Research And Development On A Hybrid Fault Diagnosis System Based On Ann And SDG Of FCCU Reactor-Regenerator System

The Fluidized Catalytic Cracking Unit (FCCU) is a key unit in the petroleum deeper processing. The process is complex with interconnected and coupling variables. In order to make the process " safety, stability, long period, full charge and optimization" , the system security, reliability and effectivity become more and more important. Therefore, studies of fault diagnosis technology for FCCU have extremely first principle significance and practical value.The neural network model is used to simulate the process of the human brain neurons. It is one kind of the parallel computing systems which is composed by a lot of simple, high interconnected neurons. It works well in domains of function approximation, pattern recognition, fault diagnosis and so on.Based on SDG (Signed Directed Graph) the qualitative simulation uses in-depth knowledge model inference mechanism. It is one kind of effective methods to reveal system potential hazard as well as the fault spread rules in the system. In recent years, SDG has made significant progress in safety evaluation in the petro-chemical industry. The fault diagnosis method and technology based on the SDG also becomes the hot spot research now.Because of the complexity of diagnosing actual questions and inherent limitation of each kind of FDD methods, it is nearly impossible for any one kind of FDD method to solve the actual diagnosis problems completely. Therefore, it is extremely essential to apply several kinds of FDD methods simultaneously and form the hybrid fault diagnosis system to carry on the integrative diagnosis.FDD studies the special behavior characteristic of the process system under malfunction. But the industry process is continuous and not allowed to carry on dangerous experiments. So establishing FCCU dynamic first principle model, developing simulation platform, producing some kinds of unusual and fault operating scenes are the effective foundation and essential conditions to carry out the FDD engineering research.In this paper the reactor - regenerator system is taken as the studying object which is the most complex and essential part of FCCU. Make use of the system dynamic simulation platform which has been established the fault diagnosis system software for neural network is developed. The SDG qualitative model and the SDG fault diagnosis system are also developed. The two unions are integrated together to form the effective and complete hybrid fault diagnosis system.The main work and achievements are as follows:1. Established and perfected the reactor - regenerator system dynamic first principle non-linear model for FCCU with highly effective coking bucket, which is not only the precision is high, but also the flexibility is good in a wide range. The model is satisfied in the normal operation (nearby stable state operating point). The model still can work under the accident and unusual operating conditions.2. Developed the system simulation platform software under the WINDOWS by programming in VC++6.0. The platform is multi-documents interface (MD1), event driving, operated with mouse and easy to operate.3. Classified the 16 kinds of possible faults and unusual operating conditions in the FCCU reactor - regenerator system into 3 types: 4 kinds for the equipment failure (For example: The raw oil pump stops.), 8 kinds for the misoperation fault (For example: The atomizing steam quantity reduced by mistake.) and 4 kinds for the process fault (For example: Catalyst activity reduces.). On the system simulation platform, 16 kinds of faults and unusual operating cases have been tested. Several kinds of fault scenes have been simulated and gained a lot of fault data. It lays the foundation for the following fault diagnosis research.4. Making use of the MATLAB neural network toolbox, the BP algorithm and RBF algorithm were studied. In view of the online and real-time application request for FDD in the chemical industry, the neural network fast training and the dynamic characteristic have been mainly studied. The improved algorithm is proposed. The MATLAB simulation results have indicated the validity and usability of the improved algorithm.5. Integrating VC++ and MATLAB programming, fault diagnosis software was developed under the WINDOWS. The operation interface, the data interface, the algorithm choice, the tasks scheduling and the management and so on are realized by programming in VC++. The neural network core algorithm is realized by the MATLAB toolbox. It takes advantage of the VC++ visualization and the MATLAB matrix operation. The software could perform off-line training of the neural network and online fault diagnosis.6. Integrating the first principle model and the experience knowledge, 64 critical variables are selected to establish SDG qualitative in-depth knowledge model for the FCCU reactor - regenerator system. The model intuitively reveals the complex interaction of the process variables in reactor - regenerator system. The catalyst circulation, the goal products output and the ESD system have been considered in the model, which has not been included in the past literatures.7. Making use of the SDG fault diagnosis software developed by the Computer Simulation and Safety Engineering Laboratory, the SDG fault diagnosis model of the reactor - regenerator system has been tested. The results of 16 kinds experiment cases show that the SDG model has high completeness and usability.8. Proposed and completed a new hybrid FDD system.The FCCU reactor - regenerator dynamic simulation system, the neural network fault diagnosis software and the SDG fault diagnosis system are integrated together through the data interface to form one kind of brand-new hybrid fault diagnosis system. Many experiment cases, like the fault "inducing" and real-time online fault diagnosis, have been successfully tested under several unusual operating conditions. On the laboratory level these experiments proved the effectivity of mathematical model and fault diagnosis methods proposed and developed in this article.This kind of integrated FDD technology has taken advantage of the SDG system "high-level" inference and the neural network "low-level" processing. The neural network guarantees diagnosis highly effective, the SDG system guarantees diagnosis complete. It is extremely suitable for the complex system diagnosis, especially for real-time diagnosis.