Predictive learning control and application to servo system of DC motor

A novel algorithm for learning control is proposed and investigated experimentally. It is assumed that the desired output is periodic and that its period is known. At each sampling time, the control input is modified so as to minimize the quadratic criterion on predicted future errors. Future errors are estimated from the following data: past errors in the last attempt, present error, control input up to the present time, and the step-response data of the system measured previously. Experimental results on the DC servo motor showed that tracking errors were reduced to ±1 pulse after only four attempts. It is concluded that the proposed method is very useful for controllers of machine tools and industrial robots because the algorithm can be easily implemented on a personal computer or a microprocessor, the learning converges quickly, and no identification is required other than measuring its step response