PSI - Issue 72

Oleh Yasniy et al. / Procedia Structural Integrity 72 (2025) 181–187

186

The stable accuracy of the model for unseen data indicates that the network was able to learn the basic physical laws that determine the crack growth rate. The error of 0.4% was obtained by the neural network method in the test sample at R = 0.03. The parameters of the constructed neural networks for two stages are summarized in Table 1. Table 1. Neural network parameters Stress ratio Algorithm of learning Error function Function of hidden activation Name of network

Function of output activation Exponential

R = 0.03, 0.1, 0.3

MLP 2-30-1 BFGS MLP 1-25-1 BFGS

SOS SOS

Logarithmic Tangential

R = 0.03

Identity

Table 1 provides the specifications of two neural networks for modeling the effect of stress ratio R on the dependence of crack length on the number of loading cycles. Network MLP 2-30-1 has more flexibility because it considers several R values together. In contrast, network MLP 1-25-1 is optimized for one specific value of R = 0.03, which allows it to achieve better accuracy in this task. 4. Conclusions This paper considers the application of neural networks to predict the durability of structural elements made of titanium alloys under cyclic loads. In particular, the proposed model based on a multilayer perceptron accurately predicts the length of a fatigue crack. Prediction error does not exceed 0.4% for test samples, which indicates the model ability to reproduce physical laws. The neural network has shown the ability to adapt to different stress ratios R =0.03, 0.1, 0.3, and loading conditions. Additionally, the absence of significant deviations between training and test data confirms that the model is not prone to overfitting and generalizes the data well. The model allows for an accurate assessment of the remaining life of the material and helps to make decisions on the maintenance and operation of structures. Thus, using neural networks makes it possible to combine traditional fracture mechanics models with machine learning methods, increasing the accuracy of predictions and reducing the dependence on the scatter of experimental data. Therefore, the proposed model is a reliable tool for assessing the behavior of structural elements under cyclic loading, reducing the risks of catastrophic failures in real-world operating conditions. References Haykin, S., 2009 “Neural Networks and Learning Machines”, Third Edition, McMaster University, Hamilton, Ontario, Canada, 938 Konovalenko, I., Maruschak, P., Brezinová, J., Viňáš, J., & Brezina, J., 2020. Steel surface defect classification using deep residual neural network. Metals, 10(6), 846 Liu, J., Chen, J., Sun, Z., Zhang, H., Yuan, Q., 2023. A Study on Fatigue Crack Closure Associated with the Growth of Long Crack in a New Titanium Alloy. Metals 13, 1377 Liu, Y., Mahadevan, S., 2005. An Artificial Neural Network-Based Algorithm for Evaluation of Fatigue Crack Growth. International Journal of Fatigue 27, 790 – 795 Mohanty, J. R., Verma, B. B., Parhi, D. R. K., Ray D. R., 2009. Application of artificial neural network for predicting fatigue crack propagation life of aluminum alloys. Archives of Computational Materials Science and Surface Engineering 1, 133 – 138 Monticeli, F. M., Neves, R. M., Ornaghi Jr, H. L., Almeida Jr, J. H. S., 2022. Prediction of Bending Properties for 3D-Printed Carbon Fibre/Epoxy Composites with Several Processing Parameters Using ANN and Statistical Methods. Polymers 14, 3668 Moussouni, A., Benachour, N., Benachour, M., 2023. Modeling of Fatigue Crack Growth by Neural Networks. International Conference on Scientific and Academic Research 1, 215 – 219 Murakami, H., Okazaki, Y., 1996. Bayesian Neural Network Analysis of Fatigue Crack Growth Rate in Nickel-Base Superalloys. ISIJ International 36, 1373 – 1378 Pidaparti, R. M. V., Palakal, M., 1995. Neural network approach to fatigue-crack-growth predic¬tions under aircraft spectrum loadings, Journal of Aircraft 32, 825-831 Chen, J., Liu, Y., 2022. Fatigue modeling using neural networks: A comprehensive review. Fatigue & Fracture of Engineering Materials & Structures, 45(4), 945-979