1 June 2026 in SSRN Electronic Journal
As the core component of inertial navigation systems, Fiber Optic Gyroscopes (FOG) are severely limited by temperature-induced zero-bias drift, which is synergistically affected by constant temperature and temperature change rate. Traditional model-driven methods (polynomial fitting, piecewise regression) rely on manual modeling and fail to characterize complex nonlinear thermal errors; reported machine learning methods (BP, CART-Bagging, Online-SVR, LSTM-RNN) have insufficient nonlinear feature extraction and temporal correlation capture capabilities, with drastically degraded performance in extreme thermal environments. This paper analyzes the FOG temperature drift mechanism, establishes a constant-variable temperature coupled drift model, and proposes a CNN-BLSTM compensation algorithm integrating CNN’s deep nonlinear feature extraction and BLSTM’s bidirectional temporal correlation mining capability. Experiments under -40℃~60℃ full temperature range with four heating/cooling rates (0.2/0.5/1/2℃/min) show that the proposed algorithm improves FOG zero-bias stability by over 91% (maximum 96.21%), reducing it to 0.2388°/h after compensation. Its performance is significantly superior to the traditional MLP model and all reported mainstream methods including BP, CART-Bagging, Online-SVR and LSTM-RNN, especially maintaining high robustness under extreme low temperature and rapid temperature change conditions, fully meeting the engineering requirements of aerospace and marine navigation.
15 August 2026 in Optical Fiber Technology
1 January 2026 in IEEE Sensors Journal