Journal of Theoretical
and Applied Mechanics

42, 3, pp. 445-460, Warsaw 2004

### Fuzzy-neural and evolutionary computation in identification of defects

Tadeusz Burczyński, Piotr Orantek, Antoni Skrobol
It is known that an elastic body contains some internal defects such as voids, cracks, additional masses, etc. This paper is devoted to a method based on computational intelligence for non-destructive defect identification. In the presented paper, an elastic body loaded statically is considered. The body contains an unknown number of internal defects. There are a lot of applications based on non-destructive methods. The Evolutionary Algorithm (EA) with the Boundary Element Method (BEM) is a very effective tool in the identification of internal defects. In this method, the fitness function is calculated for each chromosome in each generation by the BEM. The number of chromosomes in each generation is quite large, and the number of generations is also large, so the time needed to carry out the identification is very long.

Methods based on Artificial Neural Networks (ANN) find the position and shape of internal defects in a very short time. Because ANNs are usually trained using gradient methods, the risk that the solution is in a local optimum is one of disadvantages of such a method. There is also a problem when the ANN has to identify two or more different kinds of defects (cracks, voids and additional masses) in one body.

In the presented method, an EA is connected with the ANN in one system. This operation allows to avoid main disadvantages of these methods and to use their advantages. The evolutionary algorithm is applied to identify the number of defects and their parameters (position and size).

The identification of a defect in the body is performed by minimizing the fitness function which is calculated as a difference between measured and computed displacements in some sensor points on the boundary of the investigated structure. The fitness function is computed using an Artificial Neural Network (ANN).
Keywords: fuzzy neural network; evolutionary algorithm; defect; identification; boundary element method