A Novel Long-Term Prediction Model for Hemispherical Resonator Gyroscope's Drift Data

Abstract

The hemispherical resonator gyroscope (HRG) is a new vibration gyro, which has features of high accuracy, long lifespan, no wear-out, and great reliability. However, the excellent performances make it impractical to get the HRG's lifespan within whole life test, and its lifespan has not even been explored. To predict the HRG's lifespan without whole life test, one residual modified autoregressive gray model, ARGM(1,1), is proposed. It combines autoregressive process inherited from artificial neural network and support vector machine with gray model to train, model, and forecast. In this paper, this model is applied to predict multiperiod sequences with one HRG's drift data, and gray correlation analysis is used to evaluate the HRG's failure stage and get the lifespan. The experimental results show the model has good characteristics of self-adaption and low demands for modeling data. Compared with the conventional GM(1,1), back propagation neural network and support vector regression, residual modified ARGM(1,1) outperforms them in long-term prediction for the HRG's drift data. Meanwhile, the predictive result shows the HRG can work about 15.74 years. Based on the 10 global oldest spacecraft, the predictive result with the method is reliable.