Online Aggregate Modeling of Time-Varying Multiphysics Field-Coupling MEMS Gyro Bias

In modern transportation systems, the accuracy of micro-electro-mechanical system (MEMS) gyros is critical for various automotive and robotic applications. However, MEMS gyro bias exhibits time-varying multiphysics field coupling characteristics, presenting challenges for online modeling and correction. In response, we propose a candidate subfield aggregation shallow network (CSASN)-based method for gyro bias online modeling, departing from existing complex neural network methods. Our CSASN-based method is distinguished by reconstruction of gyro bias estimation and careful selection of effect aggregation factors (AFs). The gyro bias estimation is reconstructed through attitude estimation from an integrated navigation system when global navigation satellite system (GNSS) signal is available. The AFs of CSASN include time, temperature, angular rate, and acceleration, ensuring the model reflects the time-varying multiphysics field coupling characteristics of MEMS gyro bias. The MEMS gyro output is corrected by the model predicted bias for real-time attitude calculation in scenarios where GNSS is inaccessible. Field experimental results with a ground vehicle demonstrate the feasibility and effectiveness of the proposed online gyro bias modeling method.