PSI - Issue 72

Oleh Yasniy et al. / Procedia Structural Integrity 72 (2025) 181–187 183 In this study, we aimed to develop a neural network model to predict the growth of fatigue cracks by analyzing the relationship between crack length and the number of loading cycles . The model was trained on experimental data collected under different stress ratios , which is defined as the ratio of the minimum to maximum load during a loading cycle. Using machine learning allowed us to improve the accuracy of predictions and capture the underlying mechanisms that govern fatigue crack growth. 2. Material and methods Fatigue lifetime is a critical parameter that determines structural material reliability. A deep understanding of this phenomenon allows us to predict the behavior of materials under cyclic loading, improve their properties, and avoid catastrophic failures in complex technical systems. Therefore, using a neural network to solve the problem of predicting titanium alloy structural elements is important. Among all NN applications in fatigue, fatigue lifetime prediction is the most widely studied topic (Fig.1).

Fig. 1. The most common applications of neural networks in fatigue research by Chen et al. (2022)

A neural network is a computational model consisting of neurons connected to each other in a certain structure. A neural network solves complex problems, such as classification, prediction, pattern recognition, modeling, etc. Fig. 2 shows an example of a simple neural network consisting of three layers: an input layer, a hidden layer, and an output layer. Information in the NN passes from the input layer through the hidden layer to the output layer by Haykin (2009). Additionally, backpropagation is a fundamental algorithm used to train neural networks. It optimizes the weights and biases in the network by minimizing the loss function. The algorithm works by calculating the gradient of the loss function with respect to each network parameter, starting from the output layer and moving back to the input layer. Backpropagation is an important component of modern machine learning, providing flexibility and power to neural networks. In general, neural networks are one of the key tools of modern machine learning and artificial intelligence, capable of solving a wide variety of tasks with extreme accuracy. Our studies were performed in two stages. In the first stage, the test sample was randomly selected by the computer from all the experimental data at different stress ratio R = 0.03, 0.1, 0.3. And in the second, the training sample (80%) contained experimental dependencies of crack length on the number of loading cycles at R = 0.03; while the test sample (20%) at R = 0.03 was chosen to check the prediction quality.