The invention discloses a tool wear monitoring system capable of realizing self-learning in a numerical control machining state. The structure of the system is that: a tool wear rule learning library stores learned tool wear rules; a data acquisition and judgment module provides a servo drive current digital signal for a data processing module to perform format conversion and store the signal as a monitoring current signal; a characteristic extraction and selection module selects signal characteristics strongly correlated with tool wear; a fitted predicted trend curve module establishes a signal characteristic-tool machining service life relationship curve; a tool wear rule module obtains the learned tool wear rules from the tool wear rule learning library, and substitutes the trend signal characteristics into the learned tool wear rules to obtain a tool wear extent; and a tool wear compensation and tool replacement module makes a tool wear compensation and tool replacement decision according to the tool wear extent, and provides the tool wear compensation and tool replacement decision for a numerical control system interface input module. The system is applied to the condition offluctuating cutting parameters in mass production, and can achieve relatively higher accuracy under the condition of constant cutting parameters.