关键词: 混凝土裂缝, 分级标准, 机器学习, BP 神经网络

Abstract: Concrete cracks are common in concrete dams, and their classification results directly affect safety assessment and treatment decisions. To address the limitations of traditional crack classification methods, which rely heavily on expert experience and have strong subjectivity, this paper proposes a machine learning-based crack classification method. First, combining engineering practice experience with specification requirements, a classification standard system is constructed, which includes crack location, length, width, depth, and development trend. Then, the BP neural network model is adopted. By training with sample data, automatic crack classification is realized, and the accuracy and reliability of the model are verified. The study results show that the accuracy of the proposed method on the test set can reach 80%, and it is highly consistent with expert rating results, providing a fast, accurate, and efficient technical means for concrete crack classification.

Key words: concrete crack, classification standard, machine learning, BP neural network