Multivariable Quality Control of a Crude Oil Fractionator

Copyright © IFAC Dynami cs and Control of Chemical Reactors (DYCORD+'9S), Copenhagen. Denmark, 1995 MULTIVARlABLE QUALITY CONTROL OF A CRUDE OIL FRACTIONATOR M. V. de Oliveira Magalblies 1 and Darci Odloak2 1- PETROBRAS. Industrial Department. Rio de Janeiro. Brazil 2- Universidade de Sao Paulo. Chemical Eng. Department. Sao Paulo. Brazil Abstract This paper describes the industrial application of a multivariable predictive controller to a typical crude oil fractionator where jet fuel and diesel fuel are the main specified products. The controller functional specification includes the main targets of the column operation, accounting for 14 controlled variables. 6 manipulated variables and 2 disturbances. The implemented algorithm is a linear DMC. that makes use of a linear programming routine specifically developed to approach the problem of bounded variables, since the system variables are assumed to be controlled by range. The controller was successfully implemented in one of the Petrobras refineries at Paulinia (Brazil) and some practical results are presented here. Keywords-Multivariable control. predictive control. distillation columns. model based control, quality control. INTRODUCTION sophisticated with the control functions distributed in three different layers. The system structure is very similar to the one considered in this paper. In the control level. the main difference concerns the manipulation of the set-point of the overhead temperature controller instead of the reflux flow as done in this work. This single difference substantially affects the dynamic response of the whole system. The manipulation of the overhead temperature partially decouples the controlled and manipulated variables of the system. As can be seen from the transfer functions given by Muske et al ,( 1991), in open loop. the yield of a side draw does not practically affect the cut -points of the products above that draw and a triangular structure is obtained. This can be explained, since any change on the flow of the side draws is compensated by the overhead temperature controller that changes the reflux flow and tends to keep the ratio LN constant. The disadvantage of this control configuration is that the overhead temperature controller can not work at its saturation limit. If for instance the reflux valve is wide open. as can occurs when the unit is operating at its maximum feed flow, the whole set of transfer functions of the system are no longer valid and the multivariable controller may become unstable. In this work the predictive controller The atmospheric fractionator is an important equipment in a crude unit, because of the large feed flow and the amount of energy consumed. The products specifications basically follow the market demands and the economic objective of the crude unit is to keep the products the closest as possible to its quality specifications. To exceed the specifications is less critical than to violate them. Any violation on the product specifications leads to an expensive reprocessing or to degrade the stream to a less valuable pool. The experienced operator keeps a safety margin from the market specifications to prevent any violations during transient conditions. The frequent change of the feed quality and flow is the major incentive for the application of advanced control on crude units. Benefits as the minimization of off-spec products and the energy consumption in the oil heaters can amount to millions of dollars a year. The application of predictive multivariable controllers to crude fractionators has already been reported in the literature. Muske et al.(l991) describes the application of the IDCOM-M package to a crude tower of the refinery of Sakai in Japan. The proposed control configuration is very 469 manipulates the reflux flow and the overhead temperature is one of the variables of the system that are controlled by range. In particular this variable has to be kept above a minimum value to prevent the condensation of the stripping steam inside the tower. Following this configuration. in open loop. the controlled variables are affected by all the manipulated variables and the triangular structure is lost, but we find a more general approach to manipulate the reflu.x flow instead of the overhead temperature. Other similar applications of multivariable controllers are reported by Cutler and Finlayson (1988). O'Connor. Grimstad and Mckay (1991) and Hsie (1989). The Linear Dynamic Matrix Control (LDMC). was developed by Morshedi , Cutler and Skrovanek (1985) as an alternative to the conventional DMC. In the LDMC the control actions are the solution of a linear problem that minimizes the absolute value of a residue function of the error on the controlled variables. The main feature of the LDMC controller is the possibility of rigorously attending the constraints on the manipulated and controlled variables. The usual approach to solve the LP problem is to use the revised simplex algorithm, but for systems with a large number of manipulated variables may turn the solution of the problem prohibitive in terms of computation time. Since most of the controlled and manipulated variables shall be kept inside well defined ranges. we found particularly useful the "bounded variables" approach of the simplex algorithm (Murty, 1976) for the solution of the LP problem. the heavy-diesel section. a nurumum liquid flow must be kept in the flash zone of the tower. In the case here studied this liquid flow is guaranteed by a controller of the level the liquid that overflows a darn at the first tray above the flash zone. Figure 1 - Schematic of the crude tower mE CRUDE FRACTIONATOR CONTROL PROBLEM In the conventional control scheme the heavy naphtha boiling point is controlled by an overhead temperature controller on the DCS, that manipulates the reflux flow. If the tower pressure and the overhead temperature are kept constant, the operators expect that the naphtha composition will also remain constant. This is certainly not true since the naphtha end-point is also affected by other variables as the LN ratio and the feed composition. The pressure on the overhead drum is controlled by a split-range controller that admits or release gas from this drum. The distillation ranges of the side products. except the heavy diesel are manually controlled by manipulating their flows. The 85% boiling point of the diesel blend is also manually controlled by the liquid overflash above the flash zone or by the outlet temperature of the feed heater. The liquid level on both the jet-fuel and the light diesel strippers are controlled on the DCS by the manipulation of inlet liquid flow to the strippers. The level of the heavy diesel stripper is controlled by the outlet liquid flow. The heat duties of the external pumparounds can also be manipulated by the operator to impose a proper liquid vapor flow in the tower and maximize the heat recovery in the oil preheating system. The major distwbances to the atmospheric tower are related to the feed quality and flow. These distwbances demand frequent correction by the operators on the flows of the product withdraws. overhead temperature, pumparounds heat duties and heater's outlet temperatures. A multivariable predictive controller can certainly PROCESS DESCRIPTION A schematic representation of the atmospheric crude tower is shown in figure 1. The reduced crude from the bottom of the pre-flash tower is heated and partially vaporized in two parallel heaters. The heated feed is then sent to the atmospheric tower where five liquid streams are produced. At the top of the tower heavy naphtha is produced and usually used as a petrochemical feed stock. There are three side draws: jet-fuel. light diesel and heavy diesel, that are sent to three stripping towers where live steam can be injected to control the flash-point of these streams. At the bottom of the crude tower we have the atmospheric resid that is sent to the vacuum section of the crude unit to the recovery of the gasoils. The commercial diesel in Brazil is a mixture of the light diesel and heavy diesel streams and eventually the heavy naphtha is also added to the diesel pool. The studied crude tower has two pumparounds that substantially affect the heat distribution along the tower. The upper pumparound preheats the crude and the lower pumparound exchanges its heat dutyi with the fresh crude and also at the reboiler of the gasoline stabilizer. At the bottom of the crude tower live steam is injected to maximize the recovery of heavy-diesel from the resid stream. To prevent the entrainment of resid to 470 and to the linearly related to the control move ~u predicted error E' on the controlled variables: perfonn better than the human operator in these circumstances. Then in general lines the objective of the advanced control is to keep the controlled variables the closest possible to their set-points without violating the constraints on the remaining variables of the system. For the crude fractionator we can consider as controlled variables. the distillation ranges of: - the overhead heavy naphtha - the jet-fuel stream - the light diesel stream - the heavy diesel stream These variables are controlled on externally defined set-points that are selected depending on the market conditions. For the particular fractionator considered in this study. an usual demand is to specify the jetfuel stream and the diesel blend that nonnally includes the light and heavy diesel and eventually external streams from other units. We consider two possible cases: Case I. There are four controlled variables with defined set-points related to the product composition specifications. These are the ASTM 90% boiling points of the naphtha, jet-fuel and light diesel cuts and the ASTM 85% of the diesel pool. Case 11. The controlled variables with defined setpoints are only the jet-fuel 90% ASTM and the 85% ASTM of the diesel blend. All the variables related to the compositions or specifications of the products are inferred from the system measured properties (temperatures, pressures and flows) . Other variables can also be controlled but not necessarily on a set-point. These variables are controlled by range and the objective is to keep them inside maximum and minimum values. In the system here considered these variables are: - rnax. 10% ASTM of the jet-fuel - max. 50% ASTM of the jet-fuel - rnax. freezing point of the jet-fuel - min. flash-point of the jet-fuel - rnax. cloud-point of the diesel pool - max. flooding % on the jet-fuel section - rnax. flooding % on the diesel section - min. tower overhead temperature - min. liquid level in the jet-fuel stripper - min. liquid level in the diesel stripper The controller can manipulate: -the reflux flow rate at the top of the tower. -the jet-fuel flow . -the light diesel flow. -the feed heaters temperature controllers. -the stripping steam to the jet-fuel stripper. -the lower pumparound heat duty. The controller also includes as measured disturbances the feed flow and the heat duty of the upper pumparound. p = (AT.G.A + R).Du - AT.G.E' (1) T where P = [PI Pl ··· PML-I and M is the number of manipulated variables and L is the number of future control actions considered in the controller. The matrix A is the modified dynamic matrix of the system. that contains the step response coefficients of the controlled variables whose corresponding predicted error components are not zero. For those variables controlled by range with no predicted error. or the predicted value of the variable is inside its range. the corresponding rows of matrix A are also zero. The diagonal matrix G weights the importance of the controlled variables. and R is also a diagonal matrix of the moving suppression factors . The control actions are then calculated by the solution of the LP problem PMd Minimize q, = u,x.z M .L LWj.(Xj +Zj) j=l (2) Subject to x - z = (AT.G.A + R).Du - AT.G.E' ( M.L equalities) u (3) , x,z~O We also have constraints on the manipulated variables: - On their absolute values: umin ~ U~ u mu (2.M.NL inequalties) - On their moves AUmin~ Au~ Au mu (2.M.(NL-l) inequalities) This LP problem has a total of 4.M.NL-2.M +M.L equations, where NL is the number of future instants where the constraints on the manipulated variables are imposed. Usually we assume NL < L to guarantee the feasibility of the solution of the LP. The revised simplex algorithm is usually the first choice to solve this problem. and since the inequalities are transfonned to equalities by the inclusion of slack variables the number of variables of the LP problem is substantially increased. A particular approach can be followed if some or all the variables of the problem are constrained inside a limited range. This is the "bounded variables" approach (Murty. 1976) and it gives a more economic solution in terms of computation effort. The LP problem with "bounded variables" can be fonnulated as: THE CONTROLLER EQUA nONS The LDMC algorithm (Morshedi . Cutler and Skrovanek, 1985) defines a vector of residues P Minimize eT. x Subject to A.x=b 471 (4) x'J ::;U'J j EJ = {l....,nl} jet-fuel and diesel blend boiling points. have opposite signs of the responses obtained for the step change on the jet-fuel flow. For a step increase on the heat duty of the upper pumparound. the boiling points of all the products are reduced since the LN ratio along the tower is increased. At the top section of the tower this happens because the increase in the pumparound duty causes a reduction on the vapor flow to the top section of the tower and the top reflu.x flow is kept constant. Bellow the pumparound. the internal liquid flow is increased by the extra condensation of vapor at the pumparound. The net effect on the diesel blend is then similar to the effect of an increase on the top reflu.x flow. as shown in fig.2d. (5) where n) is the number of bounded variables. for any j and xj ~ 0 The LP problem defined by the LDMC can be formulated with the structure defined by eqs (4) and (5) after a suitable change of variables. In this approach. using the simplex nomenclature a bounded nonbasic variable can be on any of its bounds 0 or ~ . and this turns the selection of the nonbasic vanable to enter the basis a little more complicated than in the usual simplex algorithm. The benefit to the solution of the control problem is due to a smaller number of equations and variables. The reduction in the computation time gained with this approach depends on the size of the problem in terms of the number of manipulated variables and control horizon. For the size of problem here considered the "bounded variable" simplex algorithm can be ten times faster than the conventional revised simplex. 0 . 0. ,<>Clm3ld 0 . 03 /~ 0 . 02 0 . 01 le t-tu e I 90% / o t----,--_ _ _ _ _ _ _ _ _ _ m '" '~ -0.01 dle.eI85% -0 . 02 -0 . 03 mE OPEN-LOOP DYNAMICS OF THE ATMOSPHERIC FRACTIONATOR To gain insight on the dynamic behavior of the system consider the step responses of the crude fractionator for some of the manipulated variables. These responses correspond to pulse transfer functions obtained by linear regression of the plant data. Figure 2a corresponds to a step change on the flow of jet-fuel that reduces the internal reflu.x flow below this draw and as a consequence an heavier vapor flow goes to the top. A heavier naphtha is produced since the top reflux flow is kept constant. Hence the jet-fuel 90% ASTM increases. since the higher jet-fuel flow changes the cut-point between the light diesel and the jet-fuel. Part of the diesel components are incorporated by the jet-fuel flow that becomes heavier. Some of the light components of the diesel are lost to the jet-fuel. and the light and heavy diesel cuts also become heavier following a response with almost the same gain as the jet-fuel. We have a cut-point exchange between these two products. A different pattern is followed by the diesel blend (mixture of light dieseL heavy diesel and heavy naphtha ). represented by the 85% ASTM boiling point as also shown in fig.2a. The diesel blend boiling point decreases with an underdamped behavior. with a slower dynamics than the jet-fuel and a significative dead time. Apparently the underdamped response is related with the liquid overflash flow control. The reduction on the internal reflux below the jet-fuel draw disturbs the overflash control loop that shows the underdamped response. Other properties of the diesel blend as the cloud point follow the same type of response. The behavior of the fractionator to a step change on the reflux flow can be analogously explained. In this case an increase on the reflux flow results in a higher UV ratio and the step responses shown in fig . 2b for the 0.03 0.02 0.01 Fig.2b oC/m3/d ~% ~ O~-­ min -0.01 -0.02 -0.03 50 100 3 oC/oC diesel85% 2 % 1 min 50 10 o oC/Gcallh -10 11 \ -20 ~ 100 diesel 85% min ,-~4_0 ___ 80_ _ _ 120 jet-fuel 90% Fig .2d Figure 2. Open loop responses of the crude fractionator for step responses on (a)- Jet-fuel flow; (b)-Reflux flow; (c)-Feed heater temperature and (d)-Upper pumparound heat duty. for Case n. We also observe that the heat duty control loop seems to smooth out the effect on the internal reflux flow and the underdamped response of the diesel blend properties is not observed. It is also interesting to compare the step responses of the two controlled 472 variables of case II for the feed heater temperature. Obviously the increase in the open loop feed temperature causes an increase in the boiling point of the products since a heavier vapor is injected into the tower while the internal liquid flow remains constant. But observing fig .2c. we note that the dominant time constant of the diesel blend boiling point is larger than the jet-fuel response. This seems to be not coherent with the position of the diesel draws that are closer to the feed tray than the jet-fuel draw. Apparently. the heavier vapor flow initially affects both products with the same intensity. but the delays of the liquid flow inside the tower has an important role on the final response of the system. 235 1 oC ~ ft~ __-."/ 20 ~ _ ~_ -_ ~_ ~1-+ =_ 50 . =tnln -~ o 100 ~ 150 oC 3b . 90% B .P.ofthe light diesel 295 290 285 t-----------------1 1'-------I '--- 280~ - o PLANT RESULTS An example of the performance of the developed controller for Case I configuration in the real plant can be seen on figure 3. obtained from the crude unit at the refinery of Paulinia in Brazil. The crude tower was on a steady-state with the controller in operation. At instant 25min the set-point of the level of liquid overflash was slightly reduced and this caused an unmeasured disturbance on the system. The variable most affected by this disturbance is the diesel boiling point as shown in fig.3c. The controller compensates this effect by slightly increasing the heat duty of the lower pumparound (fig.3d) to increase the internal reflux in the diesel section. At instant 70 min the system was disturbed by a small step change in the feed flow and a step reduction on the set-point of the light diesel 90% ASTM boiling point (fig.3a). This intensifies the effects of the first disturbance and the control actions are more clearly observed. We observe that one of the variables controlled by range. the jet-fuel freezing-point (fig.3e) tend to violate its maximum constraint but is brought back to an acceptable range. The remaining variables controlled by range stay inside their range during all the time of observation. After about 50min all the product specifications are at their required set-points and the system stabilizes. 3a . 90% BP of the jet-fuel 375 50 100 150 oC ~ 370 ' l'Tfin ~- 3c . 85% B .P . of the diesel-blen j 365 360~- - o 50 100 min ~ 150 10.8 G callh 3d . Heatduty of lower PA ~ 10.7 I- - - - . J( min 10.6 + - - - - - + - - - - - + - - - - - - < o 50 100 150 oC -40 3e . Freezing point of the jet-fuel -50 ,-~= ,- -60~ -70~+ Another example of the practical performance of the controller concerns Case n where only the jet-fuel and the diesel blend are controlled with external setpoints while the remaining variables of the system are controlled by range. Figure 4 shows the plant tendencies for about 12 hOUTS of normal operation. The set-point of the diesel blend (fig.4a) was continuously changed to attend the production scheduling while set-point of the jet-fuel remained constant during all the period considered. We observe that the controller performed quite well even when the system was severily disturbed by the upper pumparound heat duty (fig.4d) that is used to preheat the crude. We note that this beat duty was increased of about 0.7Gca11h after instant 450min. o ~ 50 p <::::;, ~- min 100 150 Figure 3. Plant responses for the multivariable controller for Case I configuration. This is a measured distuIbance to the controller and had almost no effect on the controlled variables. To compensate this distuIbance the controller reduced the top reflux flow to rnantain the internal reflux constant below the pumparound. The jet-fuel flow (fig.4e) was transitorily reduced to compensate the effect of this disturbance on the jet-fuel boiling point and brought back to almost its original level after a few hOUTS. 473 375 oC allowing the application of the controller to systems of larger order while keeping the computation time at an acceptable leveL The selected control configuration uses the reflu.x flow as a manipulated variable instead of the set-point to the overhead temperature controller as usually suggested in the literature. 4a. Diesel blend 85% 370 365 1 360~ ~- o - ~- 400 200 min - 600 REFERENCES o 400 200 Cutler. C. R and S. G. Finlayson. Design Considerations for a Hydrocracker Preflash Column Multivariable Constraint Controller, IF AC Conference. Atlanta. 1988. Hsie. W. L., Modeling Simulation and Control of a Crude Tower. Ph.D. Dissertation. University of Maryland. 1989. Morshedi. A. M., C. R Cutler and T. A. Skrovanek, Optimal Solution of Dynamic Matrix Control with Linear Programming Techniques (LDMC), Proc. Am. Control Conf., Boston, pp. 199-208, 600 Gcal/h 16 15 ",upper 14 4c. Pumparound heatduties 13 lower mi 12~-n o 200 m3/d 50~d . 400 1985. Murty, K. Linear and Combinatorial Programming, John WHey & Sons, New York, 1976. Muske. K., 1. Young. P. Grosdidier and S. Tani, Crude Unit Product Qualify Control, Comp. and Chem. Engng. 15,9, 629-638, 1991. O'Connor, D. L., K. Grimstad and 1. Mckay, Application of a Single Multivariable Controller to two Hydrocracker Distillation Columns in Series, ISA Meeting, Anaheim, 600 Reflux flow 4500 1991 . min 350~-4 o 200 m3/d 180~4e 400 600 . Jet-fulOW 1500 min 120~-+ o 200 400 600 Figure 4. Plant responses with the multivariable controller for Case 11. CONCLUSION A multivariable controller was successfully applied to the quality control of a large scale industrial crude aUnospheric fractionator at the refinery of Paulinia in Brazil. The developed control algorithm shows that the LDMC formulation is adequate to large systems where several or eventually all variables are controlled by range, and the manipulated variables have minlrnax constraints. In this study it was adopted the "bounded variables" approach of the simplex algorithm that shows to be particularly useful in reducing the computation effort and 474