PSI - Issue 57

Giorgio A. B. Oliveira et al. / Procedia Structural Integrity 57 (2024) 228–235 Giorgio A. B. Oliveira et al./ Structural Integrity Procedia 00 (2023) 000 – 000

231

4

t

t

Fig. 2. Finite element model considered to simulate the fretting fatigue experiments.

3.2. ANN model

Based on the results obtained from FE numerical simulations, the equivalent stresses required by the ANN models can be calculated. The ANN model, named SHEAR, was trained and validated using a total of 132 fretting fatigue experiments from a previous study (Brito Oliveira et al., 2023). The ANN model consists of three inputs, all normalized by the material's ultimate tensile strength ( )., as illustrated in Figure 3. The first two input are based on the critical plane from the Modified Wohler Curve Method (Susmel and Lazzarin, 2002), that is: the shear stress amplitude ( ), and the maximum normal stress ( , ). To account the stress gradient effect in fretting problems, this modelis based on the Theory of Critical Distances (TCD), where the multiaxial stress inputs are computed at the material critical distance ( /2 ), which, for this material, is considered as 0.026 mm (Brito Oliveira et al., 2023). The third input parameter is the material yield stress ( ). The choice of these input parameters is well explained in earlies studies (Brito Oliveira et al., 2022 and Brito Oliveira et al., 2023). As shown in Fig. 3, the output is the normalized fretting fatigue life in the logarithmic scale, where the maximum life value ( , ) is assumed to be 10 8 . This study will employ this pre-existing model to analyze new data from Tables 1 and 2. The ANN models employ a multi-layer perceptron network architecture with one hidden layer, as shown in Fig. 3. The training process involves the Backpropagation algorithm, and cross-validation is utilized to optimize results by considering the empirical risk error (Haykin, 2008).

,

,

Fig. 3. Architecture for the SHEAR ANN-based model considered.