Machine learning-based real-time visible fatigue crack growth detection

Abstract

Many large-scale and complex structural components are applied in the aeronautics and automobile industries. However, the repeated alternating or cyclic loads in service tend to cause unexpected fatigue fractures. Therefore, developing real-time and visible monitoring methods for fatigue crack initiation and propagation is critically important for structural safety. This paper proposes a machine learning-based fatigue crack growth detection method that combines computer vision and machine learning. In our model, computer vision is used for data creation, and the machine learning model is used for crack detection. Then computer vision is used for marking and analyzing the crack growth path and length. We apply seven models for the crack classification and find that the decision tree is the best model in this research. The experimental results prove the effectiveness of our method, and the crack length measurement accuracy achieved is 0.6 ​mm. Furthermore, the slight machine learning models help us realize real-time and visible fatigue crack detection.