A real-time fault identification and correction method of sensor systems

• Main Authors: Ren-Zhi You, 游仁植
• Other Authors: Tsung-Lin Chen
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/31115101389426555763

碩士 === 國立交通大學 === 機械工程系所 === 96 === In a sensing system that was constructed by employing redundant devices (components), the conventional approach for the fault-identification was done through the “voting equations”. However, when the outputs of the devices were contaminated by noise, the conventional fault-finding measure had to set up a threshold values and an observation periods along with voting equations. Due to the setup of an observation period, the conventional approach can not be done in a real-time manner. As a consequence, the real-time fault-correction was not possible. In this thesis, we proposed a novel real-time fault-identification method to solve the problem above. Furthermore, the proposed method can combine with various feedback techniques to achieve real-time fault-correction. The proposed method uses the novel “output equations” along with “voting equations” to describe the relationship between each device output. After that, the real-time fault-identification problem was formulated into a nonlinear state estimation problem. The method of the newly added “output equations” was the key to the success of the proposed real-time fault-identification method. Furthermore, in order to handle the sensor drift (or time-varying fault) problem, we use the “Kalman filter with fading memory” techniques for the state observer. Moreover, we use state feedback techniques for the purpose of the correction of fault, and the corrected device can be kept in the “fault-tolerant of sensor system” to increase the accuracy of system output. We can estimate fault signals successfully by simulation of 3 sensors of the system, and its error standard deviation is about .For drift, its standard deviation of estimation is about .The minimum fault value that estimated is equals to 1/2 times of standard deviation of noise approximately. Besides, the state feedback technique give a fault correction to sensor output which has fault and, thus, a corrected signal is approximate to ideal signal. More facts shows in the thesis.