Multi-sensor data fusion using least square support vector regression for missing data online recovery

摘要

For many engineering and aerospace power applications, sensor fault diagnosis and recovery on-board is increasing important Consequently, the development of an efficient and real time diagnostic scheme that can accurately detect and diagnose the sensor failures will offer significant potential for maintaining the safe operating of aircraft engine. A method of multi-sensor data fusion using least square support vector regression (LSSVR) is presented to on line diagnose sensor fault and regain the signal of faulty sensor. Cross-correlation and autocorrelation to five measurable sensors in turbine engine are defined and analyzed. The autocorrelation is calcualated and used as the sensor reference, and then the sensor fault diagnosis is carried out based on the differences between the measured value and reference. The cross-correlation is used to look for the sensors with high dependence and to form the sensor subset, The fault sensor signal can recover by the LSSVR with the current information of sensor subset. Experimental results show that the engine sensor fault recognition rate is satisfied by the proposed method, and can be used to replace the fault sensor in emergency with proper tracking accuracy to the acutual state.