Copyright <0 IFAC Artificial Intelligence in Real-Time Control, Kuala Lumpur, Malaysia, 1997 An approach for Fault Tolerant Controller Design Using Kalman Filters and Fuzzy Logic Athula Rajapakse J Kazoo Furuta Shunsuke Kondo Department of Quantum Engineering and Systems Science, University of Tokyo 7-3-1, Hongo, Bunkyo-Ku, Tokyo 113, Japan J Email: athula@rokoh.gen.u-tokyo.ac.jp Abstract An autonomous control system needs to provide adequate control even under the presence of uncertainties such as actuator or component failures, with a minimum or no human intervention. We propose an approach for designing a controller that can provide higher degree of autonomy by accommodating the parametric failures occur in the plant. The controller has a two layered structure. The first layer contains the local controllers of various sub systems. The second layer contains Kalman filter fault diagnosis and fuzzy logic fault accommodation systems. A controller is implemented and applied to control a simulation model of a chemical process system. Simulation results demonstrate the ability of this control system to diagnose and recover from a heater failure in the chemical reactor heating system. Copyright © 1998 IFAC Keywords: Autonomous Control, Fault Diagnosis, Fault Accommodation, Fuzzy Logic Systems. 1. Introduction Recent advances in computer and information technology has enabled realization of more sophisticated control systems with higher degree of autonomy. The added benefits of such control systems include improved overall process system performance due to better optimization of system interactions, less downtime because of better fault accommodation, and increased safety by allowing more human response time to anomalous events. In most cases, and specifically for complex process systems, the benefits of these advanced control systems can far outweigh the additional controller complexity [1]. Complex process systems are usually controlled by decomposing the process into a number of sub systems, each having its own local controller designed to achieve a set of control objectives within the subsystem. In such a decentralized system, the coordinating or supervisory controller plays an important role by coordinating the various subsystem controllers avoiding potential conflicts between their performance objectives (2). Although the local controllers are designed to be robust and capable of meeting the local objectives, they may not be capable of handling abnormal situations such as actuator or component failures, because of the limitations imposed by the abnormality. The supervisory controller in an autonomous control system must be able to provide an adequate control even under such situations, with a minimum or no human intervention. An approach to realize a controller capable of accommodating faults occurring in the process system is presented in this 463 paper. The process considered is a chemical plant consisting of a continuously stirred tank reactor and associated heating system. The controller has a layered structure. The first layer contains the local controllers while the second layer contains fault detection and accommodation system. A set of adaptive Extended Kalman Filters (EKFs) employed for combined state and parameter estimation detect plant malfunctions from the deviations in the estimated system parameters from their respective designated values. If an abnormality is detected, the fuzzy logic fault accommodation system will modify the set points of the local controllers 'according to the information on parameter changes, in order to minimize the effects of abnormality. The remaining part of this paper is organized as following. Sections 2 of the paper briefly describes the process model. Section 3 presents the controller structure. Section 4 gives the steps involved in fault recovery process. The scheme for fault diagnosis using EKFs is presented in Section 5. In Section 6, the fault accommodation methodology is presented. Simulation results presented in Section 7 is followed by the conclusion. 2. Process Model The process model used for the demonstration of the proposed controller design approach is a section of a fictitious chemical process scheme. The system illustrated in Fig. 1 consists of a non-isothermal continuously stirred tank reactor (CSlR) and its heating system. At the normal operation, feed "A" from preceding stage is pumped into the reactor at a constant rate through the recovery heat exchanger primary side. The feed undergoes an endothermic reaction at the reactor and converts into the product. The outlet product stream from the reactor passes through the heat exchanger secondary side. Reactor temperature is controlled by a closed cycle heating water system. Heating water is circulated at a constant rate. There are two local controllers in the system: liquid level and concentration controllers. The level controller maintains the liquid level in the reactor at a desired value within the lower and upper limits of 1.2m and l.7m respectively. The concentration of 'A' in the product stream is required to be manipulated within the range 300-500 mol/ml, in order to optimize the yields of final product and byproducts in the subsequent stages of the process scheme. Concentration of 'A' in the reactor is indirectly controlled by changing the reactor temperature through adjusting the power level of the heater in the heating water system. The local controllers are fully capable to control the plant variables within their limits under normal conditions. The dynamic model of the CSlR was adopted from [3] and it is given in Appendix-A. Dynamics of