Summary: | 碩士 === 國立臺灣科技大學 === 機械工程系 === 102 === Learning-type Model Predictive Control(L-MPC) is a type of model predictive control (MPC) approach that iteratively updates the setting input in the MPC control system by using so called learning control algorithms. This control algorithm reserves the advantages of MPC in calculating the optimal control moves at all sample steps and handling system constraints, and particularly obtains the optimal reference batch command for further improved control performance. The existing literature shows that P-type learning control law is widely adopted to the L-MPC control systems. However, the stability is easily influenced by high frequency noises and the learning performance may be deteriorated with divergent error. This study proposes a PD-type L-MPC control law to tackle this problem because of the better stability offered by the PD learning control. Unlike the slow dynamic system applications presented in the previous work, the proposed L-MPC control law is applied to a linear servo stage for precision motion control. Comparative experiments for tracking dynamic motion profiles are conducted to investigate the effects of using two different learning control laws. The results show that the proposed PD-type L-MPC achieves better tracking performance than P-type L-MPC, and can effectively reduce the error magnitude occurred at the abrupt corner of a challenging motion profile.
|