An accurate cutting tool wear prediction method under different cutting conditions based on continual learning

Abstract

Cutting tool wear prediction plays an important role in the machining of complex aerospace parts, and it is still a challenge under varying cutting conditions. To overcome the limitations of the existing methods in generalization ability when dealing with cutting conditions changing largely, this paper proposed a novel cutting tool wear prediction method based on continual learning. A meta-LSTM model is firstly trained for specific cutting conditions and can be easily fine-tuned with very small number of samples to adapt to new cutting conditions. Specifically, the meta-model could be continuously updated as machining data increase by using an orthogonal weights modification method. The experiment results show that the proposed method can realize accurate prediction of tool wear under different cutting conditions. Compared with existing methods including meta-learning methods, the range of adapted cutting conditions could be expanded as the task distribution of new cutting conditions is continuously learned by the prediction model.