Optimization of Data Pre-Processing for Compensation of Temperature Dependence of FOG bias by a Neural Network

Abstract

Estimated formulas for calculating noise of the type “random walk” of algorithmic compensation for the bias of the gyroscope are obtained. An example of assessing the statistical significance of factors in calibrating the bias of a fiber-optic gyroscope in the operating temperature range and various rates of temperature change is given. It is shown, that the random error of temperature sensors can play a decisive role in the noise of the “random walk” type of algorithmic compensation of the gyroscope bias and exceed the gyroscopic noise. An example of obtaining a regression dependence of the algorithmic compensation of the bias of the gyroscope using a neural network with a multilayer perceptron is given. The factors affecting the choice of the time constant of the differentiating low-pass temperature filter are considered. The experimental dependences of the random error of algorithmic compensation for temperature sensors with various random errors are presented and the necessity of using temperature sensors with a minimum random error is demonstrated.