三阶非线性Volterra模型的自适应快速辨识

Fast adaptive identification of third-order nonlinear Volterra model

在RFID倒扣封装设备研制中,高速倒扣机械手具有很强的非线性和时变特性,线性控制方法难以满足要求,因此本文提出了一种快速辨识算法,采用三阶非线性Volterra模型对机械手进行在线实时辨识。首先,利用不同阶输入向量的结构关系,由低阶输入向量直接构建高阶输入向量。接着,根据不同阶核的相关性从低阶核加速估计高阶核。最后,把线性变步长LMS方法引入到非线性自适应算法中,并用Lyapunov全局稳定理论进行证明。对实际系统的辨识实验表明:与常规方法比较,辨识时间从100 ms缩短为30 ms,辨识速度提高了3.3倍,辨识失调降低了93.3%,同时还具有更高的辨识精度,满足了对非线性系统辨识的精度要求和实时性要求。

As part of the RFID flip chip package development, the high speed manipulator has obvious nonlinear and time-variable characters, so a nonlinear adaptive inverse control is needed. The key to this method is to identify the high speed manipulator by using a third-order Volterra nonlinear model in limited time and with sufficient accuracy. However,it is hard to satisfy real time requirement with a conventional method. This paper proposes a fast identification algorithm to resolve the problem. Firstly, a high-order input vector is constructed from a low-order input vector according to the struc- tural character. Next, it speeds up the estimates of high-order kernels based on low-order kernels ac- cording to their correlation. Finally, it uses a linear variable step-size LMS strategy in a nonlinear al- gorithm and proves convergence with the Lyapunov global stability theorem. In experiments with a manipulator based on conventional and proposed methods, respectively, the results show tha this al- gorithm reduces the identification time from 100 ms to 30 ms, improves convergent speed 3.3 times and reduces misadjustment by 93.3%, as well as having great precision. It can satisfy both requirements of real-time and identification precision .