文章摘要

文章信息/Info

Title:: Dynamically Tuned Gyroscope Closed-Loop Identification Based on Modified Maximum Likelihood Method

作者:: 王亚辉1; 李醒飞1; 2; 纪越1; 赵建远1; 1. 天津大学精密仪器与光电子工程学院，天津 300072； 2. 精密测试技术及仪器国家重点实验室（天津大学），天津 300072

Author(s):: Wang Yahui1; Li Xingfei1; 2; Ji Yue1; Zhao Jianyuan1; 1. School of Precision Instrument and OptoElectronics Engineering, Tianjin University, Tianjin 300072, China; 2. State Key Laboratory of Precision Measuring Technology and Instruments (Tianjin University), Tianjin 300072, China

Keywords:: closed-loop identification; dynamically tuned gyroscope; maximum likelihood method; colored noise

摘要:: 针对Box-Jenkins（BJ）模型辅助向量法和Newton-Raphson法计算繁杂、收敛速度慢、辨识精度不高等问题和极大似然法无法直接应用在闭环辨识的限制，把结合BJ模型的递推的极大似然（recursive maximum likelihood，RML）参数估计法应用于动力调谐陀螺仪的闭环辨识，提出了不受耦合有色噪声影响的BJ模型近似递推极大似然（BJRML）闭环辨识法，获取了动力调谐陀螺仪的参数估计值并实现陀螺仪在线性能监测.结合动力调谐陀螺仪的闭环简化模型等先验知识，通过数值仿真验证BJRML法辨识结果的无偏一致性与渐进最优性；在实验室条件下采用本方法进行动力调谐陀螺仪闭环辨识实验.仿真结果表明：在有色噪声存在的条件下，BJRML法的辨识结果是一致无偏渐进最优的；闭环辨识实验结果表明：辨识精度优于92%；辨识结果能够跟踪陀螺特性，基本实现陀螺仪性能在线监测.

Abstract:: Regarding the problems that Box-Jenkins instumental variable (BJIV) method is slow in convergence speed, that Newton-Raphson method is low in precision, and that maximum likelihood method cannot be directly applied in closedloop identification, Box-Jenkins recursive maximum likelihood (BJRML) method was proposed and applied to dynamically tuned gyroscope (DTG) closed-loop identification. DTG model parameter was obtained, and online performance monitoring was achieved. The method combined BoxJenkins model with recursive maximum likelihood method. Also, it is not affected by coupling colored noise. Firstly, the prior knowledge of simplified closedloop model of DTG was obtained. Then, the paper verified the unbiased consistency and asymptotic optimality of BJRML identification results. Finally, identification experiments were conducted on the DTG closed-loop system in laboratory. The simulation results are as follows: the estimations of BJRML method are unbiased and consistent with different noise levels, and the asymptotic variance is near-optimal. Experiment results show that the identification fitting degree is more than 92%. Identification results can track gyroscope characteristics and achieve basic online monitoring.