Neural network servo control for ultra-precision machining at extremely low feed rates.

摘要

Diamond turning of brittle materials such as glass, ceramic, germanium and zinc sulfide has been of considerable research interest in recent years due to applications in optics and precision engineering systems. When diamond turning brittle materials, material removal should be kept within the ductile regime to avoid subsurface damage. It is generally accepted that ductile regime machining of brittle materials can be accomplished using extremely low depth of cut and feed rates. Furthermore, the tool positioning accuracy of the machine must be in the nanometer range to obtain optical quality machined parts with surface finish and profile accuracy on the order of 10 nm and 100 nm respectively. Nanometric positioning accuracy of the machine tool axes is difficult particularly at low feed rates due to friction and backlash. Friction is highly nonlinear particularly at low speeds due to the transition from stiction to Coulomb friction, and as such is very difficult to model. Standard proportional-integral-derivative (PID) type controllers are unable to deal with this large and erratic friction characteristic within the requirements of ultra precision machining. In order to compensate the effects of friction in the machine tool axes, a learning motion control algorithm based on the Cerebellar Model Articulation Controller (CMAC) neural network is studied for servo control. The learning controller was implemented using 'C' language on a DSP based open architecture controller for a single point diamond turning machine.