Chatter detection in high-speed milling processes based on ON-LSTM and PBT

Chatter is a kind of self-excited vibration which frequently occurs in high-speed milling processes, which induces severe damage to both spindle tools and workpieces. In this paper, we introduce a new chatter detection technique using ordered-neurons long short-term memory (ON-LSTM) and population based training (PBT). First, we conduct a large number of milling experiments on a computer numerical control (CNC) milling machine with 4 accelerometers to get the dataset and employ vanilla LSTM for chatter detection. Then, to interpret the performance on time series of recurrent neural networks (RNN), a variation of LSTM named ON-LSTM is applied to chatter detection and a hyperparameter tuning method PBT is used for training. Finally, we compare the trained ON-LSTM with the time-frequency spectrum of the original signals obtained by short-time Fourier transform (STFT), and they show a certain degree of consistency.