Turning Chatter Detection Using Wavelet Features and Deep Learning

Chatter is a self-excited vibration of tool and workpiece that ruins the surface finish, shortens tool life and can damage the spindle. On the public turning chatter data set of Khasawneh, Yesilli et al. (Mendeley Data: 115 cutting tests of a boring bar on an engine lathe at four tool overhangs of 2 to 4.5 inch, acceleration at 10 kHz) this project classifies every 0.5-second window of a cut as stable or chatter with three methods - an FFT chatter-band rule, wavelet-packet energies with a support vector machine (SVM, the classical baseline) and a small convolutional neural network (CNN) on spectrograms - and links them to machining dynamics with a regenerative stability-lobe model. Unlike the related projects (A12 milling tool-wear regression, EE6 bearing faults and the other signal-fault projects), PR3 detects a process state that depends on spindle speed, depth of cut and overhang, evaluates by cutting test and tests transfer to unseen overhangs. On 29 unseen cutting tests the methods reach 89.8 / 88.1 / 89.6 % window accuracy - no significant difference; transfer to a new overhang is the hard part. A FastAPI web application with five interchangeable databases analyses signals, raises chatter alarms and has a stability-lobe calculator.

Category: Production Engineering

Production Engineering Project PR03 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.