Fault Detection And Isolation For Sensors On MUH Based On SVM And Wavelet Transform

Extensive research on Mini unmanned helicopter (MUH) has been carried out around the world as a result of the growing autonomous control technology. Its unique flight performance and great value in use have become new focal points of latest research. The requirements for its security and reliability are also increasing. To ensure the flight safety, the reliability of airborne sensor system serves as a premise. Airborne sensor, the basic component of flight control system, is one of the parts that are most prone to faults. In the domain of aviation, once the sensor comes across failure and outputs the incorrect data, the consequences will be very serious. As for that, the research on fault diagnosis of MUH sensor system has become a highly urgent task to increase its security and reliability.In this paper, the main research object is the sensor system of "Wings of Yuquan" MUH. The fault diagnosis methods by using wavelet transform and support vector machine are proposed. These simple methods are proved to be effective in solving basic sensor failure problems occurred during the research process; therefore can guarantee flight experiments go smoothly. The main contributions of this work are as follows:1. The history and current research status of the fault diagnosis of airborne sensors system on MUH are systematically reviewed.2. Introducing the hardware architecture design, sensor selection and information fusion and filtering. A hierarchical filtering structure is adopted for sensor fusion, in which complementary filters are applied to fuse information from sensors with different frequency characteristics.3. Wavelet transform method is introduced into the fault diagnosis of sensors on MUH. According to the variation of output signal energy, any abnormal state of sensor output signals can be discovered promptly by identifying multi-scale wavelet features, and subsequently to detect and isolate the fault sensors.4. A new method based on Support vector machine is proposed to detect the fault of MUH airborne sensors system. Least square support vector machine (LS-SVM) is applied to compute the SVM model, then a residual generator is constructed to detect faults. Furthermore, SVM is also used to build the Fault Classifier.5. A new fault detection and isolation method based on Support Vector Regression (SVR) combined with Discrete Wavelet Transform (DWT) method is presented in this paper. With its strong capabilities in self learning and nonlinear mapping, SVR is used to build a residual generator to detect faults. Then, DWT is used to isolate the faulty sensor.