- 10:15 AM

Application of a Model Predictive Control Strategy on a Fluid Catalytic Cracking Pilot Plant

George M. Bollas1, Simira A. Papadopoulou2, Angelos A. Lappas1, Spyros S. Voutetakis3, Panos Seferlis1, and Iacovos A. Vasalos1. (1) Chemical Process Engineering Research Institute (CPERI), Centre for Research and Technology Hellas (CERTH), 6th km Charilaou–Thermi Rd, PO Box 361, Thessaloniki, Greece, (2) Department of Automation, Alexander Technological Educational Institute of Thessaloniki, P.O. Box 14561, Thessaloniki, 54101, Greece, (3) Chemical Process Engineering Research Institute, Centre for Research and Technology - Hellas, 6th km Charilaou–Thermi Rd, PO Box 60361, Thessaloniki, 57001, Greece

Chemical process industry faces the potential of applying model-based predictive control strategies on the fluid catalytic cracking (FCC) unit as a very promising solution for process optimization and profit maximization. However, the cost of developing a reasonably accurate first-principles model for the FCC process is usually prohibitive. This is due to the strong interactions and the high degree of uncertainty in the integrated riser-regenerator loop. The stochastic nature of the air distribution in the regenerator, the moderately defined flow regime of the gas-catalyst mixture in the riser, and the catalyst circulation throughout the unit mainly form a complex integrated system. The different feed and catalyst qualities can be considered as typical disturbances to the system. On the other hand, operational constraints set for safe and stable operation, product specifications and environmental restrictions contribute to the challenges of an already complex control problem.
The main objective of this work was to improve the control performance in the FCC pilot plant (PP) operated in Chemical Process Engineering Institute (CPERI - Thessaloniki, Greece). This PP is operated in a mode suitable for catalyst benchmarking experiments. Therefore, there are many similarities and several differences between the operation of the pilot plant and that of a typical industrial unit. While the potential of yielding more market-oriented FCC products, increasing the capacity and the stability are the typical concerns in industrial operation practice, the main concern in the pilot plant operation is the stable operation within a narrow process window. Catalysts should be evaluated at constant conversion levels and riser reactor temperatures. Therefore, control of this pilot scale process faces several challenges:
• Riser temperature, controlled by the catalyst circulation rate in the closed loop plant operation, should satisfy a specified set-point that guarantees constant selectivity in the product slate.
• Conversion of the gas-oil feed should meet a determined value for easy comparison (testing) of variable catalyst activity and selectivity.
• Excess gas from the regenerator is subject to environmental constraints regarding the CO, SO2, NOx emissions in commercial units and this pattern should be followed, or even examined, in the pilot plant operation also.
While the PP unit was operated many years through the application of conventional control schemes based on several PID controllers, unit productivity is related to the stability and ease of operation at specified conditions. A model predictive control (MPC) strategy was therefore implemented for the tight and efficient control of the pilot process. A dynamic mathematical model has been developed and verified on the basis of steady state and dynamic experimental data of the FCC pilot plant of CPERI [1]. The simulator predicts conversion, coke yield, and heat consumed by feed vaporization and catalytic reactions in the pilot riser reactor through the use of semi-empirical models developed in CPERI [2-3]. The pilot regenerator reactor model uses the two-phase theory with a dilute phase model in account for post-combustion reactions [4]. The effective manipulated variables in the pilot plant can be the catalyst circulation rate, feed preheat temperature, combustion air flow rate (and temperature), and gas-oil feed flow rate (though in an industrial unit it is driven by the need for target production), while the gas-oil composition and the catalyst quality are considered as disturbances. The controlled variables are chosen to be the regenerator temperature, riser temperature, conversion (feed basis), and coke yield (feed basis). However, the variables that can be measured online in the pilot plant are the riser and regenerator temperature, the flue gas of regenerator, system pressure and pressure drops. The state variables were corrected through the use of an extended Kalman filter algorithm. Conversion and coke yield were therefore inferred from the updated process measurements and the modeling relations. Thus, it was possible to implement feedback control scheme based on optimization through a cost function around the desired operational point. Moreover, constraints were imposed on the emissions of the regenerator for CO, NOx, and SO2. As the pilot plant regenerator operates under full combustion mode the goal of minimum to zero CO emissions could be achieved, yet for the other two goals the effect of the optimal operating point of the pilot plant was explored.
The development of the control structure underwent two main stages. First, an original detailed simulator was used as the “virtual process” (VP) in a theoretical study, whereas an actual dynamic model was implemented in the MPC scheme. A disturbance, such as a change in feedstock or catalyst quality, was implemented in the VP and the robustness and effectiveness of the MPC scheme was explored. After the proper control structure was selected and a suitable tuning (e.g., objective function weight factors) was verified, it was applied on the actual pilot plant. Control parameter updates were obtained using multiple, variable and/or infrequent rate process measurements and the dynamic model. The optimal piecewise constant future control actions were calculated through dynamic programming over desired prediction and control horizons. The sensitivity of the performance of the online optimal nonlinear MPC with respect to the duration of the control intervals and the prediction and control horizons was examined in conjunction to the effort for the numerical solution. The implementation of the MPC scheme showed extreme robustness to changes in the feed quality and catalyst activity and selectivity. The MPC scheme allowed for an accurate targeting of the desired feed conversion with little knowledge about the catalyst properties and selectivity. In conclusion, the MPC strategy allowed for tight following of the prescribed operating conditions and the elimination of additional and/or repeat experiments with the same catalyst in catalyst evaluation tests, thus improving the overall productivity of the catalyst evaluation studies task that this PP is mainly used for.
1. Bollas, G.M. et al., (2006). Integrated riser-regenerator dynamics studied in a fluid catalytic cracking pilot plant. Submitted for publication to Chemical Engineering Science.
2. Bollas, G.M. et al., (2004). Bulk Molecular Characterization Approach for the simulation of FCC feedstocks. Industrial & Engineering Chemistry Research, 43(13), 3270-3281.
3. Bollas, G.M. et al., (2002). Modeling small-diameter FCC riser reactors. A hydrodynamic and kinetic approach. Industrial & Engineering Chemistry Research, 41(22), 5410-5419.
4. Faltsi-Saravelou, O. and Vasalos, I.A., (1991). Fbsim - A model for fluidized-bed simulation. 1. Dynamic modeling of an adiabatic reacting system of small gas-fluidized particles. Computers & Chemical Engineering, 15(9), 639-646.