CHEMICAL ENGINEERING TRANSACTIONS VOL. 32, 2013 A publication of The Italian Association of Chemical Engineering Online at: www.aidic.it/cet Chief Editors: Sauro Pierucci, Jiří J. Klemeš Copyright © 2013, AIDIC Servizi S.r.l., I SBN 978-88-95608-23-5; I SSN 1974-9791 Composition Estimator Design for Industrial Multicomponent Distillation Column Marcella Porru a , Jesus Alvarez b , Roberto Baratti* a a Dip. Ingegneria Meccanica, Chimica e dei Materiali, via Marengo, 2 09123 Cagliari, Italy b Dep. de Ingeniería de Procesos e Hidráulica, UAM, Apdo. 55534, 09340 México, D.F. Mexico roberto.baratti@dimcm.unica.it The problem of on-line estimating on the basis of temperature measurements the distillate NC4 impurity in an industrial IC4/NC4 splitter (operating at Saras refinery at Sarroch in Italy) is addressed. The application of the adjustable model-based Geometric Estimation approach yields a dynamic data processor that adequately performs the estimation task in the light of a prescribed estimation tolerance with a scheme that is considerably simpler than the Extended Kalman Filter (EKF) employed in previous studies. The implementation of the proposed estimator with experimental data shows that distillate NC4 impurity can be inferred with an uncertainty similar to the one of off-line occasional determinations. 1. Introduction Distillation columns are important energy-consuming industrial units where a mixture is separated into two or more key components. The related control problem consists in efficiently performing the component separation in the presence of disturbances, in the sense of purity within prescribed low and high limits, and non-wasteful control action. Due to high investment and maintenance costs as well as equipment reliability and measurement delay problems of on-line composition analyzers, more often than not an effluent composition controller cannot be implemented. This motivates the development of (first-principle or empirical) model-based composition estimators driven by temperature measurements for monitoring and control purposes. Basically, this estimation problem has been addressed with the first-principle (mostly EKF) (Baratti et al., 1995, 1998) and (input-output) data driven (Mejdell and Skogestad, 1993; Kano et al., 2000) models. On one hand, the EKF functions rather well over an ample set of column types and operating condition, but its implementation requires a detailed first-principle model and the on-line integration of a number of ODEs that grows rapidly (quadraticaly) with the number of stages and components. On the other hand, the data driven approach does not require a first-principle model, but the implementation requires significant model identification effort using considerable experimental input-output data, and validity is restricted to the specific column and operating condition encompassed by the data. Recently, the dimensionality problem of the model based EKF approach for multicomponent distillation column (Frau et al., 2009, 2010; Frau, 2011) has been addressed with the so called Geometric Estimation (GE) approach (Álvarez and Fernandéz, 2009). While in the EKF a complete observability is required and the model is fixed, in the GE only detectability is needed and the (possibly truncated) model is a design degree of freedom. Consequently, in multicomponent distillation column, the GE has considerably less ODEs then the EKF. The GE has been successfully implemented in binary laboratory (Fernandéz, 2007) and ternary pilot (Pulis, 2007) columns with experimental data, and tested with a six-component industrial scale column through simulations (Frau, 2011). These considerations motivate the scope of the present study: the implementation -for the first time- with industrial experimental data of a geometric estimator for a multicomponent (IC4-NC4 splitter) column. Specifically, in this paper the problem of on-line estimating, within a prescribed uncertainty value, the distillate NC4 impurity for an industrial IC4-NC4 splitter (operating at Saras refinery at Sarroch in Italy) on the basis of temperature measurements is addressed. The estimator design includes decisions on the 1975