|Ph.D Student||Rabinovich Daniel|
|Subject||Identification of Defects by Genetic Optimization Schemes|
|Department||Department of Aerospace Engineering||Supervisors||Professor Dan Givoli|
|Dr. Shmuel Vigdergauz|
|Full Thesis text|
Defect detection and identification is a crucial ingredient in non-destructive testing (NDT) of structures. Some forms of defect detection are practiced in various fields of application, e.g., in the aerospace industry.
A common methodology which is currently used for flaw detection using NDT tools is as follows. A signal is introduced which propagates through the specimen tested and produces boundary measurements. These measurements are compared by a human inspector to a reference measurement of a specimen without a flaw. If the difference between the two measurements is sufficiently large, a flaw is detected. This procedure indicates the mere presence of a flaw and gives some limited information on its location. The current work proposes that the human inspector be replaced with a computerized identification system. A model is appended to the system which simulates the wave propagation in a flawed specimen. Many possible flaws are modeled. The simulated measurements from each model are compared numerically with the true measurements generated by the NDT system for the given flawed specimen. When the difference between the computed and the actual measurement is minimal, the flaw is said to be identified.
Thus, the goal of the present work is the identification of defects in structures by simulation of a non-destructive process where interaction occurs between a wave traveling through the material and the defect. The characterization of the defect based on the measurements made is mathematically an inverse problem, which is solved in this case via a global optimization process using a Genetic Algorithm (GA). GA have been used in the past for identification of flaws in simple structures such as beams and frames either based on modal information or by using the pattern of the scattered wave. In the present work crack identification is performed in a general two-dimensional domain, using scattering data which are either time-harmonic or time-dependent.
The proposed method is applied to membranes which include defects of two kinds - a small inclusion and a crack. The membrane with the inclusion is modeled using the standard Finite Element Method (FEM), where the inclusion domain is characterized by an integer number of elements. The crack is simulated using the eXtended Finite Element Method (XFEM), which allows the solution of all direct problems with different cracks on a single fixed computational mesh, and also accounts for the singular behavior of the asymptotic field at the crack tip.