The main purpose of this paper is to present a new framework for the integration of data rectification (DR) in incipient process fault diagnosis (IPFD). Significant amount of efforts have been made to investigate DR and IPDF separately during the last two decades. Integration of DR and IPFD has seldom been addressed in detail even though DR has been recognized as a key prerequisite for successful implementation of IPFD because there exist random errors and maybe exist gross errors in measured data fed into the on-line IPFD system and therefore can generate false alarms. . Gross errors can be divided into two classes: (1) measurement bias and (2) outliers. The bias refers to the situations in which the measurement values are consistently too high or too low. Outlier maybe considered as the abnormal behavior of the measurements, process peaks or unmeasured disturbance, for example. So the data with gross errors have great negative influence on IPFD. This paper is intending to improve the effectiveness of the existing IPFD systems by using integration of DR in it. In the proposed framework, there are five major components: process state identification (PSI) [1], Gross error detection and identification, sensor fault diagnosis SFD , dynamic data reconciliation (DDR)[2-3] and IPFD. Through PSI, the process is identified as steady state or dynamic one. Under the dynamic states, the gross error detection and identification is carried out. If the bias type of gross errors exist, then the zero drift or malfunction of the instruments may occur, which presents the base of the detection of sensor fault diagnosis. The Optegrity platform which is a new commercial package from Gensym corporation designed for abnormal situation management of chemical processes with built-in fault diagnosis capability of common equipments, such as sensors, heat exchangers, heaters, pumps and so on, is used here for SFD. If no sensor fault is identified, the data containing gross errors should be firstly treated properly and then reconciled through DDR to remove random errors. The reconciled data is then fed into the IPFD model for detecting process faults. The cause-effect IPFD model is also configured based on Optegrity. The benefit of using Optegrity for IPFD is that the reasoning is done automatically by the platform once the cause-effect model is established. Therefore, the development and implementation of IPFD is greatly eased. The proposed framework is demonstrated by an industrial case study. The simulation results indicate that the number of false alarms of IPFD is significantly reduced by using the framework proposed.
展开▼
