In vibration-assisted drilling, the wear state of the drill bit affects the processing quality of the hole. The traditional method of identifying the wear state of the drill bit adopts the method of packet decomposition, ignoring the timing characteristics of the signal. In this paper, the force and acoustic emission signals in vibration-assisted drilling are used. The Gram angle field converts the onedimensional time series into a two-dimensional image, while retaining the trajectory of the time series in the high-dimensional space. Based on the Graham difference field (GADF) image of force and AE, the Inception improved convolutional neural network (IN-CNN) is used to identify the wear state. The experiment proves that compared with the traditional convolutional neural network, BP neural network and support vector machine, the recognition rate of IN-CNN drill wear state based on GADF is 93.1 %, which is increased by 2.5 %, 10.6 % and 8.1 % respectively. It provides a reliable condition monitoring method for the state identification of the drill bit in semi-closed vibration-assisted machining, and has practical engineering significance for improving the machining accuracy and efficiency of composite equal-holes.