In many chemical plants, separation of products from unconverted raw materials is usually done by a train of several distillation columns, downstream of the reactor section. Frequently, each product has its own column with the quality specification for only one stream. However, there are several reasons to control the other stream as well. The justification is competitive economic environment: balancing the energy consumption versus the loss of valuable product, increasing the throughput, or stabilizing downstream units.

Overview of a typical column operation

Consider the ethane splitter, as a part of a steam cracker. The top stream of the column, ethylene, is the main product stream of the cracker. Its impurity, primarily ethane, has to be below a certain level: exceeding that level automatically results in dumping the ethylene stream to the flare. However, there are two incentives to have the maximum allowable amount of ethane in the ethylene product stream. First it reduces the energy consumption of the column.

The bottom stream will decrease and a more valuable ethylene stream will increase. The value of the top product is much higher: in fact ethane is sold as ethylene.

Although the bottom stream, mainly the ethane, is recycled via a furnace, where it is cracked, the amount of impurity is also important for optimizing the column and even plant operation. Increasing impurities, primarily ethylene, increases recycle cost and reduces capacity. Most significant perhaps, valuable ethylene is lost because it is partly cracked into less valuable products such as hydrogen, methane, propane, and so forth.

On the other hand, as the ethylene in the bottom stream decreases, separation costs in the C2 splitter go up. So there is an impurity optimum level, which can be calculated if the separation costs, compression costs and behavior of ethylene in the ethane furnace are known.

The optimal bottom impurity is not a constant: it varies as a function of energy costs, product values, and the current plant environment.

Numerous books and papers related to the distillation column process present various aspects of mathematical modeling, simulation, optimization and control. Special attention has been concentrated on steady-state models for design purpose.

Although the distillation processes are well known, it is always a challenge to apply available knowledge in practice. Such an example is the subject of this paper. A specific dynamic mathematical model of propane propylene splitter is developed on the basis of plant dynamic test experimentation on commercial plant and mathematical model identification. The valid mathematical model has been used in multivariable controller design. The practical cases of control strategies have been studied and used in process control improvement.

Improvement approach

The object of the study, a 150 tray column, separating 70 mole percentage propylene and 30 mole percentage propane feed into 94 mole percentage of propylene in distillate and 8 mole percentage of propylene in bottom product, as an optimal process conditions is considered. The process is a major consumer of energy and a difficult separation. Since it is usually positioned at a point in the process where other lighter and heavier components have been removed in previous processing steps, it means, that disturbances from upstream units are usually frequent.

One of the primary incentives for the implementation of multivariable control is to avoid the complexity and inflexibility of single loop schemes and advanced dynamic process performance and process optimization.

The control strategy of process improvement with APC in practice: energy consumption minimization, optimum products impurity level maintaining and disturbance feedforward control is established.

Although distillation processes are inherently nonlinear, if operated over a sufficiently small region, process control systems could be based on linear input-output models. So, the multivariable model predictive control of the distillation column is developed using discrete convolution summation.

From the preceding equations, a set of future predictions is based on past moves. Therefore, to find the convolution summation equation, the experimentation investigation had to be done on commercial plant using plant tests and multivariable model identification.
The experimental plan for all important column loops and measurements has been performed.

Step 1: Pretesting

Before plant tests, every independent variable has been checked, using pretests phase, to make sure that each can be moved. The number of input variables and the time to steady state determine the actual plant test.

Step 2: Test

The protocol consists of pulses of differing duration, long, short and intermediate duration and different step magnitude. The step magnitude has been set high enough to avoid process noise magnitude.
Discrete impulse response models defining models for reflux flow, reboiler steam flow, column pressure and feed flow to overhead and bottoms composition, flood and overhead product flow have been obtained. 

Step 3: Model identification

The model quality and validation has been judged by the model uncertainty view. It has been done using frequency response and bounds on the uncertainty of 2 sigma as a function of frequency. The experimental plant test against model simulation responses has been used to compare model-predicted output and measured data. The uncertainty frequency response characteristics are shown in Figure 5 

From the step response models, uncertainty frequency response characteristics and model-predicted output vs. measured data comparison the models have been judged on three important factors:

• Does the model make physical sense?
• Does the model fit the data and?
• Is the model uncertainty reasonable?

The answers on these questions are that: models make physical sense, models fit the data and models’ uncertainties are reasonable.

Step 4: Implementation of optimization strategy

The experimentally valid multivariable model is then accepted as the valuable reference for the multivariable controller design as well as for on-line closed loop optimization and control improvement. For this purpose, few commercial cases have been established and included in the process control strategy. Dual composition target, the maximum impurity of the distillate and bottoms product are constrained, the maximum impurity of the distillate is constrained and the distillate is considered as a more valuable product on the market then bottoms product.
On the basis of the first case, the following process control strategy is established. Both composition of distillate and bottoms product are controlled by the dual composition control. The operating costs have been minimized by reduction of energy consumption, which is realized by the column pressure minimization. The pressure drop in the column is constrained by the 80% column flooding. The defined control strategy has been implemented in multivariable control design using Aspentech SMCA software package.

Step 5: APC controller results

The figure below shows multivariable controller control action from the moment controller starts acting and during next two hours. Before multivariable controller has started running the process has been in basic control. The basic control performs a difficult separation, causes variations of overhead product composition, it doesn’t allow dual composition control targets and as neither energy consumption minimization.

Against the basic control, the multivariable controller reach the control target of distillate and bottoms product composition and minimizes the column pressure moving the pressure to the lower constraint in period less than one hour.
In the case when the bottom product composition is not constrained there is enough degree of freedom to maximize the most valuable product, distillate and to minimize reboiler steam flow and column pressure. The multivariable controller has acted to move reflux flow, reboiler steam flow and column pressure to maintain the maximum distillate impurity of 6 mol%, to increase the distillate flow as much as possible and to minimize reboiler steam flow and column pressure to its’ low limit. This case is shown in the figure below:

Conclusion

The distillation process of propane-propylene, one of a difficult separation processes and a major consumer of energy, has been improved in practice on commercial plant by advanced process control: experimental plant tests, mathematical model identification and multivariable controller design.

The applied control strategies based on the cases derived from practice provided superior dynamic control performance and reduce variability in controlled variables. Setpoints or constraints can be moved closer to specification targets with corresponding operation cost reduction in energy. The designed multivariable controller provides multi-objective control handling based on objective hierarchy.

For more details, you can download the whole paper.

About the author

Darko Lukec is a mechanical engineer with Ph.D. in Chemical Engineering. During his 35 years of experience in oil, gas and petrochemical industry, he was working on modeling and simulation, process optimization, basic design projects, advanced process controI and operator training simulators. Through his scientific engagement, he published 20+ scientific and professional papers. Darko is now the owner of the company moDeL, the company he founded in 1993 and was a director for 20 years. moDeL is specialized for application of mathematical modeling and process simulation in process improvement solutions. 

©SimulateLive.com