Optimal Multivariable Control Structure Design for Chemical Plants

DAVID ZUMOFFEN; BASUALDO, MARTA

Nashville, TN, USA.

Congreso; AIChE annual meeting 2009; 2009

American Institute of Chemical Engineers

A new systematic methodology to address the plant-wide control structure selection in chemical process is presented. Additionally it can give a good decision support about which would be the best sensors location. Therefore, the integration of both topics helps to improve the investments cost as well as the overall performance of the closed loop process behavior. The approach presented in this paper is able for handling large scale process since it is independent of the system dimension. This kind of problem is generally solved by using several heuristic criteria for doing the dimensionality reduction. In this work the overall combination of different control structures are rigorously evaluated through genetic algorithms (GA)[4]. They are implemented by using a simplified steady state model obtained by system identification (SI) and closing the loops with an internal model control (IMC) strategy. Therefore, at the beginning a non-square stable open loop process (or stabilized process) together with the control requirements are accounted for the simplified linear model computation. Using the steady-state information of this linear model the optimal sensor location is carried out by evaluating the sum of square errors (SSE) index[10, 11] associated with a full IMC design. Thus, solving the combinatorial problem with GA, the control policy with equal number of input and output variables and lower interaction can be achieved. Then, the optimal control structure is computed considering the net load magnitude. In this stage, an arbitrary plant-model mismatch is introduced to reject specific disturbance and set point effects (trade off). It can be done thanks to adopt, as an objective function, a modified SSE index to minimize the net load effect along the combinatorial space. Finally, the systematic procedure can define a preliminary acceptable controller tuning if a good linear dynamic model is available. This strategy modifies and complements the variables selection and the net load evaluation presented in previous works[2, 3] representing a concrete contribution in this area. Additionally, it can be remarked that generally, publications in this area addresses the problem in separated topics. From the sensor location point of view [8, 5, 9] it is solved considering Kalman filtering techniques in steady-state, sensors precision, observability and integer optimization algorithms. Thus, the optimal sensor net is obtained by a trade off between precision and cost in a Pareto graphic. Typically, the procedure is applied to open loop plants as well as process with a given control structure, so this last topic is not discussed there. Similarly, the plant-wide control area[2, 3, 7] addresses the problem accounting the process steady-state interaction[1], several heuristic information, and some optimization routines. However it does not consider any integration with optimal sensor location issues. The methodology is applied to the well-known Shell heavy oil fractionator process[6]. A complete set of simulations is presented to evaluate the optimal control structure obtained here with a classical decentralized one. In addition, a performance and robustness analysis in the frequency domain is presented too.