Transfer Learning in CNC Milling Machines for Chatter Detection using LSTM-AutoEncoders

In this work, machine learning is applied to develop a LSTM-AutoEncoder for anomaly detection in three-axis CNC machines. This anomaly detection network is then transferred to another three-axis CNC machine for chatter detection, using significantly less data. This network is then extended to five-axis CNC machines by using the encoder from the three-axis CNC machine to develop an anomaly detection network using transfer and incremental ensemble learning. This approach is compared to a network trained from scratch, with comparable results observed. This approach demonstrates the feasibility of augmenting networks designed for three-axis CNC machines to five-axis CNC machines.