreso; The thirteenth international congress on sound and vibration; 2006

The International Institute of Acoustics and Vibration (IIAV)

The analysis of experimentally measured frequencies as a criterion for crack detection hasbeen extensively used in the last decades given its simplicity. However the inverse problem ofthe crack parameters (location and depth) determination is not straightforward. An efficientnumerical technique is necessary to obtain significant results. Two approaches are herein presented:The solution of the inverse problem with a power series technique (PST) and the useof artificial neural networks (ANNs). The free vibration problem of a Bernoulli-Euler beamwith an intermediate spring is stated and then solved with the PST. The first three flexuralfrequencies are measured and input in the algorithm. At this stage a numerical experimenton a 2D beam is employed. The crack depth is derived from a Mechanics of Fractures relationship.The ANNs technique is a different approach since it needs a training set of data. Asingle hidden layer back-propagation neural network is trained with data found with 2D fi-nite element models with more than four hundred scenarios. These data are also analyzed andsome curves are depicted to show the variables influence. The first methodology is very simpleand straightforward though no optimization is included. It yields negligible errors in thelocation and very small ones in the depth values. When using one network for the detectionof the two parameters the ANNs behave adequately. However better results are found whenan ANN is used for each parameter. Finally a combination of the two techniques is proposed:The location found with the first technique and the depth with the second one.