PSI - Issue 38

S. Häusler et al. / Procedia Structural Integrity 38 (2022) 230–237 S. Häusler et al. / Structural Integrity Procedia 00 (2021) 000 – 000

234

5

Due to visco-elastic effects of the resin and internal friction combined with poor heat conductivity of the material, high frequencies lead to self-generated heating. After an initial rise of 1-2°C the average temperature in the shear specimen increased in a parabola shape about 5°C and rapidly peaked at final failure. Especially at the shear tests with R = -1 an overlapping of growing damage with localised heating has been observed. Consequently, the heating based on damage initiation and friction of crack surfaces and later of delamination is another reason for temperature rise and overlaying the described frequency effect. The standards of ISO (2003) and ASTM (2019) appear to address only the frequency related heating with a recommended limit of 10°C in temperature rise. Beside the sudden temperature rise due to final failure, the phenomenon of damage depending friction is not discussed any further. Thus, the three data sets encountering local overheating connected to the damage growth are included in the S-N curve, Figure 2 (d) marked in orange. However, it should be noted that the average surface temperature of the locally overheated specimens exceeded the standard’s limit only close to final failure. To additionally avoid local overheating in the remaining experiments, a vast reduction in testing frequency would have been necessary. Consequently, compressed air cooling has been added to prevent overheating, so the frequencies could be kept comparable. After all, comparing the data of this dataset, the locally overheated tests are located within the scatter, Figure 2 (d). 3.2. Stiffness degradation Stiffness degradation was measured by the decrease in cyclic modulus during the fatigue loading, calculated from minimum and maximum stresses and strains of certain load cycles. Therefore, the data sets displayed in Figure 2 are analysed, excluding the blue marked ones due to technical issues. To determine errors by the DIC analysis, the anomaly finding isolation forest algorithm was used, presented by Liu et al. (2008). The advantage in applying this algorithm is the local consideration of anomalies and no filtering when a locally consistent scatter is given. Drastic changes of the slope of the examined data set state a problem in finding outliers, so the curve was flattened before the filter was applied. Therefore, a centred moving median with a small window size is subtracted from each respective point. Finally, the filtered data can be summarised to one representative degradation curve, as displayed in Figure 3. Here a Savitzky-Golay filter with a window size of 5% of the dataset’s length and with linear approximation was used, to average the data and reduce noise. For a better comparability of the trend, the data have been normalised. The cyclic modulus is normalised by the initial value and the cycles by the final cycle count.

Figure 3: Representative stiffness degradation curves averaged and smoothed by a Savitzky-Golay filter. a) Θ = 90° specimen at R = -1 and b) Θ = 45°at R = 0.1. Number of cycles and the stress amplitude are given in the legend.