Model-based Optimization of a CompactCooking™ G2 Digesting Process Stage Master’s thesis presentation for the degree of M.Sc. (Tech.) in Process Systems Engineering (Process Automation) Igor Saavedra Supervisor: Prof. Sirkka-Liisa Jämsä- Jounela Advisor: Dr.-Ing. Aldo Cipriano Instructor: D.Sc. Olli Joutsimo Tuesday January 26, 2016
41
Embed
Model-based Optimization of a CompactCooking G2 Digesting Process Stage
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Model-based Optimization of a CompactCooking™ G2 Digesting Process Stage
Master’s thesis presentation for the degree of M.Sc. (Tech.) in Process Systems Engineering (Process Automation)
Igor Saavedra
Supervisor: Prof. Sirkka-Liisa Jämsä-JounelaAdvisor: Dr.-Ing. Aldo CiprianoInstructor: D.Sc. Olli Joutsimo
Tuesday January 26, 2016
IntroductionPulp and Paper
• What is Pulp …• Fibers sources• Lignocellulosic
biomass• Market pulp
• Softwood• Hardwood
• and Paper?• Paper products• Fiber properties
Logs
Woodchips
Pulp
IntroductionKraft Pulp Mill Process
Nueva Aldea Pulp Mill 1500+1500 Adt/d pine & euca, 91% ISO, 460MWth (95 MWe), 1000 L/s water inflow
IntroductionProblem Statement
• Digesting Stage Optimization
• Key area of the Kraft pulp mill transforming woodchips into brownstock and weak black liquor by consuming steam and white liquor
• Given an scenario of operating costs and target production rate: how do we cook pulp optimally?
• CompactCooking G2 (Valmet), digesting system found in the mill, is a highly interacting process that combines liquor recycling and heat integration
IntroductionGoals, Scope and Novelty
• Main Goal• Design a process optimizer able of minimizing cost or maximizing
profit rates of a CompactCooking G2 digesting stage.
• Specific Objectives• Design and validate a dynamic model of the process stage.• Design and perform a steady-state optimization routine based on
the previously validated model.• Assess performance of theoretical optimal set-points versus
current mill set-points.
IntroductionGoals, Scope and Novelty
• Scope• Development of an applied solution for the mill.• Dynamic modeling of KPIs of stage such as
• Kappa number• Production rates • Temperatures and alkali concentrations• Pulp intrinsic viscosity and cellulose DP• “Cooking recipe” values:
• Liquor-to-wood ratios (L/W), • Alkali charges (A/W), • H-factor• Dilution factor (DF) of the wash zone
• Phenomena to be modeled are chip bed compaction, cooking reaction kinetics, and heat-exchanges within cooking liquors.
IntroductionStructure of the Thesis
Chapter 1 IntroductionLITERATURE PART
Chapter 2 The Kraft Pulp MillChapter 3 Pulp Digesting StageChapter 4 Mathematical Models on Pulp Digesters
EXPERIMENTAL PART
Chapter 5 MethodsChapter 6 Process DescriptionChapter 7 Mathematical ModelingChapter 8 Simulator DesignChapter 9 Simulation ResultsChapter 10 Optimizer DesignChapter 11 Optimization ResultsChapter 12 ConclusionsAppendix A Model-based Process Analysis
Literature ReviewMathematical Models on Pulp Digesters
• Review on first-principle modelingVroom (1957) H-Factor concept that describes the extent of delignification based on a
simple kinetic law using temperature and time as parameters
Hatton (1973; 1976) Equations relating cooking yield and kappa number with H-factor and effective alkali for softwood and hardwood species. Later he applies this work to Kraft cooking control.
Smith (1974) First version of the Purdue kinetic model. Wood solid is represented as 5 components, and parallel reaction kinetics are used to describe cooking reactions.
Christensen (1982) Improved Purdue model by search algorithm to adjust kinetics parameters for softwood and hardwood species. Liquor concentrations are also calculated.
Gustafson et al. (1983) First version of the Gustafson kinetic model. Three stages cooking: initial, bulk and residual. Wood solid is represented as 2 components: lignin and carbs
Härkönen (1987) First 2D continuous digester model with emphasis on chips and fluid flow dynamics with a simplified kinetic model. This contributed a framework for bed compaction modeling used in almost all later developments.
Literature ReviewMathematical Models on Pulp Digesters
• Review on first-principle modelingSaltin (1992) A dynamic, continuous digester model using the Purdue kinetics and a
simplified Härkönen bed compaction model. Implemented in GEMS.
Agarwal (1993) A steady-state, continuous digester model using Gustafson kinetics and implemented by the single chip approach. It also incorporated a viscosity model derived from Kubes et al. work and introduced the modelling of diffusion and chip thickness by a sphere-equivalent chip model. Implemented in GEMS.
Michelsen (1995) A dynamic, continuous digester model using a simplified Purdue-like kinetics and a modified Härkönen bed compaction model that involves solving a dynamic momentum balance for the chips phase. First modelling approach of chip level variations. Implemented in MATLAB.
Wisnewski et al. (1997) A dynamic, continuous digester model with improved Purdue kinetics but fixed bed compaction profile. It is also modelled the liquor concentration of dissolved wood substance and the chip internal porosity. Implemented in MATLAB.
He et al. (1999) First 3D model of a continuous digester based on Harkonen and Michelsen fluid dynamics assumptions with a simplified kinetics model. 3D, dyn M&E&P balances
Literature ReviewMathematical Models on Pulp Digesters
• Review on first-principle modelingBhartiya et al. (2001) Continuation of Wisnewski et al. work incorporating advances made by
Michelsen. It also contributed a modelling approach for grade transition. Implemented in MATLAB.
Andersson (2003) New kinetic model that combines Purdue and Gustafson approaches. Wood substance is represented by 5x3 components.
Kayihan et al. (2005) A dynamic, continuous digester model based on Purdue kinetics, modified Härkönen bed compaction, and Agarwal diffusion and chip thickness. It is solved by a novel cinematic approach allowing to model chip level and stochastic changes in chip size distribution. Implemented in MATLAB.
Rantanen (2006) A dynamic, continuous digester model based on Gustafson kinetics, Saltin simplified bed compaction, and Agarwal diffusion and chip thickness. It is applied to describe a LoSolids™ process (two-vessel stage) with grade transition. Implemented in MATLAB.
Nieminen et al. (2014a, 2014b)
New kinetic models of lignin and carbohydrates degradation. Delignification can be described with varying degrees of sophistication (including Donnan equilibrium); and carbs degradation is modelled based on the reaction mechanism of peeling, stopping and alkaline hydrolysis. Reactions dependence on [OH-], [HS-] and [Na+] is considered.
Literature ReviewProcess Control and Optimization
• LP optimization• Objective function as
• Cost or profit rate• Cost or profit per unit of
product or educt• Constraints on
• Flow rates, temperatures, compaction pressures, concentrations, etc.
• Linear input-output models of the process• SP-MV ( u=u(r) )• PV-MV ( y=y(u) )
Process DescriptionCompactCooking G2 System
• Physical input streams:• Woodchips• MP-steam• White liquor• Wash liquor
• Physical output streams:• Cooked pulp
Weak black liquor
Simulator DesignMethodology
• The simulator aims to capture the dynamic behavior of the system with emphasis on interaction effects
• Changes in one input variable affect several outputs in a non-linear form
• Some bias on the output is acceptable, but poor correlation between measured and simulated outputs is not.
• Simulator code builds upon parts of the Pulp Mill Benchmark Model, updating it to represent current cooking technologies
• CompactCooking G2 is a highly interacting process, thus simulation of the whole is a must for a rigorous optimization effort
Start
Process flowsheet abstraction
Conceptual model IO variables
Conceptual model states variables
Data acquisition and conditioning for
testing
Test criteria are met?
Testing runs and parameter adjustment
NO
End
Data acquisition and conditioning for
validation
Validation run
Validation criteria are met?
Validity domain definition
NO
YES
YES
Model implementation
Process historian
P&ID, PFD, DCS visualizations
Literature submodels
Open source models, code
libraries
Validated simulation
model
Process historian
Mod
el V
alid
ation
Mod
el T
estin
g
Castro, J. J., & Doyle, F. J. (2004). A pulp mill benchmark problem for control: problem description. Journal of Process Control, 14(1), 17–29.
Simulator DesignModel structure
• Logical inputs: 16 MVs 14 DVs
• Comparable outputs:
29 PVs
• Total selected outputs:
40 PV
Simulator DesignSimulink model
Simulator DesignSimulink model
Mathematical modelingMain assumptions
• Vessels are tubular moving bed reactors• Fixed levels
• Although levels were tried to be dynamically modeled, computation times increase too much and numerical stability of the model is compromised
• Two-phases reacting system• Concentrations on entrapped liquor are the same as on the free liquor phase,
thus total number of states is lowered• 1D description on the axial direction of bed compaction and reaction
kinetics phenomena• Heat-exchangers are perfectly mixed tanks
• Heat exchange occurs between hot and cold side at a given total heat transfer coefficient UA
• Liquor densities are held constant, although composition is dynamically modeled
• Liquor compositions vary solely due to retention times, no reaction kinetics take place into heat-exchangers
Mathematical modelingMain assumptions
• Woochips are composed of six mass entities• Fast lignin, slow lignin, cellulose, (galacto)glucomanan, (arabino)xylan,
and extractives
• Extractives are represent as instantaneously leached when entering the Impbin
• Liquor is composed of seven mass entities• Sodium hydroxide NaOH(aq), sodium hydrosulfide NaSH(aq), dissolved
lignin, dissolved cellulose and so on
• Consumed NaOH and NaSH are accounted for density calculations in order to keep mass balance consistency
Härkönen, E. J. (1984). A Mathematical Model for Two-Phase Flow (Doctoral dissertation). Helsinki University of Technology.Härkönen, E. J. (1987). A mathematical model for two-phase flow in a continuous digester. Tappi Journal, 70(12), 122–126.
Mathematical modelingBed Compaction
• Experimental values from literature
Lee, Q. F. (2002). Fluid flow through packed columns of cooked wood chips (Master’s thesis). University of British Columbia.
Wisnewski, P. A., Doyle, F. J., Kayihan, F. (1997). Fundamental Continuous Pulp-Digester Model for Simulation and Control. AIChE Journal, 43 (12), 3175-3192Christensen, T. (1982,). A Mathematical Model of the Kraft Pulping Process (Doctoral Dissertation). Purdue University.Smith, C. C. & Williams T. J. (1974). Mathematical Modeling, Simulation and Control of the Operation of Kamyr Continuous Digester for Kraft Process,Tech. Rep. 64, PLAIC, Purdue University.
Mathematical modelingReaction kinetics
• Experimental values from literature
Wisnewski, P. A., Doyle, F. J., Kayihan, F. (1997). Fundamental Continuous Pulp-Digester Model for Simulation and Control. AIChE Journal, 43 (12), 3175-3192Christensen, T. (1982). A Mathematical Model of the Kraft Pulping Process (PhD’s thesis). Purdue University.
Manipulated variable (simulated)Disturbance (simulated)Output (simulated)Mill data (measured)
Simulation ResultsTesting (Pine)
DCS estimate.
NO SENSOR at the mill
Cooking and bleaching yield are
actually set point
parameters
Prod. rate is assumed based on yield set points
Manipulated variable (simulated)Disturbance (simulated)Output (simulated)Mill data (measured)
Simulation ResultsValidation (Pine)
Manipulated variable (measured)Disturbance (measured)Output (simulated)Mill data (measured)
Simulation ResultsValidation (Pine)
Output (simulated)Mill data (measured)
Simulation ResultsAssessment
• In general, simulated outputs capture the main dynamic trends with reasonable agreement Model is operationally validated
• Simulated temperature signals show higher variability than measured ones• Improvements in the simulated heat-exchanger networks is required, but
this demands implementing several TI at the mill in order to estimate U coefficients for each heat-exchanger (or to estimate U within the model and to have output signals for comparison)
• Simulated blowline flow rate shows higher variability than measured • This might be generating a bias in the wash zone dilution factor calculation• One way to fix this involves using the signal as a logical input (manipulated
variable) and changing the model structure for bed compaction calculation Long-term effort
Optimizer DesignMethodology
• The routine tries to find a new cooking recipe that optimize process economics by changing following DCS setpoints:
Liquor to wood ratio (L/W) for Impbin (bottom), Digester cook zone 1 and 2
Alkali charge (EA/W) for the whole area, fresh charge to Impbin, and fresh charge to Digester
Alkali splitting as white liquor flow distributionCooking temperature (for H-Factor setpoint)Digester wash zone dilution factor (DF)
• Decision variables are taken as manipulated variables, thus optimization outputs continue to be the same as in the simulation model
• A previously validated model is a critical factor to judge the optimization results
Profit rate as o.f.Cost rate as o.f.Base case (ss)Base case (dyn)
Optimization ResultsEconomic Assessment
• Process economics can be evaluated from several point of views. This work considers 3 definitions of profit/cost: • per unit of time, • per unit of actual cooked ADt• per unit of actual woodchips m3sub consumed
• Optimized recipe for cost reduction results more attractive economically than the profit recipe• Savings per actual cooked ADt up to 4 USD/ADt• For a line aiming to produce 1500 ADt/d, this represent up to 2.19
MM annual savings
ConclusionsMain Conclusions
• CompactCooking G2 system has been dynamically modeled with fairly good results although high uncertainty on process disturbance signals.
• An LP task can be formulated around an identified mill’s steady state, thus permitting to calculate a new optimized cooking recipe (optimization direction for mill setpoint changes).
• Potential savings based on the model prediction may reach up to 4 USD/ADt, what for a modern mill (1500 – 2000 ADt/d) represent savings in the order of 1 – 3 MM/y.
THANKS FOR YOUR ATTENTION!
Simulated ContributionA novel model-based process analysis technique