Single Network Adaptive Critic for Vibration Isolation Control ?

Vibration isolation control is the critical issue to guarantee the performance of various vibration-sensitive instruments and sensors in practical engineering systems. In this paper, single network adaptive critic (SNAC) based controllers are developed for vibration isolation applications. The SNAC approach differs from the typical action-critic dual network structure in adaptive critic designs (ACDs) by eliminating the action network, which leads to substantial computational savings. Two training methods, i.e., the off-line and online methods are proposed to adapt the SNAC controllers respectively. In contrast with the existing offline SNAC training method, the off-line method proposed in this paper adopts the least mean square (LMS) training algorithm with variable learning rate to make the training procedure converge faster. Furthermore, for real-time control purpose, the online learning method is presented for tuning the weights of the critic networks along the real-time state trajectories of the isolation system. Additionally, the “shadow critic” training strategy used in the online method further improves the convergence rate. Simulation results have shown that the developed SNAC controllers using the different training methods can converge to the continuous-time optimal control solution at satisfactory speed. Moreover, the designed SNAC controllers alleviate vibration disturbance more effectively and have better control performance in comparison with the passive isolator.

[1]  Jennie Si,et al.  Online learning control by association and reinforcement , 2000, Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks. IJCNN 2000. Neural Computing: New Challenges and Perspectives for the New Millennium.

[2]  Derong Liu,et al.  Action-dependent adaptive critic designs , 2001, IJCNN'01. International Joint Conference on Neural Networks. Proceedings (Cat. No.01CH37222).

[3]  Richard S. Sutton,et al.  Neuronlike adaptive elements that can solve difficult learning control problems , 1983, IEEE Transactions on Systems, Man, and Cybernetics.

[4]  Yi-De Chen,et al.  Application of voice coil motors in active dynamic vibration absorbers , 2005, IEEE Transactions on Magnetics.

[5]  J. Spanos,et al.  A soft 6-axis active vibration isolator , 1995, Proceedings of 1995 American Control Conference - ACC'95.

[6]  Paul J. Werbos,et al.  Approximate dynamic programming for real-time control and neural modeling , 1992 .

[7]  E.H. Anderson,et al.  Satellite ultraquiet isolation technology experiment (SUITE) , 2000, 2000 IEEE Aerospace Conference. Proceedings (Cat. No.00TH8484).

[8]  S.N. Balakrishnan,et al.  Optimal control synthesis of a class of nonlinear systems using single network adaptive critics , 2004, Proceedings of the 2004 American Control Conference.

[9]  Mehdi Ahmadian,et al.  Modelling and Performance of an Experimental Active Vibration Isolator , 1995 .

[10]  Dae Sung Joo,et al.  Sliding mode neural network inference fuzzy logic control for active suspension systems , 2002, IEEE Trans. Fuzzy Syst..

[11]  George G. Lendaris,et al.  Training strategies for critic and action neural networks in dual heuristic programming method , 1997, Proceedings of International Conference on Neural Networks (ICNN'97).

[12]  Donald A. Sofge,et al.  Handbook of Intelligent Control: Neural, Fuzzy, and Adaptive Approaches , 1992 .