The nondestructive testing (NDT) of defects in structure members or geological bodies is essential in engineering survey and construction. Given the complexity of natural rock or rock masses, the NDT of defects within them is extremely challenging in geotechnical engineering. Previous studies predominantly utilized singular or limited features of waveforms, which neglected much information in the signals. In addressing this gap, a machine learning-based NDT method is proposed in this study using more features of wave signals. In this method, 13 features are extracted from 259,595 groups of signals to train the model, and the K-nearest neighbour (KNN) and support vector machine (SVM) supervised learning algorithms are applied to justify the types of the samples. This method shows high accuracy and precision in classifying the cross-section shape, integrity, and flaw geometries of the samples. The KNN and SVM models both attain a 100 % accuracy rate in classifying cross-section shapes and a 99.31 % accuracy rate in classifying sample integrity. Moreover, the accuracy rate of classifying flaw numbers and widths is 97.15 %. Furthermore, the SVM model can significantly enhance computational efficiency with high accuracy. The results prove the great potential of the machine learning-assisted NDT method in geotechnical engineering, especially in detecting complex defects or discontinuities in rocks and even rock masses.

中文翻译：

---

一种基于时域波信号的机器学习辅助无损检测方法

结构件或地质体缺陷的无损检测（NDT）在工程勘察和施工中至关重要。鉴于天然岩石或岩体的复杂性，其内部缺陷的无损检测在岩土工程中极具挑战性。以前的研究主要利用波形的单一或有限特征，忽略了信号中的许多信息。为了解决这一差距，本研究提出了一种基于机器学习的无损检测方法，利用波信号的更多特征。该方法从259,595组信号中提取13个特征来训练模型，并应用K近邻（KNN）和支持向量机（SVM）监督学习算法来证明样本的类型。该方法在对样品的横截面形状、完整性和缺陷几何形状进行分类方面显示出高精度。 KNN 和 SVM 模型在横截面形状分类方面均达到 100% 的准确率，在样本完整性分类方面达到 99.31% 的准确率。缺陷数量和宽度分类准确率为97.15%。此外，SVM模型可以显着提高计算效率且精度较高。结果证明了机器学习辅助无损检测方法在岩土工程中的巨大潜力，特别是在检测岩石甚至岩体中的复杂缺陷或不连续性方面。