94 M. Gollnick et al. Fig. 9.4 Left: Setup for the Experimental Modal Analysis on a flange. The structure with free boundary conditions is excited by a shaker between 4–12 kHz. Acquisition of the system response via 3D-LSV. Right: Setup for the Experimental Modal Analysis on a rotor blade. The structure fixed on one side was excited broadly by an automatic modal hammer (WaveHit, gfai tech GmbH). Acquisition of the system response via 3D-LSV Fig. 9.5 Left: First CMIF curve of a stainless steel flange. Right: First CMIF curve of an UAV rotor blade. Classified peaks are marked To create the training data, indicator functions with a sampling rate of 20 kHz and a frequency spacing of 0.0005 Hz were generated. The amplitude, decay and frequencies of the sinusoidal oscillation were randomly generated and white noise was added. The training set consisted of a total of 6044 peaks and 99,940 non-peaks. The classification was applied to the CMIF curves of the stainless steel flange and rotor blade. Both objects were excited at one reference point, shaker and modal hammer, respectively. This meant that only one CMIF curve was available for the evaluation of both measurement objects (see Fig. 9.5). Figure 9.5 shows the results of the classification for the stainless steel flange and rotor blade using the first CMIF curve. It becomes clear that nearly all peaks are found correctly by the applied method. Peaks with too low height and to low left or right prominence are classified as non-peak. The frequency 6043 Hz for the stainless steel flange and the frequencies 1023 Hz, 1775 Hz and 2405 Hz for the rotor blade were not detected as peaks. This result is due to the fact that these peak characteristics were not taken into account during the learning process. The amount of training must therefore be adjusted with regard to these variants. In summary, it becomes clear that peaks and non-peaks in the feature space are distinct from each other (see Figs. 9.6, 9.7 and 9.8) and that the respective features complement each other.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==