Abstract: The advancement of deep underground engineering demands a deeper understanding of the progressive failure mechanisms of surrounding rock and the refinement of instability warning methods.This study introduces a surrounding rock instability criterion based on the Volume Expansion Rate (RockVER).The method was validated through machine learning (ML) evaluation tests,numerical simulations of rock monitoring,and analysis of original observation data from the Xiaolangdi underground caverns.First,the study analyzed how RockVER characterizes the bearing capacity of surrounding rock and evaluated its effectiveness in identifying rock damage using ML techniques.Second,discrete element method simulations were conducted to investigate RockVER behavior under various deformation modes.RockVER was also applied to reproduce the safety monitoring and warning process for the Xiaolangdi caverns.The findings show that RockVER exhibits distinct patterns under different deformation conditions,including stability and instability.Its trends and thresholds qualitatively and quantitatively describe rock damage states.A RockVER-based safety control warning system was developed,incorporating its qualitative and quantitative criteria as secondary warning signals.