A new predictive filter for nonlinear alignment model of stationary MEMS inertial sensors

Abstract

This paper proposes a new approach called the Predictive Kalman Filter (PKF) which predicts and compensates model errors of inertial sensors to improve the accuracy of static alignment without the use of external assistance. The uncertain model error is the main problem in the field as the Micro Electro Mechanical System (MEMS) inertial sensors have bias which change over time, and these errors are not all observable. The proposed filter determines an optimal equivalent model error by minimizing a quadratic penalty function without augmenting the system state space. The optimization procedure enables the filter to decrease both model uncertainty and external disturbances. The paper first presents the complete formulation of the proposed filter. Then, a nonlinear alignment model with a large misalignment angle is considered. Experimental results demonstrate that the new method improves the accuracy and rapidness of the alignment process as the convergence time is reduced from 550 s to 50 s, and the azimuth misalignment angle correctness is decreased from 52" 47" to 4" 0:02".

Keywords

• predictive filter
• nonlinear alignment
• model error
• optimization
• mems inertial sensors