BO-LSTM-Based TDE Precise Estimation Model of Capacitive MEMS-Gyros Using Thermal-Induced Physical Characteristics Variation Analysis

Owing to the fact that the conventional Temperature Drift Errors (TDEs) precise estimation model of Capacitive MEMS-Gyros (CMG) has incomplete Temperature Correlated Quantities (TCQ) and an imperfect parameter identification method to reduce bias stability, a BO-LSTM-based TDE precise estimation model using thermal-induced physical characteristics variation analysis is proposed. By analyzing microstructural deformation in CMG- and Si-based materials’ stiffness deterioration caused by thermal-induced physical characteristics variation, complete TCQ are traced, including ambient temperature variation ∆T and its square root ∆T1/2 plus its higher orders (∆T2, ∆T3, ∆T4), a modified TDE precise estimation model is formed. Long Short-Term Memory (LSTM) is applied to identify the modified model’s parameters owing to the typical time series characteristics of TDE and TCQ. In addition, Bayesian Optimization (BO) is introduced in LSTM to show a good guide for LSTM’s optimal hyperparameters. The modified model is implemented with BO-LSTM and compared with the conventional model based on Radial Basis Function Neural Network (RBFNN) in bias stability. The experimental results show that the modified model can more accurately estimate the TDE of CMG in a timely and improves its bias stability by 20%, which decouples the temperature dependence of Si-based material significantly and enhances the environmental adaptability of CMG in complex conditions remarkably.