Systematic Retrofit Method for Chemical Batch ... - ACS Publications

66

Ind. Eng. Chem. Res. 2008, 47, 66-80

Systematic Retrofit Method for Chemical Batch Processes Using Indicators, Heuristics, and Process Models Levente L. Simon, Neil Osterwalder, Ulrich Fischer,* and Konrad Hungerbu1hler Institute for Chemical and Bioengineering, ETH Zurich, 8093 Zurich, Switzerland

This work presents a novel systematic indicator, heuristics, and process model based decision support framework for retrofitting chemical batch processes. The systematic framework considers the identification of improvement opportunities in a batch plant by considering first the product market situation. The batch plant analysis is structured on three levels: the plant, the process, and the unit operation level, respectively. The analysis of these levels is based on indicators which are of various types and specific for each analysis level. These indicators are linked to heuristics which are used to identify the retrofit actions. Finally, for each identified retrofit action the improvement potential is calculated in order to assess its importance. The developed method was successfully applied to a fine chemical batch production facility, and it was able to identify in a systematic way several retrofit actions with significant improvement potential. 1. Introduction All existing chemical processes have to be continuously retrofitted in order to improve their efficiency. The retrofit projects are triggered, among others, by increased competition, increasing energy costs, patent expiration, and new emission regulations. The goals of the retrofit project are various: capacity expansion, incorporation of new technology, reuse of surplus equipment units, product quality and energy efficiency improvement, operation cost, and waste volume reduction. Extensive research work has been carried out by the academic community in the past decade to address these issues.1-10 The published research work focuses on the development of screening and analysis methods for chemical plants3,5 and chemical processes. The research solutions to analyze and improve chemical processes cover many aspects: Ku¨ru¨m et al.11 focused on entrainer selection,8 Guntern et al. on reactor optimization,12 Simon et al. on batch reactor operation improvement,9 Ciric and Floudas on heat exchange networks,10 Jo¨dicke et al. on water-reuse,11 Halim and Srinivasan,4 Fonyo et al., and Dantus et al. on waste minimization,13,14 and Jenzsch et al. on bioreactor mass transfer.15 The researchers recognized that in spite of the large complexity of the retrofit problems the developed methods should be systematic4 and should incorporate empirical process knowledge1 (heuristics). The authors strived to develop systematic retrofit methods for both chemical plants1,16 and processes.17-19 The other important feature of the developed retrofit frameworks is the incorporation of process knowledge in the form of expert systems20 or heuristics.1 The process heuristics are used on the chemical plant level1,21 as well as on the chemical process level.22 The developed retrofit methods identify the improved solutions based on two approaches: algorithmic solution of numerical optimization problems16 and process heuristics.2 The development of an automated, heuristic based decision making framework is presented by Roda et al.23 and Vidal et al.24 Parallel with the development of retrofit frameworks for continuous processes, frameworks for batch process retrofitting have been developed. Extensive attention was given to the deterministic retrofitting of the batch plants in the form of MINLP problem formulation. The MINLP problems * To whom correspondence is addressed. Tel.: +41-44-632-5668. Fax: +41-44-632-1189. E-mail: [email protected].

considered the optimal addition of extra equipment to the existing pool25,26 and retrofit alternatives which consider not only new equipment addition but also existing equipment pool operation mode alteration.27 Montagna28 presents an extension of the previous MINLP formulation by proposing the inclusion of intermediate storage tanks. A further extension is a capacity expansion retrofit scenario of an integrated production and distribution system which is optimized by Lee et al.29 using an MINLP model. The objective functions used in the MINLP model optimization are extended by Goel et al.30 to account for batch plant inherent reliability and maintainability characteristics of existing and new equipment. The MINLP deterministic retrofitting approach is used by Papageorgaki and Reklaitis31 to investigate the retrofit problem of a plant which accommodates changes in product demands and addition and elimination of units. The modified model proposed by Subrahmanyam et al.32 is used to address retrofit problems with regard to market uncertainty. One of the alternatives to the retrofitting problem formulation is the genetic algorithm; in this approach the retrofit problem is formulated in the form of genes (Dedieu et al.33). Inclusion of process specific heuristics in batch plant synthesis is presented, e.g., by Lee et al.34 in the context of an MINLP model optimization. We present a novel systematic indicator, heuristics, and model-based retrofit method for batch processes. The developed decision support framework considers the identification of improvement opportunities in a batch plant by considering first the product market situation. The batch plant analysis is structured on three levels: the plant, the process, and the unit operation level, respectively. The analysis of these levels is based on indicators which are of various types and specific for each analysis level. These indicators are linked to heuristics which are used to identify the retrofit actions. Finally, for each identified retrofit action the improvement potential is calculated in order to assess its importance. After the presentation of the method its application to a case study will be presented. 2. Materials and Methods—The Retrofit Method 2.1. Characterization of Batch Operation. A chemical batch plant can be decomposed into three different levels (Figure 1). The plant level (level 1) comprises all production lines and processes, the production line or process level (level 2), and

10.1021/ie070044h CCC: $40.75 © 2008 American Chemical Society Published on Web 12/04/2007

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 67

Figure 1. Interaction between the production levels: level 1 - plant level, level 2 - process level, and level 3 - unit operation level.

the unit operation level (level 3). On level 3 the dynamic behavior of each unit operation has to be taken into account. A production plant can consist of one or more production lines which produce the same or different products. These production lines can be interconnected by using shared equipment or can function as totally parallel lines. The analysis of the batch operation is regarded as a complex task and should consider all the production levels described above. There might be a strong relation between the processing levels, i.e., a prolongation of the reaction time might increase the cycle time and thus also the campaign duration if the reaction step is the time bottleneck. 2.2. The Batch Retrofit Method Description. The developed systematic batch retrofit methodology has a top-down structure and is presented in Figure 2. The batch plant retrofit is started by gathering information about the production plant layout and recipes. First of all information is collected about the market situation in order to decide upon the business case. Evaluation of the Retrofit Potential. In order to continue the retrofit analysis of the chemical batch plant the product line with the highest retrofit potential is chosen. The highest retrofit potential is estimated using the assumption that the bottleneck unit can be completely debottlenecked and the product price is known. For a cycle time reduction or increase in batch size the retrofit potential is estimated by eq 1

RPi ) Pi*∆pi ∆pi )

mi + ∆mi mi tc,i - ∆tc,i tc,i

(1) (2)

where index i refers to a certain product i, RPi is the retrofit potential [CHF/h], Pi is the product price minus the variable costs per unit of product [CHF/kg], ∆pi is the productivity increase [kg/h], mi is the batch size [kg], ∆mi is the batch size increase [kg], tc,i is the cycle time [h], and ∆tc,i is the cycle time decrease [h]. The cycle time reduction effect is shown in Figure 3. Batch process debottlenecking has to be carried out in the case of production limited sales of mono- and multiproduct batch plants. When the market limits the sales, the production debottlenecking should be carried out for multiproduct plants. By this, equipment and resources are made available for other products. The goal of batch process debottlenecking is to increase the product throughput by increasing the batch size or reducing the cycle time (eq 2). The first debottlenecking action should be that the unit operation processes are changed so that they run with improved time and volume performance. The interaction between a single unit operation, a production line, and a production plant is shown in Figure 1. This means that first the batch size limitation should

Figure 2. The systematic retrofit framework for chemical batch processes (continuous line); the dashed lines represent chemical process information sources and routes.

be tackled because volume changes might have an influence on operation times; later on the cycle time reduction should be targeted. Afterward, the equipment capacity should be increased to 100%, whenever possible. Depending on the magnitude of the retrofit goal, batch process debottlenecking can be carried out with or without equipment changes in the production line. If the equipment cannot be debottlenecked, a second vessel should be added to operate in parallel. In the case when significant capacity increase is required, throughput increase can be achieved by adding additional processing units or buffer tanks.

68 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008

Figure 3. Cycle time reduction by reducing occupancy time of the bottleneck unit.

Evaluation of the Market Situation. The evaluation of the product business situation is an important step to be carried out because it decides on the retrofit goals. From this point of view the market can be limiting or nonlimiting. The limited market means that the product cannot be sold in larger amounts than a certain quantity, while in the situation of a nonlimiting market more product could be sold to the customers. These considerations lead to different retrofit incentives: for the first business case, the same amount should be produced the cheapest and within target constraints (e.g., safety and quality), while for the second business case, production and sales should be as high as possible, priced the cheapest, and the constraints fulfilled. Process Performance Improvement. In the case of a monoproduct plant operating under market limited sales condition production increase is not meaningful; however, under these conditions, process performance improvement should be done, e.g., equipment idle time usage, cost, and energy demand reduction. This improvement step can be carried out for all products additionally to the debottlenecking efforts, as shown in Figure 2. 3. Using Indicators in the Batch Retrofit Framework The indicators used in this framework monitor the operation or the state of a system. The use of process indicators in the chemical industry is widely spread. The indicators in their nature are of two types: quantitative and qualitative. One of the most frequent quantitative examples in the chemical engineering field are the dimensionless numbers which show a relation between two phenomena. As an example for a qualitative indicator is the reaction quality1 which shows if a chemical substance has a positive or a negative impact on the reaction yield. In continuous processes these indicators show information about the system at any time, because they are time invariant. The application of the indicators which represents phenomena that are dynamic in batch reactors is more complicated because their properties vary in time. Due to this reason it is not appropriate to use only one indicator value to characterize a batch process. The most complete picture would be given by a vector of the same indicator which is calculated with a certain time interval during the batch process. Other options would be to calculate an average, a maximum, and a minimum of the indicator vector. These would provide some of the information in a more compressed way; however, the loss of information is obvious which will result in the fact that some important state evolutions during the batch are discarded. The set of indicators proposed for this batch process analysis framework are the process performance, unit operation specific, and path flow indicators (Figure 4). The overall process indicators are calculated on process level 2, see Figure 1, and contain batch process productivity related

indicators. The general unit operations indicators can be calculated for any of the batch processing units and are time and volume utilization related. Also in this category belong indicators which assess the batch to batch variability of parameters. These indicators should be used to identify if there is a correlation between the variations of sequenced units. Such an analysis can be performed only if intermediate buffers are not in the sequence. The unit operation specific group of indicators consists of static and dynamic indicators. The static indicators comprise the parameters which are constant or are measured only at the beginning and at the end of the processing, and the dynamic indicators capture process states at different moments during the batch processing. The last category of indicators is the path flow related class and is part of the path flow based process analysis method developed by Uerdingen et al.1 The information needed to calculate the process indicators has various sources: the simplest source is the process recipe which contains information that allows for the calculation of the overall process indicators. Other information sources include the plant measurement system which provides batch to batch information and is needed to calculate the variations. Plant measurements can also be used to calculate some of the dynamic unit operation indicators. The third information source is the batch process model.9 In this context the batch model is considered an information source because it uses first principles or process knowledge. At the same time it can be regarded as a tool which transforms by inference the measured process parameters (e.g., temperature, pressure, and mass) into parameters that are not available to be measured (e.g., concentrations). 3.1. Process Performance Indicators. 3.1.1. Overall Process Performance Indicators. The equipment occupancy time to,j, [h], is the sum of the time that an equipment takes to execute the tasks assigned to it

to,j )

∑i ti,j

(3)

where i enumerates the steps in equipment j, and ti,j is the time needed to perform task i in equipment j, [h]. The limiting cycle time tc, [h], of the process is defined as

tc ) max (to,j) j

(4)

A measure of how effectively a piece of equipment is used is shown by the equipment uptime uj, [-], that is defined as

uj )

to,j tc

(5)

The target result is a value of 1 for uj and this for every equipment. The maximum production rate of a product per unit time is the batch size m, [kg], divided by the cycle time (maximal temporal productivity pt), [kg/h]:

pt )

m tc

(6)

The volume productivity pv, [kg/m3], of a process is defined as follows

pv )

m Vt

(7)

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 69

Figure 4. Indicator types in the systematic batch process retrofit method.

where Vt is the total volume of all production equipment (vessels, reactors), [m3]. Combining eqs 6 and 7, ptv, [kg/m3*h], defines overall productivity with respect to time and volume:

ptv )

m tc*Vt

(8)

3.1.2. General Unit Operation Performance Indicators. The share ui,j of each task i to the equipment occupancy time to,j is calculated as follows:

ui,j )

ti,j to,j

(9)

This way the steps with the largest contribution to the equipment usage time can be determined. This indicator can help to take a decision after which task i indicates the process should be split into several equipments. Volume utilization per step Vi,j, [-], defines the relative filling degree of an equipment

Vi,j Vo,j

Vi,j )

(10)

where Vi,j is the actual volume of the contents in step i for equipment j [m3], and Vo,j is the nominal volume of equipment j, [m3]. The volume limiting equipment in a process can be identified based on eq 11 using the Vl,j indicator:

max (Vi,j) i Vl,j ) max max(Vi,j) j i

(

)

(11)

The equipment with the largest Vl,j is the volume limiting one. Without buffer tanks, the Vl,j represents the volume proportions between the different unit operations as defined by the recipe. The total equipment filling degree Vt,j is calculated as the integral of the filling degrees in all steps, and ideally it should be close to one:

Vt,j )

1 to,j

∫0t

o,j

to a reference vector. The goal of this indicator is to quantify the similarity of the same type of process profiles (e.g., temperature, pressure, level) compared to a reference trend. By this, process trend disturbances and anomalies can be identified analyzing process data over a certain period

Vi,j(t) dt

R h2 )

1 n

∑n

(

1-

∑i

(yni - yjn)2

(13)

where i is the number of points in the reference and nth vector, n is the number of profiles to be compared with the reference n profile, yref i is the ith reference profile point, yi is the ith point, and yjn is the average of the nth vector, respectively. The reference profile is a best case or nominal profile. This indicator can be used only when the batch times are the same, thus the lengths of the profile vectors are identical. However in practice this is rarely the case; therefore, the profiles with similar batch times can be compared by removing some data points (e.g., from the beginning or end of the profile) which do not carry valuable information. In the case when the batch times differ significantly the comparison should be made using the dynamic time warping technique.35 3.2. Unit Operation Specific Indicators. 3.2.1. Static and End-Point Unit Indicators. Reactor. The selectivity, initial reactant ratio, yield, conversion, and relative volatility indicators are well established in the chemical engineering literature36 and will also be used within the retrofit framework. The minimum driving force indicator Min∆Fj, [K], or [mol/m3], is related to mass or energy transfer during processing, and it shows the driving force bottleneck in a certain equipment

Min∆Fj ) min(∆Fj(t))

(14)

where ∆Fj(t) is the time variant driving force during the process. The relative mass or energy transfer coefficient indicator, Urj , [-], shows for equipment j the relative value of the overall property transfer coefficient compared to a maximum value that is specific to the system (vessel)

(12)

Another general unit operation performance indicator is the batch to batch variability of process parameters. Different indicators are introduced for scalar and vector types of process parameters; the comparison basis is a best case or a nominal value. In case of vector parameters, e.g., temperature profiles, the R h 2 indicator, [-], calculates the average deviation compared

)

∑i (yrefi - yni )2

Urj )

max (Upj (t)) Um j

(15)

where Upj (t) is the time variant heat or mass transfer coefficient during the process [W/m2K] or [kg/m2s], and Um j is the maximum achievable property transfer for a certain system and driving force [W/m2K] or [kg/m2s]. For example, the time

70 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008

variant heat transfer coefficient of a vessel for large-scale systems can be calculated based on a mass and energy balance. The heat transfer area HTA, [-], indicator compares the minimum heat transfer area with the initial heat transfer area

HTA )

min(A(t)) A0

(16)

where A(t) is the time variant transfer area during the process, [m2], and A0 is the initial transfer area when the heating/cooling is started, [m2]. The Damkoehler number, Da, [-], represents a concept for relating different physicochemical phenomena that might be ratelimiting for the overall process

Da )

r1 r2

(17) Re )

where r1 and r2 are transfer and/or reaction rates. For example the reaction rate can be compared with a mass transfer rate.37 In the case study presented below the comparison of the reaction rate with the solid dissolution rate is considered, Dad/r. Mass and Heat Transfer Related Indicators. Maximum vapor rise speed VRS, [-], indicator is used in the case when one of the reaction products is in vapor phase. It is important when reactor swelling is analyzed and might help in the relief valve calculations

VRS )

max(Vv(t)) max(Vs(t))

operation are missing; therefore, it is almost impossible to determine the model parameters in the crystallization model. In the proposed framework it is considered that the most important and most frequently used end-point indicators for a crystallization process are the particle size and distribution (PSD), morphology, and crystallinity. The most important indicators for the centrifugation and drying process were found to be the following: the particle size and distribution, morphology, purity, and humidity. 3.2.3. Scale-Up Indicators. The scale-up indicators are proposed as similarity numbers which means that the value of the indicator should be similar on both scales in order to achieve the same process operation performance. The Reynolds number, Re, is a good indicator for the liquid dynamics in mechanically stirred vessels and is formulated as follows38

(18)

where Vv(t), [m/s], and Vs(t), [m/s], are the vapor flow rates generated by reaction and calculated using a hydrodynamic model to avoid reactor content swelling, respectively. The temperature evolution indicator ∆Te, during the evaporative process, [C], highlights heat transfer problems. Often during evaporative processing the reactor temperature decreases because the rate of heat addition is lower than the heat rate taken for evaporation. This results in a decreasing temperature profile, and it can be assessed by looking at the temperature gradient. 3.2.2. Process Dynamics Indicators. Concentration at a fixed time ct [kg/m3] or time to achieve a fixed conversion tc [min], help to compare the advancement of two batch runs by looking at a fixed time for the achieved conversions or by analyzing how long it takes to obtain the same conversion value. It can also be used to identify the time needed to achieve a target concentration value. The accumulation gradient at a certain time ∆ct, [min/%], [min*m3/kg], expands the set of dynamic indicators with the accumulation or depletion gradient of a certain static performance variable (conversion, concentration) at a certain time during the process. There are a multitude of chemical reactions and work-up operations in the batch chemical industry; therefore, the number and type of relevant indicators changes from case to case. From this point of view we propose an open framework, which means that the indicators presented are either the most important ones or the most relevant for the case study and the list can be extended based on various needs. The development of a reliable large scale model for a crystallization process is a very difficult task in the chemical industries. On one side, crystallization is a complex process to be described with first principle models, and, on the other side, the appropriate large scale process measurements during regular

FND2 µ

(19)

where F is the fluid density [kg/m3], N is the rotation speed [RPM], D is the mixer diameter [m], and µ is the dynamic viscosity [Pa s]. The ratio of stirrer pumping rate to reactor volume SPR, [kg/min*m3], calculates the stirrer pumping rate in function of reaction volume, and it should be constant for different scales. This indicator is important when a mass transfer process takes place from the inside of reaction mass to the liquid surface

SPR )

Fp Vr

(20)

where Fp is the stirrer pumping rate, [kg/min], and Vr is the reactor volume, [m3]. The ratio of boil-up mass to reactor volume BUP, [kg/m3], is typical for the batch processes where separation occurs during the boilup of the reaction mixture

BUP )

Mb Vr

(21)

where Mb is the boil-up mass [kg], which represents the reaction mass boiled during operation, and it can be regarded as the integral of the boil-up rate during the process. The heavy component is refluxed, and the light component is collected after condensation in a separate vessel. This way the removal of a light component from the reaction mixture to shift the equilibrium is directly influenced by the boil-up rate. It is important to have a constant boil-up mass to reactor volume ratio during scale-up to achieve similar operation performance. The cScalet indicator, [-], relates the product concentration ratio at a certain time between two scales. It is useful to identify at which operational stage the scale-up effect is significant

cScalet )

(ct)s (ct)l

(22)

where (ct)s and (ct)l are the concentrations at fixed time t in a small and a large scale reactor, respectively. 3.3. Path Flow Based Retrofit Method Modification for the Batch Framework. The systematic retrofit method based on the path flow decomposition developed by Uerdingen et al.1 is suitable for continuous processes; however, with some modifications it can also be applied to the batch framework.

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 71

Figure 5. Batch process representation scheme in the path flow framework.

Figure 6. Path flows with different sources passing through the bottleneck unit.

The batch process specific modifications are described in the section below. 3.3.1. Batch Process Representation in the Path Flows Framework. In the adapted retrofit methodology, the material streams are averaged and considered per batch.39 In this way the batch process is regarded as a static set of sequences, and the dynamic features are discarded. In Figure 5 batch process input-output and intermediate streams are depicted schemati(c) cally. The input into a unit is the external supply si,ip in step i, (c) specified in the input column, and/or a supply by reaction si,ir (middle column). The output of the unit is the external demand (c) (c) di,op (last column, output) and/or a demand by reaction di,or . (c) The output from a unit can be also a recycle stream ri,ir. The (c) parameter fi,step indicates that a new processing step or a transfer occurs. 3.3.2. Path Flow Analysis in Bottleneck Units. In order to analyze the bottleneck units the path flow methodology can be used. Using this strategy the sources of the component path flows which contribute to the volume or time bottleneck can be identified. By eliminating, rerouting, or reducing path flows which pass through a bottleneck unit the volume and time productivity are automatically increased. The methodology is presented in Figure 6. 3.3.3. Path Flow Indicators. The path flow indicators were formulated in the context of the systematic retrofit method developed by Uerdingen et al.1 and are summarized below. Material-Value Added (MVA). The MVA indicator is only applicable to open component path flows leaving the process boundaries in demand flows. It calculates the difference between the value they represent outside the process boundaries per batch and the costs in raw material consumption they caused such as, e.g., for a solvent, fuel credit incineration minus solvent purchase cost. Energy and Waste Cost (EWC). The EWC indicator quantitatively allocates overall process costs per batch related

to utility consumption and waste treatment to a component path flow. Reaction Quality (RQ). The RQ indicator qualitatively measures the effect of a component path flow upon reactions that occur along its path. Positive RQ values indicate a positive effect on overall plant productivity, whereas negative values identify undesirably located component path flows in the process and thus highlight potential for cost savings. Accumulation Factor (AF). The AF indicator rates the accumulative behavior in recycle flows per batch and, therefore, only applies to component cycle path flows. A large accumulation factor often indicates unfavorable buildup in a cycle and can be caused by nonoptimal separation or too low of a reaction conversion. A high accumulation factor may reduce the volume productivity considerably. Total-Value Added (TVA). The TVA value finally describes the economic impact of a given component path flow per batch on the variable process costs, i.e., energy, waste, and material cost. It is calculated as the sum of the MVA and EWC values. Usually, only negative TVA values designate process improvement potentials in the process. However, a path flow with an important positive EWC value compensated by a higher MVA value yields a positive TVA value but can still show a significant energy and waste cost reduction potential. 3.4. Process Heuristics. The presented heuristics are compiled based on chemical engineering books,36,38 scientific literature,40-42 and discussions with process engineers. The process performance heuristics, Table 1, are used to find retrofit actions on batch process level 2, as defined in Figure 1. Unit operation specific heuristics presented in Table 2 are used to find retrofit actions on batch process level 3, as defined in Figure 1. The path flow heuristics presented in Table 3 are used to find retrofit actions based on the presented path flow analysis framework; however, these heuristics are activated also by the path flow analysis in the bottleneck units.

72 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 Table 1. Process Performance Heuristics

generic to,j tc

m

uj ui,j Vt,j pv pt ptv R h2

{

Merge neighboring tasks into the same equipment if equipment requirements are compatible; unless the unit is the time bottleneck Insert storage tanks between two consecutive tasks; this allows different batch sizes and cycle times up-and downstream of the storage point; consider loss of batch entity Increase concentration whenever possible Reduce, increase occupancy time in a unit, depending on whether it is a bottleneck or not Reduce cycle time; if the time bottleneck is caused by a step that has a very long cycle time, new equipment should be operated in a staggered mode based on the cycle time of the next time bottleneck; if time bottleneck is caused by equipment, sometimes it can be eliminated by moving secondary operations into a nonbottleneck unit; consider buying new equipment, check with economic criteria Increase batch size until at least one step operates at 100% usable capacity; increase the equipment capacity for the task that imposes the batch size limitation; add a second equipment to operate in phase with the first; if process operates at maximum batch size, work to reduce plant cycle time by eliminating time bottlenecks If equipment uptime is low try to increase the number of cycles per batch for that equipment; this may create opportunities for additional increases in batch size < 10% or >90% Not necessary to divide into 2 units 10-30% or 70-90% Consider dividing up into 2 units with equal tocc only in the case of the bottleneck equipment 30-70% Divide into 2 separate units and aim for equal tocc Should be close to 1; if not, check for causes. Centrifuges and dryers present exceptions because they are not operated at full volume load. Is important if the focus is on volume efficiency. Compare different design options with each other. Should stay the same or improve. Is important if the focus is on time efficiency. Compare different design options with each other. Should stay the same or improve. Compare different design options with each other. Should stay the same or improve. Identify the reason if variability is significant

}

}

{

4. The Fine Chemical Case Study In order to show the developed retrofit framework a fine chemical case study is considered. The fine chemical batch plant consists of a single batch process; therefore, the analysis is carried out on levels 2 and 3. 4.1. Fine Chemical Process Description. The production site is a monoproduct batch plant with partially parallel production lines. The processing sequence starts with the reactor (RCT) where the fine chemical is produced and continues with the crystallizer (CRY). In this step hot-drop crystallization is carried out. After this, centrifugation (CTR) is performed, and the processing is ended with the product drying (DRY). The first buffer tank can be found between the reactor and crystallizer in order to eliminate the downstream processing waiting time and to homogenize the reaction product. The two reactors operate out of phase. The second buffer can be found between the crystallizer and the centrifuge, and their role is to free up the crystallizer after crystallization is finished. With the same goal the third set of buffers can be found between the centrifuge and dryer. The reaction chemistry consists of four catalyzed equilibrium reactions in a series. The reaction scheme is as follows

As a Al

(23)

Al + B a C + D

(24)

B+CaE+D

(25)

B+EaF+D

(26)

B+FaP+D

(27)

operation is to remove byproduct D from the liquid phase as fast as possible to shift the equilibrium reactions to the product side. In order to facilitate the removal of component D the reaction mixture is boiled, and component B is condensed (CND) and refluxed (Figure 7). In a second condenser coproduct D is initially condensed and afterward not anymore and is removed by the vacuum pump (PMP). When the reaction mixture has reached a target percentage in component F, the separation of the not reacted component B is carried out. This separation process is a one-step distillation after which the separated component B is stored in the storage tank (ST1) and later reused. The next processing step is the hot-drop crystallization. During this step, cold solvent is charged, and the reaction melt is fed in the crystallizer. After the crystallization step is completed, the mass is transferred into the buffer tanks, and the crystallizer is ready to process a new charge. During the centrifugation most of the solvent is separated from the crystal mass, and during drying the solvent removal is completed. Additionally to the main processing building the plant has solvent regeneration and waste incineration facilities. The solvents with different concentrations are stored in storage tanks (ST2, ST3 and ST4) prior to regeneration or usage. A continuous operating distillation column (DIC) regenerates the solvent by increasing its concentration (top fraction). The bottom fraction is incinerated in the final separation step. During production the plant is fed with fresh solvent; therefore, the solvent accumulation is controlled by solvent incineration. The incineration is specific to the waste stream composition and is carried out in units WI1 and WI2. 5. Results and Discussions

where As and Al represent component A in the solid and the liquid phase, respectively. Raw materials are component A and B; components C, E, and F are intermediates; and P is the desired product. Raw material A (limiting component) and B (component in excess) are charged at the beginning of the reaction step, and the reaction mixture is heated up to the set point. The reaction starts when the catalyst is dosed as solution. In the first stage of the process operation, until the complete dissolution of component A, the reactor system consists of three phases: solid, liquid, and gas. The main goal of the process

5.1. Evaluation of the Market Situation and the Retrofit Potential. After the evaluation of the product business case it was concluded that the market is not limiting and more final product could be sold. This means that the primary retrofit goal of this fine chemical case study is the production capacity expansion. The retrofit potential from the capacity expansion point of view is calculated by assuming that the current productivity bottleneck is eliminated. The retrofit potential is estimated to be a 43% plant productivity increase. In this section the indicators described in the framework part are calculated and discussed for the presented case study.

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 73 Table 2. Unit Specific Heuristics indicator reactor initial and end point indicators selectivity

initial reactant ratio yield conversion relative volatility Min∆Fj Urj Dard VRS HTA ∆Te dynamic indicators ct, tc ∆ct scale-up indicators Re SPR BUP cScalet crystallizer particle size and distribution

morphology crystalinity yield, purity

heuristics

In order to increase the selectivity for an intermediate component an optimal temperature or feeding profile should be calculated, depending on the reaction scheme40 (difference in activation energy or reaction order) If unwanted products are formed in concentrations below 1-3%, then check for thermal decomposition, decrease operating temperature or reaction time, and check the catalyst selectivity Optimize ratio with regard to selectivity, yield, reaction rate, and separation difficulty Maximize yield by minimizing side reactions Maximize conversion During reaction and separation steps, separate components using distillation when relative volatility is large enough Increase driving force by using utilities with higher/lower temperature than currently used Maximize the transfer coefficient; if low improve heat transfer conditions: stirrer type, rotation number, baffles If Dard ∼ 0 then increase dissolution by higher mixing intenisty, better surface to volume ratio (smaller particles); if Dard . 1 then increase reaction rate, e.g., by higher temperature Optimize reaction rate under swelling constraint, using temperature and pressure as control variables in optimal control strategy Increase heat transfer area by considering direct heating or an additional heat exchanger in an external loop Increase heat input, increase driving force and area, and improve heat transfer by means of mixing It is useful to compare reaction evolutions (batch to batch) and to identify target values It is useful to compare the static indicator gradient during reaction evolutions (batch to batch) and to identify target values Ensure similar process conditions during scale up and identify target values Mixing conditions can be altered by the stirrer type, speed, and baffles Choose stirrer accordingly and adjust stirrer rotation speed Supply enough energy to keep the ratio constant with respect to scale-up; consider adding external heat exchanger or direct heating Try to obtain the same performance indicators at the same time on different scales. If the large scale process is significantly slower check if the reaction rates are similar or reverse reactions are significant material for filtration (goal: large uniform particles) material for milling (goal: small particles) fine powder for reaction (goal: small particles) material as final sales form free-flowing/low dusting crystals fine, uniform powder needles and plates compact crystals crystalline amorphous pure crystals (no incorporation of by products)

5.2. Calculated Indicators. 5.2.1. Process Performance Indicators. Overall Process Indicators. The results of the overall process indicators are presented in Table 4. In the case of a buffered plant structure the production unit operations are disconnected during processing. The pv indicator shows that the best volume efficiencies are reached in the reactor and crystallizer. By analyzing the pt value calculated for the mass of two reactor charges and one independent production line (2 reactors, 2 crystallizers, 1 centrifuge, and 1 dryer) it is concluded that the dryer is the productivity bottleneck for both maximum and average productivity. The maximum productivity is calculated based on the shortest production time ever, and the average productivity was calculated based on a 2-month time span. The volume limiting equipment and the equipment with the highest up-time is the dryer. In Table 5 the general unit operation indicators are presented. In the case of the reactor it can be noticed that most of the time is taken by the reaction and component B separation step and the equipment is fully used. In the case of crystallization significant time is taken by the heating and cooling procedures,

seeding, controlled supersaturation fast precipitation fast precipitation seeding, controlled supersaturation fast, controlled precipitation suitable seed suitable seed suitable seed fast precipitation seeding, slow crystallization

and the crystallization step itself accounts only for 10% of the time. In the centrifugation most of the time takes the process itself; however, the unit is not completely filled, and an investigation might show that it is possible to process more material in one run without a loss of centrifugation performance. The same comment is valid for the dryer as well, and it should be attempted to charge more material. The drying time takes 50% of the total time, and the cooling time has again a significant contribution. In order to increase the plant productivity the cooling speed might be improved. In Table 6 the process variability indicators are calculated, and in Figures 8 and 9 the calculation of the R h 2 indicator is presented for the filling degree in a tank for the case of identical batch times. The unit operation variability indicators show that the reactor operation is performed with low variability. In the crystallizer the processing time variability is somewhat higher in spite of the fact that the feed mass variability is low. The centrifuge presents low processing time variation. The processing time variation is the highest in the dryer. It can be noticed that the dryer jacket temperature has large variations, while the inside temperature shows a somewhat lower variation; however, it is

74 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 Table 3. Path Flow Specific Heuristics component paths

generic retrofit action

all categories reduce the specific energy consumption of the path Category 1: Open Component Path, RQ < ) 0 1 remove/reduce the open path flow rate at the source 2 reroute (partially) the open path in the process 3 if RQ ) 0, then recycle the open path flow only when the recycling yields a positive effect on productivity 4 replace the open path component with a better component 5 increase the specific value of product price of the open path Category 2: Open Component Path, RQ > 0 1 optimize/reduce the open path flow rate 2 recycle (partially) the open path to the process 3 increase the specific product price value of the open path Category 3: Cycle Component Path, RQ < ) 0, AF > 1 1 remove/reduce the cycle path at the source 2 change from cycle to open path 3 reduce the accumulation factor of the cycle path 4 replace the cycle path component with a better component 5 reroute (partially) the cycle path in the process Category 4: Cycle Component Path, RQ > 0 1 Optimize the cycle path flow rate 2 reroute (partially) the cycle path in the process 3 reduce number of steps in cycle

not known if the dryer jacket temperature is the cause of the problem, and it needs investigation. The product quality deviation is acceptable. Another straightforward plant productivity increase could be achieved by decreasing the processing time variation of the dryer. 5.2.2. Unit Operation Specific Indicators. In Table 8 the reactor specific indicators are presented. It can be seen that the selectivity is not 100% due to side reactions and thermal decomposition. Due to the side reactions and reaction intermediate F the yield is around 95%. The relative volatilities during the reaction and separation step are large enough to ensure an efficient separation. The minimum driving force is low; however, the transfer coefficient is relatively high, around 50% of the maximum value, and it shows improvement potential. The Damkoehler number defined for this case study was calculated for small and large scale at the beginning of the reaction when solubilization still takes place, and it shows that the solubilization and reaction rates are approximately equal. Based on the maximum vapor rise speed indicator we can conclude that the process currently is operated near swelling conditions; however, there is still some improvement possibility. The heat transfer area indicator shows a decrease of the heat transfer area with 40% between the beginning of the reaction and the end of distillation. This fact combined with the mixture temperature rise during the distillation step and, implicitly, with the minimum driving force value highlights a serious heat transfer problem and a retrofit potential. The temperature evolution indicator during the evaporation of component D (beginning of reaction) shows a negative value which means that the transferred heat rate is not enough to keep the reactor temperature constant. Based on the dynamic indicators calculated in Table 9 we conclude that the overall reaction rate decreases with time (the time to obtain 1% more conversion increases). The scale-up indicators are presented in Table 10. The scaleup effect increases with increasing conversion. The calculated Reynolds number shows different mixing dynamics, and the stirrer pumping rate on small and large scale is almost equal. This way the component D transfer to the reaction mixture surface by stirring is similar. The boil-up ratio

could not be compared due to lack of data with the boil-up ratio on other scales; however, it is an important indicator and should be considered during scale-up calculations. 5.2.3. The Path Flow Indicators. The fine chemical batch process is presented in the form of a graph based on the work of Uerdingen et al.,1 Figure 7, and Table 11 shows the results of the path flow decomposition and assessment procedure. The path flows are sorted according to the ascending TVA values in each path flow category; therefore, path flows with highest cost impact appear at the top of each category’s list. The calculation procedure used is presented in the Appendix section of the work developed by Uerdingen et al.1 In the first two categories of Table 11 the open path flows are presented. Value to the raw materials is added in path flows O15, O16, and O21, which represent the product flow. The highest value added carries the main product P; however, some of the reaction intermediate (component F) and raw material (component B) is sold in the form of tolerated impurity. Improvement potential is shown by paths O17 and O18 which show that some component B is lost during processing (O17). The catalyst is not separated after the reaction step and is lost during the washing procedure in the centrifuge; therefore, path flow O18 shows an improvement potential regarding the separation of the catalyst. The zero value of the MVA coefficient of the solvent D path flows is due to the fact that handling costs for the used solvent are compensated by the energy costs gained during incineration. The energy related costs are small compared to the value added; this fact is typical for specialty batch plants. The path flows with relative significant energy costs all have in common high flow rates and pass through the distillation column which is used for solvent reconcentration. It is known that distillation is an energy intensive process; therefore, the obtained EC values correlate accordingly. One of the exceptions from this observation is the main product path O16 which does not pass through the distillation unit (DIC) and still has a relatively high-energy demand value due to the high flow rate and many processing units. The other exception is the open flow O19 which shows that the heating up of a large amount of component B takes a significant amount of energy. The most important message which is carried by the RQ indicator is that addition by any means of component D and solvent S2 in the reactor must be avoided. The second conclusion is that component D has to be separated as soon as possible from the reaction mixture. It is concluded that the path flow based batch process analysis methodology clearly highlights the high value of the product stream and the low-energy cost values. 5.3. Retrofit Actions. The retrofit actions are generated systematically one-by-one by relating the indicator values to the list of heuristics, and this procedure will be demonstrated below. The process analysis discusses the capacity debottlenecking and path flows based retrofit actions separately; later on the focus is on the path flows which pass through the bottleneck unit. Based on the assessment results for the fine chemical batch process, this section discusses, in order of descending improvement potential, which retrofit actions are found to be applicable. In order to increase the plant throughput the overall process productivity debottlenecking actions are proposed, see Table 12. The proposed retrofit actions tackle the first productivity bottleneck, namely the dryer and the realization of these actions might increase the plant productivity between 13 and 43%. The first productivity bottleneck is considered for the unit with the lowest productivity of all units. The first action in this class is the addition of a second dryer or a larger one. It is considered

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 75

Figure 7. Process graph of the fine chemical process (bold lines represent the product stream). Table 4. Calculated Overall Process Indicators

unit reactor crystallizer centrifuge dryer

two lines average pt ptv [kg/min] pv [kg/m3* min] [kg/m3] maximum average 2070 2070 598 1257

37.12 40.96 50.56 32.96

32.64 32 42.56 22.4

2.6 3.5 40 4.1

uj [-]

Vl,j [-]

Vt,j [-]

0.78 0.67 0.51 1

0.99 1 0.53 0.42

0.83 0.65 0.42 0.37

that the drying time remains the same. By this the dryer productivity is doubled, from 7 to 14 kg/min; however, now the process productivity bottleneck would be the crystallizer (10 kg/min) and closely followed by the reactor (10.2 kg/min). Considering these facts, the retrofit potential was calculated for a plant throughput increase from 7 kg/min to 10 kg/min. The second retrofit action has a retrofit potential of around 30% productivity increase and implies the separation of the drying and cooling processes performed in the dryer. The third retrofit action is suggested by the heuristics for the variability indicator R h 2 and considers the introduction of an alarm which signals

the termination of a dryer batch. It is concluded that if the drying time variation is reduced by half, then a productivity increase of 15% can be achieved without any major investment. Similar retrofit potential is shown by the fourth retrofit action which indicates the reduction of the cooling utility temperature; therefore, the dryer mass is cooled faster. This retrofit action is concluded based on the heuristics for the task share indicator ui,j calculated for the dryer; the implementation of this retrofit action implies the change of the cooling media and cooling capacity. Retrofit actions number 5 and 6 result from the path flow analysis in the production bottleneck unit. In the dryer the drying process itself is the bottleneck; therefore, the humidity sources should be minimized. Humidity in product P is present in the form of component D and solvent S2, and reduction of any of these sources would reduce the drying time. In Table 13 the retrofit actions based on the path flow analysis are presented. The highest retrofit potential is shown by reduction of the loss of product B during centrifugation. This retrofit action is suggested by the heuristics specific for the open

76 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 Table 5. Calculated General Unit Operation Indicators, for a 100 kg Batch task

mass [kg]

ui,j [%]

Vi,j [%]

7.6 83.6

3 5 5 1 1 41 44 2

5 98 98 99 99 90 69

4 55 9 19 9 4

51 51 96 96 100 100

4.7 5.9 2.7

7 2 3 88

19 42 53 53

47.1 -3.3 -43.7

9 53 38

42 42 39

Reactor component A dosing component B dosing heating catalyst dosing heating reaction distillation transfer

1.2 -6.9 -19.4 64.7

Crystallizer charge solvent heat up to dissolve component P and cool down charge component P melt cool down dosing solvent transfer to next buffer

72.4 63.5 4.8

Centrifuge charge wash 1 wash 2 centrifugation and transfer to next buffer Dryer vacuum probe and charge drying cooling down and transfer

Figure 9. Comparison of the level profiles (dots) with the reference profile (line). Table 7. Calculated Process Variability for Scalar Variables scalar variables minimum processing time [min]

processing time deviation [%]

2288

10.0

2000

28

product P feed minimum av batcha chargea [kg] [kg]

SD [%]

Reactor

Table 6. Calculated Process Variability for Vector Variables

Crystallizer 60.6

62.1

2.3

vector variables R h2 reactor crystallizer dryer

temperature

pressure

0.61 inside temperature 0.87 inside temperature 0.75

0.92

receiver component D 0.92 jacket temperature 0.80 jacket temperature 0.52

path flows with a positive reaction quality indicator. The retrofit action suggest that the separation of component B should be improved; however, this impacts the productivity because the last component B fractions are the most difficult to separate (most time demanding). The improvement potential of this action is calculated as the improvement gain compared to the product P path flow MVA value. The catalyst retrieval retrofit action (action number 8) is proposed by the open path flow with positive reaction quality indicator heuristics. Its effect is evaluated in the same manner as for action 7; however, this retrofit action would require another unit operation which would be used for the catalyst retrieval. The benefit of the implementation of such a unit operation has to be balanced against the investment costs.

Figure 8. Filling degree in an accumulation vessel.

scalar variables

receiver component B 0.95

minimum processing time [min]

processing time deviation [%]

13.33*60.8 ) 811

19

1.33*960 ) 1277

43

a

product quality av batch specification [kg] SD [%]

Centrifuge Dryer 97.2

97.5

0.3

Calculated for a 100 kg batch basis.

The unit operations retrofit actions are presented in Table 14 and do not affect directly the plant throughput because these units are not the productivity bottlenecks at this moment; however, these can be used to improve other unit operations with regard to some other criteria, e.g., product quality, energy cost decrease. Sometimes the improvement of a mentioned criterion implies the processing time reduction and implicitly the productivity increase. For this case study an attempt to improve the product quality in the reactor is made by reducing the reaction time. With reaction time decrease the side reactions are decreased, and the productivity is increased. In Table 14 the indicators used for assessment are presented together with the process variables which are affected. These indicators affect the batch processing level 3 as defined in Figure 1. Additionally the potential unit operation productivity increase due to the retrofit action is estimated; however, this productivity does not affect the overall plant throughput. The estimation of the retrofit potentials of the unit operations requires detailed dynamic models, which often are not available or difficult to develop. As a consequence not all of the retrofit potentials could be estimated. In the case of the reactor almost all retrofit actions impact the product quality by way of a shorter reaction time. This is due to the fact that the product decomposes during the reaction, and it should be attempted to minimize the reaction time. By reducing the reaction time the reactor productivity is increased which is considered as a potential benefit due to the

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 77 Table 8. Calculated Static and End Point Indicators reactor raw materials selectivity [%] A B 95 a

98

initial A:B ratioa

yield [%]

conversion of A [%]

1:5

95

100

relative volatility [-] b c > 500

b

Min∆Fj [K]

uj,r [-]

6

0.5

> 900

Dad/r [-] scale 1 kg 6t ∼1

∼1

VRS [-]

HTA [-]

∆Tevap [C]

0.8

0.57

-10 C

c

For stoichiometry 1:4. Components D:B, during reaction. Components B:P, during distillation.

problem. The kinetic, hydrodynamic models and the optimization results will be presented in a separate publication. Retrofit action number 11, which is proposed based on the volume utilization per step indicator heuristics, suggests that the reaction rate can be increased by adding more component B in the reactor in the semibatch mode, after the reaction volume has shrunk. The retrofit potential of this action was evaluated using the reactor model presented by Simon et al.,9 and it was found that 10% of the reaction time decrease is possible. The reaction conversion increase is also possible by removing component D from the liquid phase with a higher rate. This is possible by using a falling film evaporator which is fed with a liquid-phase reactor content. This retrofit action is proposed by analyzing the concentration at a certain time (ct), and indicator heuristics assumes that component D is separated 100% from the rest of the components in the falling film evaporator. The connection between the two units is made at the bottom of the reactor where the accumulation of product D is highest. The effect of this retrofit action is in function of the pump-around liquid streamflow rate. Intuitively one might consider that there is an optimum between the pump-around flow rate, implicitly the size of the falling film evaporator, and the reactor productivity increase. The retrofit action was evaluated by extending the hybrid-model presented by Simon et al.9 with a removal term

Table 9. Calculated Dynamic Indicators product concn [%]

ct [min]

∆ct [min/%]

25 50 70

2279 2656 2983

3.8 6.3 25

Table 10. Calculated Scale-Up Indicators Re

scale [kg]

initial

final

1 30 6300

7500 50000 660000

1500 11000 150000

SPR [kg /min*m3]

BUP [kg/m3]

cScalet [-]

0.22

0.4 0.5 0.6

93696 82320

fact that after the dryer debottlenecking the reactor is the bottleneck. Retrofit action number 9 based on the heuristics list suggests the separation of the reaction and distillation; by this a reactor productivity increase of 30% would be achieved. By implementing the retrofit action number 10 an optimal reactor temperature was calculated with a time decrease of 20% for the reaction time.43,44 This retrofit action is recommended by the heuristics formulated for the vapor rise speed (VRS) indicator. In this optimal control problem the goal was to operate at the highest possible temperature, in order to allow the fastest conversion, however without causing reactor content swelling. The reactor swelling acts as a constraint in the optimization Table 11. Path Flow Indicators Calculated for Fine Chemical Case Study

flow [kg/ batch]

MVA [CHF/ batch]

EC [CHF/ batch]

WC [CHF/ batch]

EWC [CHF/ batch]

TVA [CHF/ batch]

RQ [-]

Open Path Flows, RQ < 0 S2 467 ST3 dWI2,op D 5213 D 1488 dWI1,op S2 106 dWI1,op D 118 S2 93 S2 618 D 32 D 32 S2 141 dWI2,op S2 35 S2 122 dWI2,op S2 10 ST3 dWI2,op S2 10 F 378 P 12640

0 0 0 0 0 0 0 0 0 0 0 0 0 0 864 46976

38 112 51 10 3 10 13 13 0 3 3 3 3 0 6 192

99 0 0 22 26 19 0 0 6 0 0 0 0 0 0 0

138 112 51 32 29 29 13 0 6 3 3 3 3 0 6 192

-138 -112 -51 -32 -29 -29 -13 -13 -6 -3 -1 -3 -1 0 858 46784

-1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1 -1

Open Path Flows, RQ > 0 dWI1,op B 314 dWI1,op catalyst 48 B 12042 A 1526 B 64

-627 -496 0 0 128

83 3 90 13 3

67 10 0 0 0

150 13 90 13 3

-778 -509 -90 -13 93

1 1 1 1 1

component O1 O2 O3 O4 O5 O6 O7 O8 O9 O10 O11 O12 O13 O14 O15 O16

sCRY,ip sST4,ip sRCT,ir sST4,ip sST4,ip sCTR,ip sCRY,ip sRCT,ir sRCT,ir sST4,ip sCRY,ip sCTR,ip sCTR,ip sST4,ip sRCT,ir sRCT,ir

CTR CRY ST2 CRY CRY CTR CTR PMP ST2 CRY CTR DRY CTR CRY CRY CRY

ST2 CTR DIC CTR CTR ST2 dDRY,op,vapor dPMP,op DIC CTR ST2 dDRY,op,vapor ST2 CTR CTR CTR

O17 O18 O19 O20 O21

sRCT,ip sRCT,ip sRCT,ip sRCT,ip sRCT,ip

CRY CRY dRCT,or dRCT,or CRY

C1 C2 C3 C4 C5

CTR CTR CTR ST4 CTR

C6 C7

RKT RKT

DIC ST2 ST3 ST2 ST2 DIC

dWI1,op DIC dWI2,op DIC DIC dWI1,op

dWI1,op dDRY,op,vapor DIC

ST3

DIC ST2 dDRY,op,solid dDRY,op,solid

ST3 DIC

CTR CTR

ST2 ST2

DIC DIC

CTR

dDRY,op,solid

ST2 ST2 ST2 CRY ST2

DIC DIC DIC DRY DIC

ST3 ST3 ST3 ST4 ST3

CND ST1

RKT

CRY CTR CRY CTR

Cycle Path Flows, RQ < 0 CTR D D CTR S2 D S2

9523 6342 838 960 64

0 0 0 0 0

272 138 67 10 3

0 0 0 0 0

272 138 67 10 3

-272 -138 -67 -10 -3

-1 -1 -1 -1 -1

Cycle Path Flows, RQ > 0 B B

3910 4800

0 0

26 19

0 0

26 19

-26 -19

1 1

AF [-]

1.8 1.2 0.5 0.0 0.0 1 1

78 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 Table 12. Overall Process Productivity Debottlenecking and Path Flow Retrofit Actions

retrofit actions and ranking add a second dryer or a larger onea divide drying and cooling in separate units install an alarm to show the end point of drying decrease utility temperature

1. 2. 3. 4. 5. 6. a

indicator used for assessment uj ui,j R h2

implementation difficulty

effect on time, productivity, equipment costs productivity, equipment costs time variations, productivity, equipment costs time, productivity, utility costs (brine)

ui,j

Path Flow Analysis in the Bottleneck Unit uj time, productivity time, productivity uj

eliminate/reduce solvent S2 source eliminate/reduce solvent D source

estimated productivity retrofit potential [%]

high high low

43 30 15

medium

13

low low

By this action the production bottleneck is new in the crystallizer and reactor.

Table 13. Path Flow Assessment Based Retrofit Actions

7. 8.

retrofit actions and ranking

indicator used for assessment

separate more component B retrieve catalyst

TVA TVA

effect on

implementation difficulty

estimated monetary retrofit potential/ batch [%]

distillation productivity investment cost

medium high

1.7 1.1

Table 14. Unit Operation Improvement Retrofit Actions

retrofit actions 9. 10. 11. 12. 13. 14. 15. 16. 17. 18.

divide reaction and distillation of component B into separate units implement the optimal reactor temperature profile add component B when reactor volume shrinks add a falling film evaporator optimized reaction-distillation sequence with regard to productivity change steam utility (higher temperature) add an external heat transfer unit or internal heating coils add an external heat transfer unit or internal heating coils for scale-up add an external heat transfer unit or internal heating coils install stirrer with a higher pumping rate

indicator used for assessment ui,j

effect on Reactor productivity, equipment costs

VRS

product quality, control system cost product quality, energy cost product quality, equipment cost

Vi,j ct pt

implementation difficulty

estimated productivity retrofit potential [%]

high

30

medium

20

low high low

10 9 0 already optimal

∆Fj HTA

product quality, utility cost product quality, equipment cost

high high

∆Te

product quality, equipment cost

high

BUP

product quality, equipment cost

high

SPV

product quality

low

Note: To prolong the reactor operation time is not an option because longer reaction time decreases product quality. 19.

divide disolution of P in solvent and crystallization into separate units

ui,j

Crystallizer productivity, equipment costs

which is a function of the component D mass fraction in liquid phase and the loop flow rate.45 The loop flow rate was specified by the industrial partner, and it was found that, based on the assumptions presented above, the reaction conversion time can be shortened by around 9%. The retrofit action number 13 suggests the reactor-distillation sequence optimization. In this problem the control variable is the excess of component B. This has two adverse effects related to the productivity; therefore, the tradeoff and the need of optimization is obvious. On the one hand, more excess of component B decreases the reaction time, and, on the other hand, more excess of component B increases the separation time; therefore, an attempt was made to find the solution by considering the processes together. For this a dynamic reaction model was used together with a dynamic distillation model, and it was found that the current process settings are optimal.46 This is not surprising, since the plant has been operating during the last 30 years, and it was continuously optimized. The retrofit actions 14-16 are proposed by analyzing the specific indicator heuristics, and it is suggested that temperature increase during reaction should be higher; therefore, heat transfer related improvements are needed. The lack of an adequate energy model for the reactor did not allow the evaluation of the improvement

high

30

potential of these retrofit actions. Retrofit action number 17 is related to the reaction scale-up, and it indicates that an external heat transfer unit has to be used in order to minimize the scaleup effect. The retrofit action number 18 works on the transfer of component D from the liquid phase into the vapor phase and by this on the back reactions. It is known that stirrers produce a liquid pumping effect, similarly to pumps;38 therefore, it is considered that a stirrer with a higher pumping rate would increase the transport of component D from the bottom of the reactor to the reactor surface. The evaluation of this retrofit action was not possible due to a missing accurate mixing model. Based on retrofit action number 15 it was found that the crystallizer productivity could be increased by 30% if the dissolution of component P would be carried out in a separate vessel. This retrofit action has an important impact on the plant productivity after the dryer is debottlenecked. 6. Conclusions This work presents a systematic indicator, heuristics, and process model based framework for retrofitting chemical batch processes. The developed framework considers the identification of improvement opportunities in a batch plant by considering

Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 79

first the product market situation. The batch plant analysis is structured on three levels: the plant, the process, and the unit operation level, respectively. The analysis of these levels is based on indicators which are of various types and specific for each analysis level. These indicators are linked to heuristics which are used to identify the retrofit actions. Finally, for each identified retrofit action the improvement potential is calculated in order to assess its importance. The developed method was successfully applied to a fine chemical batch production facility, and it was able to identify in a systematic way several retrofit actions with significant improvement potential. The benefits of the proposed framework are that it guides the decision maker in a systematic manner through the process analysis toward the generation of retrofit actions. Thus, the method is capable of identifying improvement opportunities which are not straightforward and might be neglected (e.g., batch reactor operation optimization with regard to swelling). Future work will investigate the possibility of transforming identified retrofit potentials directly into monetary units in order to have a common basis to compare and rank all retrofit alternatives and their potential. However, to estimate, e.g., what is the Reynolds number improvement effect money wise, very detailed models are needed. It is also interesting to develop the method in the direction of a decision support system as a tool for process designers. The engineers should be able to use it for what-if scenarios analysis and/or sensitivity analysis in order to justify their decision. Furthermore, it will be investigated which extensions are needed to be made in order to improve the generalization capabilities of the method by analyzing other batch processes. Nomenclature (ct)l ) concentration at fixed time t in a large reactor [mol/m3] (ct)s ) concentration at fixed time t in a small scale reactor [mol/m3] A(t) ) transfer area during the process [m2] A0 ) transfer area when the heating/cooling is started [m2] BUP ) ratio of boil-up mass to reactor volume [kg/m3] cScalet ) relates the product concentration ratio at a certain time between two scales [-] ct ) concentration at a fixed time [mol/m3] D ) mixer diameter [m] Da ) Damkoehler number [-] (c) di,op ) external demand [kg/h] (c) di,or ) demand by reaction [kg/h] (c) fi,step ) transfer stream [kg/h] Fp ) stirrer pumping rate, [kg/min] HTA ) compares the minimum heat transfer area with the initial heat transfer area [-] M ) batch size [kg] Mb ) boil-up mass [kg] n ) number of profiles to be compared with the reference profile [-] N ) rotation speed [RPM] nD ) coproduct D mole number in liquid phase [mol] NS ) number of stages [-] Nµf ) dimensionless viscosity number [-] Pi ) product price minus the variable costs per unit of product [CHF/kg] pt ) maximal temporal productivity [kg/h] pv ) volume productivity [kg/m3] R h 2 ) average deviation compared to a reference vector [-] Re ) Reynolds number [-]

(c) ) recycle stream [kg/h] ri,ir RPi ) retrofit potential [CHF/h] (c) ) external supply [kg/h] si,ip (c) si,ir ) supply by reaction [kg/h] SPR ) stirrer pumping rate to reactor volume [kg/(min m3)] T ) reactor temperature [K] t )time [s] tc ) limiting cycle time [h] tc ) time to achieve a fixed conversion [min] tc,i ) cycle time [h] tf ) final time [h] ti,j ) time needed to perform task i in equipment j [h] ui,j ) share of each task to the equipment occupancy time [-] uj ) equipment uptime [-] 2 Um j ) maximum achievable property transfer [W/m K] or [kg/ m2s] Upj (t) ) heat or mass transfer coefficient during the process [W/m2K], [kg/m2s] r Uj ) relative mass or energy transfer coefficient indicator [-] V ) reactor mass volume [m3] Vi,j ) volume utilization per step [-] Vi,j ) actual volume of contents in step i for equipment j [m3] Vo,j ) nominal volume of equipment j [m3] VRS ) maximum vapor rise speed [-] Vs(t) ) vapor flow rate calculated using a hydrodynamic model [m/s] Vt ) total volume of all production equipment (vessels, reactors) [m3] Vt,j ) total equipment filling degree [-] yref i ) ith reference profile point [-] yni ) ith point of the nth vector [-]

Greek symbols 1/∆ct ) accumulation gradient at a certain time [min/%], [min m3/kg] ∆Fj(t) ) driving force during the process [-] ∆mi ) batch size increase [kg] ∆pi ) productivity increase [kg/h] ∆tc,i ) cycle time [h] ∆Te ) temperature evolution indicator during evaporative process [K] AbbreViations AF ) accumulation factor [-] EC ) energy cost [CHF/batch] WC ) waste cost [CHF/batch] EWC ) energy and waste cost [CHF/batch] MVA ) material-value added [CHF/batch] MINLP ) mixed-integer nonlinear programming [-] RQ ) reaction quality [-] TVA ) total-value added [CHF/batch] Literature Cited (1) Uerdingen, E.; Fischer, U.; Hungerbu¨hler, K.; Gani, R. Screening for profitable retrofit options of chemical processes: A new method. AIChE J. 2003, 49 (9), 2400. (2) Uerdingen, E.; Fischer, U.; Gani, R.; Hungerbu¨hler, K. A New Retrofit Design Methodology for Identifying, Developing, and Evaluating Retrofit Projects for Cost-Efficiency Improvements in Continuous Chemical Processes. Ind. Eng. Chem. Res. 2005, 44 (6), 1842. (3) Rapoport, H.; Lavie, R.; Kehat, E. Retrofit design of new units into an existing plant: Case study: Adding new units to an aromatics plant. Comput. Chem. Eng. 1994, 18 (8), 743. (4) Halim, I.; Srinivasan, R. Systematic Waste Minimization in Chemical Processes. 3. Batch Operations. Ind. Eng. Chem. Res. 2006, 45 (13), 4693.

80 Ind. Eng. Chem. Res., Vol. 47, No. 1, 2008 (5) Fisher, W. R.; Doherty, M. F.; Douglas, J. M. Screening of process retrofit alternatives. Ind. Eng. Chem. Res. 1987, 26 (11), 2195. (6) Halasz, L.; Nagy, A. B.; Ivicz, T.; Friedler, F.; Fan, L. T. Optimal retrofit design and operation of the steam-supply system of a chemical complex. Appl. Therm. Eng. 2002, 22 (8), 939. (7) Liu, J.; Fan, L. T.; Seib, P.; Friedler, F.; Bertok, B. Holistic Approach to Process Retrofitting: Application to Downstream Process for Biochemical Production of Organics. Ind. Eng. Chem. Res. 2006, 45 (12), 4200. (8) Ku¨ru¨m, S.; Heinzle, E.; Hungerbu¨hler, K. Plant optimisation by retrofitting using a hierarchical method: Entrainer selection, recycling and heat integration. J. Chem. Technol. Biotechnol. 1997, 70 (1), 29. (9) Simon, L. L.; Fischer, U.; Hungerbu¨hler, K. Modeling of a ThreePhase Industrial Batch Reactor Using a Hybrid First-Principles NeuralNetwork Model. Ind. Eng. Chem. Res. 2006, 45 (21), 7336. (10) Ciric, A. R.; Floudas, C. A. A retrofit approach for heat exchanger networks. Comput. Chem. Eng. 1989, 13 (6), 703. (11) Jo¨dicke, G.; Fischer, U.; Hungerbu¨hler, K. Wastewater reuse: a new approach to screen for designs with minimal total costs. Comput. Chem. Eng. 2001, 25 (2-3), 203. (12) Guntern, C.; Keller, A. H.; Hungerbu¨hler, K. Economic Optimization of an Industrial Semibatch Reactor Applying Dynamic Programming. Ind. Eng. Chem. Res. 1998, 37 (10), 4017. (13) Fonyo, Z.; Kurum, S.; Rippin, D. W. T. Process development for waste minimization: The retrofitting problem. Comput. Chem. Eng. 1994, 18, S591. (14) Dantus, M. M.; High, K. A. Economic Evaluation for the Retrofit of Chemical Processes through Waste Minimization and Process Integration. Ind. Eng. Chem. Res. 1996, 35 (12), 4566. (15) Jenzsch, M.; Lange, M.; Ba¨r, J.; Rahfeld, J. U.; Lu¨bbert, A. Bioreactor retrofitting to avoid aeration with oxygen in Pichia Pastoris cultivation processes for recombinant protein production. Chem. Eng. Res. Des. 2004, 82 (A9), 1144. (16) Harsh, M. G.; Saderne, P.; Biegler, L. T. A mixed integer flowsheet optimization strategy for process retrofits-the debottlenecking problem. Comput. Chem. Eng. 1989, 13 (8), 947. (17) Jegla, Z.; Kohoutek, J.; Stehlik, P. Global algorithm for systematic retrofit of tubular process furnaces. Appl. Therm. Eng. 2003, 23 (14), 1797. (18) Marcoulaki, E. C.; Kokossis, A. C. On the development of novel chemicals using a systematic optimisation approach. Part II. Solvent design. Chem. Eng. Sci. 2000, 55 (13), 2547. (19) Ben-Guang, R.; Fang-Yu, H.; Kraslawski, A.; Nystro¨m, L. Study on the Methodology for Retrofitting Chemical Processes. Chem. Eng. Technol. 2000, 23 (6), 479. (20) Crowe, C.; Douglas, P. L.; Glasgow, J. S. K. M. Development of an Expert System for Process Synthesis. Chem. Eng. Res. Des. 1992, 70 (2), 110. (21) Kudva, G.; Elkamel, A.; Pekny, J. F.; Reklaitis, G. V. Heuristic algorithm for scheduling batch and semi-continuous plants with production deadlines, intermediate storage limitations and equipment changeover costs. Comput. Chem. Eng. 1994, 18 (9), 859. (22) Jain, S.; Moharir, A. S.; Li, P.; Wozny, G. Heuristic design of pressure swing adsorption: a preliminary study. Sep. Purif. Technol. 2003, 33 (1), 25. (23) Roda, I. R.; Sanchez-Marre, M.; Comas, J.; Cortes, U.; Poch, M. Development of a case-based system for the supervision of an activated sludge process. EnViron. Technol. 2001, 22 (4), 477. (24) Vidal, N.; Banares-Alcantara, R.; Rodriguez-Roda, I.; Poch, M. Design of wastewater treatment plants using a conceptual design methodology. Ind. Eng. Chem. Res. 2002, 41 (20), 4993. (25) Vaselenak, J. A. G. I. E.; Westerberg, A. W. Optimal Retrofit Design of Multiproduct Batch Plants. Ind. Eng. Chem. Res. 1987, 26, 718. (26) Fletcher, R.; Hall, J. A. J.; John, W. R. Flexible retrofit design of multiproduct batch plants. Comput. Chem. Eng. 1991, 15 (12), 843. (27) Yoo, D. J.; Lee, H.-K.; Ryu, J.-H.; Lee, I.-B. Generalized retrofit design of multiproduct batch plants. Comput. Chem. Eng. 1999, 23 (6), 683.

(28) Montagna, J. M. The optimal retrofit of multiproduct batch plants. Comput. Chem. Eng. 2003, 27 (8-9), 1277. (29) Lee, H.-K.; Lee, I.-B.; Reklaitis, G. V. Capacity expansion problem of multisite batch plants with production and distribution. Comput. Chem. Eng. 2000, 24 (2-7), 1597. (30) Goel, H. D.; Weijnen, M. P. C.; Grievink, J. Optimal Reliable Retrofit Design of Multiproduct Batch Plants. Ind. Eng. Chem. Res. 2004, 43 (14), 3799. (31) Papageorgaki, S.; Reklaitis, G. V. Retrofitting a General Multipurpose Batch Chemical Plant. Ind. Eng. Chem. Res. 1993, 32, 345. (32) Subrahmanyam, S.; Pekny, J. F.; Reklaitis, G. V. Design of Batch Chemical Plants under Market Uncertainty. Ind. Eng. Chem. Res. 1994, 33, 2688. (33) Dedieu, S.; Pibouleau, L.; Azzaro-Pantel, C.; Domenech, S. Design and retrofit of multiobjective batch plants via a multicriteria genetic algorithm. Comput. Chem. Eng. 2003, 27 (12), 1723. (34) Lee, H.-K.; Jung, J. H.; Lee, I.-B. An evolutionary approach to optimal synthesis of multiproduct batch plant. Comput. Chem. Eng. 1996, 20 (9), 1149. (35) Gins, G.; Espinosa, J.; Smets, I. Y.; Van Brempt, W.; Van Impe, J. F. M. Data alignment via dynamic time warping as a prerequisite for batch-end quality prediction. In AdVances in Data Mining; 2006; Vol. 4065, p 506. (36) Biegler, L. T.; Grossmann, I. E.; Westerberg, A. W. Systematic Methods of Chemical Process Design; Prentice Hall, Inc.: NJ, 1997. (37) Kelkar, V. V.; Ng, K. M. Design of reactive crystallization systems incorporating kinetics and mass-transfer effects. AIChE J. 1999, 45 (1), 69. (38) Paul, E. L.; Atiemo-Obeng, V. A.; Kresta, S. M. Handbook of industrial mixing; John Wiley & Sons, Inc.: Hoboken, NJ, 2004. (39) Osterwalder, N. Retrofit Design of Chemical Batch Processes for the Improvement of Production Cost-Efficiency, Diploma Thesis, Swiss Federal Institute of Technology, Institute for Chemical- and Bioengineering, Safety and Environmental Technology Group: Zurich, Switzerland, 2004. (40) Rippin, D. W. T. Simulation of single- and multiproduct batch chemical plants for optimal design and operation. Comput. Chem. Eng. 1983, 7 (3), 137. (41) Koulouris, A.; Calandranis, J.; Petrides, D. P. Throughput analysis and debottlenecking of integrated batch chemical processes. Comput. Chem. Eng. 2000, 24 (2-7), 1387. (42) Petrides, D.; Koulouris, A.; Siletti, C. Throughput analysis and debottlenecking of biomanufacturing facilities - A job for process simulators. Biopharm. - Appl. Technol. Biopharmaceutical DeV. 2002, 15 (8), 28. (43) Simon, L. L.; Introvigne, M.; Fischer, U.; Hungerbu¨hler, K. Batch reactor optimization under liquid swelling safety constraint. Chem. Eng. Sci. 2007, http://dx.doi.org/10.1016/j.ces.2007.08.076. (44) Simon, L. L.; Introvigne, M.; Fischer, U.; Hungerbu¨hler, K. In Optimal temperature control of an industrial batch reactor with regard to swelling, ESCAPE 17; Elsevier: Bucharest, Romania, 2007; p 793. (45) Simon, L. L. Systematic retrofit method for chemical batch processes using indicators, heuristics and process models, Ph.D. Thesis no. 17092, ETH Zurich: Zurich, Switzerland, 2007. (46) Favre, M. Optimization of a Batch Reaction-Distillation Sequence Using a Combined First Principles-and Neural Network Model, Diploma Thesis, Swiss Federal Institute of Technology, Institute for Chemical- and Bioengineering, Safety and Environmental Technology Group: Zurich, Switzerland, 2006.

ReceiVed for reView January 9, 2007 ReVised manuscript receiVed September 7, 2007 Accepted September 19, 2007 IE070044H