Pose-dependent tool tip dynamics prediction using transfer learning

Machining chatter has a great influence on the producing efficiency and surface quality. Frequency response function at the tool tip is a crucial input for constructing accurate milling stability model. However, the tool tip dynamics usually change with the continuously varying postures of the machine tool axes during the whole machining process. How to predict the pose-dependent tool tip dynamics precisely has become one of the most challenging tasks in chatter suppression in both research and industry. Compared to traditional finite element analysis or kinematic modeling based methods, this paper proposes a data-driven method using transfer learning to predict the pose-dependent tool tip dynamics for different tool-holder assemblies. Firstly, a tool-holder assembly is selected as the source tool and its pose-dependent tool tip dynamics are obtained as the source data by sufficient impact tests. For a new tool-holder assembly, namely the target tool, only few impact tests are required to measure the tool tip dynamics as the target data. Then both the target and source data are used to train a regression model for predicting the target pose-dependent tool tip dynamics based on transfer learning by integrating domain adaptation and adaptive weighting. Furthermore, a detailed experimental validation with a five-axis machine tool is carried out to verify the accuracy and efficiency of the proposed method.