Milling Chatter Monitoring based on Improved Deep Residual Network and Transfer Learning Approach*

The multi-cavity and thin-wall parts commonly utilized in aerospace have suffered from chatter during cavity milling. This paper proposes an online chatter monitoring method for the cavity milling process based on an improved deep residual network and transfer learning. First, input the time-frequency maps obtained by the short-time Fourier transform approach into the deep residual network, and fuse the trained deep features with artificial chatter features to build the chatter monitoring model for low-speed processes. Then, using the established model and small sample data, a new model suitable for high-speed operations is quickly set up through the parameter transfer learning method. The experimental results demonstrate that the classification accuracy of the above two models is 99.06% and 99.38%, respectively, which is superior to the threshold model based on artificial features and can be applied to the chatter monitoring of cavity milling under complex cutting paths and variable process parameters.