PSI - Issue 75

Jan Schubnell et al. / Procedia Structural Integrity 75 (2025) 94–101 Schubnell / Structural Integrity Procedia (2025)

98

5

Artificial Neural Networks (ANNs) are increasingly employed for fatigue assessment in engineering materials. As computational models mimicking human brain functionality, ANNs can learn from data and accurately predict fatigue life based on parameters like loading type, material composition, and stress cycles (Awad and Khanna, 2015). This not only enhances material design and maintenance efficiency but also reduces costs and augments safety measures. Characteristics of ANNs such as the number of layers and neurons, activation functions, learning rate, and loss function, play a pivotal role in determining the network's performance and accuracy of predictions. Thus, ANNs serve as a valuable tool for material fatigue prediction, leading to more reliable engineering designs and is used as an standard approach for fatigue analysis by ML (Chen and Liu, 2022)

{1024}

(a)

(b)

{256}

{21}

… … {3}

HV SCF Rz …

As a measure for the prediction accuracy of the ML approaches the 2 score, according to equation 3, and the Root Square Mean Error (RSME), according to equation 4 were used. The prediction quality is calculated as the coefficient of determination (R2-score). The optimal score is 1.0, and it has the potential to be negative (as the model could be arbitrarily worse) [33]. 2 =1− (3) where the is the sum of the squared differences between the predicted values (obtained from the regression model) and the actual values. The is the sum of the squared differences between each data point and the mean of the dependent variable. = √∑ ( − ) = 1 (4) The RSME represents the square root of the average squared differences between predicted and observed results and has the advantage that it has the same unit than the resulting value. An overview about the displayed investigations in this work is given in Table 3. Fig. 2. (a) Illustration of decision tree of Random Forest approach, (b) Structure of applied artificial neural network

Table 3. Overview about investigations in this work Target parameter DC1 DC2

(

,

DC3

DC4 ANN ANN

DC5

DC6

, ) -

RF*

- -

- -

- -

- -

RF ANN RF = Random Forrest approach, ANN = artificial neural network, ( , ) -approach see Figure 1 (a) and -approach (Figure 1 (b)) * (Fliegener, J. Rosenberger, M. Luke, J. Domínguez, J. Morgado, H.U. Kobialka, T. Kraft, 2024) RF RF, ANN RF, ANN ANN