Precursory processes and short-term forecasting model of mode I and mode II single-crack failure in rocks
Jian‐Zhi Zhang, Wentao Wu, Ting Zhang
Abstract
Short-term failure forecasting of rocks is a critical scientific issue in predicting major geohazards such as rockbursts, landslides, and earthquakes. This study conducted Mode I and Mode II single-crack failure tests on rock specimens, combined with real-time acoustic emission (AE) monitoring, to investigate the precursors to local instability and their multidimensional damage evolution. A failure forecast model (FFM) based on the time-reversed Omori law, along with a long short-term memory (LSTM) neural network failure forecast model, was proposed to quantitatively predict the single-crack failure time of rocks. The results indicate that the precursory processes manifested as power-law damage acceleration in the time dimension, damage localization in the spatial dimension, and high-energy damage in the energy dimension, which demonstrate physical synchronicity. The traditional FFM, improved FFM, and LSTM neural network failure forecast model all accurately predict the single-crack failure time of rocks, exhibiting good statistical reliability and robustness, with the predictive advantage of the LSTM model being particularly notable. Utilizing the AE event rate and AE amplitude rate as predictive factors can enhance prediction quality. However, the local time series formed by integrating the cracking nature and energy mechanism does not fully represent the complete energy consumption process, resulting in a short early warning time. Additionally, this study shows that the localized zone, which plays a controlling role in rock strength, serves as a reliable precursor source for failure prediction, providing computational technical support for short-term failure forecasting of rocks.