An enhanced MEMS error modeling approach based on Nu-Support Vector Regression.

Area of Science:

• Engineering
• Sensor Technology
• Navigation Systems

Background:

• Micro Electro Mechanical System (MEMS)-based inertial sensors enable low-cost civilian land vehicle navigation.
• Accurate modeling of MEMS sensor errors (biases, drift, noise) is crucial but challenging.
• Conventional methods like Gauss-Markov (GM) and Neural Networks (NN) have limitations for MEMS units.

Purpose of the Study:

• To develop an enhanced Support Vector Machine (SVM) based error model for MEMS inertial sensors.
• To overcome the drawbacks of GM and NN methods in modeling MEMS sensor errors.
• To improve the accuracy and real-time implementation of low-cost navigation systems.

Main Methods:

• Developed an enhanced Support Vector Machine (SVM) algorithm for MEMS sensor error modeling.
• Compared the SVM approach against Gauss-Markov (GM) and Neural Network (NN) models.
• Evaluated performance based on noise reduction and positional error drift.

Main Results:

• The SVM approach reduced noise standard deviation by 10-35% for gyroscopes and 61-76% for accelerometers.
• Positional error drifts under static conditions improved by 41% (vs. NN) and 80% (vs. GM).
• SVM avoids local minimization and overfitting issues inherent in NN, providing a reliable global solution.

Conclusions:

• The proposed SVM-based error model offers a superior solution for MEMS inertial sensor error modeling.
• SVM provides a more accurate, efficient, and reliable method compared to GM and NN.
• This advancement contributes to the development of more robust and precise low-cost navigation systems.