The Novel Methods of Constructing the Sensitive Factor for Vibration Sensor Network of Aero-Engine Based on Multi-Feature Fusion and Transformation

Abstract

To date, the concept and indicator of vibration sensor networks sensitive factors haven't been reported in literature, and thus, it is impossible to assess the degree of importance of each node of the sensor and to solve the consistency of decision-making fusion of multi-sensor. We propose two methods for sensor networks sensitive factor calculation, namely, multi-feature fusion method and kernel principal component analysis (KPCA) weighted nonlinear feature transformation method. Both of them are based on the calculation of time and frequency domain feature sensitivity of the vibration signal. The first method is obtained by the weighted fusion of the most sensitive feature of each sensor. The second method is obtained by the nonlinear feature transformation of all features. These methods are verified by the rotor fault simulation data obtained by multi-sensor. The results showed that: the sensitive factor obtained in the two method can both reflect changes in failure or abnormal state, but the KPCA feature transformation method works better, because for the same fault or abnormal condition it has higher sensitive factor and stronger sensitivity; sensitive factors obtained through two methods can both effectively measure the sensitivity of different sensor nodes for the same fault or abnormal condition in the network. The resulting sequencing of the sensitivity of sensors is consistent, and both can be used to calculate the degree of importance of the sensor nodes.