A sound-based machine learning method for crack-type recognition in hard rock

The recognition of tensile and shear cracks during hard rock cracking is critical for early warning of rockbursts in deep rock engineering. However, direct observation of cracking inside hard rocks by imaging equipment is difficult. A sound-based machine learning method for crack-type recognition in hard rock is therefore proposed. First, the sound signals of tensile and shear cracks in granite are obtained by Brazilian tension and shear tests, respectively. Then, the spectrogram conversion of the two kinds of signals is conducted to build a dataset. Next, a deep learning network EfficientNet is used to automatically extract the features of the spectrograms. Last, these deep learning–based features are used to construct a classification model of the crack types by a shallow machine learning method CatBoost. The experiments showed that the combination of two learning methods achieves high accuracy. We further validated the performance of the proposed method in laboratory cases involving biaxial and triaxial compression, as well as in real-world cases. The results indicate that the proposed method can accurately analyze the failure process of rocks by recognizing crack types. The proposed method is straightforward to implement and can provide a sound basis for making informed decisions on early warning of rockbursts.