基于信赖域方法的MR阻尼器瞬时最优半主动控制实验研究

Experiment study of trust-region method based instantaneous optimal semi-active control for MR dampers

通过对一个悬臂梁的磁流变(MR)阻尼器控制试验研究了一种基于信赖域方法的MR阻尼器瞬时最优半主动控制算法。该控制算法不以主动最优控制力为目标出力,而是从MR阻尼器的性能出发,提出了MR阻尼器瞬时最优半主动控制的指标函数;同时将下一时刻阻尼器的可能出力范围作为半主动控制问题的限界约束条件,应用精细积分法将MR阻尼器半主动控制问题转化为一组带有限界约束的最小值优化问题;最后采用信赖域方法在限界约束内求解得到半主动最优控制力。悬臂梁控制试验通过dSPACE控制平台实现,试验中采用NARX神经网络建立了MR阻尼器模型并在每个时间步长对阻尼器的出力范围进行预测。在不同强度的随机激励下对所提出的控制算法的控制效果进行了验证,并与不同控制算法的控制效果进行了比较。试验结果表明,时间步长的选择对基于信赖域方法的瞬时最优半主动控制算法的控制效果有很大的影响。通过与Passive-off,Passive-on被动控制及Clipped-optimal控制算法的比较,也证明了所提出的控制算法的控制效果,特别在对结构反应均方值的控制效果上的优越性。

A trust-region based instantaneous optimal semi-active control algorithm is proposed and verified through a control experiment on a cantilever beam installed with MR damper. Rather than taking the active optimal control force as target control force, an instantaneous performance index is first set up for semi-active control system in terms of the characteristics of MR damper. Then, the maximum and minimum control forces for the next time step are obtained from MR damper forward model and considered as constraints for the semi-active control system. Finally, by the application of precise integration method, the semi-activecontrol problem is described as a typical single target optimization problem subjected to a bound constraint and trust-region method is utilized to find the optimal semi-active control force within the bound constraint at each time step. Control experiment is conducted through dSPACE control platform. The dynamic behavior of the MR damper is accurately formulated by NARX-based approach and neural network technique. During experiment, both low and high level＇ s random excitations are used. The performances of two passive control algorithms, clipped-optimal control algorithm and the proposed control algorithm are studied and compared. The experiment results show that time step has great influence on the performance of the proposed controller and should he carefully designed. The comparison results also demonstrate the better control efforts achieved by the proposed controller, especially its superiority on suppressing root-mean-square responses over other controllers.