Milling force prediction model based on transfer learning and neural network

In recent years, the growing popularity of artificial neural networks has urged more and more researchers to try introduce these methods to the machining field, with some of them actually producing good results. The acquisition of cutting data often means higher cost and time, limiting the application of neural network in the machining sector, to a certain extent. In this paper, for the task of cutting force prediction, a “transfer network” was established, based on data obtained by simulation, combined with the theory and method in the field of transfer learning. Compared to “ordinary network”, that is, traditional back-propagation neural network based on experimental samples alone, transfer network exhibits obvious performance advantages. On one hand, this means that, using the same experimental samples, the prediction error of transfer network will be controlled; while on the other hand, when the same prediction error is achieved, the number of experimental samples required by the transfer network will be less.