期刊
出版社
ASME
DOI: 10.1115/1.4048001
关键词
online chatter detection; long short-term memory neural networks; machine learning; soft-computing techniques; milling operation; ball screw drive; advanced materials and processing; computer-integrated manufacturing; machining processes; sensing; monitoring and diagnostics
资金
- UMGF Graduate Fellowship from University of Manitoba
- Natural Sciences and Engineering Research Council (NSERC) of Canada [RGPIN-2015-04173]
This study proposes an online chatter detection method based on current signals applied to the ball screw drive of CNC machine to detect self-excited vibrations of milling tools. The method eliminates the need for additional sensors and utilizes neural networks to achieve a high level of accuracy in experiments.
For certain combinations of cutter spinning speeds and cutting depths in milling operations, self-excited vibrations or chatter of the milling tool are generated. The chatter deteriorates the surface finish of the workpiece and reduces the useful working life of the tool. In the past, extensive work has been reported on chatter detections based on the tool deflection and sound generated during the milling process, which is costly due to the additional sensor and circuitry required. On the other hand, the manual intervention is necessary to interpret the result. In the present research, online chatter detection based on the current signal applied to the ball screw drive (of the CNC machine) has been proposed and evaluated. There is no additional sensor required. Dynamic equations of the process are improved to simulate vibration behaviors of the milling tool during chatter conditions. The sequence of applied control signals for a particular feed rate is decided based on known physical and control parameters of the ball screw drive. The sequence of the applied control signal to the ball screw drive for a particular feed rate can be easily calculated. Hence, costly experimental data are eliminated. Long short-term memory neural networks are trained to detect the chatter based on the simulated sequence of control currents. The trained networks are then used to detect chatter, which shows 98% of accuracy in experiments.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据