PSI - Issue 32

O.V. Bocharova et al. / Procedia Structural Integrity 32 (2021) 299–305 Bocharova O.V., Andzjikovich I.E., Sedov A.V., Kalinchuk V.V. / Structural Integrity Procedia 00 (2021) 000 – 000

303

5

method is based on the use of optimal orthogonal decompositions of signals in a basis that is specially tuned for the maximum possible sensitivity to the type, size and location of a particular defect. The output measured value is a scalar function of the response of the medium     * , 0,  н f t t T to the probe impact F . The time interval н Т is sufficient for stable detection of emerging diagnosed states. The training sample * f is formed from separate graphs   * i f t obtained for various types of defects in the structure, but for one type of probing non-destructive action. Each graph   * i f t can be considered as a vector of real values * * * * * 1 2 3 , , , ,       R T N i i i i iN f f f f f of response change in time. A preliminary procedure for estimation of amplitude spectra 1 2 3 , , , ,       R T N i i i i iN f f f f f based on the discrete Fourier transform is applied to the measured vectors * i f : were decomposed according to an orthonormal adaptive-tunable basis 1 2 3 , , , , m     , determined as a result of solving a set of optimization problems (Bocharova et al. 2016; Sedov et al. 2017): 0    i i f A A , where 0 A is constant component of the transformation. Using basis 1 2 3 , , , , m     , vectors  R N i f of responses are transformed into images   1 2 , , ...,   R m i i i im A a a a . For simplicity of the physical interpretation, we have realized the defectoscopy of samples by images   2 1 2 ,   R i i i A a a in two-dimensional feature space. As the two coordinates of image, we choose those that have the maximum spread. As investigation have shown, this dimension of feature space is sufficient for conducting qualitative diagnostics of samples. Proposed approach maximizes the difference of signal images, corresponding to various types of defects, obtained after processing. It allows to significantly increase the efficiency of inhomogeneity recognition. 3. Experimental investigation and results A series of experiments on the numerical signal processing has been carried out. The possibility of the method to determine the character of the inhomogeneity has been investigated. 1  N * 0 2 exp           ik i f f j k N  . Then the vectors  R N i f

Fig. 4. Location of images in the space of recognition for samples with various types of defects (accelerometer 1).