Research On Inertial Sensor Calibration Method For Train Integrity Monitoring System

GNSS/INS based train integrity monitoring system obtains real-time acceleration, speed and position information of the train by using a variety of sensors, and it determines the train integrity based on the train length and velocity difference between head and tail of the train. Being an important part of the system, inertial sensors are used to measure the acceleration and angular velocity information of the head and tail of the train. The accuracy of the inertial sensor data directly affect the accuracy and reliability of the train integrity verdict from the train integrity monitoring system based on GNSS/INS. The signal of inertial sensors are subject to the effects of train body shaking, engine vibration, electromagnetic interference from peripherals under the train operation environment. Therefore, it is necessary to calibrate the inertial sensors.This thesis analyses the error characteristics of the inertial sensor signal under the train operation environment and then establishes the signal error model. By applying wavelet packet denoising, zero speed detection, ARMA model and other technologies, a dynamic calibration scheme is proposed for inertial sensors during the train operation. Dynamic calibration scheme consists of three steps:signal noise reduction, error modeling and data calibration.The contents of the thesis are as follows:(1) Denoising Method Based on best wavelet packet. Since the inertial sensor signal has more random error component in train runtime environment, this paper proposes the method of best wavelet packet denoising as the first step of the calibration scheme. Shannon entropy is adopted as the cost function to calculate best wavelet packet decomposition tree, and hard thresholding is used to handle wavelet packet decomposition coefficients. Real time denoising of inertial sensor signal is achieved by sliding data window.(2) Stochastic signal error modeling of inertial sensor by ARMA. According to the random error characteristics of inertial sensor signal after denoising, ARMA time series model is selected for error modeling. The optimal order of ARMA model is calculated by AIC method, and the model parameter identification methods are derived in this paper.(3) ARMA-Kalman dynamic calibration method based on zero velocity compensation. A zero velocity detection method based on generalized likelihood ratio is proposed in this paper to identify the train movement and to correct the signal error of inertial sensors. At the meantime, the Kalman filter using ARMA model of inertial sensor is established to calculate the real-time correction of random errors in order to denoise the signal.In the end, the paper builds a simulation platform based on MATLAB to test and verify the proposed methods in different scenarios using the field data collected from Qinghai-Tibet Railway Line. The experiment results shows that the methods discussed in this thesis can not only effectively achieve the dynamic calibration of inertial sensors under train operation environment but also improve the accuracy and availability of the inertial sensor data.