A novel chatter detection method for milling using deep convolution neural networks

Regenerative chatter is harmful to machining operations, and it must be avoided to increase production efficiency. The recent success of deep learning methods in many fields also presents an excellent opportunity to advance chatter detection and its wider industrial adoption. In this work, a chatter detection method based on deep convolutional neural network (DCNN) is presented. The method uses a cardinal model-based chatter solution to precisely label regenerative chatter levels. During milling, vibration data are collected via a non-invasive data acquisition strategy. Considering nonlinear and non-stationary characteristics of chatter, continuous wavelet transform (CWT) is used as the pre-processing technique to reveal critical chatter rich information. Afterward, the images are used for training and test of the developed DCNN. The validation of the method revealed that when cutting parameters are also included as input features to the DCNN, average accuracy reached to 99.88%. [Display omitted] •Regenerative chatter is detected using wavelet transfer and deep convolutional neural networks.•Stability lobes are used for accurate labeling of experimental cuts.•RPM and DOC are added as input scalars to fully connected layers.•The custom designed DCNN predicts with 99.88% average accuracy.•A low-cost and non-invasive accelerometer is used to enable rapid industry adoption.