This paper proposes an autonomous method for detection and classification of fatigue crack damage and risk assessment in polycrystalline alloys. In this paper, the analytical and computational tools are developed based on convolutional neural networks (CNNs), where the execution time is much less than that for visual inspection, and the detection and classification process is expected to be significantly less error-prone. The underlying concept has been experimentally validated on a computer-instrumented and computer-controlled MTS fatigue testing apparatus, which is equipped with optical microscopes for generation of image data sets. The proposed CNN classifier is trained by using a combination of the original images and augmented images. The results of experimentation demonstrate that the proposed CNN classifier is able to identify the images into their respective classes with an accuracy greater than 90%.

本文提出了一种用于检测和分类多晶合金疲劳裂纹损伤和风险评估的自主方法。在本文中，基于卷积神经网络（CNN）开发了分析和计算工具，其执行时间比目测检查的执行时间短得多，并且检测和分类过程预计将大大减少出错的可能性。基本概念已在计算机仪表和计算机控制的MTS上经过实验验证疲劳测试设备，该设备配备了用于生成图像数据集的光学显微镜。通过使用原始图像和增强图像的组合来训练提出的CNN分类器。实验结果表明，提出的CNN分类器能够以大于90％的精度将图像识别为各自的类别。