PSI - Issue 23

Hynek Lauschmann et al. / Procedia Structural Integrity 23 (2019) 107–112 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

109

3

expressing the dependence of the mean crack growth rate y , features f i and their quadratic errors 2 the value of x . Now x is understood as continuous along the interval  1, n  of former indices 1,..., n . Then a locality of a fracture surface characterized with a set of features  i may be assessed crack growth rate corresponding with minimum of weighted distance of the vector of features  i  from the trained model courses of features ( ) i f x :   2 2 1 ˆ ( ), where argmin ( ( )) ( ) f n i i i X i y y f x s x          . (1) Crack growth rates for both training and testing data are estimated and compared with experimental values. Good results on testing data show that the algorithm of classification is not contaminated by individual random fluctuations of training data. So it may be applied on any other crack surfaces to estimate an unknown course of crack growth rate. 3. Experiment Three CT specimens (Fig. 1a) from the steel AISI 304L were loaded in air under room temperature with force control by 3 types of loading:  Specimen A1: constant amplitude loading ( F min = 200 N, F max = 3600 N, R = 0.04),  Specimen A2: block of 199 identical cycles ( F min = 200 N, F max = 3600 N) followed by an overload ( F min = 200 N, F max = 5200 N),  Specimen A3: block of 1000 cycles with log-normally distributed maximum ( F min = 200 N, F max   200, 5250 N  ). Dynamic testing machine INOVA ZUZ 50 in the Dept. of Materials, FNSPE, was used for fatigue tests. Crack development was measured and recorded in detail. Local crack growth rates were estimated from these records. 2 i i s f f   , on ( ) i

fatigue

a b Fig. 1: a - CT specimen, b - areas of 3D reconstructions (a schematic plot)

4. Fractography Description of fracture surfaces in 3D. Fracture morphologies were documented by 3D reconstructions in about 110 fields of size 210 x 315  m located along the central axis of fracture surface (Fig. 1b). Areas of reconstruction were oriented with shorter dimension in crack growth direction, neighbouring areas overlapped by 10  m. 3D reconstructions were computed by in house-developed software DFFL from sets of equidistantly focused images provided by a modified metallographic microscope Reichert ME F2. An example is shown in Fig. 2. Let us denote a single 3D structure z ( x,y ), where x and y are coordinates in projection plane (rectangle), and z is height of fracture surface measured from an arbitrary level. The real resolution of 3D reconstructions in plane ( x,y ) is about 5  m, the resolution in axis z is about 0.5  m. The mean local crack growth rates were assigned to 3D reconstructions according to their locations. Dimensional decomposition. 3D reconstructions of fracture surfaces were decomposed into 8 sections by means of sequential wavelet transformation (Jirou šková, 2018) . On each level, the resolution is about one half of the previous, and present surface shapes are consequently greater. An example is shown in Fig. 3.