Controle Avançado para Máquina de Papel / Advanced Control for Paper Machine

Esse artigo é bastante interessante, pois mostra uma solução de controle avançado para máquina de papel, área tradicionalmente atendida pelo QCS (quality control system). Aproveitem.

 

Wet End Control Applications using a Multivariable Model Predictive Control Strategy

 

Stephen Chu, P.Eng, Honeywell

500 Brooksbank Avenue, North Vancouver, B.C. V7J 3S4 Canada

 

INTRODUCTION

 

The wet end of the paper machine is an extremely dynamic and interactive process. With infinite combinations of furnish compositions, chemical additives flows, vacuum box pressure levels and machine speeds, the papermaker refers to papermaking as more art than science. Regulatory control loops (single input / single output – SISO) and new wet end gauges (such as SpectraFoil MD which measures stock consistency on the wire) are useful but possible changes or upset to the paper machine will affect several other variable at the same time. During machine break recoveries and grade changes, the papermaker is constantly looking at multiple monitors to determine if each regulatory loop and gauge measurement is within target.

 

The Profit Multivariable MPC (Model Predictive Control) controller models the interaction of the key wet end variables and is used in a control scheme discussed in this paper. A significant reduction in variation in the key wet end variables were observed while the Profit Multivariable MPC controller was on. By stabilizing the wet end, the pre size press ash variability was minimized, yielding a more consistent product from grade to grade. Once the wet end variables were stabilized, the pre size press ash setpoint was increased incrementally. The economic goal was to replace fiber with filler while not affecting the paper machine run-ability and paper quality.

 

IDENTIFYING THE KEY WET END VARIABLES

 

A large US fine paper manufacturer installed a SpectraFoil MD drainage measurement system on their paper machine to monitor the stock consistency on the Fourdrinier prior to the Dandy Roll. The papermaker uses this measurement (Dandy Consistency) as a guideline to monitor the dry line that is located downstream of the Dandy Roll. The dry line is an indication of paper machine run-ability and paper quality – if the dry line is too far upstream, the paper machine handles process upsets (i.e. changes in broke ratio, refining and/or base stock fiber) better but the paper quality (formation) gets worse.

 

The dry line location is manipulated by manually changing the slice lip opening and / or the vacuum box before the Dandy Roll (Vacuum Box 1). Along with this balancing act, the operators must be aware of the wet end retention aid (Silica) and filler (CaCO3) flows that also affect the dry line location and formation. The challenge for the papermaker is to adjust all these variables, which are in manual control or in simple regulatory loop control (SISO) to get the desired machine run-ability and paper quality.

 

The Dandy Consistency is measured by the Honeywell MD SpectraFoil sensor located under the Fourdrinier wire. The Floc Intensity and Floc Size are measured by the Honeywell formation sensor and the Ash is measured by the Honeywell ash sensor. Both formation and ash sensors are located in the size press scanner. The Tray Solids PV and Ash Retention PV are measured using a Chemtronics system. Figure 1 shows a schematic of the stock approach flow, Fourdrinier and dry end of the paper machine. Also shown are the locations of each Profit variable and how they interact with the Profit Multivariable MPC controller to yield the desired run-ability and paper quality.

APC for Paper Machine Scheme

PROFIT MULTIVARIABLE MPC CONTROL – ON CONTROL VERSUS OFF CONTROL

 

Normal Operation – Profit Multivariable MPC Control OFF

 

During normal operations when the Profit Multivariable MPC controller is off, there is only one regulatory control loop (SISO). The MV is Silica flow SP and the CV is Tray Solids PV.

 

The remainder of the variables are in manual control: The papermaker manually adjusts the CaCO3 flow SP to get ash into target range and also manually adjusts the Vacuum Box 1 SP to get the desired dry line location (using the Dandy Consistency as a guideline). Floc Intensity and Floc Size are secondary measurements and are generally overlooked but the papermaker relies on visual tests at the dry end to confirm good formation.

 

Profit Multivariable MPC Control – On Control

 

 

All CVs have High Limits (HL) and Low Limits (LL). For example, the Ash HL = 16.0% and the Ash LL = 15 %. This is the target operating range for Ash as determined by the papermaker and past targets. The advantage of controlling to a range is that the Profit Multivariable MPC controller will less likely be constrained and the MVs will have more freedom to go after other CVs that are outside its range.

 

All MVs have High Limits and Low Limits. The purpose of this was to ensure paper machine run-ability – the papermaker does not want the Profit Multivariable MPC Controller to control to a setpoint outside an operating range. The operating ranges are determined by the papermaker and past ranges.

 

Figure 2 shows the comparison of Ash and CaCO3 flow SP between on and off Profit Multivariable MPC control. During off control, it is clear that the papermaker is adjusting the CaCO3 flow SP manually to get to the target Ash. While on control, the CaCO3 flow SP is smoothly ramped up by the controller to achieve the target ash range. The two-sigma Ash variation was reduced from 2.2 to 0.4 – an 81.2% reduction by going on control.

 

 

Table 1 shows the summary of improvements while on Profit Multivariable MPC control. There is a significant reduction in two-sigma variations in these CVs. With these CVs stabilized, the papermaker can make more consistent product from grade to grade and from shift to shift. Further, there is now a potential to optimize CVs (such as increase ash) in a systematic and scientific manner.

APC Benefits

INCREASING ASH CONTENT

In Figure 3, the Ash HL and LL were increased by 0.5% and the Profit controller reacted accordingly. After this change, the average Ash increased to 15.05%. In manual control, the Ash was 14.67%. So the difference between Profit control and Ash manual control = 0.38% While increasing Ash content, all CVs (Dandy Consistency, Floc and Size Intensity, Tray Solids PV and Ash Retention PV) were all within their target ranges. Therefore paper machine run-ability and quality did not suffer while increasing Ash content.

Increasing Ash Content

Cost Savings attributed to by the Profit Multivariable MPC Controller

Replacing 1% fiber with 1% Ash saves the paper mill about $0.50 US/Ton and a 0.38% increase in Ash, yields a cost savings of $82,000/year.

 

CONCLUSION

 

The desire to save money by reducing raw material costs and not sacrificing paper machine run-ability or paper quality continues. Because of the high interaction of key wet end variables, SISO control loops and new sensors such as the SpectraFoil MD are adequate but not enough to produce consistent product when increasing Ash content for the purposes of reducing raw material costs. The Profit Multivariable MPC controller takes into account all the interactions of the key wet end variables. This allows the papermaker to incrementally increase the ash target setpoint without sacrificing paper machine run-ability and paper quality.

 

REFERENCES

[1] R. Rauch, W Falkenberg and D. Watzig, “Optimized retention: the key to process, quality and productivity improvements”, Paper Technology, October 2004.

 

Use of process simulation to calculate the Benefits of application of Advanced Process Control to decanters in series

Pessoal, beleza?

Vou publicar aqui um artigo de um grande especialista em soluções avançadas, Mr. Ian Craw.

Mr Ian Craw possui mais de 25 anos de experência em soluções avançadas, principalmente na parte de simulação de processos e controle avançado. Atualmente, é Consultor de Negócios da Honeywell Internacional e eu tenho o prazer de participar de alguns de seus estudos, aqui no Brasil. Aproveitem!

Ian Alistair Craw, P.E. – Honeywell Process Solutions

Introduction

Starting in the early 1980’s, companies in the petroleum refining and petrochemical industries developed and began to implement new process control strategies to improve and optimize the real-time control of industrial facilities.  Collectively, these new process control strategies are known in these industries as Advanced Process Control (APC).  These strategies are now starting to be accepted in the minerals processing industries such as alumina and copper manufacturing. Fiske ⁽¹⁾ reports that in a recent survey of APC Best Practices, 8.25% of respondents were companies in the Metals and Mining industries compared to more than 80% from the Oil and Gas and Petrochemical industries.

The most widely-applied APC applications in the oil refining and petrochemical industries are model-based, predictive control strategies such as Honeywell’s Robust Multivariable Predictive Control Technology (RMPCT) that are implemented as supervisory controls on top of base-level regulatory controls.  Regulatory controls, for example flow controls, level controls, and temperature controls, in general use the traditional control strategy known as the proportional-integral-derivative (PID) algorithm which is taught in most process engineering curricula at universities.  The shortcoming of this traditional control strategy is that it only addresses the control problem based on current observations.  This strategy does not take into account the history of control actions or predictions of the future state of the variables we are interested in controlling (known as controlled variables or CVs).  APC strategies, especially the model-based, predictive algorithm, do take into account the recent history of control actions, the current observations, and predictions of the future state of the CVs to improve control performance.

The benefits of applying these advanced strategies are that it results in significant reduction of the variability of the CVs.  It is an accepted industry benchmark that advanced control strategies will reduce the variability of CVs, as measured by the standard deviation, by 50% over the performance that can be achieved by traditional regulatory control ⁽²⁾.  The figure below illustrates this concept.

Variability of Controlled Variable (CV)

 

 

 

 

 

 

 

In this figure, the hatched area represents the probability function of the value of the CV under only regulatory control.  The colored area represents the probability function of the value of the CV under APC control.  Since the variability (or standard deviation) is reduced under APC, the probability of exceeding the Operating Constraint is reduced.  If we are allowed to maintain the same probability of exceeding the operating constraint as before the application of APC, we can shift the setpoint towards the operating constraint, thus producing tangible benefits.  This is illustrated in the following figure.

 

New setpoint value of Controlled Variable

 

 

 

 

 

 

 

 

 

In this figure, we illustrate that with APC, we can move the setpoint of the CV closer to a plant constraint while maintaining the same probability of exceeding it.  This, generally, is a more profitable operating point. For example if the CV we are controlling is related to plant feed, and we can operate closer to the plant feed limit, we can produce more product which, in most cases, results in higher profitability.  This is the incentive for applying APC to processing facilities.

 

One of the challenges that engineers face is the decision whether to apply APC to their process and, if so, what will be a realistic estimate of the benefits to be achieved.  This is especially important in today’s economic environment, where expenditures must be justified from a financial point of view.  Usually, this means that a payback or return on investment must be calculated as a justification of the expenditure.  This requires a cost figure and also a quantification of the expected benefit.  The cost figure can be determined in a straightforward manner.  This paper presents a methodology to calculate the benefits using process simulation. 

Example Process

The example used is a train of decanters that are recovering a valuable component from a waste mud stream before the mud is discarded in a tailings pond.  This particular example is based on a real plant but the flows and data have been modified to protect the confidentiality of the customer.

The process is a thickener followed by six decanters in series, with the wash water (decantate) flowing counter-current to the mud flow and a final polishing unit that provides a final wash with clean water.  The waste mud feed stream is fed to the thickener where the objective is to increase the density of the mud stream by taking out a rough liquid cut.  This liquid cut is returned to the process.  The underflow from this unit is then fed to the first stage decanter where it is washed by the decantate from the second stage. The decantate from the first decanter is returned to the process as it contains important quantities of the recovered component.  The second through fifth stages are identical.  The sixth stage is where the primary washing solution is fed.  The underflow from this sixth stage undergoes a final polishing with clean water before being discarded.  The decantate from the polishing unit is fed to the sixth stage together with the primary washing solution.  The figure below shows a simplified process flow diagram.

 

 

 

 

 

Example process showing decanters in series for recovering valuable components from mud (mud shown in red)

 

 

 

 

 

 

 

 

 

 

 

In this process, the content of the valuable component in the mud being fed to the thickener is 1200 – 1300 units and the final content in the waste mud is less than 12 units.

 

Methodology

The methodology to calculate the expected APC benefits using numerical simulation is as follows:

 

1. Review the process with plant personnel to understand all of the flows within the scope to be considered.  Of special importance is the flow topology since the model will require the correct topology in order to replicate the heat and material balance properly.
2. Review all of the operating objectives and constraints with the plant operators to ensure a clear understanding of the operating targets and limitations.
3. Collect historical data for all of the flows, densities, temperatures, pressures and compositions within the scope of interest.  In general, it is recommended that one years’ worth of one hour averages be used.  This granularity of data minimizes the information loss due to averaging, without generating an overwhelming number of points and also provides information as to the impact of seasonal weather variations or of different feedstocks or products on the plant.
4. Using the historical data collected, calculate the average and standard deviations for all of the variables of interest.  If there are significantly different operating modes due to seasonal impacts or different feedstocks or products, then the data may have to be segmented and several cases considered.  The need to consider several scenarios becomes evident upon visual inspection of the data.
5. Using the known flow topology, and the calculated averages of the variables of interest as model input values, build a representative simulation model of the process.  Check that the model’s dependent variables match the plant data. This model is known as a tuned plant model since we are using real plant data to “tune” the model and is the Base Case against which we will compare the improved operation that can be expected with the implementation of APC.
6. Once the Base Case model is established, new values for the input variables of the model can be implemented and the model can then solve for all of the unknown or dependent variables.  The model solution will take into account all of the effects to each of the operating equipment in the model.  This is important because for realistic benefits calculations, you must not only predict all of the primary effects but also the secondary effects on the process.

 

 

The benefits were estimated by calculating the reduction in valuable component losses in the underflow mud from the polishing unit by use of a numerical simulation.  The valuable component losses can be reduced by: a) increasing the primary wash solution flow rate, b) reducing the valuable component content of the washing solution or c) increasing the underflow density of each decanter such that the washing efficiency of the whole washing train is improved.  Since the operators currently try to maximize the underflow densities, subject to various operating constraints, it is reasonable to apply advanced control strategies that mimic the operator’s strategy.

 

The calculation of the economic benefits of application of APC resulting in improved control requires knowledge of the relationship between the improved underflow densities and the valuable component content in the underflow of the Polishing unit.  Any valuable component contained in the mud from the last decanter is lost from the process.  Any reduction in the valuable component content of the mud will, by material balance, provide additional valuable component production.

The relationship between underflow densities and valuable component losses was established by use of a simulation model of the washing train.  Honeywell’s UniSim Design rigorous simulation software was utilized for this numerical model.  The model utilizes a simplified representation of the dissolved valuable component in the washing train but it does include the hydraulic mixing effects in each decanter and decanter split factors.  The model was tuned to the plant data and was found to represent the process well-enough to provide an estimate of the relationship between the underflow densities and the valuable component losses.

 

Graphical User Interface (GUI) of commercial simulation package

 

 

 

 

 

 

 

 

 

The figure above shows the graphical user interface (GUI) in Honeywell’s UniSim Design simulation package.  This screen capture shows the representation of front part of the example process, the Thickener and First Wash stage.  A full view of the simulated process is not possible due to space limitation of this paper.  However, the tables showing some of the variables of interest are evident in this screen shot.  Detailed information on any stream is available by double-clicking on its respective icon.

The procedure used in matching the plant data was to let the model converge on the calculated average underflow densities in each decanter, by varying the split factors in each decanter and also letting the wash solution to the sixth decanter, and all flows in between the decanters to vary.  An optimization function in UniSim Design was used for this purpose.  The predicted valuable component losses in the underflow from Polishing unit were in good agreement with the values observed in the plant data.  All of the flows predicted were also within the ranges specified by plant personnel.  Thus, the use of the model to calculate changes in valuable component losses was acceptable for the purpose of this study.

 

The following chart illustrates the results of the data analysis and the estimated improvements due to improved control.  As indicated, the first row is the calculated average underflow density in each of the decanters over the period covered by the data.  The second row is the standard deviation of the data.  The third row is one half of the standard deviation, and the fourth row is the improved underflow densities after implementation of APC to the decanters.

Table 1: Results of data analysis and estimated improvements

	Thickener	1	2	3	4	5	6	Polishing
Average Density	1678	1713	1648	1618	1645	1640	1645	1657
Std. Dev.	6.26	24.9	11.7	32.1	22.5	9.67	6.94	30.3
1/2 SD	3.13	12.5	5.85	16.1	11.3	4.84	3.47	15.2
Improved Average Density	1681	1726	1653	1634	1656	1645	1649	1673

 

 

 

 

 

 

 

 

 

 Original From: Honeywell process Solutions Study Report for Confidential Client

The calculation of the improved average density is based on standard industry practice that application of APC will permit a reduction in variability of one-half of the standard deviation.  The average can then be moved towards operating constraints by that one-half standard deviation while maintaining the same probability of exceeding the constraints.  Thus, our conservative expectation is that application of APC will permit the underflows to operate at a higher density by one half of the standard deviation of current operation, as shown on the fourth row of the table above.

 

This results in a reduction of average losses of valuable component from 23.4 units per hour to 21.9 units per hour resulting in a significant economic benefit.

 

Plant personnel indicated that the underflow pumps could handle underflow densities as high as 1750 before experiencing problems.  With improved control, additional benefits can be achieved by increasing the underflow densities target to operate at higher levels than current operation without concern for exceeding the 1750 density upper limit.  The potential benefits associated with this mode of operation were arrived at by letting the model calculate the operating conditions if all the decanters were allowed to run with a 1700 underflow density target which still leaves a safety margin of 50 units of density.  In this mode of operation, the average valuable component losses would be further reduced to 15.8 units per hour, resulting in a greater economic benefit per year compared to the Base case.  This is results in an incremental benefit over the benefits associated with just improved control.  The feasibility of this type of operation is dependent on other constraints, such a rake torque, capacity of flocculant pumps, etc.  The controller would maximize the underflow densities subject to these constraints on a real-time basis as the algorithm takes into consideration constraints in real-time.

 

The manner in which this would be implemented is to configure the underflow densities as targets for the APC strategy.  The controller would then seek to increase the underflow density of each decanter, up to its limit, whenever there are no constraints active.  The controller has a predictive model that will predict values for all of the important CVs as a function of the past control moves, as well as the current state of the variables.  The controller is designed to accept hard targets or ranges both for the CVs as well as the constraint variables.

One of the insights provided by the model was that significant additional benefits can be achieved by maximizing the flow of the primary Wash Water to the sixth decanter.  Maximization of the primary Wash water is an example of an APC strategy known as a constraint pushing application.  Constraint Pushing is a mode of operation in a multivariable application wherein the controller monitors all of the relevant constraints within its scope and, if no constraints are active, maximizes a flow until all of the constraints are met.  Typically, this type of application results in significant increases in throughput and, thus, increased revenue.

A rough estimate of the benefits that could potentially be achieved by a constraint pushing APC application on the primary Wash Water loop was arrived at by allowing the model maximize this flow subject to constraints.  Plant personnel indicated that the maximum achievable flow of Primary Solution is 460 units per hour.  By operating at the maximum underflow densities of 1700 in each decanter with a maximum primary Wash Water flow of 460 units per hour results in valuable component losses of only 1.53 units per hour in the underflow of Polishing unit.  So, APC in a constraint pushing mode could reduce the valuable component losses by a factor of ten.

 

CALCULATION of Estimated Benefits

The table below is a summary of the estimated benefits under the conditions described above.  The figures shown are each compared to the base case therefore the benefits are not cumulative.  The incremental benefits can be calculated by the difference between them.

Table 2: Estimated benefits

	Units of Measure	Base case	Improved Control	Max U/F Densities	Maximize Wash Water
Polishing unit valuable component content in U/F	Composition units	12.64	12.22	9.32	0.9
Volumetric U/F from Polishing Unit	Flow units per hour	252.4	246.5	237	237.1
Mass density U/F from Polishing Unit	Density units	1657	1673	1700	1700
Liquid Volume Fraction in Mud from Polishing Unit	Fraction	0.733	0.727	0.716	0.716
Valuable component Loss from Polishing Unit	Mass units per hour	23.4	21.9	15.8	1.53
Economic Incentive compared to Base Case (Base Case = 1)		1	2.94	10.89	29.56

Original From: Honeywell Process Solutions Study Report for Confidential Client

NOTE:  These data have been modified but are correct relative to each other

Acknowledgements

I would like to thank my colleagues at Honeywell Process Solutions for their support and assistance in writing this paper.  I would especially like to thank Bob Jonas for sharing his knowledge of mining processes with me and Maria Carolina Baeta for her assistance with the graphics and formatting.

 

References

(1)  TOM FISKE, ARC Best Practice, ARC Advisory Group, ARCWeb.com, January 2008, page 10

(2)  G.D. MARTIN, L.E. TURPIN, R.R. CLINE, Estimating Control Function Benefits, Hydrocarbon Processing, June 1991, page 58 – 59

 

 

 

 

The benefits analysis indicates that application of Advanced Process Control (APC) and Advanced Regulatory Control (ARC) applications to the train of Counter Current Decanters will result in significant benefits in reduced valuable component losses in the underflow of the Polishing unit.  This can be achieved by a combination of Improved Control and operating the decanters at Maximum Underflow Densities.

O que é um Step Test?

Pessoal, tudo bem? Hoje, vou blogar a respeito de como fazer um bom step test. Todos nós, que trabalhamos na área, sabemos o quão importante para a modelagem são os dados de um bom step test.

Primeiramente, o que é um step test?

Segundo Olsen, “Step test is when you place a controller in manual and change the output a small percentage so as not to significantly upset the process and then observe the response in the process variable. The response can be the most common type, a first-order plus dead time response. There are many types of responses that could be observed such as integrating, second order response, inverse response and others”

E quais seriam as etapas de um step test?

1. Coleta de dados históricos, telas de operação, fluxogramas de processo e informação de engenheiros e operadores, através de entrevistas
2. Revisão do status de todos os instrumentos envolvidos na aplicação, além da configuração/estratégia de controle e parâmetros de sintonia.
3. Após análise de toda informação coletada, definição de uma pré-estrutura para a matriz do controle avançado. Esta matriz irá conduzir os testes iniciais de planta (pré-testes)
4. Antes do início dos testes, a planta deve ser levada ao máximo de estabilidade possível, de forma que cada modificação (step) realizado em possíveis variáveis manipuladas seja percebido pelas variáveis controladas relacionadas. Estas modificações, primeiramente, devem ser percebidas unicamente, de forma a ser identificada a dinâmica e o tempo para o estado estacionário do processo.
5. Na fase de pré-teste, são determinadas as dimensões dos movimentos a serem realizados nas variáveis manipuladas, bem como um esboço dos tempos para as variáveis controladas atingirem o estado estacionário. Além disso, também são identificados de forma primária os efeitos que as variáveis distúrbio geram no processo.
6. Posteriormente ao Pré-Teste, o “Step Test” é realizado, aplicando-se “steps” nas variáveis consideradas manipuladas. Os steps são degraus aplicados, com tempos e ordens de grandeza variáveis, de forma a identificar-se as interações que construirão a matriz de controle. O número de degraus será determinado pela precisão necessária para a modelagem.
A participação dos operadores nessa fase é primordial, uma vez que são eles que operam a planta e também são os únicos a possuir o poder de manter a planta o mais estável possível para o teste. Por isso, a etapa de preparação para o Step Test (na qual são definidas as variáveis a serem movidas, além de definir o tamanho dos movimentos e conscientizar os operadores da importância) é muito importante para o processo.

É isso aí. Até a próxima!

Artigo Premiado ABTCP 2008

 Pessoal, tudo tranquilo?

Desculpem a demora em blogar de novo, mas é que eu estava na feira da ABTCP (Associação Brasileira de Tecnologia em Celulose e Papel). Por isso, minha semana foi muito corrida.

 

Felizmente, a correria da semana rendeu frutos: meu artigo foi um dos mais bem avaliados pela comissão julgadora. Dessa forma, vou publicá-lo aqui, para dividir com vocês. Saudações e aproveitem.

 

Multivariable Process Control Applications for the Pulp Industry

 

 

Authors:

Rafael Lopes, Rick Van Fleet, Daniel Figueiredo

1. INTRODUCTION

1.1 Overview

The production of cellulose pulp is a highly intensive process involving the interactions of numerous processes. The ability to make this process economically viable relies on good process design together with an understanding of the process dynamics. The present economic climate requires that cellulose producers are able to make good quality cellulose at the right price and in a timely manner. Best in class pulp and paper producers are aware of this and embrace every opportunity to gain an advantage.  Multivariable control and site-wide optimization ultimately are an excellent choice to enable this.

 

1.2 MPC (Multivariable Predictive Control)

One of the effective and leading optimization and advanced control technologies is Multi-Variable Predictive Control (MPC). The philosophy of MPC is to predict the plant behavior, based on heuristic models, in order to take more timely preventive and corrective actions to ensure better regulation and plant stability. These models consider the interactions between the key plant variables.. In addition, the predictive capability of model-based MPC further reduces plant variability and allows operating closer to the real plant limits, enabling more productivity and cost reductions than any other advanced control technology. 

 

While MPC is different than PID and Expert Systems, a common perception is that it is more difficult to apply. Actually, the reverse is true. The complexity of application is internal to the software, thus the implementer needs only to define models and then configure desired operation targets. Model definition has been made much easier through use of automated modelling software, process stepper applications, and on-line and on-control modelling software. In early methods such as Expert Rules, Fuzzy Logic, and complex PID strategies, the implementer needed to be a process and control expert, and careful attention to designing the system was very important. It becomes very important that the expert considers how to delivery the majority of benefits given the limitations of these control technologies. For example, expert system and complex PID strategies have difficulties in the following areas: 

• Time-based dynamics require considerable skill and effort to implement, and thus are often eliminated or simplified to reduce cost and effort.
• Interactions from controls and processes require some skill and effort to implements, and thus are often eliminated or simplified to reduce cost and effort.
• Design, building, testing, commissioning, and maintaining a complex customized application, and the resulting complexity (cost) vs benefits achieved often is a challenge. Sustainability is general has also been an issue.
• Poor operator acceptance and utilization due to elimination and simplification of dynamics and interactions. Simply stated, the advanced controls do not properly respond to many situations. It is not uncommon to have less than 50% control utilization

MPC is widely accepted in the petrochemical industries, where traditionally there are high amount of base level automation and the potential benefits are large. Other industries outside of Petrochemical have been slow to adopt MPC, but that trend has been changing. The reliability and amount of automation in industries such as pulp and paper, chemical manufacturing and mining and metals have improved. In addition, in some markets original equipment manufacturers and local automation consultants have introduced and gained some acceptance of simple expert systems that are low-cost to implement for the supplier, but offer lower benefits for the customer. Today, MPC has been utilized in all these markets, but the widespread use and full acceptance is more a cultural issue rather than a technical issue.

In the MPC philosophy, the variables that have to be maintained inside a range or in a target value are called Controlled Variables (CVs). In order to achieve the operational objectives for the CVs, the application adjusts the variables called Manipulated Variables (MVs). These are sometimes referred to as the “handles”. In a “single output-single input”, SISO control, there is only one CV (the process value or controller input) and one MV (process output – valve opening). With MPC technology, there multiple MVs and CVs, as well as Disturbance Variables (DVs), which are measured disturbances that influence in the process. This type of controller is called a multiple input, multiple output or MIMO controller.

1.3 Robust Multivariable Predictive Control Technology

RMPCT technology represents an advance of the traditional MPC technologies. Like the others, this technology models the process, make the necessary predictions and uses multivariable control movements in order to: optimize the process, maintain the variables inside operational limits and respect the process and plant constraints. The performance gain and robustness is due to a patented feature called “range control algorithm” (RCA), which makes that the disturbances and prediction errors inherent to the process, considered in the future movement plan. The picture sketches how the RCA technology works. 

 

 

Figure 1 – RCA Technique Controlling a CV Inside Limits

RCA minimizes the effects of the model uncertainties while determining the smallest process moves required to simultaneously meet control and optimization objectives. The correction horizon concept is that CV errors are reduced to zero at the correction horizon in the future. Prior to the correction horizon, the controller is free to determine any trajectory for the CV as long as the CV is brought within limits or to setpoint at the correction horizon. Because no trajectory is imposed on the controller, the controller has the freedom to determine a trajectory that requires minimum MV movement and is least sensitive to model error.

 

However, the correction horizon by itself does not say anything about what happens to the CV prior to the horizon. It is important that the controller does not transiently move a CV farther outside a limit while correcting other CV errors, even though all CVs are brought to zero error by their correction horizons. Limit funnels are used to prevent the controller from introducing transient errors prior to the correction horizons, by defining constraints on the CVs that are imposed at intervals from the current interval out to the horizon.

2. APPLYING ROBUST MULTIVARIABLE CONTROL TO PULP

 

The challenge to any pulp process is to minimise the cost of production per tonne of cellulose, consistent with quality, safety and environmental considerations.  This translates to maximising the production of cellulose (plant flow and yield), rationalizing the chemical usage and perhaps minimising the energy costs per tonne of product.

 

There are a number of areas within a typical cellulose pulp plant that are well suited to the application of advanced process control. Significant benefits can be derived from these by continuously pushing the various process limits. The benefits can be described bellow:

 

Table 1 – Typical MPC Benefits for Pulp

 

MPC Solution	Typical $/Ton Savings	*Typical $/yr (Savings & Prod Increase)	Other Typical Benefits
Digester – Continuous	$ 1.00 – 2.00	$ 600k – $1.5M	Production increase 1-3%, Pulp Quality, Bleaching Cost
Digester – Batch	$ 2.00 – 4.00	$ 700k – $3M	Production increase 2-8% , Pulp Quality, Bleaching
O2 Delignification	$ 0.50 – 2.50	$150k – $300k	Environmental, Bleaching Cost, Helps de-bottleneck the evaporators
Brownstock Washing	$ 0.50 – 2.50	$300k – $600k	Environmental, Bleaching Cost, Pulp Quality
CLO2 Generation	$ 0.30 – 2.00	$ 200K – $600K	Product Quality, Productivity
Bleach Plant	$ 1.50 – 3.50	$ 350k – $750k	Bleach Cost, Pulp brightness, Productivity
Causticizing	$ 0.40 – 2.00	$150k – $300k	Stable %CE and white liquor strength, Productivity, Reduce dead-load
Lime Kiln	$ 0.50 – 1.50	$200k – $500k	Energy Savings, Lime Quality (e.g. residual carbonate), Productivity, Environmental
Evaporator	$ 0.25 – 1.00	$250k – $1M	Energy Savings, Productivity, BL Solids Stable to Recovery Boiler
Recovery Boiler	$ 2.50 – $4.0	$400k – $1.5M	This solution is made up of 8 different modules with benefits associated with each.  This ranges from increased steam production, increased liquor burning, reduced soot-blowing steam, reduced plugging, etc…
TMP	$ 2.00 – 5.00	$300k – $2M	Energy Savings, Purchased Kraft savings, Pulp Quality
De-Inking	$ 0.20 – 2.00	$100k – $350k	Pulp variability reduction in brightness, Productivity

 

* These values are dependant on plant size and configuration but are typical.

2.1 ROBUST MULTIVARIABLE CONTROL Solution for Continuous Cooking

 

The main objective of this area is to safely produce pulp of the desired quantity and quality, with the minimum consumption of chips, chemicals and energy. The typical main variable interactions in this area are described in the following table.

 

Table 2 – Main Variables in the Continuous Digestion for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Chip Bin Level	Wood Flow
Production Rate	Chip Meter / Blow Flow
Alkali-to-Wood	White Liquor Flow
Liquid-to-Wood	Black Liquor Flow
IV Chip Level (Dual Vessel)	Chip Meter / IV Sluice Flow
Digester. Chip Level	Chip Meter / Blow Flow
Digester. Liquor Level (Vap. Phase)	Extraction Flow
Residual Alkali	Alkali-to-Wood
Dilution Factor	Extraction Flow / Cold Blow
Blow Consistency	Counter Wash Flow
Kappa Number	UH Temp / LH Temp
Coordinated Rate Change	All
Coordinated Grade Change	All

 

In order to implement the MPC in this area, these interactions are considered in the control matrix. It is necessary to excite the process, often referred to as step tests or bump test, to gain information about the relationships. Through step tests, it is possible to model the interactions, which are also important to be considered.

 

In this case, the proven benefits achieved were:

 

• 30 – 40% Reduction in K/KAPPA Variation
• 0.5 – 2%  Increase in Yield
• 2 – 5% Increase in Throughput
• 3 – 6% Reduction in Cooking Chemical Usage
• Reduction in Bleaching Chemical Usage
• Improved Process Visibility

2.2 MPC for Brownstock Washing

The main objectives of this area in a pulp mill are:

 

• Produce clean pulp for bleaching or papermaking by separating black liquor from blown digester pulp
• Accomplish with an acceptable economic balance of:
  • Minimal carryover of spent cooking liquor and dissolved lignin in pulp (minimal “soda loss”)
  • Maximum solids in weak liquor (H2O must be evaporated- $$)
  • Minimal discharge to sewer (environmental permit issue)
  • Minimal use of hot process wash water (energy-$$)

For this case, the main interactions in this area are:

 

Table 3 – Main Variables in the Brown Stock Washing for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Filtrate Tanks Level 1st stage	Shower Flow 1st stage
Filtrate Tank Level 2nd stage	Shower Flow 2nd stage
Filtrate Tank Level 3rd stage	Shower Flow 3rd stage
COD (PV)	Dilution Factor (SP)
% Solids (PV)	Dilution Factor (SP)

 

In this case, the proven benefits achieved were:

 

• Optimized washing to clean pulp and reduce soda losses
• Minimized defoamer usage
• Optimized refining to improve fiber properties and reduce energy
• Maintained screening effectiveness throughout operating range by adjustment of reject and dilution flows
• Maintained production through automatic wiping of screens, knotters, or fibrilizers

2.3 MPC for Bleaching

Bleaching is a chemical process applied to pulp fibers in order to increase their brightness. It has to be accomplished, subject to the following operational constraints:

 

• Without compromise strength properties of the bleached pulp
• With minimum cost
• Without expose personnel to toxic chemicals
• With minimum impact on the environment

 

The main bleaching chemicals used are:

 

Table 4 – Bleaching Chemicals

 

Oxidant	Function	Advantages	Disadvantages
Chlorine (C )	Oxidize and chlorinate lignin	Effective, economical delignification; good particle removal	Organochlorine formation (environmental concern)
Hypochlorite (H)	Oxidize, decolorize, and solubilize lignin	Easy to make and use; low cost	Can cause loss of pulp strength; chloroform formation
Chlorine dioxide (ClO2) (D)	1) Oxidize, decolorize, and solubilize lignin   2) In small amounts with chlorine, protects against cellulose degradation.	Achieves high brightness without loss of pulp strength; good particle bleaching. (Better than chlorine for environment)	Must be made on-site; cost; some organochlorine formation; highly corrosive
Oxygen (O2) (O)	Oxidize and solubilize lignin	Low chemical cost; provides chloride-free effluent for recovery (good for environment)	Requires significant capital equipment when used in large amounts; potential loss of pulp strength
Hydrogen peroxide (H2O2) (P”)	Oxidize and decolorize lignin	Easy to use; low capital cost (good for environment)	High chemical cost; poor particle bleaching; can cause loss of pulp strength
Ozone (O3) (“Z”)	Oxidize, decolorize, and solubilize lignin	Effective; provides chloride-free effluent for recovery (good for environment)	Must be made on-site; cost; poor particle bleaching and pulp strength
Alkali “Extraction” (NaOH)	Hydrolyze chlorolignin and solubilize lignin	Effective and economical	Darkens pulp
Chelants (added to P) EDTA or DTPA	Remove metal ions	Improves peroxide selectivity and efficiency	Cost

 

For this application, some of the control matrix variables could be

 

Table 5 – Main Variables in the Bleaching for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Residual Chemicals	Chloride Dioxide Addition
pH’s	Caustic Soda Addition
Final Brightness	Sulphuric Acid Addition
Tower Temperatures	Oxygen Addition
Tower Pressure Differentials	Peroxide Addition
	Steam Additions

 

In this case, the proven benefits achieved were:

 

• Reduction of chemical usage: 2 – 4%
• Energy savings, due to a better tower temperature control
• Increase in plant stability à Directly translated on Production Increase (around 1 to 5%, depending on how big is the plant variability)

2.4 MPC for Evaporation

The main objectives for this area are:

 

• Remove water from black liquor to match input / recovery demand with minimum steam
• Maximize % Total Dissolved Solids (TDS) to Recovery Boiler
• Minimize fouling effects / Operate at maximum uptime
• Control evaporators levels

 

The evaporator plant consists in multiple heat exchangers (effects) in series. It removes water from Weak Black Liquor (WBL), which comes from brownstock washers, digesters and/or O₂ delignification to WBL storage(s). Typically, there are around 14-18% Total Dissolved Solids (TDS) in the stream and at the end of the operation, the strong black liquor has 68-72% of TDS.

 

The main interactions in this area to be considered in the RMPCT are:

 

Table 6 – Main Variables in the Evaporation for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Production Rate Control	Feed liquor control, Intermediate liquor control, Semi-strong liquor control, Steam control
Strong Black Liquor Dry solids control	Steam control
Feed liquor control	Intermediate liquor control
Evaporation Stages Levels
Evaporation Stages Pressures

 

The main proven benefits for the evaporation system are:

 

• Production increase
  • Increased liquor flows by operating closer to constraints, including: Levels, Condensate quality, Pump Loads
  • Minimise Bypass flows
  • Typical Number: 0.25 to 0.5% decrease in soda losses

 

• Increased yield
  • improvements in the amount of evaporation allows more production
  • improvements in the control of the liquor (digestion) temperature: maximize recirculation flows and limit liquor temperate
  • Typical Number: 0.25 to 0.5% throughput increase (digestion yield)

 

• 0.25^oC to 0.5^oC temperature for evaporation decrease, due to reduced variability. This diminution on the temperature is straightly translated on steam saving.

2.6 MPC for Slaker/Causticizing

The main objective of this area is to convert green liquor from recovery to white liquor for the digester operation. The main reactions are:

 

CaO      +     water       ====>   Ca(OH)₂    +    HEAT (slaking)

 

Ca(OH)₂  +   Na₂CO₃   ====>  2 NaOH    +    CaCO₃  (causticizing)

 

The main problems found on this area are:

 

• Not efficient clarification à Generally caused by the lime feed quality
• Filters plugging à As above
• Slow reaction to Causticizing Efficiency à The plant has a dead time from 2 to 4 hours, depending on the plant. This time is the time to get a reaction from the operator
• Overboiling on the slakers à Lime addition over the operational limits

 

For this case, the main variables should be:

 

Table 7 – Main Variables in the Slaking/Causticizing for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Slaker Temperature	Reburned Lime Addition
Causticizing Efficiency	Fresh Lime Addition
Temperature Differential through the causticizing process	Liquor Addition
	Water Addition

The main proven benefits for this system are:

 

• Maximize causticizing efficiency
• Increase liquor and pulp production, decrease liquor bottlenecks
• Maintain good lime mud settling / filterability
• Provide safe operation:
• Thorough, robust controls
• Better operator visibility to process
• Better operator response to process upsets
• Better operator visibility to process

2.7 MPC for Lime Kiln

The main objective for this area is to convert lime mud from causticizing to quick lime for the slaker. The reaction is the following

 

CaCO3 + Heat à CaO + CO₂

 

The main interactions in this case are:

 

Table 8 –  Main Variables in the Lime Kiln for Pulp and Paper

 

Controlled Variable	Manipulated Variable
Residual O2	Id Fan Speed
Cold End Temperature	Energy Flow
Hot End Temperature	Energy Flow

 

The proven main benefits for this application are:

 

• Production
  • Energy Usage dropped by 8%
  • Lime Requirement reduced by 4%

 

• Quality Improvements
  • Residual Carbonate Variability reduced by 35%
  • Lime Availability Variability reduced by 47%
  • Lime Reactivity Variability reduced by 43%

 

• Kiln Operational Stability
  • Cold end Temperature standard deviation reduced by 65%
  • Excess Oxygen standard deviation reduced by 60%
  • Hot End Temperature standard deviation reduced by 10%

 

 

 

3. CONCLUSIONS

 

The fact that the MPC requirements are always “study and understand deeply the process, get operators and engineers experience, find the key variables and solve the control problem” is driving other industries to look for this technology. Some examples showed that in the mining industry successful applications of MPC have been successfully implemented in ore grinding, smelting, and in all parts of alumina refining. In the pulp industry, MPC also is now a reality and is has been implemented successfully lime kilns and bleach plants.

 

But, the main requirement for a good multivariable control implementation in this field is to have a MPC controller and model that is robust to changing plant conditions. An MPC is needed that can handle with large and sudden disturbances, varying transport delays, the anomalies of sensors (e.g.; ore slurry densities and pulp consistencies), non-linear temperature behaviours, and many of the unique issues associated with a given process.

 

The conclusion is that, even though MPC is used most widely in Petrochemical industries and its implementation and benefits in other industries are proven, MPC could be deployed more frequently and widely in other industries in order to deliver higher productivity and lower costs. Finally, in order to successfully implement multivariable control, is necessary to have a good methodology (Assess à Define à Execute à Deliver à Sustain) and selection of MPC software that is robust and can therefore sustain more time on-control and generate more benefits.

APC – O que é isso?

Fala, pessoal, beleza?

Hoje, vou falar um pouco sobre controle avançado multivariável. Quero falar aqui um pouco sobre os conceitos básicos dessa ferramenta e como ela se encaixa no contexto de solução avançadas. Então, vamos lá.

MPC (model predictive control) é muitas vezes definido como uma família integrada de controladores, onde aplica-se (ou obtém-se) um modelo explícito e separadamente identificável das interações entre as mais diversas variáveis. A capabilidade principal desta matriz de modelos é prever as respostas do processo com relação a mudanças futuras em variáveis manipuláveis e a possíveis distúrbios. Na prática, o MPC é caracterizado por sua habilidade em lidar com restrições tanto nas variáveis manipuladas, quanto nas controladas. Técnicas de MPC fornecem a única metodologia para lidar com restrições de uma forma sistemática, durante o design e a implementação do controlador. Além disso, a otimização se dá através de uma função objetivo, que considera os parâmetros econômicos (preços) de cada variável. A otimização se dá através da minimização (ou maximização, dependendo da abordagem de cada algoritmo) da função objetivo.

As variáveis em um processo que devem ser mantidas dentro de um range operacional  ou em um valor-alvo são chamadas variáveis controladas (CVs). Para manter os valores das CVs dentro do objetivo definido pelo controlador, a aplicação ajusta o valor das variáveis manipuladas (MVs). Em um controle “single output-single input”, há apenas uma CV (a entrada do controlador ou o valor do processo) e uma MV (saída do controlador – abertura de válvula). Com a técnica de MPC, há múltiplas MVs e múltiplas CVs. O controlador observa todas as variáveis e as toma juntas em um mesmo sistema. Com isto, a tecnologia considera as interações entre todas as MVs, CVs e DVs (variáveis distúrbio) simultaneamente.

Dando um exemplo simples, digamos que, nesse frio, você queira tomar um banhozinho quente para relaxar um pouquinho…Concorda que, para que você controle a temperatura da água, é necessário mexer nas 2 torneiras (de água quente e fria)?

Dessa forma, teríamos:

MV’s: Vazão de água quente e água fria

CV’s: Temperatura da água

DV’s: Temperatura da água a ser aquecida. Se a água estiver muito fria, o aquecimento poderá ser ineficiente e você terá que abrir menos a torneira de água fria para compensar essa temperatura mais fria.

Bem, é isso! Espero que tenha esclarecido alguma coisa pra vocês!

Até a próxima!