Efficient stability prediction of milling process with arbitrary tool-holder combinations based on transfer learning

Chatter occurring in the milling process can seriously deteriorate the machining efficiency and surface quality. The stability diagram predicted using the tool tip frequency response functions (FRFs) is an effective approach to avoid the chatter vibration. The tool tip FRFs highly depend on the characteristics of the tool-holder-spindle-machine tool frame assembly. Thus, when the tool-holder assembly or only the tool overhang length changed, the FRFs will be reobtained to plot the stability diagrams. Considering this time-consuming situation, this paper introduces the transfer learning to efficiently predict the milling stability of arbitrary tool-holder combinations. First, a source tool-holder assembly is selected to measure sufficient overhang length-dependent tool tip FRFs and then predict the limiting axial cutting depth a(plim) values under different process parameters for forming the source data. For a new tool-holder assembly, impact tests are only performed under a few key tool overhang lengths to measure the tool tip FRFs and then predict the a(plim) values to form the target data. Combining the target data and the source data, the transfer learning containing the domain adaptation and adaptative weighting is introduced to train an overhang length-dependent milling stability prediction model of a target tool-holder assembly. A case study has been performed on a vertical machine tool with four different tool-holder assemblies to validate the feasibility of the proposed transfer learning-based milling stability prediction method.