Micro electromechanical system (MEMS) gyroscopes are applied across various fields due to their low power consumption and high level of integration. However, temperature variations significantly affect their performance, introducing issues like increased noise, unstable bias. This study aims to focus on the temperature behavior of multi-frame vibrating MEMS gyroscopes and develops a model for temperature-related discrepancies.
A parallel processing framework based on robust local mean decomposition (RLMD) is introduced to separate gyro’s output into trend, periodic and random components, which are then processed through weighted Whittaker filter (WWF) and reconstruction. Temperature-induced drift is compensated through a bidirectional long short-term memory network, which models temporal dependencies bidirectionally and extracts latent patterns from the data set.
Through tests, the method attained a gyro bias stability of 7.11°/h and an angular random walk of 16.74°/h/vHz in a changing environment from −40 to 100 °C.
The model exhibits strong resistance to interference, with RLMD using robust envelope estimation and adaptive filtering, while WWF adjusts weights dynamically based on local variance. In addition, the parallel framework based on RLMD enhances the extraction of temperature-related features, preserving static characteristics and improving the accuracy of the compensation network.
