The destructive behavior of rocks and the evolution behavior of cracks are highly correlated. With the continuous development process of crack, the stress state of rock is constantly broken until entirely failed, so it is necessary to study the spatial and temporal behavior characteristics of the crack in the process of rock destruction. In this paper, the destruction process of phyllite specimens is analyzed by thermal imaging technology, and the temperature evolution process of the crack is studied to explore the infrared characteristics of the crack evolution process. Furthermore, a model for predicting rock destruction time is proposed based on Bi-LSTM recurrent neural network model combined with Attention mechanism. The results show that: (1) During the development of rock cracks, the rock surface shows a stable dynamic infrared response, and shows different evolutionary characteristics in different stages, mainly including temperature reduction in the compaction stage, temperature rise in the elastic and plastic stages, and temperature peaks in the failure stage; (2) During the evolution of the crack, rock destruction has a significant control effect on the IRT field along the fracture tangential and normal direction, and its distribution has the volatility controlled by the time; (3) The recurrent neural network method is used to predict the rock failure time, the results can be used as a method to predict the time of rock destruction, and it can be further put forward the corresponding protective measures accordingly, to maintain the long-term stability of the rock mass.
© 2023. The Author(s).