Few-Shot Learning for Smart Manufacturing: Tool Wear Prediction Using Out-of-Domain Data Based on Meta-Learning

Tool wear has a significant impact on many fields of the machining process, consisting of tool life, quality of the workpiece, and machining efficiency. Over the past few years, although numerous data-driven models have been proposed to accurately predict tool wear, they usually require abundant historical data. Meanwhile, predicting tool wear with limited historical data under various cutting conditions, especially with large variations is still a challenge due to the domain shift problem. To fill the research gap, this paper proposes a novel model-agnostic domain generalization (MADG) technique to effectively predict tool wear with small data under various cutting conditions. Through iterative training on randomly constructed subtasks, MADG can provide promising initial parameters for models to generalize among domains. Experimental results on NASA Ame milling data prove the validity and practicability of this technique.