Chatter detection in the turning process using acoustic wavelet scattergrams and transfer learning

Chatter is an un wan ted vibration in m achin in g processes that degrades surface quality, reduces productivity, an d shorten s tool life. Reliable chatter detection is essen tial, particularly un der varyin g m achin in g conditions where traditional vibration-based approaches are often in trusive, complex, an d require expert in terpretations. This study in troduces a n ovel tim e–frequen cy represen tation of acoustic sign als usin g wavelet scattergram s for chatter detection in the turn in g process. These wavelet scattergram s gen erated through a wavelet scatterin g n etwork (WSN) are used as an in put to fin e-tun e a deep n eural n etwork (DNN) usin g a tran sfer learn in g fram ework. To en sure robust evaluation, experim en ts were conducted with five differen t overhan g workpiece configurations, where the acoustic sign als from the first four configurations were used for train in g an d validation, an d a completely n ew workpiece configuration was used to test the m odel perform an ce. Am ong the evaluated m odels, ResNet50V2 achieved the highest classification accuracy of 99.2% while m ain tain in g a low detection laten cy of about 62 m s, outperform in g MobileNet an d Den seNet121. The results dem onstrate that the proposed m ethod can reliably detect chatter usin g a n on-in trusive m icrophone sen sor an d lim ited experimen tal data while m ain tain in g robustn ess across varyin g m achin in g conditions, m akin g it suitable for accurate chatter detection in in dustrial applications. © 2026 Elsevier Ltd Author