multivariable model predictive controller improves … · 1 multivariable model predictive...

11
1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process Management KEYWORDS: demethanizer, multivariable predictive control, model predictive control, MPC, advance process control, APC, ethane recovery, control performance, embedded MPC ABSTRACT Since its application to the NGL process in 1964, the turbo-expander based demethanizer process has been the workhorse of many gas plants. The turbo-expander process isentropically expands the gas through a high speed turbine to create cryogenic temperatures facilitating the separation of ethane and heavier components from the methane. This process also uses significant heat integration to reduce energy consumption. These aspects combined with the need to match the demethanizer capacity with the incoming feed rate, create a complex control problem. The use of traditional independent PID feedback control loops on this process often results in poor performance and excessive operator intervention. The application of an advanced multivariable model predictive controller (MPC) to this complex process provides significant improvement in the process performance and greatly reduces the need for manual operator intervention. We will describe how to apply MPC to the turbo-expander based demethanizer process and show the benefits of its application in actual projects. INTRODUCTION In 2010, there were 493 active natural gas processing plants in the United States with a combined processing capacity of 77 billion cubic feet per day [1]. Most of these facilities include distillation unit operations. Compared to a typical petrochemical facility, these natural gas processing plants are relatively small. However, the natural gas processing plants have the same control challenges such as process interaction, feedstock variation, etc. that a large petrochemical plant has. Control of processes with these characteristics can benefit from Multivariable-Model Predictive Control (MPC). Typically, MPC is expensive to implement and maintain and cannot be justified and supported by these small facilities. However, modern MPC systems that are embedded in the control system, include tools for process model identification and commissioning, and automatically generate the operator interface can significantly lower the implementation and maintenance cost. This allows MPC to be economically justified in many of the typical natural gas processing plants. MPC DESCRIPTION Figure 1 shows the basic difference between conventional single variable “Proportional, Integral, Derivative (PID) controllers and MPC. PID controllers typically have a single control variable (CV) and a single manipulated variable (MV). A PID controller is used for each CV/MV pair and is only concerned with its own CV/MV. In a low percentage of PID loops, a “feed forward” algorithm is utilized when an independent disturbance variable (DV) has an impact on the CV. The feed forward scheme is used to take action directly on the MV when the DV changes. An MPC, however, can include many CV’s , MV’s and DV’s and takes in to account the relationship or “model” between all of these variables. In addition, by using high and low limits, rather than single value for the set point of the CV, the CV effectively become a constraint variable, sometimes called a limit variable (LV). Most MPC’s include some type of optimizer that can be used not only to control the process but to control it in a manner that also achieves a desired process or economic “objective”. Examples of optimizer objectives are to maximize production, minimize energy and to maximize profit.

Upload: truongkhue

Post on 11-May-2018

218 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

1

MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO

EXPANDER DEMETHANIZER COLUMN PERFORMANCE

James Beall Emerson Process Management

KEYWORDS: demethanizer, multivariable predictive control, model predictive control, MPC, advance process control, APC, ethane recovery, control performance, embedded MPC

ABSTRACT

Since its application to the NGL process in 1964, the turbo-expander based demethanizer process has been the workhorse of many gas plants. The turbo-expander process isentropically expands the gas through a

high speed turbine to create cryogenic temperatures facilitating the separation of ethane and heavier components from the methane. This process also uses significant heat integration to reduce energy

consumption. These aspects combined with the need to match the demethanizer capacity with the incoming

feed rate, create a complex control problem. The use of traditional independent PID feedback control loops on this process often results in poor performance and excessive operator intervention. The application of an advanced multivariable model predictive controller (MPC) to this complex process provides significant

improvement in the process performance and greatly reduces the need for manual operator intervention. We

will describe how to apply MPC to the turbo-expander based demethanizer process and show the benefits of its application in actual projects.

INTRODUCTION

In 2010, there were 493 active natural gas processing plants in the United States with a combined processing capacity of 77 billion cubic feet per day [1]. Most of these facilities include distillation unit

operations. Compared to a typical petrochemical facility, these natural gas processing plants are relatively small. However, the natural gas processing plants have the same control challenges such as process

interaction, feedstock variation, etc. that a large petrochemical plant has. Control of processes with these

characteristics can benefit from Multivariable-Model Predictive Control (MPC). Typically, MPC is expensive to implement and maintain and cannot be justified and supported by these small facilities. However, modern MPC systems that are embedded in the control system, include tools for process model identification and

commissioning, and automatically generate the operator interface can significantly lower the implementation

and maintenance cost. This allows MPC to be economically justified in many of the typical natural gas processing plants.

MPC DESCRIPTION

Figure 1 shows the basic difference between conventional single variable “Proportional, Integral, Derivative (PID) controllers and MPC. PID controllers typically have a single control variable (CV) and a single

manipulated variable (MV). A PID controller is used for each CV/MV pair and is only concerned with its own

CV/MV. In a low percentage of PID loops, a “feed forward” algorithm is utilized when an independent disturbance variable (DV) has an impact on the CV. The feed forward scheme is used to take action directly

on the MV when the DV changes.

An MPC, however, can include many CV’s , MV’s and DV’s and takes in to account the relationship or

“model” between all of these variables. In addition, by using high and low limits, rather than single value for

the set point of the CV, the CV effectively become a constraint variable, sometimes called a limit variable (LV). Most MPC’s include some type of optimizer that can be used not only to control the process but to control it in a manner that also achieves a desired process or economic “objective”. Examples of optimizer

objectives are to maximize production, minimize energy and to maximize profit.

Page 2: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

2

PROCESS DESCRIPTION

Many natural gas processing plants include gas separation systems where the gas is compressed,

cryogenically separated and then fractionated to produce methane, ethane, propane, butane and natural

gasoline products. One of the most important sections in the separation systems is the inlet gas cooling system and demethanizer column. These units are often a “turbo-expander” cryogenic demethanizer system which is the topic of this paper. In this section, the main inlet gas is cooled by the low pressure residue gas

and a propane type chiller. A trim heater raises the temperature of the side sidestream and the demethanized liquid bottoms product and adds heat to the bottom and side reboilers. A turbo-expander cools

the feed gas to approximately -150 Deg. F before the gas enters the demethanizer distillation column.

Typically the bottom liquid product from the demethanizer is further separated by additional fractionating columns. The significant use of heat integration for efficiency and the typical characteristics of distillation columns result in a highly interactive system. Also, due to the complexity of the process and process

equipment, there are many constraints that must be managed. Figure 2 shows a process diagram of the

turbo-expander type demethanizer system which is a typical arrangement for these types of units.

Page 3: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

3

MULTIVARIABLE-MODEL PREDICTIVE CONTROL (MPC)

MPC provides the basic control functionality of reducing the process variability and keeping the CV’s and LV’s at the desired operating point. However, utilizing allowable set point “ranges” instead of a single value, MPC’s often include an optimizer that can adjust the actual target set points for the CV’s and keep the LV’s

within constraints to achieve a desired operating object such as to maximize production, minimize energy and to maximize profit.

MPC uses “dynamic matrix control” [2] techniques and process models from the MV’s, DV’s to the CV’s and

LV’s to provide these control capabilities. The following capabilities of MPC provide the improved control capability and certainly match the requirements to control the turbo-expander type cryogenic demethanizer system.

Handle Process Interactions

By including all the related variables in the MPC, and using the process models between these variables, the

MPC can predict and therefore mitigate the interactions between the variables. The use of extensive

decoupling techniques would be required for the same capability with multiple PID controllers.

Accommodate DV’s

MPC can include DV’s whose impact on the CV’s and LV’s is predicted through the corresponding models. Using these models, the MPC can take anticipatory action to minimize the impact of the DV’s. PID controllers

utilize a “feed forward” variable to provide a similar action but this becomes overly complex to implement and maintain as the number of PID controllers that interact increases.

Page 4: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

4

Control CV’s or LV’s that are Measured Infrequently

Sometimes a CV or LV may not be updated very frequently. For example, a stream composition that is measured by a process analyser may have an “update” time of 10-45 minutes. If the measurement update

time is a significant proportion on the process response time, the use of the measurement will limit control

performance. However, since the MPC has the model of the response on the CV’s and LV’s, the MPC can actually predict and control from the expected value of the measurement in between measurement updates.

When the new measurement update is received by the MPC, it can correct the prediction. This feature allows a higher level of control performance when CV’s or LV’s are measured infrequently.

Maintain LV’s Within Constraints

LV’s are similar to CV’s except that rather than being controlled to a desired set point, they are controlled to be within a specified range. In other words, the LV is controlled between upper and lower constraints. In

some cases it is quite acceptable for the LV to be anywhere between the upper and lower constraint values.

In other cases, it might be desirable for the LV to be closer to the upper or lower constraint value and the MPC optimizer can be used to accomplish this objective.

Provide Optimization

Optimization is the process of selecting the “best” solution for a problem when there are multiple possibilities.

In mathematics, optimization is the discipline which is concerned with finding the maxima and minima of functions, possibly subject to constraints. These techniques can be applied in an industrial manufacturing

environment. An example of an optimization problem in a manufacturing environment is the following:

maximize the profit of a manufacturing operation while ensuring that none of the resources exceed certain limits and also satisfying as much of the demand faced as possible. The MPC used on this project is a “Linear Program” type of optimizer.

PROJECT IMPLEMENTATION

Emerson Process Management’s DeltaV PredictPro MPC was applied on turbo-expander type demethanizer units in two west Texas gas plants. The DeltaV PredictPro MPC was chosen because it was embedded in

the process controller and had integrated tools to identify process models, tune and commission the MPC, and automatically generate the operator interface. These features reduced the cost to implement and maintain MPC.

Process Objectives

There can be several operating objectives for the plant based on such things as the feedstock supply, the

current market conditions, equipment problems etc. It is very important to understand the various process objectives of the plant and to design the MPC to accomplish these objectives. For example, if the demand for

ethane is low, the MPC should control to a lower ethane recovery target and maximize the ethane in the

methane product up to the product specification. The optimizer in the embedded MPC chosen for this project can have multiple objective functions from which the operator can choose based on the current plant objective.

Control Performance Improvement

A control loop performance audit was performed and all loops associated with the demethanizer systems

were re-tuned to have a coordinated response using the Lambda tuning technique. Poorly performing control

valves were identified and were repaired or replaced. ISA standard ANSI-ISA-TR-75-25-0-2-2000, “Control Valve Response Measurement from Step Inputs” explains how to test the valve response [3],[4]. The EnTech

Control Valve Dynamic Specification helps specify valve performance requirements based on process performance requirements [5].

This is a very important portion of an advanced control project (APC) and will typically produce 25-50% of

the overall benefit of an APC project as well as increase the overall benefit. [6] Figure 3 shows an example of a poorly performing control valve exhibiting excessive dead band. Figure 4 shows a step test, process dynamics and Lambda Tuning for a key control loop in the demethanizer system.

Page 5: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

5

Page 6: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

6

MPC Design

Based on the process objects, the MPC variables, CV’s, LV’s, DV’s and MV’s, were selected. The process response dynamics found in the control performance project are useful in the design of the MPC and

selection of the variables. Table 1 shows the selection and types of variables included in the MPC for the

demethanizer.

Table 1 – Turboexpander Demethanizer MPC Variable Selection

Control Variables Constraint Variables Disturbance Variables Manipulated Variables

Btms C1 %LV Btms C1/C2 %LV Total Feed Column Pressure (cascades to turboexpander speed)

Lower Column Temp (Pressure Compensated)

Btms Vapor Pressure C1/C2 of Inlet Feed Reboiler Heat Duty

C2 Recovery Feed Temperature

MPC Implementation

The embedded MPC uses the same configuration system, history collection system and database as the control system. This significantly reduces the amount of time required for configuration of the MPC. Figure 5 shows the configuration method for the embedded MPC.

The next step is to perform process tests to determine the process models from each DV and MV to each

CV and LV. The embedded MPC has an automated process testing for the MV’s. The MPC cannot change the DV’s so they must be changed by operator or other means. Once the step process test have been

completed, the embedded MPC has a model identification application that includes a verification function to verify the model against the test data as well as historical data from another time period . Figure 6 shows the display for the automated process testing program and Figure 7 shows the model verification display.

The embedded MPC has a built in process and MPC simulator that can be used to observe and tune the response of the MPC under various conditions. Figure 8 shows the MPC Simulation display. Once the tuning is adjusted to provide the desired response, the MPC can be commissioned. An operator interface display is

automatically generated to be used to operate the MPC. With the click of one button, the operator can switch

from the normal regulatory control scheme to the MPC scheme. Figure 9 shows the MPC operator display that is automatically generated for the embedded MPC.

Page 7: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

7

Page 8: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

8

Page 9: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

9

PROJECT RESULTS

The MPC reduced variability of the process significantly and also reduced the number of manual operator interventions required for process disturbances. On one project, the variability of the methane in the bottoms

flow from the turboexpander demethanizer, a key product specification, was reduced by 50%. The ethane recovery, a key performance index, was increased by an average of about 5% points. Figure 10 shows a

trend of these variables before and after MPC was applied.

Another plant that used the embedded MPC on a turboexpander demethanizer reported that the ethane recovery was increased by 4% points and the “methane in the bottoms flow” had significantly less variability and was controlled closer to the stream specification.

Page 10: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

10

SUMMARY

The turbo-expander demethanizer process is complex, interactive and subject to many process disturbances and constraints. In addition, the process objectives for the unit changes due to such things as variations in

demand for the various hydrocarbon products product and equipment issues. MPC technology has characteristics that are beneficial to good control and optimization of a turbo-expander demethanizer process but the traditional MPC project is often too costly and difficult to support in gas plants which are usually in

remote locations. However, MPC technology that is embedded in the DCS, with easy to use tools for configuring, developing process models, commissioning, operating and maintaining have significantly reduced the cost of implementation and maintenance of the applications. The typical ROI for an embedded

MPC project on a turbo-expander demethanizer is about 6 months. The embedded MPC reduces process variability, controls key product composition closer to the specification, increases capacity and reduces

operator intervention during process disturbances.

C1 in Bottoms Flow, %LV

2 Sigma = 0.31

Ethane Recovery Avg. = 88.7

C1 in Bottoms Flow, %LV

2 Sigma = 0.12

Ethane Recovery Avg. = 94.2

Without MPC With MPC

Figure 10 Improvements of Key Performance Indices with Embedded MPC

Page 11: MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES … · 1 MULTIVARIABLE MODEL PREDICTIVE CONTROLLER IMPROVES TURBO EXPANDER DEMETHANIZER COLUMN PERFORMANCE James Beall Emerson Process

11

REFERENCES

1. US Energy Information Agency, Natural Gas Processing Plants in the United States: 2010 Update, http://www.eia.gov/pub/oil_gas/natural_gas/feature_articles/2010/ngpps2009/ .

2. Camacho, E. F., and Bordens, C., Model Predictive Control in the Process Industry, Springer-Verlag,

London, 1995, ISBN3-540-19924-1.

3. ANSI-ISA-TR75-25-02-2000 Control Valve Response Measurement from Step Inputs

4. ANSI-ISA-75-25-01-2000 Test Procedure for Control Valve Response Measurement from Step Inputs

5. EnTech Control Valve Dynamic Specification V3.0, 1998.

6. Tolliver, Terry, “Process Analysis for Improved Operation and Control”, 1996, Fisher-Rosemount Systems Advanced Control Seminar