关键词: 多轴伺服系统, 后置输入整形器, 递归最小二乘, 遗忘因子, 神经网络

Abstract: A widely applicable post-adaptive input shaper algorithm was proposed to solve the residual vibration of multi-axis servo system caused by the flexibility of the system in the emergency stop section of high-speed motion. The algorithm does not need to identify the system modal parameters and it is based on the recursive least square method (RLS). The residual vibration signal was used as the input of the algorithm to optimize the shaper coefficient vector with the best vibration suppression effect under the current trajectory. The adaptive forgetting factor updating algorithm was introduced to improve the tracking performance of the shaper in non-stationary environment. A full connection multilayer neural network model was established and trained by using multiple sets of excitation tra-jectory samples. It solves the problem of the original algorithm that the cost of time is significantly increased due to multiple changes of trajectory and reoptimization. The experimental results show that the residual vibration amplitude of the post adaptive input shaper with adaptive forgetting factor is reduced by 28.3% averagely and 36.9% maximally, and the convergence time of residual vibration is reduced by 28.4%, compared with the traditional post adaptive input shaper. The residual vibration amplitude of the input shaper predicted by the multilayer perceptron model is reduced by 21.6% averagely and 29.8% maximally, and the convergence time of residual vibration is reduced by 23.7%, compared with the ordinary post adaptive input shaper. The algorithm proposed in this paper has certain application significance for improving the positioning accuracy of multi-axis servo system and shortening the positioning waiting time. The introduction of multilayer neural network model improves the overall work efficiency under the condition that the expected trajectory changes frequently.

Key words: multi-axis servo system, post input shaper, recursive least square method, forgetting factor, neural network