| 摘要 |
PURPOSE:To detect a fault with high sensitivity and to assume quickly the area where the fault occurs by using a statistic inspection of an error of estimation to detect the generation of the fault and presuming the area and the time of the fault generation as well as the type of the fault from the variation of the error of estimation produced when the fault is detected. CONSTITUTION:A multi-variable AR model is identified from the fluctuation components of plural signals obtained from a plant and by means of a structure matrix. Then the time series data on the error of estimation is calculated from the difference between the value estimated by the AR model and the observed value of signals. Then the generation of a fault is detected by a statistic inspection of said error of estimation. The area and the time of the fault generation as well as the type of the fault are presumed from the variation of the error of estimation when the fault is detected. In such a way, a fault of a plant can be detected at an extremely early stage with high sensitivity. Furthermore the faulty area can be automatically diagnosed among many signals. In this way, the reliability and the working efficiency of the plant are improved. |
