Shen Qiang, Yang Dengfeng, Zhou Jie, Wu Yixuan, Zhang Yinan, Yuan Weizheng
Research & Development Institute of Northwestern Polytechnical University in Shenzhen, Shenzhen 710000, China.
MOE Key Laboratory of Micro and Nano Systems for Aerospace, Northwestern Polytechnical University, Xi'an 710072, China.
Micromachines (Basel). 2019 Apr 30;10(5):294. doi: 10.3390/mi10050294.
This paper first presents an adaptive expectation-maximization (AEM) control algorithm based on a measurement-data-driven model to reduce the variance of microelectromechanical system (MEMS) accelerometer sensor under multi disturbances. Significantly different characteristics of the disturbances, consisting of drastic-magnitude, short-duration vibration in the external environment, and slowly-varying, long-duration fluctuation inside the sensor are first constructed together with the measurement model of the accelerometer. Next, through establishing a data-driven model based on a historical small measurement sample, the window length of filter of the presented algorithm is adaptively chosen to estimate the sensor state and identify these disturbances simultaneously. Simulation results of the proposed AEM algorithm based on experimental test are compared with the Kalman filter (KF), least mean square (LMS), and regular EM (REM) methods. Variances of the estimated equivalent input under static condition are 0.212 mV, 0.149 mV, 0.015 mV, and 0.004 mV by the KF, LMS, REM, and AEM, respectively. Under dynamic conditions, the corresponding variances are 35.5 mV, 2.07 mV, 2.0 mV, and 1.45 mV, respectively. The variances under static condition based on the proposed method are reduced to 1.9%, 2.8%, and 27.3%, compared with the KF, LMS, and REM methods, respectively. The corresponding variances under dynamic condition are reduced to 4.1%, 70.1%, and 72.5%, respectively. The effectiveness of the proposed method is verified to reduce the variance of the MEMS resonant accelerometer sensor.
本文首先提出了一种基于测量数据驱动模型的自适应期望最大化（AEM）控制算法，以降低微机电系统（MEMS）加速度计传感器在多干扰情况下的方差。首先构建了干扰的显著不同特征，包括外部环境中幅度剧烈、持续时间短的振动，以及传感器内部缓慢变化、持续时间长的波动，并结合加速度计的测量模型。接下来，通过基于历史小测量样本建立数据驱动模型，自适应选择所提算法滤波器的窗口长度，以同时估计传感器状态并识别这些干扰。将基于实验测试的所提AEM算法的仿真结果与卡尔曼滤波器（KF）、最小均方（LMS）和常规期望最大化（REM）方法进行了比较。在静态条件下，KF、LMS、REM和AEM估计等效输入的方差分别为0.212 mV、0.149 mV、0.015 mV和0.004 mV。在动态条件下，相应的方差分别为35.5 mV、2.07 mV、2.0 mV和1.45 mV。与KF、LMS和REM方法相比，基于所提方法在静态条件下的方差分别降低到1.9%、2.8%和27.3%。在动态条件下，相应的方差分别降低到4.1%、70.1%和72.5%。验证了所提方法在降低MEMS谐振加速度计传感器方差方面的有效性。
