Gibaek Lee

Robust fault diagnosis based on clustered symptom trees

Abstract This study suggests a new methodology for fault diagnosis, based on the signed digraph (SDG), in developing the fault-diagnostic system of a boiler plant. The suggested methodology uses a new model, the clustered symptom tree (CST). The CST utilizes the advantage of the SDG to represent the causal relationship between process variables and/or the propagation paths of faults in a simple and graphical way, and is therefore easy to understand. It also covers the problems, such as symptom variation, that conventional SDG-based methods cannot handle. The advantages of the presented method were confirmed through case studies.

Process Monitoring of an Electro-Pneumatic Valve Actuator Using Kernel Principal Component Analysis

Abstract In this paper, an approach for process monitoring using a multivariate statistical technique, namely kernel principal component analysis is studied. Kernel principal analysis has recently been proposed as a new method for performing a nonlinear form of principal component analysis (PCA). The basic idea of kernel PCA is to first map the input space into a feature space via a nonlinear map and then compute the principal components in that feature space. For the process monitoring application, reconstructed input patterns can be obtained by approximating the pre-image of scores in feature space. An application study of an electro-pneumatic valve actuator in a sugar factory is described. The results show that the kernel PCA approach can detect several actuator faults earlier than linear PCA This study indicates the great potential of Kernel PCA for process monitoring.

Detection of High Impedance Faults Using the Randomness of Even Harmonic Currents

Abstract In this paper, we propose a new method for the detection of high impedance faults, which use the randomness of even order harmonic currents. From the analysis of the staged fault test data, it was found that there exists an intermittent arcing phenomenon in most high impedance fault and the waveform of this arcing fault current has an unsymmetric shape in each cycle. From these facts, we present new criterions for fault detection, all of which are based on changes of even order harmonic powers in fault currents. The proposed detection method under these criterions is compared with existing detection methods for high impedance faults through the analysis of a staged fault data and a normal switching event data. By this comparison, it will be shown that the detection method presented in this paper is superior to existing methods and that high impedance faults can be easily distinguished from normal switching events. Microprocessor based protective relays, which detect high impedance faults by using the proposed methods, have been constructed and installed on KEPCO substations and tested during last two years.

Qualitative Fault Diagnosis with System Decomposition: Application to a Large-Scale Boiler Plant

Abstract This study suggests a systematic method to decompose a large-scale process into sub-processes and then diagnose them. It is based on qualitative fault diagnosis of fault-effect tree model and for the minimization of knowledge base and flexible diagnosis. The new node, called a gate-variable, is introduced to connect the cause-effect relationships of each sub-process. Off-line analysis is performed to construct fault-effect trees of gate-variables. And, diagnosis strategy is modified to get the same result without system decomposition. The method is illustrated with a fault diagnosis system for a boiler plant.

Development of structural controllability evaluation procedure for chemical processes with disturbance propagation

Abstract A structural controllability measure based on relative order is introduced in this work. Because the structural controllability measure requires only structural information, and semi-quantitative information about processes if available, it will be useful in the integration of process design and control in early design stage. The effectiveness of the proposed method is demonstrated in HEN synthesis problem, and the result is validated by dynamic simulation.

Evaluation of Structural Controllability in Chemical Processes Using Relative Order

Abstract The control performance of a chemical process is determined by process structure as well as the performance of controllers. Therefore, the concept of “controllability” should be introduced in the early design stage to maximize the control performance. Structural information makes controllability evaluation possible by giving insights into the pathways of disturbances in the process. In this study, a simple controllability evaluation procedure is suggested to screen out design alternatives using relative order analysis and structural decomposition. The effectiveness of the proposed method was validated by comparing the results with the case of dynamical simulation.

The methodology for knowledge base compression and robust diagnosis: Application to a steam boiler plant

Abstract The new diagnosis approach, which is based on SDG to use the advantages of SDG and covers the problems that conventional SDG-based methods cannot handle, is employed for robust diagnosis. Two compression methods are suggested to prevent the drawbacks of the approach which a very large knowledge base gives in large-scale processes. The clustering of measured variables and the system decomposition enables minimization, easy construction and maintenance of the knowledge base and flexible diagnosis throughout the operational change of the process. To show the advantages of the proposed methods, the fault diagnosis system for a steam boiler plant, ENDS (ENergy Diagnosis System) was developed using the expert system shell G2. In the case study, the size of the diagnostic rules is reduced to 0.75% of that of the case without compression, and the system is verified to give fast and robust diagnosis results for the real system.

A DYNAMIC PROCESS SIMULATOR BASED UPON THE CLUSTER-MODULAR APPROACH

The objectives of this work are to present the dynamic simulation strategy based on clustermodular approach and to develop a prototype simulator. In addition, methods for the improvement of computational efficiency and applicability are studied. A process can be decomposed into several clusters which consist of strongly coupled units depending upon the process dynamics or topology. The combined approach of simultaneous and sequential simulation based on the cluster structure is implemented within the developed dynamic process simulator, MOSA (Multi Objective Simulation Architecture). Dynamic simulation for a utility plant is presented as a case study in order to prove the efficiency and flexibility of MOSA.

DETECTION OF HIGH IMPEDANCE FAULTS USING THE RANDOMNESS OF EVEN HARMONIC CURRENTS

In this paper, we propose a new method for the detection of high impedance faults, which use the randomness of even order harmonic currents. From the analysis of the staged fault test data, it was found that there exists an intermittent arcing phenomenon in most high impedance fault and the waveform of this arcing fault current has an unsymmetric shape in each cycle. From these facts, we present new criterions for fault detection, all of which are based on changes of even order harmonic powers in fault currents. The proposed detection method under these criterions is compared with existing detection methods for high impedance faults through the analysis of a staged fault data and a normal switching event data. By this comparison, it will be shown that the detection method presented in this paper is superior to existing methods and that high impedance faults can be easily distinguished from normal switching events. Microprocessor based protective relays, which detect high impedance faults by using the proposed methods, have been constructed and installed on KEPCO substations and tested during last two years.