Towards a model-based fault accomodation system

In this work; a general model based fault accommodation system is proposed and demonstrated. Detailed modeling and accommodation strategies are presented for different types of abnormal situations. The proposed accommodation system includes five major components: Data acquisition and validation, monitoring and diagnosis; identification and estimation; optimization and supervisory control, and control. Given the complex nature of any real process, no single diagnostic method developed so far is adequate for the task of fault diagnosis. A combination of different diagnostic techniques; such as Signed Directed Graph (SDG) based methods, Qualitative Trend Analysis (QTA) and Probability Density Functions (PDF) based statistical classifier, involved in collective problem solving and parallel ways of reasoning; is used in the diagnosis module. In the fault identification and state estimation block, a pseudo-measurements based Extended Kalman Filter (EKF) is developed to perform the state and parameter estimation task for general differential and algebraic equation systems (DAEs). Based on the results from the identification block, a dynamic model based optimizer is then utilized to perform the corrective action proposing, or control redesign. Generally this is done in two steps, first the optimizer tries to drive the process into the feasible operation region and then move the process to an optimal state. The dynamic optimization problems arising, from this formulation are solved using a model decomposition strategy, an efficient numerical method described and demonstrated for solving general dynamic optimization problems. The proposed system can handle both single fault and multiple simultaneously occurring faults. Another advantage of this method lies in its ability to simultaneously achieve economic optimality and operation safety due to its capability to handle both hard equipment constraints and operational constraints. The proposed approach is applied to the Amoco Model IV Fluid Catalytic Cracking Unit (FCCU). The simulation results show that more than 85% of the modeled faults in CATSIM, a dynamic simulator of the FCCU, can be effectively accommodated through this model based approach. The remaining 15% of the faults are controller faults. Due to the complexity of the joint dynamics of the process and the controllers, these are not simulated in this work