Camilo A. Monroy and Carlos AM Riascos Department of
Transcription
Camilo A. Monroy and Carlos AM Riascos Department of
2013 AIChE Annual Meeting SIMULTANEOUS DYNAMIC DATA RECONCILIATION AND MODEL UPDATING FOR DETERMINATION OF KINETIC PARAMETERS OF BIOMASS PYROLYSIS USING NSPSO BIO-INSPIRED OPTIMIZATION ALGORITHM [1] [2] METHODOLOGY Biomass pyrolysis is a thermochemical conversion process that is of both industrial and environmental importance. Understanding kinetics of solid state pyrolysis is imperative for multiples purposes, from designing and operating of industrial biomass conversion systems to reactors modeling. Biomass pyrolysis involves a complicated network of multiple reactions, and it is not easy to determine the reaction mechanism in the necessary detail to formulate the reaction kinetics. One simplified kinetic model of three reactions was proposed for cellulose, hemicellulose and lignin pyrolysis. This kinetic model involves lumping the complicated multiple reactions as a single first-order (3Rxn_1) or nth-order reaction (3Rxn_n). This work only consider kinetic model 3Rxn_n because previous investigations showed maximum likelihood whit this model. The thermal decomposition of lignocellulosic biomass materials and their components obtained by enzymatic process was developed for generating experimental information that supports formulation of the kinetic model that describes the pyrolysis process. A kinetic analysis of biomass pyrolysis was conducted using either a thermogravimetric (TGA) curve. TGA analysis shows 3 peaks that represent the pyrolysis of biomass component (hemicellulose, cellulose and lignin). Data reconciliation (DR) as filtering technique that improves the reliability and accuracy of process data was employed; it reduces impacts of random errors and biases, and generates estimations in tune with the process model by forcing them to satisfy material and energy balance constraints. Dynamic data reconciliation (DDR) is a natural extension of DR to dynamic processes and provides a set of reconciled estimates in some time horizon. DDR combines state, parameter and unknown input estimation, and is typically formulated as a dynamic optimization problem restricted by a process model. In the present work, a maximum likelihood function for dynamic measurements and kinetic model of pyrolysis process is formulated as the objective for dynamic data reconciliation and model updating problems. Kinetic parameters were estimated by using a bio-inspired algorithmic NonDominated Sorting Particle Swarm Optimization (NSPSO) and D-Optimal experiment. For analyzing results quality, the confidence interval for estimates were determined too. Finally, model validation was developed by Fisher-Snedecor test. Kinetic Model Parameter estimation The following equations were used as kinetic model for pyrolysis of biomass: Dynamic Data Reconciliation considers and optimization problem that allows model updating Maximum-likelihood equation where wbm is the organic fraction of biomass, t is the time of reaction, i is the number of reaction, Ai is the pre-exponential factor, Ei is the activation energy, T is the temperature and k the rate of heating. Model Validation dy(t ) sa f y(t ), u (t ), (t ), t 0 dt h y(t ), u (t ), (t ), t 0 Model error g y(t ), u (t ), (t ), t 0 Fisher-Snedecor have two test: If the first test is not true but the second is true, where y(t), ӯ(t) are vectors of measures and estimates at time step t, parameter bi represents you have a generalized behavior model. If both tests are true, it can say that the the slope of linear regression of error model, c is the current time, k is the sampling time, Q model is common knowledge, however, if the two tests are not true you cannot is the covariance matrix, f, h are differential and algebraic equality constraints, and g are say that the model is not validated (model is considered validated in default of). inequality constraints that include upper and lower bounds. It was found that DDR is useful when the pyrolysis process involves only 3 thermal peaks of reaction, which are included in the kinetic model 3Rxn_n. When thermal analysis presents 4 peaks it is required to determinate previously the kinetic parameters of the additional reaction. With DDR the error between experimental data and model prediction was reduced. In the Figures 1 and 2 it is observed that DDR eliminated unusual behavior in experimental data of pyrolysis. TGA for lignin presents 2 peaks of reactions (Figure 3), but the kinetic model considers only one reaction for this component, due to this topological difference between the model and the behavior of the true system, DDR method does not generate adequate results. RESULTS INTRODUCTION Camilo A. Monroy and Carlos A. M. Riascos Department of Chemical and Environmental Engineering, National University of Colombia, Bogota/COLOMBIA a) b) a) Figure 2. Pyrolysis of lignin a) with out DDR and b) with DDR Figure 1. Pyrolysis of palm shell a) with out DDR and b) with DDR Similar values for pyrolysis kinetic parameters of palm shell were obtained when used a) NSPSO algorithm with out DDR and b) NSPSO algorithm plus DDR (See Table 1 and Table 2). The second methodology (NSPSO-DDR) decreased the maximum likelihood objective function (MLOF) from –35173.6 to –35173.9. This value is into confidence interval (+/- 10 from the objective function value), therefore is not a significant improvement. The validation of the model and kinetic parameters was realized for NPSO and NPSO-DDR methodologies. Both methodologies approved the first test (validation) of Fisher-Snedecor tests but do not approved the second test (Repetition). ACKNOWLEDGMENT This work was possible thanks to the teachings and knowledge in modeling and simulation of Profs. Hermes Rangel, Carlos Riascos, Sonia Rincón and Alejandra Guzman. We also thanks the National University of Colombia for all the resources used for developing this work. [1] Camilo Antonio Monroy Peña, Colombian, Che Eng., Student at Chemical Engineering PhD Program, Universidad Nacional de Colombia, Bogota. [email protected]. [2] Carlos Arturo Matinez Riascos, Colombian, Che Eng, Msc and PhD. Professor, Universidad Nacional de Colombia, Bogota. [email protected]. b) Figure 3. Pyrolysis of lignin with two peaks Table 1. Kinetic parameter identificated with NSPSO algorithm and with out DDR, and with NSPSO algorithm and DDR NSPSO algorithm and NSPSO algorithm and with out DDR DDR Rxn1 Rxn2 Rxn3 Rxn1 Rxn2 Rxn3 E [KJ/mol] 258.8 172.4 31.2 257.6 172.1 31.2 Log A 20.48 14.79 -0.78 20.37 14.76 -0.78 ni 1.91 2.54 2.23 1.9 2.54 2.23 MLOF -35173.6 35173.9 CONCLUSIONS NSPSO-DDR methodology is a good technic for identification of kinetic parameter of biomass pyrolysis and data reconciliation NSPSO-DDR result satisfactory with 3Rxn_n kinetic model describe pyrolysis process. When in pyrolysis of biomass appear a fourth peak, the kinetic model required incorporating another reaction for the new lignin component. Kinetic parameters identified with NSPSO-DDR methodology are similar to the ones identified with NSPSO with out DDR. There is intersection in the confidence intervals for the parameters, therefore with both methodologies, results have not significant difference.