FUZZYFCC: Fuzzy logic control of a fluid catalytic ... - Semantic Scholar
Computers and Chemical Engineering 30 (2006) 850–863
FUZZYFCC: Fuzzy logic control of a fluid catalytic cracking unit (FCCU) to improve dynamic performance ¨ Harun Tas¸kin, Cemalettin Kubat, Ozer Uygun ∗ , Seher Arslankaya Department of Industrial Engineering, Faculty of Engineering, Sakarya University, Turkey Received 16 February 2005; received in revised form 19 December 2005; accepted 22 December 2005 Available online 15 March 2006
Abstract In this paper, fuzzy logic control of a fluid catalytic cracking unit (FCCU) is proposed. Fluid catalytic cracking (FCC) process is a unit that converts heavy distillates like gas oil or residues to gasoline and middle distillates using cracking catalyst. About 45% of worldwide gasoline production comes from FCC processes and its ancillary units. Since a typical FCC unit can process a large amount of the feedstock into more valuable products, the overall economic benefits of a refining could be considerably increased if proper control and optimization strategies are implemented. FCC processes are known to be very difficult to model and control because of the large process scale, complicated hydro-dynamics and complex kinetics of both cracking and coke burning reactions. One of the more heavily investigated terms of nonlinear control, in the field of “intelligent control”, is that due to fuzzy logic controllers (FLCs). FLCs have been successfully applied to a stream of difficult, nonlinear dynamical process such as FCC. Here, with an application to a Turkish refinery FCC unit of FLC fuzzy results obtained using Matlab-Fuzzy Logic Toolbox version 6.5 were found to be acceptable. The paper indicates how fuzzy logic control (FLC), as a promising control technique, would be effectively used for improved process control of FCC in refinery process industry. © 2006 Elsevier Ltd. All rights reserved. Keywords: Chemical process; Intelligent control; Fuzzy logic; FCC; Gasoline; LPG
1. Introduction Fluid catalytic cracking (FCC) is an important oil refinery process, which converts high molecular weight oils into lighter hydrocarbon products. It consists of two interconnected gas–solid fluidized bed reactors: the riser reactor, where almost all the endothermic cracking reactions and coke deposition on the catalyst occur, and the regeneration reactor, where air is used to burn off the coke deposited on the catalyst. The heat produced is carried from the regenerator to the reactor by the catalyst. Thus, in addition to reactivating the catalyst, the regenerator provides the heat required by the endothermic cracking reactions. Industrial FCC units are designed to be capable of using a variety of feed stocks, including straight run distillates, atmospheric and vacuum residua and vacuum gas oils. They produce a range ∗
Corresponding author. Tel.: +90 2643460353; fax: +90 2643460351. E-mail addresses: [email protected] (H. Tas¸kin), [email protected] (C. Kubat), ¨ Uygun), [email protected] (O. [email protected] (S. Arslankaya). 0098-1354/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.compchemeng.2005.12.016
of products, which must adapt to seasonal, environmental, and other changing demand patterns. Since FCC units are capable of converting large quantities of heavy feed into valuable lighter products, any improvement in design, operation or control can result in substantial economic benefits (Bollas et al., 2003). A fluidized catalytic cracking unit (FCCU) consists of reactor–regenerator, riser reactor, main fractionator, absorber– stripper–stabilizer, main air blower, wet gas compressor, etc. The FCCU converts heavy oil into a range of hydrocarbon products, including LPG, fuel gas, gasoline, light diesel, aviation kerosene, slurry oil, among which high octane number gasoline is most valuable. But their values are all market driven, so it is one of the control goals to maximize the production of one or more in different seasons. Since the catalyst circulates through a closed loop consisting of riser, regenerator and reactor, these three main parts are of particular interests both in industrial and research circles. Numerous papers have been published concerning different modelling approaches and control strategies for the FCC process, which deal with the strong interactions and many constrains from the operating, security and environmental point of view. The potential of yielding more market-oriented
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
oil products, increasing production rate and stabilizing the operation become the major incentives to search for more accurate and practical models, high performance, and cost effective and flexible control strategies (Chunyang, Sohrab, & Arthur, 2003). 2. Optimization and control of FCC: a literature review As known, FCC is the most important refinery unit. Since 1940s, a lot of modelling, optimization and control studies of the FCC have been realized. There is a rich literature about FCC, for example: Voerhies’s reaction kinetics model, Fan and Fan’s stability (steady state) specification of fluidized catalytic bed reactor. Optimization studies examples are: Schrake’s linear programming with gradient algorithm, Savas’s and Nicholson’s local linearization of nonlinear model. Kurihara has used optimal control (Pontryagain Maximum Principle) regulatory control algorithm for the first time. Lee and Kugelman’s Local Stability and Open-Loop Strategies, Cadman and Kugelman’s Nonlinear Feedforward Control, Lee and Weekman’s Advanced Control practices, Bromley, and Ward’s Structural Analysis based on multiple loop and feed forward control. Taskin uses Kurihara’s optimal control model of FCC for a Turkish refinery unit (Taskin, 1983). Ramirez et al.’s model is Robust Regulation of Temperature in Reactor-Regenerator (Ramirez, Aguilar, & Isunza, 1996). Here, temperature regulation is a basic control objective impressed to guarantee a safe process operation. Lu et al.’s paper starts with a development of fuzzy modelling and supervisory expert optimization control. A real-time application of fuzzy optimization control has been realized in a fluidized catalytic cracking unit (FCCU) at an oil refinery (Lu, He, & Xu, 1997). Yescas and Isunza used two models: each one using a different kinetic scheme, a different formulation for the catalyst activity and a different conception of the regenerator are compared when simulating the same FCC unit (Yescas & Isunza, 1997). Alhumaizi and Elnashale’s paper is about the effect of control loop configuration on the bifurcation behaviour (Alhumaizi & Elnashaie, 1997). Ali and Elnashaie’s paper addresses the problem of stabilizing the operation of an industrial type IV fluid catalytic cracking unit (FCCU) around an unstable high gasoline yield operating steady state (Ali & Elnashaie, 1997). Chitnis and Corripio present the application of an on-line constrained optimization algorithm, Supervisory Multivariable Constrained Optimization, to the simulating of a Model to Fluid Catalytic Crackling Unit (Chitnis & Corripio, 1998). Yescas et al.’s another modelling approach is nonlinear dynamics which explains open-loop simulation, stability, steady state multiplicity and zero dynamics of the FCC adiabatic systems (Yescas, Bogle, & Isunza, 1998). Abasaeed and Einashaie’s investigation concentrates on the effect of facing on the chaotic behaviour as well as on the gasoline yield and amplitude of temperature oscillations of the forced systems at selected farcing amplitudes (Abasaeed & Elnashaie, 1998). Zeydan applies an other fuzzy modelling and control of FCC in a Turkish refinery unit (Zeydan, 1999). Due to the presence of hot air, carbon deposited on the catalyst is burned off by
851
reacting with oxygen in the air, which also heats up regenerated catalyst. The flue gas travels up the regenerator in to cyclones where entrained catalyst is removed and returned to the bed (Loeblin & Perkins, 1999). Sarma and Rengaswarny use a relatively standard and general fuzzy logic control (FLC) construction heuristic which was followed for sub-structures such as the fuzzy logic variable (FLV) partitioning equi-grids and geometries (Sarma & Rengaswamy, 2000). Zanin et al.’s paper focuses on the modelling of the practical implementation of an optimizing controller in a FCC unit (Zanin, de Gouvea, & Odlak, 2001). Sebzalli and Wang presented an industrial case study which uses principal component analysis and fuzzy c-means clustering to identify operational spaces and develop operational strategies for desired products of the FCC (Sebzalli & Wang, 2001). Skogestad’s goal is to find a set of controlled variables which, when kept at constant set points, indirectly lead to near-optimal operation with acceptable loss. This is denoted “self-optimizing” control (Skogestad, 2000). Han and Chung’s study is to develop a detailed dynamic model of a typical FCC unit that consist of the reactor regenerator, and catalyst transfer lines (Han & Chung, 2001). Alaradi and Rohani’s paper present Identification and Control of A Riser-Type FCC Unit Using Neural Networks (Alaradi & Rohani, 2002). Adebiyi and Corripio propose a model of dynamic neural networks partial least squares (DNNDLS) as a strategy for open-loop identification of multivariable chemical processes that circumvent some of the difficulties associated with multivariable process control (Adebiyi & Corripio, 2003). Cristea et al. use simulation and model predictive control of FCC (Cristea, Agachi, & Marinoiu, 2003). Ramirez et al. developed a combined multivariable control for composition regulation in FCC units (Ramirez, Valencia, & Puebla, 2004). Azeem, Ahmad, and Hanmandlu (in press) paper deals with the fuzzy system identification of reactor–regenerator–stripper–fractinator’s (RRSF) section of a fluidized catalytic cracking unit (FCCU). 3. Intelligent control of chemical process Chemical processes include manufacturing phases such as, filling or emptying a reactor, heating or mixing a product. Intelligent control is a control system with the ultimate degree of autonomy in terms of self-learning, self-reconfigurability, reasoning, planning and decision making, and the ability to extract the most valuable information from unstructured and noisy data from any dynamically complex system and/or environment (Shoureshi, 1993). Complex industrial processes such as a batch chemical reactors; blast furnaces, cement kilns and basic oxygen steel making are difficult to control automatically. This difficulty is due to their nonlinear; time varying behaviour and the poor quality of available measurements. In such cases automatic control is applied to those subsidiary variables, which can be measured and controlled, for example temperatures, pressures and flows. The overall process controls objectives, such as the quality and quantity of product produced, has been left in the hands of the human operators in the past (King & Mamdani, 1997).
852
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
The research and development of math-model-based modern control techniques have been significantly progressing in both theoretical and practical aspects in the past two decades. However, it is still difficult to design and implement a real-time optimization control for some complex industrial processes, if they are highly nonlinear, high dimensional, seriously coupled, and significantly uncertain. It has been an attractive area in the control industry to explore the novel control strategies, which can be designed and implemented with limited deep (process principle and mathematical) knowledge of the controlled environment. With the rapid progresses of the studies on fuzzy logic, neural networks, genetic algorithms, and rule-based expert systems, the intelligent system techniques with remarkable capabilities of dealing with system’s imprecise and/or incomplete information and knowledge have been recognized as one of the most important measures in solving complex industrial control problems. As noted by Kosko that adaptive math-model free estimation is an intelligent system that adaptively estimates continuous function from data without specifying mathematically how outputs depend on inputs (Lu et al., 1997). In conventional hard controllers, the knowledge, which is pertinent to the process being controlled, and the methods for using this knowledge are interrelated and can be expressed in analytical form. To apply such techniques, explicit knowledge of the microscopic behaviour of the process is essential. In contrast, in intelligent or soft control, there is a clear demarcation between the knowledge and the information about the process dynamics, i.e., the macroscopic behaviour of our process, is not essential (Manesis, Sapidis, & King, 1998). Modern control techniques, e.g., parameter estimation, stochastic control, and optimal control are used in either model identification or design of math able control law. However, some industrial processes are too complicated to be modelled and/or controlled by math-algorithms, because they are highly nonlinear and significantly uncertain with unknown structure and imprecise information (Lu, 1996). Intelligent control is therefore particularly attractive when the expertise to control a process is available in the form of linguistic rules acquired from normal operational experience. A fundamental attribute of intelligent control is its ability to work with symbolic, inexact and vague data that human operators comprehend best. Indeed, its ability to deal with incomplete and ill-defined information, an inherent characteristic of wastewater treatment plants, permits implementation of human-like control strategies which have hitherto defied solution by any of the conventional hard control techniques. Certainly for processes, which are known microscopically, hard control is clearly the
methodology to be preferred. Modern technology techniques have, however, in general failed to solve industrial problems and the thousands of industrial plants worldwide which rely on three-term (PID) controllers attest to the limitations of these techniques. Fuzzy logic and artificial neural networks are two examples of soft computing, which have migrated into the realm of industrial control over the last two decades. Chronologically, fuzzy control was the first and its application in the process industry has led to significant improvements in product quality, productivity and energy consumption. Fuzzy control is now firmly established as one of the leading advanced control techniques in use in industry (Manesis et al., 1998). 3.1. Fuzzy control A common definition of a fuzzy control system is that it is a system which emulates a human expert. In this situation the knowledge of the human operator would be put in the form of a set of fuzzy linguistic rules. These rules would produce an approximate decision, just as a human would. Consider Fig. 1 where a block diagram of this definition is shown. As seen, the human operator observes quantifies by making observation of the inputs, i.e., reading a meter or measuring a chart, and performs a definite action, e.g., pushes a knob, turns on a switch, close, a gate, or replaces a fuse, i.e., leading to a crisp action, shown here by the output variable y(t). The human operator can be replaced by a combination of a fuzzy rule-based system (FRBS) and a block called defuzzifier. The sensory (crisp or numerical) data is fed into FRBS where physical quantities are represented or compressed into logistic variables with appropriate membership functions. These linguistic variables are then used in the “antecedents” (IF-Part) of a set of fuzzy rules with an inference engine to result in a new set of fuzzy linguistic variables or “consequents” (THEN-Part) as variables are then denoted in this figure by z, combined and changed to a crisp (numerical) output y* (t) which represents an approximation to actual output y(t) (Jamshidi, 1997). 4. FCC process modelling and control application: FUZZYFCC The FCC Unit (Plant 7) is a UOP (an international supplier and licensor of process technologies, catalysts, adsorbents, process plants to the petroleum refining industries) design (FCC, 2001). The main parts of the FCCU are catalyst and fractionator. Catalyst circulation takes place in the catalyst part. The catalyst
Fig. 1. A conceptual definition of a fuzzy control system (Jamshidi, 1997).
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
853
Fig. 2. Schematic description of the FCC Unit-Plant 7.
consists of reactor and regenerator units which are connected with each other with riser and stand-pipe. Catalyst in the catalyst circulation rises from riser to reactor and from there flows to regenerator by stripper and stand-pipe and then rises up again from riser to reactor. Fig. 2 illustrates a schematic drawing of FCC unit-Plant 7 (An FCC plant of Turkish refinery unit). Some detailed information about FCCU operating system of our FCC is also given in introduction.
4.1. Database for FCCU-Plant 7 The database for the FCCU-Plant 7 is obtained from the reports that are produced periodically (once a week) after the operations. Fig. 3 shows an example of those reports. Each input and output variables and their corresponding data used in this study were taken from the reports produced in the year 2000 and 2001. From all of the reports of the selected
Fig. 3. Sample summary report of the FCCU-Plant 7.
854
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
period, only 22 reports were selected that represent relatively better operating conditions in order to obtain data. Table 1 shows input (independent) and output (dependent) variables and their values for corresponding period. 4.2. FCC modelling FCCU operating variables have been termed independent and dependent variables and all these variables are directly controlled usually with a classical control system (PID control). Fig. 4 lists the major variables in a FCCU. Those variables are going to be used in this study as input and output variables. Among all of the factors affecting the FCC unit 10 input and 10 output variables are selected to examine since they are the most significant ones. 4.3. Knowledgebase for FCC-Plant 7 The knowledgebase in Table 2 is obtained from FCC UOP Process-Plant 7 Reference Manual of the selected unit (FCC UOP Process, 2000). The knowledgebase is formed in two columns in which the left one is for reactor part of the unit and the column at the right is for regenerator part of the unit. This knowledgebase is used in order to obtain fuzzy rules. 4.4. Clustering of input and output variable values Intervals and linguistic variables of values for selected inputs and outputs of the FCCU are obtained by using NCSS (Number Cruncher Statistical System) software. NCSS is statistical software and has clustering feature. In this study NCSS software is used for clustering data of related input and output variables under three clusters (L, low; M, medium; H, high) and for iden-
tifying linguistic variables’ intervals. Linguistic variables and their intervals are shown in Table 3. According to the yields obtained form clustering with NCSS software, membership functions are drawn in Matlab-Fuzzy Logic Toolbox. Fig. 5 displays screens of Membership Function Editor giving the intervals of selected input and output variables used in this study. Membership functions of the variables are drawn (Fig. 5) after all input and output variables are entered through the FIS Editor interface. Fig. 6 shows FIS Editor Screen in which the FCCU input and output variables can be seen. 4.5. Fuzzy rules of the FCCU After membership functions are determined, the inference is performed in accordance with 59 fuzzy linguistic rules (see Fig. 7). Those rules are obtained from the knowledgebase of the FCC Unit. Some other rules are also included heuristically in terms of comparing output values in accordance with input values. All the rules, entered to the system through the Matlab-Fuzzy Logic Toolbox Rule Editor are given in Table 4. Number of rules given to the system as fuzzy rules was restricted such that only the rules affecting the FCC process were taken into account. After giving the rules to the system, defuzzified results and graphical outputs can be derived. Fig. 8(a and b) and Fig. 9 illustrate an example of Surface Viewer screen obtained from Fuzzy Logic Toolbox. Two- or three-dimensional graphic results of variables can be plotted and compared. Fig. 10 shows the results of applied rules and their corresponding outputs according to the mass center of variables. Using the interface, defuzzified values for output variables can be derived changing input values manually. Different output val-
Fig. 4. FCC Unit input and output variables.
Table 1 Database for the FCC Unit variables Independent variables
49829 53289 48888 52963 58999 57617 59174 58136 57145 58001 58001 58394 57904 58295 58973 56696 51764 53791 50957 49654 40977 51303
CFeed
CCR
TFR
CTOR
FFR
CSS
TRG
CBL
Spec gravity
GOC
COC
Gasoline
LPG
Coke
CO
TRC
Pressure
RFR
CO2 /CO
266 223 241 247 249 255 230 217 234 221 178 209 212 216 194 244 243 226 223 246 246 238
14.09 11.31 13.78 14.16 16.08 14.37 15.76 14.41 15.45 15.23 16.00 16.15 15.05 15.11 15.03 14.95 12.85 16.02 16.31 14.42 11.24 12.81
2417 2474 2441 2372 2465 2374 2016 2248 2417 2396 2404 2389 2483 2505 2435 2439 2494 2554 2422 2459 1952 2500
9.18 7.16 8.91 9.29 10.35 9.40 12.21 10.10 10.25 10.30 10.77 10.54 9.76 9.77 9.91 9.86 8.44 10.14 10.72 9.46 9.60 8.12
2271 2237 2254 2206 2294 2228 1844 2059 2207 2175 2274 2247 2273 2277 2283 2293 2266 2326 2253 2276 1724 2272
1375 1680 1575 1500 1575 1788 925 1400 1300 1125 975 1175 1450 1825 1863 1625 1575 1425 1450 1350 1650 1500
658 664 656 670 660 686 667 669 661 662 654 660 661 659 656 660 656 661 663 647 633 657
2.16 1.64 2.33 2.44 2.33 2.16 2.30 2.66 0.76 2.47 2.33 1.99 3.24 3.14 2.93 3.00 1.42 1.44 1.44 1.81 1.82 1.76
0.758 0.726 0.664 0.611 0.656 0.716 0.702 0.670 0.677 0.660 0.665 0.711 0.701 0.733 0.738 0.669 0.701 0.704 0.725 0.736 0.720 0.686
79.95 78.08 76.83 78.01 80.93 76.72 84.40 81.07 85.55 87.33 85.27 72.97 81.72 85.65 80.15 83.65 89.81 88.02 81.73 83.23 92.59 76.73
0.098 0.107 0.092 0.077 0.096 0.094 0.077 0.098 0.090 0.078 0.089 0.121 0.092 0.076 0.092 0.091 0.066 0.075 0.073 0.074 0.095 0.107
49.62 53.74 52.01 47.60 50.13 46.77 51.65 49.08 58.01 53.82 53.32 45.80 50.76 53.47 51.51 52.03 58.65 56.92 51.13 51.21 53.71 50.09
20.44 13.82 15.22 19.56 20.91 19.47 19.98 22.09 19.85 25.20 22.07 16.79 20.53 22.47 20.56 22.00 21.30 21.57 18.78 22.76 27.12 17.21
5.65 6.59 5.12 5.08 6.31 5.76 7.26 6.42 5.97 6.45 5.54 5.89 5.72 5.87 5.96 6.06 5.54 5.17 4.93 4.46 6.91 5.45
6.6 8.4 5.0 5.0 5.0 4.8 4.4 4.8 4.0 5.0 3.2 4.8 3.6 3.8 3.6 3.8 3.8 3.6 2.8 2.0 8.2 5.0
535 520 526 534 530 537 536 531 530 532 525 534 525 524 521 531 517 517 525 520 519 516
2.10 2.20 2.10 2.15 2.15 2.00 1.72 1.90 2.10 2.20 2.10 2.15 2.20 2.25 2.30 2.05 2.15 2.17 2.15 2.00 1.65 2.20
146 237 187 166 171 146 172 189 210 221 130 141 210 228 153 146 228 228 169 182 228 228
1.67 1.30 2.12 1.76 1.76 2.13 2.50 2.20 2.80 2.20 3.40 2.20 3.10 2.90 3.30 3.40 3.20 2.90 4.10 5.50 1.30 2.20
ATR (m3 /h), air to regenerator; GOC (wt.%), gas oil conversion; CFeed (C), combined feed temperature; COC, coke on catalyst; CCR (t/min), catalyst circulation rate; gasoline (wt.%), gasoline; TFR (m3 /d), total feed rate; LPG (wt.%), liquefied petroleum gas; CTOR (t/t), catalyst to oil ratio; coke (wt.%), coke; FFR (m3 /d), fresh feed rate; CO (vol.%), CO in flue gas; CSS (kg/h), catalyst stripping steam; TRC (C), riser outlet (temperature of reactor); TRG (C), temperature of regenerator; pressure (kg/cm2 ), pressure of reactor; CBL (m), catalyst bed level; RFR (m3 /d), recycle feed rate; spec gravity, specific gravity; CO2 /CO (mol/mol), CO2 /CO in flue gas.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
ATR
Dependent variables
855
856
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 5. Membership functions of the input and output variables.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
857
Table 2 Knowledgebase of the FCCU-Plant 7 Reactor
Regenerator Low regenerator temperature increases conversion and coke formation
Feed rate Effects of 0.1 step increase of combined feed rate are Gasoline yield increases Conversion increases 1–2% Coke formation increases 0.3%
Feed specifications If API gravity is increases then Conversion rate increases Gasoline conversion rate increases
Feed temperature If total feed temperature is increased 38 ◦ C then Conversion decreases by 0.4% Regenerator temperature increases 11 ◦ C Coke formation decreases 0.5% Catalyst/oil ratio decreases
Regenerator bed level Regenerator productivity is better when bed level is higher, since catalyst residence time would be more. Low bed level decreases oxygen usage efficiency
Recycle composition and temperature Increasing recycle temperature has the same effect of increasing feed temperature Reactor temperature If reactor temperature is increased then conversion rate, gasoline and LPG conversion increases Reactor level The effects of the raising reactor level Conversion rate increases Gasoline conversion rate decreases for fixed conversion Coke formation increases Reactor pressure The effects of raising reactor pressure Conversion rate increases Regenerator temperature increases
Catalyst circulation rate Increasing catalyst circulation rate affects regeneration negatively Air to regenerator If air to regenerator is not enough then coke on catalyst increases Bed temperature When decreasing catalyst/oil ratio coke formation decreases. Process variables increasing regenerator temperature are Increasing specific gravity of feed Decreasing feed quality (high carbon level) Increasing total feed temperature Increasing reactor bed level Increasing reactor temperature Increasing reactor pressure
Stripping steam Increasing stripping steam occurs if Recycle feed increases Reactor pressure increases Total feed temperature is higher Reactor temperature is lower Catalyst/oil ratio Catalyst/oil increases when Reactor temperature is increased Regenerator temperature is increased Combined feed temperature is decreased If catalyst/oil ratio is increased then conversion and coke formation increases
Fig. 6. Fuzzy inference system (FIS) editor screen (inputs and outputs of the FCCU).
858
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 7. Rule editor of Matlab-Fuzzy Logic Tool Box.
Fig. 8. (a) Gasoline graphic according to CFeed and TFR interaction and (b) gasoline graphic according to TRG and CBL interaction.
ues can be get trough the Rule Viewer according to the given input values. To get defuzzified output values for all the real input values is not flexible using the interface. For that reason a program is written using Matlab codes to drive defuzzified output results in accordance with real input values. Fig. 11 gives a Table 3 Linguistic variables’ intervals of inputs and outputs for the FCCU Variable
Low
Medium
High
ATR (m3 /h) CFeed (C) CCR (t/min) TFR (m3 /d) CTOR (t/t) FFR (m3 /d) CSS (kg/h) TRG (C) CBL (m) Spec gravity GOC (wt.%) COC Gasoline (wt.%) LPG (wt.%) Coke (wt.%) CO (vol.%) TRC (C) Pressure (kg/cm) RFR (m3 /d) CO2 /CO (mol/mol)
40977–52963 178–217 11.24–14.42 1952–2396 7.16–9.77 1724–2207 925–1400 633–660 0.76–2.33 0.611–0.701 72.97–80.93 0.659–0.922 45.80–51.65 13.82–20.44 4.46–5.72 2.00–5.00 516–525 1.65–2.10 130–172 1.30–2.13
49654–57617 194–243 12.50–15.76 2016–2483 8.9–10.45 1844–2277 1125–1680 656–669 1.64–3.00 0.665–0.733 76.83–86.00 0.0762–0.1072 47.60–56.92 16.79–22.47 5.12–6.45 3.20–8.20 519–533 1.72–2.20 146–221 1.76–3.10
52963–59174 223–266 14.75–16.31 2404–2554 9.77–12.21 2215–2326 1425–1863 661–686 2.44–3.24 0.704–0.758 81.73–92.59 0.922–0.1210 51.65–58.65 20.44–27.12 5.89–7.26 5.00–8.40 525–537 2.10–2.30 182–237 2.20–5.50
comparison between real values and defuzzified values of output variables. Table 5 is also gives a comparison between real output values and derived defuzzified output values for each period. The last three rows of the table give average, maximum and minimum values of both real and defuzzified values. Each application steps were obtained using Matlab-Fuzzy Logic Toolbox version 6.5. Comparison of average for defuzzified results with average for real values is shown in Table 6.
Fig. 9. LPG graphic according to TRG and ATR interaction.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
859
Table 4 Fuzzy rules of the FCCU entered to the system 1. If (TFR is L) then (GOC is H) (COC is M) (LPG is L) (RFR is M) 2. If (TFR is M) then (COC is M) (gasoline is L) (LPG is M) (RFR is L) 3. If (TFR is H) then (GOC is H) (COC is L) (gasoline is H) (LPG is H) 4. If (TFR is H) and (FFR is L) then (RFR is H) 5. If (FFR is L) then (GOC is H) (pressure(reac) is L) 6. If (FFR is H) then (GOC is L) (TRC is M) (pressure(reac) is M) 7. If (CFeed is H) then (GOC is L) (gasoline is L) (coke is M) (TRC is H) (pressure(reac) is M) (RFR is H) 8. If (CFeed is M) then (GOC is M) (coke is M) (pressure(reac) is H) (RFR is H) 9. If (CFeed is L) then (GOC is M) (gasoline is H) (pressure(reac) is H) (RFR is L) 10. If (CBL(reac) is H) then (GOC is M) (gasoline is L) (coke is M) 11. If (CBL(reac) is M) then (GOC is M) 12. If (CBL(reac) is L) then (GOC is L) (gasoline is H) (coke is M) 13. If (CSS is H) then (RFR is H) 14. If (CSS is H) then (pressure(reac) is H) 15. If (CSS is H) then (TRC is M) 16. If (CTOR is H) then (TRC is H) 17. If (CTOR is L) then (coke is M) 18. If (CTOR is H) then (GOC is H) 19. If (TRG is H) then (CO is M) (CO2 /CO is L) 20. If (TRG is H) then (pressure(reac) is H) 21. If (TRG is L) then (GOC is L) (gasoline is L) (LPG is M) 22. If (TRG is M) then (GOC is H) (gasoline is H) (LPG is M) 23. If (TRG is H) then (GOC is L) (gasoline is L) (LPG is M) 24. If (TRG is H) then (coke is M) (RFR is H) 25. If (TRG is H) then (TRC is H) 26. If (CCR is H) then (CO is L) (CO2 /CO is H) 27. If (CCR is H) then (coke is H) 28. If (ATR is H) then (CO is M) (CO2 /CO is M) 29. If (ATR is L) then (COC is M)
30. If (ATR is H) then (coke is H) 31. If (spec gravity is H) then (GOC is H) (gasoline is M) 32. If (spec gravity is L) then (CO is H) (CO2 /CO is L) 33. If (spec gravity is H) then (coke is H) (RFR is H) 34. If (spec gravity is H) then (TRC is L) (pressure(reac) is M) 35. If (CTOR is M) then (coke is M) 36. If (TRG is L) then (coke is L) 37. If (CCR is L) then (coke is L) 38. If (CCR is M) then (coke is M) 39. If (ATR is M) then (coke is L) 40. If (CTOR is H) then (coke is H) 41. If (TRG is L) then (coke is L) 42. If (ATR is M) then (COC is L) 43. If (TRG is M) then (CO is M) 44. If (TRG is L) then (CO is L) 45. If (CCR is M) then (CO is L) 46. If (CCR is L) then (CO is H) 47. If (CSS is M) then (TRC is M) 48. If (TRG is L) then (TRC is L) 49. If (FFR is M) then (pressure(reac) is M) 50. If (TRG is L) then (pressure(reac) is M) 51. If (TRG is M) then (pressure(reac) is H) 52. If (CSS is M) then (RFR is H) 53. If (CSS is L) then (RFR is L) 54. If (TRG is M) then (RFR is H) 55. If (spec gravity is L) then (RFR is L) 56. If (TRG is M) then (CO2 /CO is M) 57. If (CCR is L) then (CO2 /CO is L) 58. If (ATR is M) then (CO2 /CO is M) 59. If (ATR is L) then (CO2 /CO is L)
Fig. 10. Rule viewer screen to obtain defuzzified results for the FCCU.
860
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 11. Defuzzified output values compared with real output values.
Deviation column is calculated by real value minus defuzzified value. Deviation ratio is similarly calculated by dividing deviation value to corresponding average of real data. As can be seen from Table 6, the minimum deviation ratio is 0.0006 and the maximum deviation ratio is (ignoring the negativity) 0.08 which shows that derived defuzzified results are not so different form the real data. 5. Results and discussions There is no general and mathematically optimal solution for the control if plant (unit) to be controlled is complex and highly nonlinear. The modelling of the process and its solution become even more difficult if a sufficiently precise model is unknown or cannot be identified. It is well known however that in many cases a human can master the performance of such a plant using linguistic control algorithms that represent the operator knowledge and experience about the plant/unit by using If-Then rules. First, we quantified the qualitative character of linguistic rules and represented real data as fuzzy modelling by using fuzzy logic software (Matlab-Fuzzy Logic Toolbox, version 6.5). In the next stage, a development and application of dynamic model-based optimization and fuzzy control were performed. The idea behind this kind of fuzzy control has been to take the respective advantages in math-based control, especially in chemical process such as refinery processes-catalyst cracking, hydro-cracking, and distillation. FCCU control of the plant we examined is carried out by a program (PID control) which is based on consultancy of the
experts or experienced engineers/operators. However, this control system used in that plant is not fully integrated with other units (no real-time, on-time control) and sometimes manual interventions are required by the operator when the plant operations are disturbed. Time and product losses are caused by this intervention during the cracking process. The most repeated results were selected as data set (the most optimal values are derived from weekly summary report between January 2001 and December 2001) from the empirical FCCU data to appoint to use as rules (rules are obtained from knowledgebase of the FCCU-Plant 7 reference manual). Although some variables were deviated from the optimal (average) values, most of the fuzzy results obtained from this study were found to be acceptable and reasonable level of compatibility were achieved compared to the empirical FCCU data and targeted gasoline and LPG conversion rate. Deviation of output results such as COC and CO (vol.%) could be rectified and optimized by the further research on flue gas analysis and catalyst contents. However, GOC (wt.%), Gasoline (wt.%), LPG (wt.%) and other outputs/results were acceptable and in good agreement with optimal values. In this work, the results obtained using fuzzy control is found to be very closer to the plant’s real results. This study can be extended using long time period, data different raw oil contents—gravity, viscosity and conradson carbon, UOP K factor, Gravity (API) and Octan RON and the soft computing algorithms, and AI techniques such as neural nets, genetic algorithm, and chaos theory. Another approach may be a comparison between PID control, optimal control and neuro-fuzzy control.
Table 5 Comparison of real (crisp) and defuzzified output variables of the FCCU-Plant 7 GOC
Gasoline
LPG
Coke
CO
TRC
Pressure
RFR
CO2 /CO
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
79.95 78.08 76.83 78.01 80.93 76.72 84.40 81.07 85.55 87.33 85.27 72.97 81.72 85.65 80.15 83.65 89.81 88.02 81.73 83.23 92.59 76.73
82.799 82.942 81.325 78.480 82.883 81.262 83.367 81.377 83.546 85.434 82.839 84.139 82.992 82.942 82.942 82.552 83.282 82.954 82.937 83.259 83.255 83.349
0.0984 0.1072 0.0922 0.0772 0.0958 0.0944 0.0770 0.0980 0.0897 0.0783 0.0888 0.1210 0.0920 0.0762 0.0916 0.0911 0.0659 0.0747 0.0730 0.0741 0.0952 0.1071
0.0903 0.0775 0.0861 0.0836 0.0789 0.0916 0.0918 0.0917 0.0903 0.0916 0.0913 0.0917 0.0767 0.0761 0.0871 0.0860 0.0806 0.0762 0.0857 0.0837 0.0918 0.0821
49.62 53.74 52.01 47.60 50.13 46.77 51.65 49.08 58.01 53.82 53.32 45.80 50.76 53.47 51.51 52.03 58.65 56.92 51.13 51.21 53.71 50.09
51.1112 53.3729 51.2217 48.0296 52.4325 49.6041 50.4656 48.0724 52.4982 52.3077 52.4982 52.1430 52.4982 52.3389 52.3389 52.1514 52.4982 55.6235 52.3481 52.1517 51.8575 52.7802
20.44 13.82 15.22 19.56 20.91 19.47 19.98 22.09 19.85 25.20 22.07 16.79 20.53 22.47 20.56 22.00 21.30 21.57 18.78 22.76 27.12 17.21
20.4931 22.8234 21.9529 19.6418 22.6798 19.6493 17.3286 18.8455 20.4931 19.8108 19.9737 19.7313 22.5910 22.8520 21.6090 21.9450 22.5923 22.5527 20.7832 22.6886 16.8690 22.7216
5.65 6.59 5.12 5.08 6.31 5.76 7.26 6.42 5.97 6.45 5.54 5.89 5.72 5.87 5.96 6.06 5.54 5.17 4.93 4.46 6.91 5.45
6.0101 5.8666 5.3002 5.3269 6.3288 6.3072 6.3599 6.3090 6.3372 6.3253 5.8966 6.3744 6.3236 6.1816 5.8666 6.2774 5.3088 5.8740 6.0336 5.9332 5.7936 5.3204
6.6 8.4 5.0 5.0 5.0 4.8 4.4 4.8 4.0 5.0 3.2 4.8 3.6 3.8 3.6 3.8 3.8 3.6 2.8 2.0 8.2 5.0
4.3554 6.2555 5.1816 5.1353 5.2050 4.5518 4.5661 5.2050 5.5099 5.7027 5.2050 4.5428 4.7839 4.8346 4.5430 5.3281 5.0248 4.5549 4.5168 3.5129 5.8863 5.1309
535 520 526 534 530 537 536 531 530 532 525 534 525 524 521 531 517 517 525 520 519 516
526.3274 524.5537 526.3728 529.3402 529.2806 527.6673 532.3396 529.2452 529.0842 532.0483 526.3342 529.1735 525.6386 522.7887 523.7430 529.2543 526.3315 528.2685 526.2901 526.3158 526.7578 526.5170
2.10 2.20 2.10 2.15 2.15 2.00 1.72 1.90 2.10 2.20 2.10 2.15 2.20 2.25 2.30 2.05 2.15 2.17 2.15 2.00 1.65 2.20
2.0463 2.0774 2.0608 2.0772 2.0745 2.0773 1.9545 2.0362 2.0774 2.0709 2.0774 2.0784 2.0774 2.0774 2.0774 2.0772 2.0608 2.0783 2.0774 2.0072 1.9726 2.0270
146 237 187 166 171 146 172 189 210 221 130 141 210 228 153 146 228 228 169 182 228 228
184.162 213.307 183.947 184.439 184.431 184.103 180.111 184.155 184.428 184.427 146.867 181.410 212.476 218.796 184.428 184.428 212.476 214.938 184.534 203.638 197.492 190.842
1.67 1.30 2.12 1.76 1.76 2.13 2.50 2.20 2.80 2.20 3.40 2.20 3.10 2.90 3.30 3.40 3.20 2.90 4.10 5.50 1.30 2.20
1.9240 1.9958 1.6042 2.0014 3.3766 2.1348 3.4365 1.9972 3.1504 2.6454 3.3763 3.7024 2.6966 2.8739 2.7073 2.1547 2.0570 3.6884 3.4902 1.6238 1.6038 2.0454
82.29 92.59 72.97
82.766 85.434 78.480
0.0890 0.1210 0.0659
0.0856 0.0918 0.0761
51.87 58.65 45.80
51.8338 55.6235 48.0296
20.44 27.12 13.82
20.9376 22.8520 16.8690
5.82 7.26 4.46
5.9843 6.3744 5.3002
4.60 8.40 2.00
4.9787 6.2555 3.5129
526.6 537.0 516.0
527.4397 532.3396 522.7887
2.09 2.30 1.65
2.0564 2.0784 1.9545
187.1 237.0 130.0
190.902 218.796 146.867
2.63 5.50 1.30
2.5585 3.7024 1.6038
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Avrg Max Min
COC
861
862
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Table 6 Comparison between averages of defuzzified and real values of the outputs and deviations between them Output
Variation intervals of empirical data (optimal range)
Average of real data
Average of defuzzified results
Deviation (real-defuzzy)
Deviation ratio
GOC (wt.%) COC Gasoline (wt.%) LPG (wt.%) Coke (wt.%) CO (vol.%) TRC (C) Pressure (kg/cm) RFR (m3 /d) CO2 /CO (mol/mol)
72.97–92.59 0.0659–0.121 45.8–58.65 13.82–27.12 4.46–7.26 2.00–8.40 516–537 1.65–2.30 130–237 1.3–5.5
82.290 0.089 51.865 20.441 5.823 4.600 526.591 2.090 187.091 2.634
82.766 0.086 51.834 20.938 5.984 4.979 527.440 2.056 190.902 2.558
−0.4757 0.0035 0.0312 −0.4967 −0.1611 −0.3787 −0.8487 0.0340 −3.8107 0.0752
−0.00578 0.03909 0.00060 −0.02430 −0.02767 −0.08233 −0.00161 0.01629 −0.02037 0.02855
Finally our goal is to achieve such comparison in the further research paper.
Excellence (Contract No. 500273). http://ww.iproms.org .
6. Conclusion The following conclusions can be drawn from the application of fuzzy logic to the control of the FCC as described in this paper: • A fuzzy logic-based control system was developed to estimate the variables in a FCCU. By careful selection of the input variables (here 10 variables) and designing the rules (here 30 rules) for the system and their statistical analysis, 97.53% of control accuracy can be obtained. • The results were generally in compliance with empirical FCCU data. However, in most cases even though limited of rules and inputs were applied the results obtained indicated a very high accuracy. This clearly shows that by increasing the number of inputs (for example advanced lab data about raw oil-gravity, viscosity and conradson carbon and fractionatortop temperature and bottom temperature) by increasing data for each variables for example by extending yearly data and periods, and by improving the rules used in the MATLAB package, more proper and accurate results could be expected. • Since the assumptions and very high nonlinear behaviours of the cracking reaction, there could be a deviation between the results obtained in this study and the FCCU-Plant 7’s empirical values. • As a final remark fuzzy logic is a promising control technique and would be effectively used for improved process control of the FCCU and other units in the petroleum refinery production processes. The study will be continued to increase the effectiveness of the proposed model by increasing and manipulating the content of the rules, data and variables (sensitivity and stability analysis) and comparison with other classical and intelligent techniques. Acknowledgement Sakarya University is a partner of the Innovative Production Machines and Systems (I*PROMS) Network of
References Abasaeed, A. E., & Elnashaie, S. S. E. H. (1998). On the chaotic behavior of externally forced industrial fluid catalytic cracking units. Chaos, Solutions & Fractals, 9(3), 455–470. Adebiyi, O. A., & Corripio, A. B. (2003). Dynamic neural networks partial least squares (DNNDLS) identification of multivariable processes. Computers & Chemical Engineering, 27(2), 143–155. Alaradi, A. A., & Rohani, S. (2002). Identification and control of a riser-type FCC unit using neural networks. Computers & Chemical Engineering, 26(3), 401–421. Alhumaizi, K., & Elnashaie, S. S. E. H. (1997). Effect of control loop configuration on the bifurcation behaviour and gasoline yield of industrial fluid catalytic cracking (FCC) units. Mathematical and Computer Modelling, 25(3), 37–56. Ali, E. E., & Elnashaie, S. S. E. H. (1997). Nonlinear model predictive control of industrial type IV fluid catalytic (FCC) units for maximum gasoline yeald. Industrial & Engineering Chemistry Research, 36(12), 389– 398. Azeem, M. F., Ahmad, N., & Hanmandlu, M. (in press). Fuzzy modeling of fluidized catalytic cracking unit. Applied Soft Computing [Corrected proof, available online September 26, 2005]. Bollas, G. M., Papadokonstadakis, S., Michalopoulos, J., Arampatzis, G., Lappas, A. A., Vasalos, I. A., et al. (2003). Using hybrid neural networks in scaling up an FCC model from a pilot plant to an industrial unit. Chemical Engineering and Processing, 42(8–9), 697–713. Chitnis, U. K., & Corripio, A. B. (1998). Online optimization of a models fluid catalytic cracking unit. ISA Transactions, 37(3), 215–226. Chunyang, J., Sohrab, R., & Azrthur, J. (2003). FCC unit modeling, identification and model predictive control, a simulation study. Chemical Engineering and Processing, 42(4), 311–325. Cristea, M. V., Agachi, S. P., & Marinoiu, V. (2003). Simulation and model predictive control of a UOP fluid catalytic cracking unit. Chemical Engineering and Processing, 42(2), 67–91. FCC (2001). Plant 7 The Year 2001 Data in a Turkish Refinery Unit. FCC UOP Process (2000). Plant 7 Reference Manual in a Turkish Refinery Unit. Han, I. S., & Chung, C. B. (2001). Dynamic modeling and simulation of a fluidized catalytic cracking process. Part I. Process modelling. Chemical Engineering Science, 56(5), 1951–1971. Jamshidi, M. (1997). Large-scale systems: Modeling and control with fuzzy logic. New Jersey: Prentice-Hall. King, P. J., & Mamdani, E. H. (1997). The application of fuzzy control system to industrial processes. Automatica, 13(3), 235–242.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863 Loeblin, C., & Perkins, J. D. (1999). Structural design for on-line process optimisation. II. Application to a simulated FCC. AIChE Journal, 45(5), 1030–1040. Lu, Y. Z. (1996). Industrial intelligent control. New York: John-Wiley and Sons. Lu, Y. Z., He, M., & Xu, C. W. (1997). Fuzzy modeling and expert optimization control for industrial processes. IEEE Transaction on Control Systems Technology, 5(1), 2–11. Manesis, S. A., Sapidis, D. J., & King, R. E. (1998). Intelligent control of wastewater treatment plants. Artificial Intelligence in Engineering, 12(3), 275–281. Ramirez, J. A., Aguilar, R., & Isunza, F. L. (1996). Robust regulation of temperature in reactor–regenerator fluid catalytic cracking units. Industrial & Engineering Chemistry Research, 35, 1652–1659. Ramirez, J. A., Valencia, J., & Puebla, H. (2004). Multivariable control configurations for composition regulation in a fluid catalytic cracking unit. Chemical Engineering Journal, 99(3), 187–201. Sarma, P., & Rengaswamy, R. (2000). Multivariable gain-scheduled fuzzy logic control of a fluidized catalytic cracker unit. Computers & Chemical Engineering, 24(2–7), 1083–1089. Sebzalli, Y. U., & Wang, X. Z. (2001). Knowledge discovery from process operational data using PCA and fuzzy clustering. Engineering Applications of Artificial Intelligence, 14(5), 607–616.
863
Shoureshi, R. (1993). Intelligent control systems: Are they for real? Transactions of the ASME, 115, 392–400. Skogestad, S. (2000). Plantwide control: The search for the self-optimizing control structure. Journal of Process Control, 10(5), 487–507. Taskin, H. (1983). Process modeling, optimization and control: An application to a petroleum refinery unit (FCCU). Ph.D. Thesis [Turkish], Istanbul Technical University, Institute of Science and Technology, Faculty of Management. Yescas, R. M. Y., & Isunza, F. L. (1997). Comparison of two dynamic models for FCC units. Catalysis Today, 38(1), 137–147. Yescas, R. M. Y., Bogle, D., & Isunza, F. L. (1998). Approach to the analysis of the dynamics of industrial FCC units. Journal of Processes Control, 8(2), 89–100. Zanin, A. C., de Gouvea, M. T., & Odlak, D. (2001). Implementation of a real time optimization strategy on the FCC catalytic converter. Latin American Applied Research, 31, 301–308. Zeydan, M. (1999). Fuzzy modeling and control in a petroleum refinery unit (FCC). Ph.D. Thesis [Turkish], Sakarya University, Institute of Science and Technology, Department of Industrial Engineering [Superviser, Tas¸kin, H.].
FUZZYFCC: Fuzzy logic control of a fluid catalytic cracking unit (FCCU) to improve dynamic performance ¨ Harun Tas¸kin, Cemalettin Kubat, Ozer Uygun ∗ , Seher Arslankaya Department of Industrial Engineering, Faculty of Engineering, Sakarya University, Turkey Received 16 February 2005; received in revised form 19 December 2005; accepted 22 December 2005 Available online 15 March 2006
Abstract In this paper, fuzzy logic control of a fluid catalytic cracking unit (FCCU) is proposed. Fluid catalytic cracking (FCC) process is a unit that converts heavy distillates like gas oil or residues to gasoline and middle distillates using cracking catalyst. About 45% of worldwide gasoline production comes from FCC processes and its ancillary units. Since a typical FCC unit can process a large amount of the feedstock into more valuable products, the overall economic benefits of a refining could be considerably increased if proper control and optimization strategies are implemented. FCC processes are known to be very difficult to model and control because of the large process scale, complicated hydro-dynamics and complex kinetics of both cracking and coke burning reactions. One of the more heavily investigated terms of nonlinear control, in the field of “intelligent control”, is that due to fuzzy logic controllers (FLCs). FLCs have been successfully applied to a stream of difficult, nonlinear dynamical process such as FCC. Here, with an application to a Turkish refinery FCC unit of FLC fuzzy results obtained using Matlab-Fuzzy Logic Toolbox version 6.5 were found to be acceptable. The paper indicates how fuzzy logic control (FLC), as a promising control technique, would be effectively used for improved process control of FCC in refinery process industry. © 2006 Elsevier Ltd. All rights reserved. Keywords: Chemical process; Intelligent control; Fuzzy logic; FCC; Gasoline; LPG
1. Introduction Fluid catalytic cracking (FCC) is an important oil refinery process, which converts high molecular weight oils into lighter hydrocarbon products. It consists of two interconnected gas–solid fluidized bed reactors: the riser reactor, where almost all the endothermic cracking reactions and coke deposition on the catalyst occur, and the regeneration reactor, where air is used to burn off the coke deposited on the catalyst. The heat produced is carried from the regenerator to the reactor by the catalyst. Thus, in addition to reactivating the catalyst, the regenerator provides the heat required by the endothermic cracking reactions. Industrial FCC units are designed to be capable of using a variety of feed stocks, including straight run distillates, atmospheric and vacuum residua and vacuum gas oils. They produce a range ∗
Corresponding author. Tel.: +90 2643460353; fax: +90 2643460351. E-mail addresses: [email protected] (H. Tas¸kin), [email protected] (C. Kubat), ¨ Uygun), [email protected] (O. [email protected] (S. Arslankaya). 0098-1354/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.compchemeng.2005.12.016
of products, which must adapt to seasonal, environmental, and other changing demand patterns. Since FCC units are capable of converting large quantities of heavy feed into valuable lighter products, any improvement in design, operation or control can result in substantial economic benefits (Bollas et al., 2003). A fluidized catalytic cracking unit (FCCU) consists of reactor–regenerator, riser reactor, main fractionator, absorber– stripper–stabilizer, main air blower, wet gas compressor, etc. The FCCU converts heavy oil into a range of hydrocarbon products, including LPG, fuel gas, gasoline, light diesel, aviation kerosene, slurry oil, among which high octane number gasoline is most valuable. But their values are all market driven, so it is one of the control goals to maximize the production of one or more in different seasons. Since the catalyst circulates through a closed loop consisting of riser, regenerator and reactor, these three main parts are of particular interests both in industrial and research circles. Numerous papers have been published concerning different modelling approaches and control strategies for the FCC process, which deal with the strong interactions and many constrains from the operating, security and environmental point of view. The potential of yielding more market-oriented
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
oil products, increasing production rate and stabilizing the operation become the major incentives to search for more accurate and practical models, high performance, and cost effective and flexible control strategies (Chunyang, Sohrab, & Arthur, 2003). 2. Optimization and control of FCC: a literature review As known, FCC is the most important refinery unit. Since 1940s, a lot of modelling, optimization and control studies of the FCC have been realized. There is a rich literature about FCC, for example: Voerhies’s reaction kinetics model, Fan and Fan’s stability (steady state) specification of fluidized catalytic bed reactor. Optimization studies examples are: Schrake’s linear programming with gradient algorithm, Savas’s and Nicholson’s local linearization of nonlinear model. Kurihara has used optimal control (Pontryagain Maximum Principle) regulatory control algorithm for the first time. Lee and Kugelman’s Local Stability and Open-Loop Strategies, Cadman and Kugelman’s Nonlinear Feedforward Control, Lee and Weekman’s Advanced Control practices, Bromley, and Ward’s Structural Analysis based on multiple loop and feed forward control. Taskin uses Kurihara’s optimal control model of FCC for a Turkish refinery unit (Taskin, 1983). Ramirez et al.’s model is Robust Regulation of Temperature in Reactor-Regenerator (Ramirez, Aguilar, & Isunza, 1996). Here, temperature regulation is a basic control objective impressed to guarantee a safe process operation. Lu et al.’s paper starts with a development of fuzzy modelling and supervisory expert optimization control. A real-time application of fuzzy optimization control has been realized in a fluidized catalytic cracking unit (FCCU) at an oil refinery (Lu, He, & Xu, 1997). Yescas and Isunza used two models: each one using a different kinetic scheme, a different formulation for the catalyst activity and a different conception of the regenerator are compared when simulating the same FCC unit (Yescas & Isunza, 1997). Alhumaizi and Elnashale’s paper is about the effect of control loop configuration on the bifurcation behaviour (Alhumaizi & Elnashaie, 1997). Ali and Elnashaie’s paper addresses the problem of stabilizing the operation of an industrial type IV fluid catalytic cracking unit (FCCU) around an unstable high gasoline yield operating steady state (Ali & Elnashaie, 1997). Chitnis and Corripio present the application of an on-line constrained optimization algorithm, Supervisory Multivariable Constrained Optimization, to the simulating of a Model to Fluid Catalytic Crackling Unit (Chitnis & Corripio, 1998). Yescas et al.’s another modelling approach is nonlinear dynamics which explains open-loop simulation, stability, steady state multiplicity and zero dynamics of the FCC adiabatic systems (Yescas, Bogle, & Isunza, 1998). Abasaeed and Einashaie’s investigation concentrates on the effect of facing on the chaotic behaviour as well as on the gasoline yield and amplitude of temperature oscillations of the forced systems at selected farcing amplitudes (Abasaeed & Elnashaie, 1998). Zeydan applies an other fuzzy modelling and control of FCC in a Turkish refinery unit (Zeydan, 1999). Due to the presence of hot air, carbon deposited on the catalyst is burned off by
851
reacting with oxygen in the air, which also heats up regenerated catalyst. The flue gas travels up the regenerator in to cyclones where entrained catalyst is removed and returned to the bed (Loeblin & Perkins, 1999). Sarma and Rengaswarny use a relatively standard and general fuzzy logic control (FLC) construction heuristic which was followed for sub-structures such as the fuzzy logic variable (FLV) partitioning equi-grids and geometries (Sarma & Rengaswamy, 2000). Zanin et al.’s paper focuses on the modelling of the practical implementation of an optimizing controller in a FCC unit (Zanin, de Gouvea, & Odlak, 2001). Sebzalli and Wang presented an industrial case study which uses principal component analysis and fuzzy c-means clustering to identify operational spaces and develop operational strategies for desired products of the FCC (Sebzalli & Wang, 2001). Skogestad’s goal is to find a set of controlled variables which, when kept at constant set points, indirectly lead to near-optimal operation with acceptable loss. This is denoted “self-optimizing” control (Skogestad, 2000). Han and Chung’s study is to develop a detailed dynamic model of a typical FCC unit that consist of the reactor regenerator, and catalyst transfer lines (Han & Chung, 2001). Alaradi and Rohani’s paper present Identification and Control of A Riser-Type FCC Unit Using Neural Networks (Alaradi & Rohani, 2002). Adebiyi and Corripio propose a model of dynamic neural networks partial least squares (DNNDLS) as a strategy for open-loop identification of multivariable chemical processes that circumvent some of the difficulties associated with multivariable process control (Adebiyi & Corripio, 2003). Cristea et al. use simulation and model predictive control of FCC (Cristea, Agachi, & Marinoiu, 2003). Ramirez et al. developed a combined multivariable control for composition regulation in FCC units (Ramirez, Valencia, & Puebla, 2004). Azeem, Ahmad, and Hanmandlu (in press) paper deals with the fuzzy system identification of reactor–regenerator–stripper–fractinator’s (RRSF) section of a fluidized catalytic cracking unit (FCCU). 3. Intelligent control of chemical process Chemical processes include manufacturing phases such as, filling or emptying a reactor, heating or mixing a product. Intelligent control is a control system with the ultimate degree of autonomy in terms of self-learning, self-reconfigurability, reasoning, planning and decision making, and the ability to extract the most valuable information from unstructured and noisy data from any dynamically complex system and/or environment (Shoureshi, 1993). Complex industrial processes such as a batch chemical reactors; blast furnaces, cement kilns and basic oxygen steel making are difficult to control automatically. This difficulty is due to their nonlinear; time varying behaviour and the poor quality of available measurements. In such cases automatic control is applied to those subsidiary variables, which can be measured and controlled, for example temperatures, pressures and flows. The overall process controls objectives, such as the quality and quantity of product produced, has been left in the hands of the human operators in the past (King & Mamdani, 1997).
852
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
The research and development of math-model-based modern control techniques have been significantly progressing in both theoretical and practical aspects in the past two decades. However, it is still difficult to design and implement a real-time optimization control for some complex industrial processes, if they are highly nonlinear, high dimensional, seriously coupled, and significantly uncertain. It has been an attractive area in the control industry to explore the novel control strategies, which can be designed and implemented with limited deep (process principle and mathematical) knowledge of the controlled environment. With the rapid progresses of the studies on fuzzy logic, neural networks, genetic algorithms, and rule-based expert systems, the intelligent system techniques with remarkable capabilities of dealing with system’s imprecise and/or incomplete information and knowledge have been recognized as one of the most important measures in solving complex industrial control problems. As noted by Kosko that adaptive math-model free estimation is an intelligent system that adaptively estimates continuous function from data without specifying mathematically how outputs depend on inputs (Lu et al., 1997). In conventional hard controllers, the knowledge, which is pertinent to the process being controlled, and the methods for using this knowledge are interrelated and can be expressed in analytical form. To apply such techniques, explicit knowledge of the microscopic behaviour of the process is essential. In contrast, in intelligent or soft control, there is a clear demarcation between the knowledge and the information about the process dynamics, i.e., the macroscopic behaviour of our process, is not essential (Manesis, Sapidis, & King, 1998). Modern control techniques, e.g., parameter estimation, stochastic control, and optimal control are used in either model identification or design of math able control law. However, some industrial processes are too complicated to be modelled and/or controlled by math-algorithms, because they are highly nonlinear and significantly uncertain with unknown structure and imprecise information (Lu, 1996). Intelligent control is therefore particularly attractive when the expertise to control a process is available in the form of linguistic rules acquired from normal operational experience. A fundamental attribute of intelligent control is its ability to work with symbolic, inexact and vague data that human operators comprehend best. Indeed, its ability to deal with incomplete and ill-defined information, an inherent characteristic of wastewater treatment plants, permits implementation of human-like control strategies which have hitherto defied solution by any of the conventional hard control techniques. Certainly for processes, which are known microscopically, hard control is clearly the
methodology to be preferred. Modern technology techniques have, however, in general failed to solve industrial problems and the thousands of industrial plants worldwide which rely on three-term (PID) controllers attest to the limitations of these techniques. Fuzzy logic and artificial neural networks are two examples of soft computing, which have migrated into the realm of industrial control over the last two decades. Chronologically, fuzzy control was the first and its application in the process industry has led to significant improvements in product quality, productivity and energy consumption. Fuzzy control is now firmly established as one of the leading advanced control techniques in use in industry (Manesis et al., 1998). 3.1. Fuzzy control A common definition of a fuzzy control system is that it is a system which emulates a human expert. In this situation the knowledge of the human operator would be put in the form of a set of fuzzy linguistic rules. These rules would produce an approximate decision, just as a human would. Consider Fig. 1 where a block diagram of this definition is shown. As seen, the human operator observes quantifies by making observation of the inputs, i.e., reading a meter or measuring a chart, and performs a definite action, e.g., pushes a knob, turns on a switch, close, a gate, or replaces a fuse, i.e., leading to a crisp action, shown here by the output variable y(t). The human operator can be replaced by a combination of a fuzzy rule-based system (FRBS) and a block called defuzzifier. The sensory (crisp or numerical) data is fed into FRBS where physical quantities are represented or compressed into logistic variables with appropriate membership functions. These linguistic variables are then used in the “antecedents” (IF-Part) of a set of fuzzy rules with an inference engine to result in a new set of fuzzy linguistic variables or “consequents” (THEN-Part) as variables are then denoted in this figure by z, combined and changed to a crisp (numerical) output y* (t) which represents an approximation to actual output y(t) (Jamshidi, 1997). 4. FCC process modelling and control application: FUZZYFCC The FCC Unit (Plant 7) is a UOP (an international supplier and licensor of process technologies, catalysts, adsorbents, process plants to the petroleum refining industries) design (FCC, 2001). The main parts of the FCCU are catalyst and fractionator. Catalyst circulation takes place in the catalyst part. The catalyst
Fig. 1. A conceptual definition of a fuzzy control system (Jamshidi, 1997).
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
853
Fig. 2. Schematic description of the FCC Unit-Plant 7.
consists of reactor and regenerator units which are connected with each other with riser and stand-pipe. Catalyst in the catalyst circulation rises from riser to reactor and from there flows to regenerator by stripper and stand-pipe and then rises up again from riser to reactor. Fig. 2 illustrates a schematic drawing of FCC unit-Plant 7 (An FCC plant of Turkish refinery unit). Some detailed information about FCCU operating system of our FCC is also given in introduction.
4.1. Database for FCCU-Plant 7 The database for the FCCU-Plant 7 is obtained from the reports that are produced periodically (once a week) after the operations. Fig. 3 shows an example of those reports. Each input and output variables and their corresponding data used in this study were taken from the reports produced in the year 2000 and 2001. From all of the reports of the selected
Fig. 3. Sample summary report of the FCCU-Plant 7.
854
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
period, only 22 reports were selected that represent relatively better operating conditions in order to obtain data. Table 1 shows input (independent) and output (dependent) variables and their values for corresponding period. 4.2. FCC modelling FCCU operating variables have been termed independent and dependent variables and all these variables are directly controlled usually with a classical control system (PID control). Fig. 4 lists the major variables in a FCCU. Those variables are going to be used in this study as input and output variables. Among all of the factors affecting the FCC unit 10 input and 10 output variables are selected to examine since they are the most significant ones. 4.3. Knowledgebase for FCC-Plant 7 The knowledgebase in Table 2 is obtained from FCC UOP Process-Plant 7 Reference Manual of the selected unit (FCC UOP Process, 2000). The knowledgebase is formed in two columns in which the left one is for reactor part of the unit and the column at the right is for regenerator part of the unit. This knowledgebase is used in order to obtain fuzzy rules. 4.4. Clustering of input and output variable values Intervals and linguistic variables of values for selected inputs and outputs of the FCCU are obtained by using NCSS (Number Cruncher Statistical System) software. NCSS is statistical software and has clustering feature. In this study NCSS software is used for clustering data of related input and output variables under three clusters (L, low; M, medium; H, high) and for iden-
tifying linguistic variables’ intervals. Linguistic variables and their intervals are shown in Table 3. According to the yields obtained form clustering with NCSS software, membership functions are drawn in Matlab-Fuzzy Logic Toolbox. Fig. 5 displays screens of Membership Function Editor giving the intervals of selected input and output variables used in this study. Membership functions of the variables are drawn (Fig. 5) after all input and output variables are entered through the FIS Editor interface. Fig. 6 shows FIS Editor Screen in which the FCCU input and output variables can be seen. 4.5. Fuzzy rules of the FCCU After membership functions are determined, the inference is performed in accordance with 59 fuzzy linguistic rules (see Fig. 7). Those rules are obtained from the knowledgebase of the FCC Unit. Some other rules are also included heuristically in terms of comparing output values in accordance with input values. All the rules, entered to the system through the Matlab-Fuzzy Logic Toolbox Rule Editor are given in Table 4. Number of rules given to the system as fuzzy rules was restricted such that only the rules affecting the FCC process were taken into account. After giving the rules to the system, defuzzified results and graphical outputs can be derived. Fig. 8(a and b) and Fig. 9 illustrate an example of Surface Viewer screen obtained from Fuzzy Logic Toolbox. Two- or three-dimensional graphic results of variables can be plotted and compared. Fig. 10 shows the results of applied rules and their corresponding outputs according to the mass center of variables. Using the interface, defuzzified values for output variables can be derived changing input values manually. Different output val-
Fig. 4. FCC Unit input and output variables.
Table 1 Database for the FCC Unit variables Independent variables
49829 53289 48888 52963 58999 57617 59174 58136 57145 58001 58001 58394 57904 58295 58973 56696 51764 53791 50957 49654 40977 51303
CFeed
CCR
TFR
CTOR
FFR
CSS
TRG
CBL
Spec gravity
GOC
COC
Gasoline
LPG
Coke
CO
TRC
Pressure
RFR
CO2 /CO
266 223 241 247 249 255 230 217 234 221 178 209 212 216 194 244 243 226 223 246 246 238
14.09 11.31 13.78 14.16 16.08 14.37 15.76 14.41 15.45 15.23 16.00 16.15 15.05 15.11 15.03 14.95 12.85 16.02 16.31 14.42 11.24 12.81
2417 2474 2441 2372 2465 2374 2016 2248 2417 2396 2404 2389 2483 2505 2435 2439 2494 2554 2422 2459 1952 2500
9.18 7.16 8.91 9.29 10.35 9.40 12.21 10.10 10.25 10.30 10.77 10.54 9.76 9.77 9.91 9.86 8.44 10.14 10.72 9.46 9.60 8.12
2271 2237 2254 2206 2294 2228 1844 2059 2207 2175 2274 2247 2273 2277 2283 2293 2266 2326 2253 2276 1724 2272
1375 1680 1575 1500 1575 1788 925 1400 1300 1125 975 1175 1450 1825 1863 1625 1575 1425 1450 1350 1650 1500
658 664 656 670 660 686 667 669 661 662 654 660 661 659 656 660 656 661 663 647 633 657
2.16 1.64 2.33 2.44 2.33 2.16 2.30 2.66 0.76 2.47 2.33 1.99 3.24 3.14 2.93 3.00 1.42 1.44 1.44 1.81 1.82 1.76
0.758 0.726 0.664 0.611 0.656 0.716 0.702 0.670 0.677 0.660 0.665 0.711 0.701 0.733 0.738 0.669 0.701 0.704 0.725 0.736 0.720 0.686
79.95 78.08 76.83 78.01 80.93 76.72 84.40 81.07 85.55 87.33 85.27 72.97 81.72 85.65 80.15 83.65 89.81 88.02 81.73 83.23 92.59 76.73
0.098 0.107 0.092 0.077 0.096 0.094 0.077 0.098 0.090 0.078 0.089 0.121 0.092 0.076 0.092 0.091 0.066 0.075 0.073 0.074 0.095 0.107
49.62 53.74 52.01 47.60 50.13 46.77 51.65 49.08 58.01 53.82 53.32 45.80 50.76 53.47 51.51 52.03 58.65 56.92 51.13 51.21 53.71 50.09
20.44 13.82 15.22 19.56 20.91 19.47 19.98 22.09 19.85 25.20 22.07 16.79 20.53 22.47 20.56 22.00 21.30 21.57 18.78 22.76 27.12 17.21
5.65 6.59 5.12 5.08 6.31 5.76 7.26 6.42 5.97 6.45 5.54 5.89 5.72 5.87 5.96 6.06 5.54 5.17 4.93 4.46 6.91 5.45
6.6 8.4 5.0 5.0 5.0 4.8 4.4 4.8 4.0 5.0 3.2 4.8 3.6 3.8 3.6 3.8 3.8 3.6 2.8 2.0 8.2 5.0
535 520 526 534 530 537 536 531 530 532 525 534 525 524 521 531 517 517 525 520 519 516
2.10 2.20 2.10 2.15 2.15 2.00 1.72 1.90 2.10 2.20 2.10 2.15 2.20 2.25 2.30 2.05 2.15 2.17 2.15 2.00 1.65 2.20
146 237 187 166 171 146 172 189 210 221 130 141 210 228 153 146 228 228 169 182 228 228
1.67 1.30 2.12 1.76 1.76 2.13 2.50 2.20 2.80 2.20 3.40 2.20 3.10 2.90 3.30 3.40 3.20 2.90 4.10 5.50 1.30 2.20
ATR (m3 /h), air to regenerator; GOC (wt.%), gas oil conversion; CFeed (C), combined feed temperature; COC, coke on catalyst; CCR (t/min), catalyst circulation rate; gasoline (wt.%), gasoline; TFR (m3 /d), total feed rate; LPG (wt.%), liquefied petroleum gas; CTOR (t/t), catalyst to oil ratio; coke (wt.%), coke; FFR (m3 /d), fresh feed rate; CO (vol.%), CO in flue gas; CSS (kg/h), catalyst stripping steam; TRC (C), riser outlet (temperature of reactor); TRG (C), temperature of regenerator; pressure (kg/cm2 ), pressure of reactor; CBL (m), catalyst bed level; RFR (m3 /d), recycle feed rate; spec gravity, specific gravity; CO2 /CO (mol/mol), CO2 /CO in flue gas.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
ATR
Dependent variables
855
856
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 5. Membership functions of the input and output variables.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
857
Table 2 Knowledgebase of the FCCU-Plant 7 Reactor
Regenerator Low regenerator temperature increases conversion and coke formation
Feed rate Effects of 0.1 step increase of combined feed rate are Gasoline yield increases Conversion increases 1–2% Coke formation increases 0.3%
Feed specifications If API gravity is increases then Conversion rate increases Gasoline conversion rate increases
Feed temperature If total feed temperature is increased 38 ◦ C then Conversion decreases by 0.4% Regenerator temperature increases 11 ◦ C Coke formation decreases 0.5% Catalyst/oil ratio decreases
Regenerator bed level Regenerator productivity is better when bed level is higher, since catalyst residence time would be more. Low bed level decreases oxygen usage efficiency
Recycle composition and temperature Increasing recycle temperature has the same effect of increasing feed temperature Reactor temperature If reactor temperature is increased then conversion rate, gasoline and LPG conversion increases Reactor level The effects of the raising reactor level Conversion rate increases Gasoline conversion rate decreases for fixed conversion Coke formation increases Reactor pressure The effects of raising reactor pressure Conversion rate increases Regenerator temperature increases
Catalyst circulation rate Increasing catalyst circulation rate affects regeneration negatively Air to regenerator If air to regenerator is not enough then coke on catalyst increases Bed temperature When decreasing catalyst/oil ratio coke formation decreases. Process variables increasing regenerator temperature are Increasing specific gravity of feed Decreasing feed quality (high carbon level) Increasing total feed temperature Increasing reactor bed level Increasing reactor temperature Increasing reactor pressure
Stripping steam Increasing stripping steam occurs if Recycle feed increases Reactor pressure increases Total feed temperature is higher Reactor temperature is lower Catalyst/oil ratio Catalyst/oil increases when Reactor temperature is increased Regenerator temperature is increased Combined feed temperature is decreased If catalyst/oil ratio is increased then conversion and coke formation increases
Fig. 6. Fuzzy inference system (FIS) editor screen (inputs and outputs of the FCCU).
858
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 7. Rule editor of Matlab-Fuzzy Logic Tool Box.
Fig. 8. (a) Gasoline graphic according to CFeed and TFR interaction and (b) gasoline graphic according to TRG and CBL interaction.
ues can be get trough the Rule Viewer according to the given input values. To get defuzzified output values for all the real input values is not flexible using the interface. For that reason a program is written using Matlab codes to drive defuzzified output results in accordance with real input values. Fig. 11 gives a Table 3 Linguistic variables’ intervals of inputs and outputs for the FCCU Variable
Low
Medium
High
ATR (m3 /h) CFeed (C) CCR (t/min) TFR (m3 /d) CTOR (t/t) FFR (m3 /d) CSS (kg/h) TRG (C) CBL (m) Spec gravity GOC (wt.%) COC Gasoline (wt.%) LPG (wt.%) Coke (wt.%) CO (vol.%) TRC (C) Pressure (kg/cm) RFR (m3 /d) CO2 /CO (mol/mol)
40977–52963 178–217 11.24–14.42 1952–2396 7.16–9.77 1724–2207 925–1400 633–660 0.76–2.33 0.611–0.701 72.97–80.93 0.659–0.922 45.80–51.65 13.82–20.44 4.46–5.72 2.00–5.00 516–525 1.65–2.10 130–172 1.30–2.13
49654–57617 194–243 12.50–15.76 2016–2483 8.9–10.45 1844–2277 1125–1680 656–669 1.64–3.00 0.665–0.733 76.83–86.00 0.0762–0.1072 47.60–56.92 16.79–22.47 5.12–6.45 3.20–8.20 519–533 1.72–2.20 146–221 1.76–3.10
52963–59174 223–266 14.75–16.31 2404–2554 9.77–12.21 2215–2326 1425–1863 661–686 2.44–3.24 0.704–0.758 81.73–92.59 0.922–0.1210 51.65–58.65 20.44–27.12 5.89–7.26 5.00–8.40 525–537 2.10–2.30 182–237 2.20–5.50
comparison between real values and defuzzified values of output variables. Table 5 is also gives a comparison between real output values and derived defuzzified output values for each period. The last three rows of the table give average, maximum and minimum values of both real and defuzzified values. Each application steps were obtained using Matlab-Fuzzy Logic Toolbox version 6.5. Comparison of average for defuzzified results with average for real values is shown in Table 6.
Fig. 9. LPG graphic according to TRG and ATR interaction.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
859
Table 4 Fuzzy rules of the FCCU entered to the system 1. If (TFR is L) then (GOC is H) (COC is M) (LPG is L) (RFR is M) 2. If (TFR is M) then (COC is M) (gasoline is L) (LPG is M) (RFR is L) 3. If (TFR is H) then (GOC is H) (COC is L) (gasoline is H) (LPG is H) 4. If (TFR is H) and (FFR is L) then (RFR is H) 5. If (FFR is L) then (GOC is H) (pressure(reac) is L) 6. If (FFR is H) then (GOC is L) (TRC is M) (pressure(reac) is M) 7. If (CFeed is H) then (GOC is L) (gasoline is L) (coke is M) (TRC is H) (pressure(reac) is M) (RFR is H) 8. If (CFeed is M) then (GOC is M) (coke is M) (pressure(reac) is H) (RFR is H) 9. If (CFeed is L) then (GOC is M) (gasoline is H) (pressure(reac) is H) (RFR is L) 10. If (CBL(reac) is H) then (GOC is M) (gasoline is L) (coke is M) 11. If (CBL(reac) is M) then (GOC is M) 12. If (CBL(reac) is L) then (GOC is L) (gasoline is H) (coke is M) 13. If (CSS is H) then (RFR is H) 14. If (CSS is H) then (pressure(reac) is H) 15. If (CSS is H) then (TRC is M) 16. If (CTOR is H) then (TRC is H) 17. If (CTOR is L) then (coke is M) 18. If (CTOR is H) then (GOC is H) 19. If (TRG is H) then (CO is M) (CO2 /CO is L) 20. If (TRG is H) then (pressure(reac) is H) 21. If (TRG is L) then (GOC is L) (gasoline is L) (LPG is M) 22. If (TRG is M) then (GOC is H) (gasoline is H) (LPG is M) 23. If (TRG is H) then (GOC is L) (gasoline is L) (LPG is M) 24. If (TRG is H) then (coke is M) (RFR is H) 25. If (TRG is H) then (TRC is H) 26. If (CCR is H) then (CO is L) (CO2 /CO is H) 27. If (CCR is H) then (coke is H) 28. If (ATR is H) then (CO is M) (CO2 /CO is M) 29. If (ATR is L) then (COC is M)
30. If (ATR is H) then (coke is H) 31. If (spec gravity is H) then (GOC is H) (gasoline is M) 32. If (spec gravity is L) then (CO is H) (CO2 /CO is L) 33. If (spec gravity is H) then (coke is H) (RFR is H) 34. If (spec gravity is H) then (TRC is L) (pressure(reac) is M) 35. If (CTOR is M) then (coke is M) 36. If (TRG is L) then (coke is L) 37. If (CCR is L) then (coke is L) 38. If (CCR is M) then (coke is M) 39. If (ATR is M) then (coke is L) 40. If (CTOR is H) then (coke is H) 41. If (TRG is L) then (coke is L) 42. If (ATR is M) then (COC is L) 43. If (TRG is M) then (CO is M) 44. If (TRG is L) then (CO is L) 45. If (CCR is M) then (CO is L) 46. If (CCR is L) then (CO is H) 47. If (CSS is M) then (TRC is M) 48. If (TRG is L) then (TRC is L) 49. If (FFR is M) then (pressure(reac) is M) 50. If (TRG is L) then (pressure(reac) is M) 51. If (TRG is M) then (pressure(reac) is H) 52. If (CSS is M) then (RFR is H) 53. If (CSS is L) then (RFR is L) 54. If (TRG is M) then (RFR is H) 55. If (spec gravity is L) then (RFR is L) 56. If (TRG is M) then (CO2 /CO is M) 57. If (CCR is L) then (CO2 /CO is L) 58. If (ATR is M) then (CO2 /CO is M) 59. If (ATR is L) then (CO2 /CO is L)
Fig. 10. Rule viewer screen to obtain defuzzified results for the FCCU.
860
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Fig. 11. Defuzzified output values compared with real output values.
Deviation column is calculated by real value minus defuzzified value. Deviation ratio is similarly calculated by dividing deviation value to corresponding average of real data. As can be seen from Table 6, the minimum deviation ratio is 0.0006 and the maximum deviation ratio is (ignoring the negativity) 0.08 which shows that derived defuzzified results are not so different form the real data. 5. Results and discussions There is no general and mathematically optimal solution for the control if plant (unit) to be controlled is complex and highly nonlinear. The modelling of the process and its solution become even more difficult if a sufficiently precise model is unknown or cannot be identified. It is well known however that in many cases a human can master the performance of such a plant using linguistic control algorithms that represent the operator knowledge and experience about the plant/unit by using If-Then rules. First, we quantified the qualitative character of linguistic rules and represented real data as fuzzy modelling by using fuzzy logic software (Matlab-Fuzzy Logic Toolbox, version 6.5). In the next stage, a development and application of dynamic model-based optimization and fuzzy control were performed. The idea behind this kind of fuzzy control has been to take the respective advantages in math-based control, especially in chemical process such as refinery processes-catalyst cracking, hydro-cracking, and distillation. FCCU control of the plant we examined is carried out by a program (PID control) which is based on consultancy of the
experts or experienced engineers/operators. However, this control system used in that plant is not fully integrated with other units (no real-time, on-time control) and sometimes manual interventions are required by the operator when the plant operations are disturbed. Time and product losses are caused by this intervention during the cracking process. The most repeated results were selected as data set (the most optimal values are derived from weekly summary report between January 2001 and December 2001) from the empirical FCCU data to appoint to use as rules (rules are obtained from knowledgebase of the FCCU-Plant 7 reference manual). Although some variables were deviated from the optimal (average) values, most of the fuzzy results obtained from this study were found to be acceptable and reasonable level of compatibility were achieved compared to the empirical FCCU data and targeted gasoline and LPG conversion rate. Deviation of output results such as COC and CO (vol.%) could be rectified and optimized by the further research on flue gas analysis and catalyst contents. However, GOC (wt.%), Gasoline (wt.%), LPG (wt.%) and other outputs/results were acceptable and in good agreement with optimal values. In this work, the results obtained using fuzzy control is found to be very closer to the plant’s real results. This study can be extended using long time period, data different raw oil contents—gravity, viscosity and conradson carbon, UOP K factor, Gravity (API) and Octan RON and the soft computing algorithms, and AI techniques such as neural nets, genetic algorithm, and chaos theory. Another approach may be a comparison between PID control, optimal control and neuro-fuzzy control.
Table 5 Comparison of real (crisp) and defuzzified output variables of the FCCU-Plant 7 GOC
Gasoline
LPG
Coke
CO
TRC
Pressure
RFR
CO2 /CO
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
Real
Defuzzy
79.95 78.08 76.83 78.01 80.93 76.72 84.40 81.07 85.55 87.33 85.27 72.97 81.72 85.65 80.15 83.65 89.81 88.02 81.73 83.23 92.59 76.73
82.799 82.942 81.325 78.480 82.883 81.262 83.367 81.377 83.546 85.434 82.839 84.139 82.992 82.942 82.942 82.552 83.282 82.954 82.937 83.259 83.255 83.349
0.0984 0.1072 0.0922 0.0772 0.0958 0.0944 0.0770 0.0980 0.0897 0.0783 0.0888 0.1210 0.0920 0.0762 0.0916 0.0911 0.0659 0.0747 0.0730 0.0741 0.0952 0.1071
0.0903 0.0775 0.0861 0.0836 0.0789 0.0916 0.0918 0.0917 0.0903 0.0916 0.0913 0.0917 0.0767 0.0761 0.0871 0.0860 0.0806 0.0762 0.0857 0.0837 0.0918 0.0821
49.62 53.74 52.01 47.60 50.13 46.77 51.65 49.08 58.01 53.82 53.32 45.80 50.76 53.47 51.51 52.03 58.65 56.92 51.13 51.21 53.71 50.09
51.1112 53.3729 51.2217 48.0296 52.4325 49.6041 50.4656 48.0724 52.4982 52.3077 52.4982 52.1430 52.4982 52.3389 52.3389 52.1514 52.4982 55.6235 52.3481 52.1517 51.8575 52.7802
20.44 13.82 15.22 19.56 20.91 19.47 19.98 22.09 19.85 25.20 22.07 16.79 20.53 22.47 20.56 22.00 21.30 21.57 18.78 22.76 27.12 17.21
20.4931 22.8234 21.9529 19.6418 22.6798 19.6493 17.3286 18.8455 20.4931 19.8108 19.9737 19.7313 22.5910 22.8520 21.6090 21.9450 22.5923 22.5527 20.7832 22.6886 16.8690 22.7216
5.65 6.59 5.12 5.08 6.31 5.76 7.26 6.42 5.97 6.45 5.54 5.89 5.72 5.87 5.96 6.06 5.54 5.17 4.93 4.46 6.91 5.45
6.0101 5.8666 5.3002 5.3269 6.3288 6.3072 6.3599 6.3090 6.3372 6.3253 5.8966 6.3744 6.3236 6.1816 5.8666 6.2774 5.3088 5.8740 6.0336 5.9332 5.7936 5.3204
6.6 8.4 5.0 5.0 5.0 4.8 4.4 4.8 4.0 5.0 3.2 4.8 3.6 3.8 3.6 3.8 3.8 3.6 2.8 2.0 8.2 5.0
4.3554 6.2555 5.1816 5.1353 5.2050 4.5518 4.5661 5.2050 5.5099 5.7027 5.2050 4.5428 4.7839 4.8346 4.5430 5.3281 5.0248 4.5549 4.5168 3.5129 5.8863 5.1309
535 520 526 534 530 537 536 531 530 532 525 534 525 524 521 531 517 517 525 520 519 516
526.3274 524.5537 526.3728 529.3402 529.2806 527.6673 532.3396 529.2452 529.0842 532.0483 526.3342 529.1735 525.6386 522.7887 523.7430 529.2543 526.3315 528.2685 526.2901 526.3158 526.7578 526.5170
2.10 2.20 2.10 2.15 2.15 2.00 1.72 1.90 2.10 2.20 2.10 2.15 2.20 2.25 2.30 2.05 2.15 2.17 2.15 2.00 1.65 2.20
2.0463 2.0774 2.0608 2.0772 2.0745 2.0773 1.9545 2.0362 2.0774 2.0709 2.0774 2.0784 2.0774 2.0774 2.0774 2.0772 2.0608 2.0783 2.0774 2.0072 1.9726 2.0270
146 237 187 166 171 146 172 189 210 221 130 141 210 228 153 146 228 228 169 182 228 228
184.162 213.307 183.947 184.439 184.431 184.103 180.111 184.155 184.428 184.427 146.867 181.410 212.476 218.796 184.428 184.428 212.476 214.938 184.534 203.638 197.492 190.842
1.67 1.30 2.12 1.76 1.76 2.13 2.50 2.20 2.80 2.20 3.40 2.20 3.10 2.90 3.30 3.40 3.20 2.90 4.10 5.50 1.30 2.20
1.9240 1.9958 1.6042 2.0014 3.3766 2.1348 3.4365 1.9972 3.1504 2.6454 3.3763 3.7024 2.6966 2.8739 2.7073 2.1547 2.0570 3.6884 3.4902 1.6238 1.6038 2.0454
82.29 92.59 72.97
82.766 85.434 78.480
0.0890 0.1210 0.0659
0.0856 0.0918 0.0761
51.87 58.65 45.80
51.8338 55.6235 48.0296
20.44 27.12 13.82
20.9376 22.8520 16.8690
5.82 7.26 4.46
5.9843 6.3744 5.3002
4.60 8.40 2.00
4.9787 6.2555 3.5129
526.6 537.0 516.0
527.4397 532.3396 522.7887
2.09 2.30 1.65
2.0564 2.0784 1.9545
187.1 237.0 130.0
190.902 218.796 146.867
2.63 5.50 1.30
2.5585 3.7024 1.6038
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Avrg Max Min
COC
861
862
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863
Table 6 Comparison between averages of defuzzified and real values of the outputs and deviations between them Output
Variation intervals of empirical data (optimal range)
Average of real data
Average of defuzzified results
Deviation (real-defuzzy)
Deviation ratio
GOC (wt.%) COC Gasoline (wt.%) LPG (wt.%) Coke (wt.%) CO (vol.%) TRC (C) Pressure (kg/cm) RFR (m3 /d) CO2 /CO (mol/mol)
72.97–92.59 0.0659–0.121 45.8–58.65 13.82–27.12 4.46–7.26 2.00–8.40 516–537 1.65–2.30 130–237 1.3–5.5
82.290 0.089 51.865 20.441 5.823 4.600 526.591 2.090 187.091 2.634
82.766 0.086 51.834 20.938 5.984 4.979 527.440 2.056 190.902 2.558
−0.4757 0.0035 0.0312 −0.4967 −0.1611 −0.3787 −0.8487 0.0340 −3.8107 0.0752
−0.00578 0.03909 0.00060 −0.02430 −0.02767 −0.08233 −0.00161 0.01629 −0.02037 0.02855
Finally our goal is to achieve such comparison in the further research paper.
Excellence (Contract No. 500273). http://ww.iproms.org .
6. Conclusion The following conclusions can be drawn from the application of fuzzy logic to the control of the FCC as described in this paper: • A fuzzy logic-based control system was developed to estimate the variables in a FCCU. By careful selection of the input variables (here 10 variables) and designing the rules (here 30 rules) for the system and their statistical analysis, 97.53% of control accuracy can be obtained. • The results were generally in compliance with empirical FCCU data. However, in most cases even though limited of rules and inputs were applied the results obtained indicated a very high accuracy. This clearly shows that by increasing the number of inputs (for example advanced lab data about raw oil-gravity, viscosity and conradson carbon and fractionatortop temperature and bottom temperature) by increasing data for each variables for example by extending yearly data and periods, and by improving the rules used in the MATLAB package, more proper and accurate results could be expected. • Since the assumptions and very high nonlinear behaviours of the cracking reaction, there could be a deviation between the results obtained in this study and the FCCU-Plant 7’s empirical values. • As a final remark fuzzy logic is a promising control technique and would be effectively used for improved process control of the FCCU and other units in the petroleum refinery production processes. The study will be continued to increase the effectiveness of the proposed model by increasing and manipulating the content of the rules, data and variables (sensitivity and stability analysis) and comparison with other classical and intelligent techniques. Acknowledgement Sakarya University is a partner of the Innovative Production Machines and Systems (I*PROMS) Network of
References Abasaeed, A. E., & Elnashaie, S. S. E. H. (1998). On the chaotic behavior of externally forced industrial fluid catalytic cracking units. Chaos, Solutions & Fractals, 9(3), 455–470. Adebiyi, O. A., & Corripio, A. B. (2003). Dynamic neural networks partial least squares (DNNDLS) identification of multivariable processes. Computers & Chemical Engineering, 27(2), 143–155. Alaradi, A. A., & Rohani, S. (2002). Identification and control of a riser-type FCC unit using neural networks. Computers & Chemical Engineering, 26(3), 401–421. Alhumaizi, K., & Elnashaie, S. S. E. H. (1997). Effect of control loop configuration on the bifurcation behaviour and gasoline yield of industrial fluid catalytic cracking (FCC) units. Mathematical and Computer Modelling, 25(3), 37–56. Ali, E. E., & Elnashaie, S. S. E. H. (1997). Nonlinear model predictive control of industrial type IV fluid catalytic (FCC) units for maximum gasoline yeald. Industrial & Engineering Chemistry Research, 36(12), 389– 398. Azeem, M. F., Ahmad, N., & Hanmandlu, M. (in press). Fuzzy modeling of fluidized catalytic cracking unit. Applied Soft Computing [Corrected proof, available online September 26, 2005]. Bollas, G. M., Papadokonstadakis, S., Michalopoulos, J., Arampatzis, G., Lappas, A. A., Vasalos, I. A., et al. (2003). Using hybrid neural networks in scaling up an FCC model from a pilot plant to an industrial unit. Chemical Engineering and Processing, 42(8–9), 697–713. Chitnis, U. K., & Corripio, A. B. (1998). Online optimization of a models fluid catalytic cracking unit. ISA Transactions, 37(3), 215–226. Chunyang, J., Sohrab, R., & Azrthur, J. (2003). FCC unit modeling, identification and model predictive control, a simulation study. Chemical Engineering and Processing, 42(4), 311–325. Cristea, M. V., Agachi, S. P., & Marinoiu, V. (2003). Simulation and model predictive control of a UOP fluid catalytic cracking unit. Chemical Engineering and Processing, 42(2), 67–91. FCC (2001). Plant 7 The Year 2001 Data in a Turkish Refinery Unit. FCC UOP Process (2000). Plant 7 Reference Manual in a Turkish Refinery Unit. Han, I. S., & Chung, C. B. (2001). Dynamic modeling and simulation of a fluidized catalytic cracking process. Part I. Process modelling. Chemical Engineering Science, 56(5), 1951–1971. Jamshidi, M. (1997). Large-scale systems: Modeling and control with fuzzy logic. New Jersey: Prentice-Hall. King, P. J., & Mamdani, E. H. (1997). The application of fuzzy control system to industrial processes. Automatica, 13(3), 235–242.
H. Ta¸skin et al. / Computers and Chemical Engineering 30 (2006) 850–863 Loeblin, C., & Perkins, J. D. (1999). Structural design for on-line process optimisation. II. Application to a simulated FCC. AIChE Journal, 45(5), 1030–1040. Lu, Y. Z. (1996). Industrial intelligent control. New York: John-Wiley and Sons. Lu, Y. Z., He, M., & Xu, C. W. (1997). Fuzzy modeling and expert optimization control for industrial processes. IEEE Transaction on Control Systems Technology, 5(1), 2–11. Manesis, S. A., Sapidis, D. J., & King, R. E. (1998). Intelligent control of wastewater treatment plants. Artificial Intelligence in Engineering, 12(3), 275–281. Ramirez, J. A., Aguilar, R., & Isunza, F. L. (1996). Robust regulation of temperature in reactor–regenerator fluid catalytic cracking units. Industrial & Engineering Chemistry Research, 35, 1652–1659. Ramirez, J. A., Valencia, J., & Puebla, H. (2004). Multivariable control configurations for composition regulation in a fluid catalytic cracking unit. Chemical Engineering Journal, 99(3), 187–201. Sarma, P., & Rengaswamy, R. (2000). Multivariable gain-scheduled fuzzy logic control of a fluidized catalytic cracker unit. Computers & Chemical Engineering, 24(2–7), 1083–1089. Sebzalli, Y. U., & Wang, X. Z. (2001). Knowledge discovery from process operational data using PCA and fuzzy clustering. Engineering Applications of Artificial Intelligence, 14(5), 607–616.
863
Shoureshi, R. (1993). Intelligent control systems: Are they for real? Transactions of the ASME, 115, 392–400. Skogestad, S. (2000). Plantwide control: The search for the self-optimizing control structure. Journal of Process Control, 10(5), 487–507. Taskin, H. (1983). Process modeling, optimization and control: An application to a petroleum refinery unit (FCCU). Ph.D. Thesis [Turkish], Istanbul Technical University, Institute of Science and Technology, Faculty of Management. Yescas, R. M. Y., & Isunza, F. L. (1997). Comparison of two dynamic models for FCC units. Catalysis Today, 38(1), 137–147. Yescas, R. M. Y., Bogle, D., & Isunza, F. L. (1998). Approach to the analysis of the dynamics of industrial FCC units. Journal of Processes Control, 8(2), 89–100. Zanin, A. C., de Gouvea, M. T., & Odlak, D. (2001). Implementation of a real time optimization strategy on the FCC catalytic converter. Latin American Applied Research, 31, 301–308. Zeydan, M. (1999). Fuzzy modeling and control in a petroleum refinery unit (FCC). Ph.D. Thesis [Turkish], Sakarya University, Institute of Science and Technology, Department of Industrial Engineering [Superviser, Tas¸kin, H.].
