Temperature compensation for RLG based on neural network

Abstract

ABSTRACT Several static tests indicate that the Ring Laser Gyro (RLG) bias inside of the Strap-down Inertial Navigation System (SINS) varies remarkably as long time working. Further experiments and analyzing results show that the SINS external metal shell could insure the inside temperature rising gently and evenly, and the RLG drifts could be viewed mainly affected by RLG inside temperature field. In order to ac hieve better RLG stability character istic within full temperature range, investigated the BP and RBF Artificial Neural Networks (ANN) nonlinear modeling and compensation technology. Firstly, introduced two typical structures for BP an d RBF neural networks, and then, take a set of static tests data from 25 $ C to 55 $ C as training samples, separately built up four-l ayer BP and two-layer RBF neural networks for RLG drifts. In order to compare the compensation effects, first-order and second-order piecewise Least Square (LS) fitting technologies are also implemented here. Four new expe rimental data were adopted to check the modeling validity. The compensation results show that the RLG drifts stability could be improved by 20%-40%; the precision of BP network modeling method is better than th at of first-order linear piecewise LS f itting, and the precision of RBF is better than that of second-order piecewise LS fitting. Keywords: Strap-down Inertial Navigation System (SINS), Ring Laser Gyro (RLG), drifts, Artificial Neural Network (ANN), Least Square (LS)