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 structural character. Next, it speeds up the estimates of high-order kernels based on low-order kernels according to their correlation. Finally, it uses a linear variable step-size LMS strategy in a nonlinear algorithm and proves convergence with the Lyapunov global stability theorem.In experiments with a manipulator based on conventional and proposed methods, respectively, the results show that this algorithm 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 .