Abstract
In this paper, a new detection and estimation method of a sensor fault, applicable to both linear static systems and linear dynamic systems, is proposed.
The method is based on vector gradient.
To the static systems, a sensor error vector, which indicates the fault of the sensor, is calculated from the outputs of the sensors and the mathematical models of the systems. And from its behavour in the vector space, the dection and the estimation of the gain parameter fault of the sensor are performed, that is, the fault detection by the magnitude of the vector and the estimation, which identifies a faulty sensor and calculates its gain deviation, by the magnitude and the vector angles.
To the dynamic systems, by transformation of state space equations a gain parameter jump is treated as a bias jump in a observer equation. The fault detection is performed by a generalized likelihood ratio test (GLR test) applied to the residual of a Kalman filter. The faulty sensor and the jump value of its gain parameter are determined by the vector gradient method same as the static systems. By combining the vector gradient method and the GLR test, the algorithm for diagnoses is simplified, keeping the excellent performance of the GLR test.
Satisfactory results are obtained at the simulation of the sensor fault in the 5th order motor drive system.
