一种MEMS加速度计的噪声处理与参数训练方法

Noise Processing and Parameter Training Method for MEMS Accelerometer

为进一步提高微机电系统(MEMS)加速度计的测量精度,建立以测量值为输入、真实值为输出的MEMS加速度计误差补偿模型,利用Allan方差和最小均方(LMS)自适应滤波算法对加速度计在6个位置下的多组实际测量数据进行噪声分析和预处理,处理后的全部测量数据作为样本训练模型参数,利用最小二乘和批量梯度下降相结合的方法获得样本数据对真实模型参数的最优拟合,并利用该模型对加速度计进行误差补偿,实现MEMS加速度计的高精度标定。实验验证表明,利用该模型对MEMS加速度计进行误差补偿后,输出值的均值误差为(0.72~1.19)×10^-4 g,标准差为(0.75~1.61)×10^-4 g,相对于补偿前,均值误差降低了2个数量级,标准差降低了1个数量级,有效提高了MEMS加速度计的测量精度和稳定性。

In order to improve the measurement accuracy of the microelectromechanical system(MEMS)accelerometer,an error compensation model of the MEMS accelerometer was established with the measured value and the true value as the input and the output,respectively.Noise analysis and pre-processing of the measurement data for accelerometer at six locations were performed by the Allan variance and least mean square(LMS)adaptive filtering algorithm.Using the processed data as the parameters of the training model,the optimal fitting of the sample data to real parameters was obtained by combing the least square method and the batch gradient descent to obtain.The raining model was used to compensate the error of the accelerometer to realize the fairly high-accuracy calibration.Experimental investigations demonstrate that the mean error and the standard deviation of the output were(0.72~1.19)×10^-4 g and(0.75~1.61)×10^-4 g.Compared with the pre-compensation,the mean error and the standard deviation were reduced by 2 orders and 1 order of magnitude,respectively,which improved the measurement accuracy and the stability of the MEMS accelerometer effectively.