Fault Detection And Estimation Algorithms For A Class Of Systems With Multiplicative Noise

In modern society, with the development of production engineering, it’s veryimportant to raise the reliability of control systems, while the fault detection technol-ogy is an effective way.Now, for the classical linear systems, there have been lots of research achieve-ments about fault detection and estimation algorithms, such as methods of analyticalmodel, knowledge and signal processing. While in actual systems, like the oil seismicexploration, underwater target tracking and so on, the system models are usuallynon-linear, and the signal is affected by delay, distortion, and attenuation in the trans-mission process, which are called the multiplicative noise, whose characteristics can’tbe described by the additive noise. So research on the fault detection and estimationfor the system with multiplicative noise is of broader significance.The system with multiplicative noise described in this thesis refers to a class ofsystems whose innovation is approximately normal. For these systems, the main studyis introduced as follows:1. The judgment whether the observed data or innovation is approximately nor-mal is provided and accordingly, the fault detection and amplitude estimation methodis developed for the system with multiplicative noise on the condition of approxi-mately normal innovation.2. Filter-based fault detection and estimation algorithm for systems with multi-plicative noise whose innovation is approximately normal is proposed. The time offault generation is detected according to the statistical characteristics changed beforeand after the fault generation. After that, the amplitude of fault is estimated by use ofthe state filtering algorithm for systems with multiplicative noise and generalized like-lihood ratio.3. The fault detection and estimation for systems with multiplicative noise basedon the smoothing algorithm is proposed. Although the fault detection algorithm basedon filtering algorithm can realize the real-time judgment, the observed data is inade-quate to guarantee the accuracy. In order to improve the estimation accuracy, the faultdetection based on the smoothing algorithm is presented further. Likewise, the time offault is detected by the change of innovation mathematical statistics. Then the faultamplitude is estimated according to the smoothing algorithm and GLR method. 4. Much simulation of the algorithms has been done in this dissertation in use ofcomputer simulation software MATLAB R2010b. The simulation results show theeffectiveness of the proposed algorithms.