ADVANCED CONTROLS PAY OUT IN 6 WEEKS AT TEXAS REFINERY

. Included were two crude units with a combined charge capacity of 70,000 b/cd. The objective of the project was to use a recently installed Bailey Network 90 distributed control system (DCS) to achieve maximum economic benefits.

The advanced controls were based on "inferred" properties supplied by an outside vendor, who also provided consulting and assistance during the implementation phases. (Inferred properties are on-line computations for estimating laboratory test properties, such as ASTM boiling point and flash point, that are used for product quality control.)

In 1991, the controls were expanded to include inferred properties and advanced controls for the main fractionator of the fluid catalytic cracking unit (FCCU), which has a charge capacity of 32,500 b/cd.

In 1993, Marathon developed and installed fuzzy logic controls to improve the advanced pumparound controls for the crude units, and implemented a statistical quality control (SPC) program. The SPC program was designed to increase the benefits from advanced control of the FCCU main fractionator.

INFERRED PROPERTIES

Inferred-property computations use on-line measurements of temperature, pressure, and flow to produce a continuous estimate of product properties. Fig. 2 (95239 bytes) shows the process locations of the inferred property calculations provided for this project's advanced controls.

The inferred-property controls yield a 46% average reduction in variability of laboratory measurements (the variability reduction ranges from 34 to 68%). This represents a significant improvement in quality control.

This variability reduction results from eliminating operator delays in taking corrective action thus preventing overcontrol and undercontrol and from eliminating guesswork in deciding what corrective action to take.

The reduction in product variability has enabled Marathon to:

• Push closer to profit-constraining specifications

• Minimize upsets and missed targets

• Change the focus of operations from meeting specifications to optimizing unit performance.

The inferred properties use a proprietary calculation method provided by the vendor. This method combines fundamental chemical engineering principles with statistical analysis.

The chemical engineering principles include heat and material balances, widely accepted refinery engineering correlations, and methods from petroleum refining literature. Statistical analysis is used to evaluate parameters and customize the inferred-property calculations to fit a particular process.

The dynamics of the proprietary inferred properties are similar to the process dynamics of distillation column tray temperatures.

Process and laboratory data from 60-90 days of operation are needed to develop and customize an inferred property. To develop these calculations, the vendor requested hourly averages for the process data for the 60-90 day period.

For the laboratory data, sampling times had to be recorded carefully so that the data could be property synchronized with the Process data. Marathon collected and supplied the required data as process logs generated by the DCS, and as daily lab sheets.

BIAS UPDATING

Deviations can develop between an inferred property and the laboratory measurement. Causes of such deviations include drifting instrument calibration and tray fouling in a distillation column.

The user must correct this deviation. A simple correction method is to have the operator compare the inferred property with the laboratory results, then adjust the set point of the inferred-property controller to keep the laboratory measurement on target.

Another method is to update the inferred-property bias, which is a corrective term added to the inferred-property calculation to align the inferred property with the laboratory measurement. Marathon uses bias updating to ensure consistent control action among several process operators.

The DCS calculates the bias from the difference between the inferred-property estimate and the laboratory measurement. The operators use a special graphics display to enter the required data.

The bias update display contains:

• An historical trend of hourly averages of the inferred property

• A point for entering the laboratory measurement

• A point for entering the inferred property (Fig. 3)(80085 bytes).

The operator enters the new laboratory measurement. He or she then checks the historical trend of the inferred properties and enters the inferred property corresponding with the time the laboratory sample was taken.

The DCS then adjusts the inferred-property bias. The adjustment typically accounts for about 30% of the difference between the inferred and lab values. The amount of the adjustment can be tuned by the Marathon process control engineer.

CONTROL STRATEGIES

For the crude units and FCCU fractionator, the economic objective of the advanced controls is maximum total product value. Marathon did not choose to increase charge rates; therefore, maximum total product value is obtained by increasing production of higher-value products while reducing production of lower-value products.

INTERRED PROPERTIES

Reducing product: variability through the use of inferred properties requires "closing the loop" around each property (i.e., putting the system on automatic control).

Typically, a reflux or product draw rate is manipulated to control the inferred property. Once the loop is closed, the product specification target can be explicitly set as the set point to the controller, instead of requiring the operator to guess the product flow rate or draw temperature necessary to meet the target.

Again, eliminating guesswork reduces variability and permits product properties to be "pushed" closer to the specification limits.

The loop around most of the inferred properties was closed in the DCS using a standard proportional integral derivative (PID) control block. To the operator, these inferred-property controllers look and act like any other controller, requiring no special activities.

Inferred properties can be used in more complicated controls. An example is control of the FCCU main fractionator bottom-level, inferred light cat cycle oil (LCCO), and inferred slurry API gravity.

As shown in Fig. 4,(74552 bytes) there are only two streams-LCCO and slurry draws-available to control these variables. The result is the simple so-called 3X2 multivariable controller shown in Fig. 4 (74552 bytes).

Economics determine the design of this controller; LCCO has a much higher product value than does slurry.

To maximize profit, the controls minimize loss of light cycle oil to the slurry product. This is done by minimizing slurry production against two product constraints: maximum LCCO end point and minimum slurry API gravity.

The control strategies used to achieve this are:

• To control inferred slurry API gravity by adjusting slurry product flow

• To override the preceding with high LCCO end point control

• To control FCCU main fractionator bottom level by adjusting LCCO flow.

Stripper-level overrides are included as a precautionary measure to prevent the controls from "pulling" the stripper dry.

PUMPAROUND

To maximize total product value, the general optimizing control strategy is to maximize product separation by maximizing vapor-liquid traffic through the tower. The pumparounds are minimized against process constraints with priority given to minimizing lower pumparounds.

Reducing the pumparounds increases vapor-liquid traffic, which reduces overlap in the boiling-point curves of the adjacent product streams. This reduced overlap, coupled with control of inferred boiling point, permits increased production of higher-value products.

Tower flooding typically is the limit for minimizing pumparounds. Tower flooding is measured by tower pressure differentials and calculated indications of vapor-liquid traffic.

The pumparound controls have eliminated tower flooding incidents and associated production of off-spec product.

CONTROL IMPLEMENTATION

Marathon participated in implementing the advanced controls, having extensive experience working with the DCS during the changeover to that system. The "hands on" approach to implementation led to several benefits, including:

• Project implementation time was reduced. Many controls were commissioned by Marathon at its convenience. This minimized schedule conflicts between the refinery and the vendor and reduced the vendor's on site time.

• Operator acceptance was excellent. The advanced controls were integrated into the existing graphics and followed Marathon's DCS standards; therefore, the operators were comfortable with the look and operation of the new controls from the start.

• Hand-over of the controls to Marathon was seamless. No training of Marathon engineering personnel was required.

• On-line time was excellent immediately after implementation. Marathon personnel identified problems early and the vendor corrected them quickly.

IMPLEMENTATION

During the implementation process, Marathon provided:

• Historical process and lab data (90 days) for use by the vendor in developing the inferred properties

• DCS installation of the programs that calculate inferred properties

• The operator interface (graphics and group displays)

• An on-line system for computing and keeping records of the on-line factors for the advanced control loops

• The on-line system for bias-updating the inferred properties

• Documentation and operator training

• Modification of the basic DCS controls to accommodate the advanced controls

• Installation of the advanced controls

• Commissioning and loop tuning.

The vendor provided:

• Project planning and scheduling

• Development of the inferred-property calculations

• On-line programs for the inferred-property calculations

• An introductory, class on inferred-property control for the process operators

• Functional specifications for the advanced controls

• Assistance in configuring the advanced controls

• Assistance in commissioning and tuning some of the advanced control loops.

The on-line programs for the inferred-property calculations were written in C language for installation in the DCS.

POST-PROJECT WORK

Inferred-property controls stabilize the process and provide a foundation for additional control improvements. Marathon pursued two follow-up projects to the inferred-property projects.

Both follow-up projects use the inferred properties as a foundation for further advanced control, and both address problems that an inferred-property model does not.

FUZZY LOGIC

Fuzzy logic controls use so-called fuzzy set theory and operator experience to control a project. Fuzzy logic control is an excellent technique for controlling systems that are difficult to model, and for multivariable control problems for which a wealth of operating experience is available.

A fuzzy logic controller has two parts: a state estimator that uses fuzzy set theory and an "expert system rule base" that determines the controller output based on the estimated state of the process.

In this way, fuzzy-logic controls model the best process operator, incorporating his or her ability to draw conclusions from experience and noisy data, and from his or her ability to work through a set of "if/then" rules before making a control decision.

Marathon used fuzzy logic to upgrade three existing crude unit pumparound-constraint controls from conventional PID control, and to implement three new constraint controls:

• Crude-heater outlet temperature maximization

• Crude charge maximization

• Distillate end point maximization.

As mentioned, the general strategy is to maximize product separation by operating the fractionator against its flooding constraints.

This project was undertaken to eliminate problems experienced with the existing pumparound constraint controls. Because the existing controls used traditional PID logic, they could not handle the inherent nonlinearity and multivariable nature of operating this crude unit against flooding limits. In addition, fuzzy logic was expected to further optimize operation of the unit.

The problems with the existing pumparound constraint controls manifested as low controller utilization, frequent manual intervention by operators while the controllers were operating, operator complaints, and considerable control tuning.

Fuzzy logic was chosen because:

• It can handle nonlinear, multivariable control problems.

• There was significant operator experience

• It was inexpensive.

• It offers quick set-up and is easily tuned.

• It is simple to add, delete, or modify rules, thus providing an evolutionary path to control improvement.

• It requires no complex mathematics.

• It permits programmable controller response. (It responds quickly to constraint violations but moves slowly toward constraints.)

These controls operate the crude unit fractionator 35% closer to flooding limits. This allows Marathon to swing an additional 2% of charge to distillate, rather than FCCU feed, during the heating oil season. It also enables Marathon to swing an additional 10-150 F. of naphtha to kerosine.

The benefits from the controller derive from the equivalent of having the best operator available at all times. This eliminates overcontrol, undercontrol, and response delays, and minimizes the operator comfort zone. (Operators tend to control the unit very loosely, with minimum intervention.)

The controls were developed by Marathon's operations research department at Findlay, Ohio. The concept was developed using information from numerous published sources. The operations research department provided the source code and developed the expert rule base from interviews with operators and other process experts.

SPC

Within the first year of operation of the slurry API gravity inferred property, it became apparent that significant unexplained variability remained in the lab results. This variability confused operators and supervisors. It manifested as overcontrol, through manual interventions to "help the controls," and as a loss of confidence in the controls.

A review of the process revealed:

• Significant process lags existed between corrective actions and visible results (greater than the interval between samples). Therefore, the inferred property and the lab results frequently differed greatly, causing doubt regarding the accuracy of the inferred property. Operators and supervisors tended to ignore the inferred property during these times--sometimes with justification-and take action.

• Known process upsets were the primary source of variability in the slurry API gravity.

  It was apparent that there was an opportunity for further process optimization. Marathon concluded:

• A single laboratory data point is inadequate for decision making.

• Decision making guidelines were required regarding whether the inferred-property controller is in or out of control.

SPC meets these needs.

SPC was implemented in 1993 in the form of a zone control chart. Interpretation instructions provide a simplified and consistent approach to determining whether the inferred slurry API gravity property is accurate (Fig. 5)(81946 bytes).

Zone charts were used because of their simplicity and ability to detect most out-of-control situations. Each zone represents one standard deviation; the center of the chart is the target (Fig. 5)(81946 bytes). Process data from 6 months of operation were used to determine the standard deviation.

SPC has reduced product variability by an additional 25% through the near elimination of overcontrol and the positive identification of true out-of-control situations that require adjustment of the inferred property. The slurry API gravity target was reduced to capture the benefits available from reducing variability.

BENEFITS

Annual payback for the advanced controls was calculated as $ returned per year/$ invested (Fig. 6)(52575 bytes). For the various portions of the project, Marathon determined this ratio to be:

• 1990 crude inferred-property control, 3.06

• 1991 FCCU inferred-property control, 3.60

• 1993 crude pumparound fuzzy logic control, 1.60

• 1993 FCCU statistical process control, 0.40

• Total annual payback, 8.66.

The advanced controls pay for themselves almost 9 times per year, or every 6 weeks. (Total costs include external and internal costs.)

Benefits are achieved by:

• Reducing product quality variability or process variability

• Operating closer to product quality limits or process constraints.

  The inferred-property controls reduced average product quality variability by 46%. This reduction permitted Marathon to maximize swing stream economics between:

• For the crude units, gasoline and naphtha, naphtha and fuel oil, and fuel oil and FCC feed.

• For the FCCU tower, gasoline and LCCO, and LCCO and slurry.

These benefits were calculated by estimating the changes in production rates resulting from moving the quality targets. The benefits have been verified using production data; for example, slurry production was reduced from 4.5% of FCCU charge to 3.0% of charge.

The fuzzy logic controls on the crude units reduced process variability, thus permitting operation of the crude unit 35% closer to its tower flooding limits. This allows further manipulation of the naphtha/fuel oil swing and, through better fractionation, better recovery of fuel oil from FCC feed. These benefits were calculated directly from production data.

FCCU statistical process control reduced slurry API gravity variability an additional 25%. This permitted additional upgrade of slurry to LCCO through pushing the quality target closer to the specification limit. These benefits were calculated by estimating production rate changes resulting from moving the product targets.

Finally, advanced controls have significantly changed Marathon's operations. Before control, considerable effort was expended to keep products on-specification. Since control, product specifications are met.

Marathon's efforts are now directed toward optimizing the units.

UTILIZATION

Thirty-eight advanced control loops have been installed since 1990. On-line time for the advanced controls was 97% in 1993, 2-3 years after installation.

High utilization and, therefore, realization of benefits is achieved through management's , commitment to the system, and through proper advanced-control maintenance.

Marathon focuses considerable attention on advanced control utilization. Advanced control on-line factors are computed on the plant-wide data record keeping system.

On-line factor is reported weekly, and is included in the operating team's annual goals.

Naturally, control problems receive top priority from the control group.

ACKNOWLEDGMENT

The authors wish to thank Joseph James, John Matt, and the late Dr. Gyan Jain, for their contributions.