FLAWS DETECTION IN THIN PLATE STRUCTURES BASED ON SBFEM AND MACHINE LEARNING

Combining Scaled Boundary Finite Element Method (SBFEM) and machine learning algorithm, the flaw information in the structure is quantitatively reversed based on the dynamic responses of the structure when the Lamb wave propagates in the thin plate structure with flaws. By changing the scaled center position of the subdomain where the crack is located to reflect different crack information, the SBFEM minimize the remeshing processes and improve the computational efficiency. It can provide enough training data reflecting flaw characteristics. The extreme learning machine, a machine learning model based on artificial neural network, avoids the iterative process of minimizing the objective function in the traditional inversion analysis, and can guarantee the precision of flaw detection with extremely fast learning speed. Therefore, the proposed inversion model greatly reduces the time-cost for flaw detection. The several numerical examples show that for a thin plate structure with crack-like flaws, the proposed model can accurately detect the location and size of the crack on the grounds of dynamic response signal of the observation point during the propagation of Lamb wave. The method also has high inversion efficiency and it is robust against noise.