Study On Modeling And Compensation Methods For Deviation Of The Inertial Unit In An Optic-electrical Tracking Platform

The inertial measurement unit is used as the sensor in the servo control system of the optic-electrical tracking platform to measure the rotation rate in the inertial space and provide the rate feedback.Therefore,improving measurement precision of the inertial measurement unit is essential to improve performance of the optic-electrical measurement system.And fiber optic gyro is widely used as the inertial measurement unit because its performance is highly accurate and stable.Because of limitation of the manufacturing and deviation of the mounting process,the angle errors occur between sensing direction of the inertial measurement and the axis direction of the mounting base.To solve this problem,we propose a calibration method based on rate test for the tri-axis gyro.The first step is to set up the mathematical equation of the gyro rate output.Then,the rate test is designed to calibrate the mounting error of the inertial measurement unit,which used the tri-axis precision inertial navigation test platform.In the calibration test,the rate signal is sampled at a series of certain rates of the three axis.The last step is calculating the parameters of error compensation mathematical model according to the test data.To compensate random drifts of the fiber optic gyro signal,we propose a compensation method combined ARMA model and Kalman filter.And this is a parameter-identification method.First,using the autocorrelation function and the partial correlation function to examine the stability of the original signal and using the deferential calculation to do the stabilization process.Then the AR(3)model is chosen as the signal model.And based on the AR(3)model,the state equations of the Kalman filter are deduced.To overcome the drawback that the conventional Kalman filter needs accurate prior noise statistics,we combine the Sage-Husa method.Afterwards,the fiber optic gyro signal is filtered by the adaptive Kalman filter.The different kinds of noises is analyzed by calculating the parameters of Allan covariance using the least square mean method.The calculation results demonstrates that after compensation,the mean square error of gyro signal decreases from 0.0034(°/s)~2 to 1.4339e-04(°/s)~2.We also propose a fiber optic gyro drift model based on wavelet analysis and wavelet neural networks.And this is a non-parameter-identification method,which has high fitting accuracy for nonlinear models.The first step of this method is using the Mallat algorithm to extract the main trend term and rebuilding the rest term.Then the wavelet neural network model uses the rebuilt signal as the ideal output and original signal as the input.During the training process,to improve the neural network's training speed and fitting accuracy,the momentum factor is added and learning rate is designed to be adaptive.According to the generalization validation,the trained wavelet neural network model has a very good estimation ability for the fiber optic gyro deviation.The final results show that after compensated by the wavelet neural network,the mean square error of gyro signal decreases to 3.7636e-05(°/s)~2,which is superior to other conventional methods.