Real-time chatter detection during turning operation using wavelet scattering network

Chatter vibration is an undesired phenomenon in machining operations. Chatter can lead to reduced machining quality, productivity, and tool life. The main cause of chatter is the dynamic instability between the cutting tool and the workpiece. Several attempts have been made to detect chatter. Some manual feature extraction methods used to detect the chatter involve wavelet packet transform (WPT), ensemble empirical mode decomposition (EEMD), local mean decomposition (LMD), and variational mode decomposition (VMD). These methods require human expertise for manual feature extraction. In recent time, convolution neural network (CNN) has been evolved as one of the techniques for automatic features extraction. However, CNN uses images and depends on the initial weights and hyperparameters. Creating images from acquired signals for CNN input is still cumbersome and adds one more step in signal processing making it computationally heavy. In this study, a simple, accurate, and robust online chatter detection method is proposed based on wavelet scattering network (WSN). This deep network iterates over standard wavelet transform, nonlinear modulus, and averaging operators of the acoustic signals. Experiments are performed to collect the acoustic signal during the turning operation to train and validate the algorithm. The automatic extracted chatter features are then used in supervised machine learning (ML) algorithms. The results clearly show that the WSN featurization method with SVM algorithm can significantly reduce the complexity of the existing online chatter detection systems without compromising the accuracy.