This study focuses on predicting fatigue crack paths and fatigue life in modified compact tension specimens, under mixed mode and variable amplitude loading conditions, using Machine Learning techniques. Mixed-mode conditions were induced by using specimens that incorporated holes with different radii and center coordinates. Initially, multiple Finite Element Method (FEM) simulations were conducted to determine the fatigue crack path for different configurations. Subsequently, several configurations were selected for experimental fatigue testing, in which the fatigue crack path was monitored and recorded. The final phase of the study involved Machine Learning (ML) techniques, specifically Artificial Neural Networks (ANN) and k-Nearest Neighbors (kNN), to predict fatigue crack propagation. The models were trained using different numerical and experimental data. Predicted results were then compared with experimentally tested data, and the behavior and accuracy of the models were evaluated. Overall, the implemented models demonstrated the ability to predict fatigue crack path with average deviations (ANN – 1.19 mm; kNN – 1.10 mm) closely resembling results obtained through Finite Element simulations (1.65 mm). The models were also able to predict fatigue life with average errors of 10.1 % (ANN) and 16.7 % (kNN), all achieved with a reduction of computational costs greater than 90 %.

中文翻译：

---

使用机器学习研究变幅载荷下改良 CT 试件的疲劳裂纹扩展

本研究的重点是使用机器学习技术在混合模式和变幅加载条件下预测改进的紧凑拉伸样本的疲劳裂纹路径和疲劳寿命。通过使用包含不同半径和中心坐标的孔的样本来诱导混合模式条件。最初，进行了多次有限元法 (FEM) 模拟以确定不​​同配置的疲劳裂纹路径。随后，选择了几种配置进行实验疲劳测试，其中监测和记录疲劳裂纹路径。研究的最后阶段涉及机器学习 (ML) 技术，特别是人工神经网络 (ANN) 和 k 最近邻 (kNN)，以预测疲劳裂纹扩展。使用不同的数值和实验数据训练模型。然后将预测结果与实验测试数据进行比较，并评估模型的行为和准确性。总体而言，所实施的模型证明了预测疲劳裂纹路径的能力，其平均偏差（ANN – 1.19 mm；kNN – 1.10 mm）与通过有限元模拟获得的结果（1.65 mm）非常相似。该模型还能够预测疲劳寿命，平均误差为 10.1% (ANN) 和 16.7% (kNN)，所有这些都是在计算成本降低超过 90% 的情况下实现的。