Inferential Composition Control

Many industrial columns use temperatures for composition control because direct composition analyzers can be expensive and unreliable. Although temperature is uniquely related to composition only in a binary system (at known pressure), it is still often possible to use the temperatures on various trays up and down the column to maintain approximate composition control, even in multicomponent systems. Probably 75 percent of all distillation columns use temperature control of some tray to hold the composition profile in the column. This prevents the light-key (LK) impurities from dropping out the bottom and the heavy-key (HK) impurities from going overhead. 

How do we sel ect the best tray location for this temperature control The next section outlines the normal procedure and illustrates its application with a specific example. 

---

We started this section on inferential composition control by using a single tray temperature. Then we used two temperatures to control a temperature difference. Then we used multiple temperatures for average temperature control. The logical extension of this approach is to use whatever measurements are available to estimate product compositions. There are several types of composition estimators. Steady-state... 

Tray temperatures correlate very well with product compositions for many distillation columns therefore, inferential control of distillation product composition is a widely used form of inferential control. Figure 15.58 shows the arrangement for inferential temperature control of the bottoms product composition for this column. Note that the tray temperature controller is cascaded to a flow controller. 

Figure 2.2 Three different systems for the distillate composition control of a simple distillation column (a) feedback (b) feedforward (c) inferential.

The second column supplies benzene of high purity. One-point composition control is more robust in a recycle system. In inferential mode the heat input in reboiler controls a sensitive temperature in the column. If the sampling point is placed in the stripping section, this control loop ensures good composition control of the bottom product. The reflux is set in ratio with the feed flow rate. For moderate disturbances, this allows good purity of distillate simultaneously with high recovery. The levels in the flash drum and in the base are hold by manipulating distillate rate and bottom product, respectively. 

Temperature profiles after a stepwise increase of the heating rate are shown on the right side of Eig. 10.29. These profiles show distinct nonlinear wave characteristics as discussed in the previous section. Therefore the process is sensitive to disturbances and composition control is required. Again focus is on an inferential control scheme. Measured variables are the temperatures on trays 4 and 60, which are located within the upper and the lower ware front. Hence, they show good sensitivity to disturbances. Manipulated variables are the heating rate and the reflux ratio. Eor this process neither input nor output multiplicities occur. 

There are a number of ways in which this inferential may be applied. At this stage, replacing the temperature controller with a virtual composition controller is not an option since the inferential is developed on the assumption that the tray temperature is held constant. 

In general, the optimization of polymerization processes [2] focuses on the determination of trade-offs between polydispersity, particle size, polymer composition, number average molar mass, and reaction time with reactor temperature and reactant flow rates as manipulated variables. Certain approaches [3] apply nonhnear model predictive control and online, nonlinear, inferential feedback control [4] to both continuous and semibatch emulsion polymerization. The objectives include the control of copolymer composition. 

Figure 12.36 shows a control structure in which the temperature on tray 3 is controlled instead of direct composition control of the bottoms composition. Figures 12.37-12.41 show that this inferential control scheme does a fairly good job of maintaining product purity and keeps reactant losses low for throughput changes. However, for feed composition disturbances, both the distillate and bottoms compositions move away fairly significantly... 

The main focus of this chapter, however, is on the tools that can use this sensor information for real-time control. These tools are still largely developmental because inferential control is relatively new to the composites industry and change is dependent on both technical merit and on other changes in the culture of the industry. 

One solution to this problem is to employ inferential control, where process measurements that can be obtained more rapidly are used with a mathematical model to infer the value of the controlled variable, as illustrated in Figure 12. For example, if the overhead product stream in a distillation column cannot be analysed on-line, measurement of a selected tray temperature may be used to infer the actual composition. If necessary, the parameters in the model may be updated, if composition measurement become available, as illustrated by the second measuring device in Figure 12 (dashed lines). 

We might be tempted to control reflux drum level with one of the fresh reactant feeds, as done above. The problem with this is that the material in the drum can contain a little of component C mixed with either A or B, Simply looking at the level doesn t tell us anything about component inventories within the process and which might be in excess. The sj stem can fill up with either. Some measure of the composition of at least one of the reactants is required to make this system work. Compositions in the reactor or the recycle stream indicate an imbalance in the amounts of reactants being fed and being consumed. If direct composition measurement is not possible, inferential methods using multiple trays temperatures in the column are sometimes feasible (Yu and Luyben, 1984). 

Properties to be controlled may not be measurable online fast enough to allow for a timely action by the manipulated variable. Such properties may have to be inferred from other measured properties. A column product purity or composition, for example, could be inferred from measured column temperatures on a number of trays. The required property is related to the measurements by inferential property correlations whose parameters must be determined. In the composition-temperature example, the correlation parameters are evaluated from measured temperatures and laboratory composition analysis, and are updated every time laboratory analyses become available. 

A pilot-scale distillation column located at the University of Sydney, Australia is used as the case study [60]. The 12-tray distillation column separates a 36% mixture of ethanol and water. The following process variables are monitored temperatures at trays 12, 10, 8, 6, 4, and the reflux stream, bottom and top levels (condenser), and the flow rates of bottoms, feed, steam, distillate and reflux streams. The column is operated at atmospheric pressure using feedback control. Three variables are controlled during the operation top product temperature, condenser level, and bottom level. Temperature at tray 8 is considered as the inferential variable for top product composition. To maintain a desired product composition, PI controllers cascaded on flow were used to manipulate the reflux, top product and bottom product streams. 

FIGURE 15.58 Schematic for inferential control of the bottoms product composition of a distillation column. 

In chemical process control the variable that is most commonly inferred from secondary measurements is composition. This is due to the lack of reliable, rapid, and economical measuring devices for a wide spectrum of chemical systems. Thus inferential control may be used for the control of chemical reactors, distillation columns, and other mass transfer operations such as driers and absorbers. Temperature is the most common secondary measurement used to infer the unmeasured composition. 

Since the feed and overhead compositions are considered unmeasured, we can only use inferential control. The secondary measurement employed to infer the overhead composition is the temperature at the top tray. Let us now examine how we can develop and design the inferential control mechanism. 

Recall that the effectiveness of an inferential control scheme depends heavily on the goodness of the estimator, which in turn depends on the model that is available for the process. Assume that the overhead composition can be measured intermittently, either by taking samples manually and analyzing them or even better using a gas chromatograph on-line. From the composition measurements we can take the useful information needed to judge how effective the inferential control has been. Thus if the measured... 

Consider two processes one (process A) slow with time constant 5 hours and another (process B) faster with time constant 1 hour. The composition of the output streams from the two processes is measured every 2 to 3 hours. Which of the two process outputs can be controlled by conventional feedback and which one will require inferential control ... 

Show that the inferential control employed for process A or B in item 13 (above) can be improved through an adaptive mechanism that uses the direct composition measurement every 2 to 3 hours. (Consult Example 22.5.)... 

Suppose that the overall heat transfer coefficient between the cooling water and the reacting mixture drops with time due to fouling. Construct an adaptive scheme which uses intermittently exit composition measurements to adjust the parameters of the inferential controller. 

Since the catalyst activity decays, add an adaptation mechanism to the inferential control system. The adaptive system will adjust the parameters of the inferential controller using as information exit composition measurements which are taken periodically with a gas chromatograph. 

V.32 (a) Construct an inferential control system to regulate the distillate composition of a distillation column when the following transfer functions are provided ... 

Composition (inferential control) with reboiler duty. 

Possible candidates for manipulated variables are essentially the same as in non-RD, including reflux, distillate flow rate or reflux ratio at the top of the column and heating rate, bottoms flow rate, or reboil ratio at the bottom of the column. In addition, dosing of the reactants can be an interesting choice in RD, if this is compatible with the upstream processing of the plant. Possible candidates for measured variables are either product compositions or column temperatures. However, online measurement of concentrations is usually slow, expensive, and often not very reliable. Therefore inferential control schemes are preferred, where the product compositions are inferred from temperature measurements. However, the relationship between product compositions and column temperatures is frequently non-unique in RD [26, 98] and this can lead to severe problems as will be filustrat-... 

There are columns which do not lend themselves to tray temperature control, for example because temperature is insensitive to composition. Under these circumstances pressure compensation may be applied directly to what would otherwise be the MV of the temperature controller. The pressure compensation factor can be determined empirically. The approach is similar to that described for quantifying dT/dP. It is based on the assumption that an inferential can be developed based on pressure (P), the manipulated flow (F) and other independent variables. The manipulated flow may be reboiler duty, reflux, distillate or bottoms - depending on the choice of level control strategy. 

Feedforward on feed composition can be a valuable enhancement but may not be practical. Firstly it requires an on-stream analyser on feed. Few plant owners would install this as standard and there may not be sufficient economic justification to add it later. Secondly it may not be possible to acquire an analyser that responds quickly enough. If the change in feed composition affects tray temperatures and/or inferentials before being reported by the analyser then the feedback controller(s) will take corrective action. A delayed measurement of feed composition would then be less valuable than having no measurement. 

Although many process variables are easily measured, lack of on-line sensors for key polymer properties renders quality control of polymer plants difficult. Process control schemes based on process variables p, T, flow-rate and feedstock compositions) alone are no longer sufficient, because these cannot reveal all material property variations. Significant efforts are being spent on improvements to process control systems, as exemplified by the numerous attempts to monitor polymer properties during processing, such as composition, density, viscosity and dispersion of a minor phase, etc., all of which are somehow difficult to measure. The development of an on-line inferential system for polymer property is a very active research area of polymerisation reactor control [1]. A schematic of inferential systems is illustrated in Fig. 7.1. For highest quality... 

One conld also try to estimate the quality ( inferential control ) by using a suitable algorithm that uses conventional measured values of for example temperatme, pressure, etc. A simple example is the pressure compensation of temperatme measurements on the tray of a distillation column. For binary mixtures this yields a measure of the composition, for a multi-component system it is an approximate measure of the composition. 

In practice, a temperature controller is often used as a simple form of quality control. According to the phase rule of Gibbs, for binary mixtures the composition is fixed when the pressirre and temperature are constant. Obviously, this does not hold for multi-component mixtures. One could also try to estimate the product quality from some tray temperature measirrements and other easily measurable variables. This is called an inferential measurement. 

---