Prediction of Fatigue Crack Growth of Aluminium Alloy 7075T6 using Machine Learning Techniques

Abstract

As the fatigue cracks are initially invisible to the human eye, detecting them can be complicated. Large scale disruption and risk are directly caused by these cracks if not observed early or left unattended. Since the fatigue crack is challenging to locate, routine inspections or the use of ultrasonic instruments are the traditional methods for detection. Therefore, an image processing method enabled with computer vision and deep learning can be used to detect a fatigue crack. Provided there is an image data input, cracks are identified using a convolutional neural network described in this paper. Edge detector and line detection techniques are used to determine the crack's length and crack angle. Finally, the estimated values using the regression process are compared to the analytical values. At first, the values computed using both methods have excellent correlation. A minor variation can be seen at the final stage when the crack is in a transition stage where it is evolving from a stable crack to an unstable crack.