Dynamic Modelling and Experimental Validation of Mass Flow in a Pilot-Scale Pretreatment Continuous Tubular Reactor

Dynamic Modelling and Experimental Validation of Mass Flow in a Pilot-Scale Pretreatment Continuous Tubular Reactor

Antonio Espuña, Moisès Graells and Luis Puigjaner (Editors), Proceedings of the 27th European Symposium on Computer Aided Process Engineering – ESCAPE...

561KB Sizes 0 Downloads 84 Views

Antonio Espuña, Moisès Graells and Luis Puigjaner (Editors), Proceedings of the 27th European Symposium on Computer Aided Process Engineering – ESCAPE 27 October 1st - 5th, 2017, Barcelona, Spain © 2017 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/B978-0-444-63965-3.50422-0

Dynamic Modelling and Experimental Validation of Mass Flow in a Pilot-Scale Pretreatment Continuous Tubular Reactor Felicia Rodrígueza, Ismael Jaramilloa, Ricardo Jaraa, Arturo Sancheza* a

Laboratorio de Futuros en Bioenergía. Centro de Investigación y de Estudios Avanzados del IPN. Av. Del Bosque 1145, C.P. 45019, Jalisco, México [email protected]

Abstract This work presents a dynamic model of mass flow in a pilot-scale lignocellulosic biomass (LB) pretreatment continuous tubular reactor (CTR). In this CTR the biomass is subject to three processes: extrusion, autohydrolysis and steam explosion. The production of desired amounts of oligosaccharides, monosaccharides and degradation compounds during isothermal autohydrolysis in the CTR can be achieved by controlling the residence time. The model was validated with experimental data of two different LB showing that residence time can be manipulated by the screw motors speeds, which directly impact in the input and output mass flowrates. Parameters for each type of LB were estimated by PSO. Keywords: Tubular reactor, flowrate, modelling, pretreatment, PSO.

1. Introduction Lignocellulosic biomass is considered a potential feedstock that may help to meet the future world demands for bioethanol and high value-added products production. Pretreatment of LB, as the first step in biorefineries, is a key stage to attain high efficiencies in enzymatic hydrolysis and fermentation steps due to physical and chemical modifications of the lignocellulosic structure (Klinke et.al., 2004). Continuous reaction schemes are usually employed in large and medium scale facilities for LB pretreatment. Changes in physiochemical characteristics of pretreated LB result from a synergy of chemical reactions (e.g. catalyzed or uncatalized hydrolysis) and physical phenomena (steam explosion) promoted by the high pressure steam to which the LB is exposed. Autohydrolysis and steam explosion without catalysis are considered as one of the most environmentally sustainable pretreatments. Both pretreatments together increase the availability of cellulose to enzymatic attack due to hemicellulose depolymerisation weakening the lignin matrix. However, under severe temperature and large reaction times, degradation of products of interest may occur. Since autohydrolysis is a first order reaction and exhibit Arrhenius temperature dependence, the main autohydrolysis control variables are residence time, temperature and concentration. These variables were considered in the modelling of tubular pretreatment reactors reported by Prunescu et.al (2015) and Lopez-Arenas et.al. (2013). Although the modelling was based on rigorous mass and energy balances, important aspects of mass and energy transfer were not considered. Additionally, the flowrate in continuous pretreatment reactors is directly related to mechanical characteristics of the screw conveyor and process conditions

2522

F. Rodríguez et al.

(temperature and pressure) as well as LB rheological and mechanical properties. Some biomass characteristics such as shape, density, viscosity, moisture, particle size and adhesion effects as well as reactor design produce the formation of dead zones inside the reactor, causing non ideal flows. In order to predict flow patterns in a continuous pretreatment reactor, Berson et.al. (2006) and Sievers et.al. (2009) used computational fluid dynamic (CFD) and residence time distribution (RTD) techniques. However, these tools were only used for reactor design purposes and its experimental validation was carried out under unrealistic operational conditions. This paper focuses on the empirical modelling of mass flow behaviour through the tubular pilot scale pretreatment reactor based on experimental results. The modelling variables were selected such that the model can be used for control purposes. Model parameters were estimated using Particle Swarm Optimization (PSO) method (Rodríguez and Sanchez, 2014) for two different LB.

2. Process description The horizontal pilot-scale tubular continuous reactor is 1.65 m long and 0.1 m internal diameter with a nominal rate of 3 to 5 kg/h (dry basis). The reactor is equipped with steam-supply, an external heating jacket, a screw conveyor driven by a 1 HP motor (Motor-01) as well as biomass feeding and discharge systems (see Fig. 1). During the passage through the reactor, the LB is subjected to 3 pretreatments: extrusion, autohydrolysis and steam explosion. The LB extrusion is carried out in the feeding system, which consists of a feeding hopper connected to a 4:1 extruder driven by a 2 HP motor (Motor-02). Then, the LB is transported with a screw conveyor through the reactor body. In this stage, the LB is exposed to saturated steam promoting hemicellulose depolymerisation by autohydrolysis reaction. Finally steam explosion occurs in the discharge system by the asynchronous opening of two globe valves, discharging the biomass to a flash tank (biomass collector).

Figure 1: Schematic of Pilot-scale continuous tubular reactor configuration.

3. Model development The depolymerisation yield under isothermal conditions is directly related to residence time, which may be controlled by manipulating the input and output mass flowrates. In

Dynamic Modelling and Experimental Validation of Mass Flow in a Pilot-Scale Pretreatment Continuous Tubular Reactor

2523

order to develop mathematical tools to control the residence time, a dynamic model of mass flows based on experimental results was devised. The mass flow model was divided into two parts: feeding and process. Based on experimental evidence, a first order linear dynamic model was proposed to describe the input mass flow rate:

‫ݔ‬ሶ ଶ ൌ െܿ‫ݔ‬ଶ ൅ ܾ߱ଶ ൅ ݂‫ݔ‬ଶ ߱ଶ

(1)

Where ‫ݔ‬ଶ represents the output mass flow from the feeding stage, ߱ଶ is the feeding motor speed,ܾ and ܿ are scalar gains dependent of biomass characteristics and ݂ is a scalar related to the influence of screw speed. The output mass flow from the process is affected by the LB characteristics, process motor speed, the reactor's length and the feeding stage input as modelled by Eq. (2).

‫ݔ‬ሶ ଵ ൌ െܽ‫ݔ‬ଵ ሺ‫ݐ‬ሻ ൅ ܽ‫ݔ‬ଶ ሺ‫ ݐ‬െ ݄ሻ 

߱ଵ ሺ‫ݐ‬ሻ ݂݅߱ଵ ൐ Ͳ ߱ଵ ሺ‫ ݐ‬െ ݄ሻ

(2)

Where ‫ݔ‬ଵ is the output mass flow from the process stage, ߱ଵ is the process screw speed, ܽ is a scalar gain dependent on LB characteristics. ݄ is the delay time, modelling the thime that the LB remains inside the reactor. Equations (1) and (2) were discretized using Euler approximation since the LB discharges are carried out in discrete time:

‫ݔ‬ଵሺ௞ାଵሻ ൌ ሾͳ െ ܽܶሿ‫ݔ‬ଵሺ௞ሻ ൅ ܽܶ‫ݔ‬ଶሺ௞ି௛ሻ

߱ଵሺ௞ሻ ߱ଵ ሺ݇ െ ݄ሻ

‫ݔ‬ଶሺ௞ାଵሻ ൌ ൣͳ െ ܿܶ െ ݂ܶ߱ଶሺ௞ሻ ൧‫ݔ‬ଶሺ௞ሻ ൅ ܾܶ߱ଶሺ௞ሻ

(3) (4)

Where ܶ is the sampling period. The delay time was modelled as a process screw speed function according to equation (5), where ‫ݐݎ‬ଵ଴଴Ψ is the experimental LB residence time at full process screw speed:

݄ ൌ

‫ݐݎ‬ଵ଴଴Ψ  ߱ଵሺ௞ሻ

(5)

The values of parameters ܽ, ܾ, ܿ and ݂ were estimated for each type of biomass applying the Particle Swarm Optimization (PSO) algorithm. The data were fitted using the mean square error (MSE) described by Eq. 6. ௡

ͳ  ൌ  ෍ሺ‫ݕ‬௜ െ ‫ݕ‬ො௜ ሻଶ  ݊

(6)

௜ୀଵ

ො ݅ are the experimental and simulated data at the sample point ݅, and ݊ is Where ‫ ݅ݕ‬and ‫ݕ‬ the total number of experimental points. 4. Experimental set-up The experiments to study the feeding stage were carried out using only the reactor’s feeding system at room temperature. Accurately weighed LB was soaked in water with a ratio of 1:10 (w/v) for 2 hrs. Then, the solid fraction was manually fed into the hopper,

2524

F. Rodríguez et al.

where the screw conveyor transports and extrudes the LB. These experiments were performed at two different motor speeds (100 and 50 % of the nominal motor speed). The extruded biomass was collected and weighted every minute. The experiments to analyse the output mass flowrate were performed with the complete pilot-scale tubular continuous reactor at 150 psi with a discharge rate of 2 min (discrete sampling time). Two motors speeds were used in each experiment to gather data under different operating conditions. Additionally, two different LB were tested for both feeding and process experiments: wheat straw (WS) and sugar cane bagasse (CB).

5. Experimental results and model validation Typical experimental runs of the feeding and the process stages using the two different LB are shown in Figures 2 and 3 with black diamonds. Fifth-order low-pass filtered data are presented with gray continuous lines and the model simulation with gray semicontinuous lines. In Figures 2 and 3, F1 and F2 sections indicate that feeding motor was working at 100 and 50 % of its nominal speed, respectively. P1 and P2 areas show that the process motor was working at 100 and 50 % of its nominal speed, respectively. The estimated parameters by PSO, minimum MSE (MMSE) and the average output mass flow from the process stage (AOMF) at 100 % of the feeding and process motor speed obtained in each experiment are shown in Table 1. 5.1. Feeding stage Figure 2 shows typical experimental data related to the output mass flow of the feeding stage of WS and CB. Stationary states at F1 and F2 are identified in F-SS1 and F-SS2 time interval, respectively. The average output mass flow in F-SS1 was slightly higher in WS than CB. The average mass flowrate of WS decreased proportionally with motor speed. However, the average mass flowrate of CB decreased only to 64 % when the motor speed decreased to 50 % of its nominal speed. Furthermore, F-SS1 and F-SS2 of WS and CB were reached 8 and 14 min after changing the motor speed, respectively. These behaviours suggest that the feeding stage is highly influenced by the physical and rheological characteristics of the LB, as the large difference between the values of the estimated parameters in each type of LB suggests. The model of the feeding output mass flow (‫ݔ‬ଶ ) describes adequately the phenomena mentioned above. 5.2. Process stage Experimental data corresponding to the output mass flow from the process stage of WS and CB are shown in Figure 3. Filtered data showed the delay of the mass flow response to step changes in both the feeding and the process motor speeds. The mean delay for the two biomasses was 18 min (”–ଵ଴଴Ψ ሻ. Additionally, filtered data showed stationary states for each operating condition of the motors. In the Figure 3, P-SS1 is the stationary state at F1 and P1, P-SS2 is the stationary state at F2 and P1, and P-SS3 is the stationary state at F1 and P2. Process motor speed had a greater influence in output mass flow from the process stage than the feeding motor speed. The output mass flow response to step changes in the process motor speed was similar in both cases. The mass flow decreased from 50 g/min to around 40 g/min when the process motor speed was decreased to 50 % of its nominal value. As in the feeding stage, the output mass flow response to step changes in the feeding motor speed was different for each biomass. Simulation results show good agreement between the simulation and filtered data for WS and CB. The model is capable of predicting the effect of changes in both feeding and process motor speeds on the output mass flow in the CTR.

Dynamic Modelling and Experimental Validation of Mass Flow in a Pilot-Scale Pretreatment Continuous Tubular Reactor

2525

Figure 2: Experimental and simulated data of the output mass flowrate from the feeding stage of a) WS and b) CB in a pilot-scale CTR.

Figure 3: Experimental and simulated data of the output mass flowrate from the process stage of a) WS and b) CB in a pilot-scale CTR.

2526

F. Rodríguez et al. Table 1: Estimated parameters used in the simulation of the model. Estimated Parameters

Biomass Type

ܽ

ܾ

ܿ

݂

MMSE

AOMF 100 %

WS

0.00199

0.0837

0.0065

0.00083

255.91

56.16 (gr/min)

CB

0.00620

0.4510

0.0187

0.00796

91.60

50.71 (gr/min)

6. Conclusions In this work, an empirical discrete dynamic model is presented for predicting the output mass flow of a pilot-scale CTR based on experimental data of two raw materials. This model involves the mass balance, response delay as well as rheological and mechanical characteristics of the biomass. The parameters of the model were calculated from experimental data, so that the model consider operational conditions of the CTR such as high pressure and biomass discharge frequency (discrete sampling time). The model characteristics and the simulated results show that the proposed model is suitable for control purposes in the biomass pretreatment stage.

Acknowledgements Partial financial support is acknowledged from SENER-CONACYT Fondo de Sustentabilidad Energética, Grant 245750 and Centro Mexicano de Innovación en Bioenergía, Grant 249564.

References H.B. Klinke, A.B. Thomsen, B.K. Ahring, 2004, Inhibition of ethanol-producing yeast and bacteria by degradation products produced during pre-treatment of biomass, Applied Microbiology and Biotechnology, 66, 1, 10-26. R. M. Prunescu, M. Blanke, J. G. Jakobsen, G. Sin, 2015, Dynamic modeling and validation of a biomass hydrothermal pretreatment process - a demonstration scale study, AIChE Journal, 61, 12, 4235-4250. T. López-Arenas, M. Sales-Cruz, J. Alvarez, A. Schaum, 2013, Modelling, design and operation of a pretreatment reactor for lignocellulosic biomass, Computer Aided Chemical Engineering, 32, 0, 37-42. R. E. Berson, R. K. Dasari, T. R. Hanley, 2006, Modeling of a Continuous Pretreatment Reactor Using Computational Fluid Dynamics, Applied Biochemistry and Biotechnology, TwentySeventh Symposium on Biotechnology for Fuels and Chemicals, 129, 132, 621-630. D. A. Sievers, R. T. Elander, E. M. Kuhn, N. J. Nagle, M. P. Tucker, N. D. Weiss, 2009, Investigating Residence Time Distribution (RTD) and Effects on Performance in Continuous Biomass Pretreatment Reactor Designs, NREL/PO-510-45809. F. Rodríguez, A. Sánchez, 2014, Particle swarm optimization (PSO): A method to improve kinetic parameters estimation of lignocellulosic biomass autohydrolysis, Lignobiotech III 3er Symposium on Biotecnology applied to lignoceluloses. Concepción, Chile, Octuber 26-29.