Application of deep learning algorithms for crack detection after magnetic particle testing

In the metal and materials industry today, crack detection holds critical importance for structural integrity and safety. Magnetic Particle Testing (MPT) is a widely used method to detect surface defects in metal components. Analyzing materials using accurate methods is crucial for ensuring material longevity and safety. Failure to use proper techniques to identify invisible internal damages in materials can negatively impact quality assurance and long-term durability. Traditional MPT techniques rely on human interpretation and manual analysis, making it challenging to detect cracks accurately and efficiently. This study aims to examine the use of deep learning models in magnetic particle testing images. Deep learning algorithms, particularly Convolutional Neural Networks (CNN) and similar models, have demonstrated success in recognizing and classifying complex patterns due to their ability to learn from large datasets. The primary objective of this study is to evaluate and implement the performance of deep learning models for magnetic particle testing. In this thesis, a dataset comprising two classes crack and non-crack images was utilized. This dataset was classified using well-known deep learning approaches, and the results were compared based on key performance metrics. Among the methods employed, the VGG16 approach achieved the best result with an accuracy of 89.04%. This thesis study represents a significant step toward automating crack detection processes in industrial applications and minimizing human error. Moreover, it demonstrates the potential of deep learning models to enhance the accuracy and efficiency of magnetic particle testing methods, thereby opening new avenues for research and application in this field. As no similar study has been conducted previously, this thesis is expected to make a significant contribution to the literature.