Abstract: An online prediction method for tool wear based on feature transfer is proposed to predict the wear state of new cutting tools in the unprocessed stage of batch processing. Firstly, the complete cutting force, vibration, and acoustic emission signals during the tool life cycle of the historical tool, as well as a small amount of sensor data from the early stage of new tool processing, are collected. Use recursive feature elimination and cross validation to select sensor features that are highly correlated with tool wear. Secondly, based on the high similarity of batch machining processes, a relationship model of signal features selected for historical and new tool machining processes is constructed to generate feature data for the complete tool life cycle of the new tool machining process. Evaluation of the migration model's effectiveness is based on an analysis of the probability density distribution of signal features before and after the migration process. Finally, deep forest is used to predict the tool wear status during the new tool machining process. Using the PHM2010 milling cutter public dataset for validation, a method was proposed to achieve an accuracy of over 90% in identifying tool wear status during the machining process of new cutting tools, verifying the effectiveness of the method.