PRECISION GEAR GRINDING USING SELF-LEARNING CONTROL STRATEGY (MANUFACTURING, CNC MACHINE TOOL, CAM)

This thesis presents a self-learning control strategy which imparts human expertise and newly developed gear grinding methods to a CNC gear grinding machine. Thus the machine has the capability to "learn" from current conditions and to make decisions by itself during the grinding process. These decisions allow the machine to adjust control programs between operations as well as to compensate contouring errors in-process.; The self-learning control strategy is comprised of four control tactics, namely, machine setting adjustment, eccentric blank lay-out, Constant Grinding Force (CGF) control and contouring error Forecasting Compensatory Control (FCC).; The machine setting adjustment tactic updates the grinding parameters according to the calibration of the pressure angle of the grinding wheel. The eccentric blank lay-out tactic estimates the eccentricity of each workpiece, and determines the stocks to be removed from the flanks. The CGF control tactic simulates the motion trajectory of the grinding wheel strokes constrained by a constant chip area removal criterion, and generates a series of constant grinding force NC programs for every tooth flank. These three tactics "learn" from on-machine measurement, and control by computer simulation and NC programming.; The contouring error FCC tactic compensates the disturbance-caused contouring error by on-line monitoring, recursive modeling, 2-step-ahead forecasting, and real-time compensatory control. Its "learning" capability is built into the algorithm of parameter updating for the Recursive Auto Regressive (RAR) model, and the algorithm for contouring error forecasting.; Experimental results show that the grinding efficiency can be increased 3.6 to 5.8 times as compared with conventional gear grinding methods. The single pass grinding profile accuracy can be as high as 4.9(mu)m or 3.9(mu)m (AGMA 14 or 15) for a particular tooth flank.