Controlling Diesel Hydrotreaters
How APC solved a chronic and burdensome operating problem on an ultra-low sulfur Diesel hydrotreater.
Over the last few years, the refining industry has invested billions of dollars to meet increasingly stringent EPA-regulated limits for sulfur content in motor fuels. This burden falls on hydrotreating units where specially tailored catalysts promote desulfurization at high severity (high temperature) in the presence of hydrogen, yielding hydrogen sulfide and a small amount of light hydrocarbons as by-products. The primary operating target is the final sulfur content, typically 5-10 ppm for Diesel fuel. There is a severe penalty for violating the maximum specification limit for any extended period of time, and off-spec material has to be re-run or blended with more valuable low-sulfur material. The main variable the operator modulates to hit the target is the reactor inlet temperature.
Frequent lab tests are run on the product to let the operator know how well he or she is doing. However, lab test results are usually reported several hours after the sample is taken from the unit. In the meantime, process changes may have occurred that dramatically impact the degree of desulfurization. As an improvement, many units have been equipped with process analyzers that measure the product sulfur content on a continuous basis. Unfortunately, this measurement is still delayed by a process dead time of two to three hours. Under these conditions, even the best operators have a difficult time hitting the product sulfur target.
While operators may be tempted simply to raise the reactor temperature high enough so that virtually all the sulfur is reacted, there are side effects that make this impractical. The very high reactor temperatures required deactivate the catalyst very quickly while consuming excessive amounts of hydrogen and energy, and lowering final product yields. The optimum operating point is very close to, but not exceeding, the spec limit. It’s a classic control engineering challenge that one medium-size U.S. refinery met with a classic solution: advanced process control (APC). This refinery solved the problem on a 20,000 BPD Diesel hydrotreater equipped with a reliable continuous sulfur analyzer on the product stream.
Deep catalytic desulfurization requires very severe reactor conditions because a small portion of the sulfur-containing hydrocarbon molecules is very stable, requiring high desulfurization activation energy. Most desulfurization occurs rapidly in the top of the catalyst bed or in the first bed in a multi-bed configuration, while the rest is needed to eliminate the desulfurization-resistant, sulfur-containing hydrocarbon molecules.
Reaction kinetics and dynamics
According to data supplied by the catalyst manufacturer, the variables that have the most impact on the reaction kinetics are:
Feed rate—affects reactor residence time;
Reactor temperature—affects reaction kinetic rate; and
Feedstock sulfur content.
On the other hand, above a certain minimum limit, changes in hydrogen partial pressure have very little effect on the final product sulfur content.
As is the case with hydrotreating catalysts in general, reactor conversion correlates best with a variable that takes into account that temperature changes in the catalyst bed as the reaction proceeds. Hydrotreating is exothermic, so the temperature increases down the catalyst bed(s). The variable often used is weighted average bed temperature (WABT), where the weighting is by catalyst volume. This variable is calculated using temperature measurements at several locations in the beds.
Both theoretical considerations and practical operating experience indicate that reactor operation can be stabilized significantly if WABT is kept constant under changing operating conditions. From a process control standpoint, WABT is best controlled by adjusting the reactor inlet temperature which, as mentioned earlier, is the one handle the operator uses to control final product sulfur content. The complete APC solution for controlling product sulfur content is then concerned with how to adjust the inlet temperature setpoint and WABT target dynamically for changes in identified disturbance variables and on feedback from the product sulfur analyzer in order to maintain constant product sulfur content near the specification limit.
Under high-conversion conditions (sulfur removal greater than 99%), the final reactor product sulfur content is highly non-linear as a function of both feed rate and reactor temperature. The manufacturer-supplied operating guidelines suggest correlating logarithmic transforms of these key operating variables as follows:
Log of final sulfur content linear with log of the feed rate; and
Log of final sulfur content linear with WABT.
These transforms were applied to a large amount of operating data. A thorough process analysis of this data yielded the following observations and conclusions:
Reactor temperature measurements used for calculation of WABT require variable dead-time compensation. As an example, for the case of one reactor bed where only the bed inlet and outlet temperature are available, the inlet temperature measurement must be delayed so that it lines up dynamically with the outlet temperature.
Unless WABT is held constant, the sensitivity between feed and product sulfur content may be negative—when the feed sulfur content decreases, the product sulfur content may increase! This is because the reactorΔT decreases with less sulfur in the feed, decreasing WABT and average reaction severity. There may be competing reactions, such as saturation of olefinic molecules in coker Diesel that also impact reactor ΔT and that may further complicate how the control solution is structured to compensate forsuch interactions. The control solution is to feed forward changes in feed sulfur content (and other known disturbances) directly to the reactor inlet temperature so that the average reaction severity remains constant.
Variable tuning as a function of feed rate is needed to control both WABT and product sulfur content. As feed rate increases, dead time between a change in WABT and the effect on product sulfur will decrease, and vise versa.
Control of reactor inlet temperature by adjustment of reactor pre-heater fuel gas requires feedforward for changes in feed rate and inlet temperature.
Feedstock sulfur content is an important disturbance variable. While it is rarely measured in real time, changes can be approximated often from other process measurements. In the case of this hydrotreater, four different middle-distillate product streams were fed to and mixed in the hydrotreater feed header. Sulfur content of each feed stream remained fairly constant, while the relative and total feed rate changed frequently. Using measured flow rates of the individual feed streams, net sulfur content of the mixed feed could be estimated continuously with a simple calculation. This provided dynamic feedforward information about changing sulfur content of the reactor feed and proved to be quite valuable.
The diagram illustrates the control solution hierarchy. A nice feature is that WABT control can be used by itself during periods when the analyzer measurement is unavailable due to routine maintenance or failure. WABT control alone provides a significant reduction in process variance when compared to no APC at all. Another feature is that lab data may be entered by the operator to allow timely adjustment of the bias that may exist between the lab analyzer and process analyzer.
Board operators have little trouble understanding what the controls are designed to do and how they work. To them they appear as a triple cascade, only slightly more complicated that the single cascades they use in other parts of the unit and refinery. Prior to these controls, maintaining sulfur content within some reasonable range was considered to be the most aggravating operator task in the entire refinery. Since successful commissioning, this set of APC tools provides automatic control of product sulfur content, while the operators’ job is reduced to monitoring the tool’s performance.
Performance of the hydrotreater over a two-month period following commissioning is shown on the graph that plots the daily average of product sulfur analyzer measurements on a log scale and includes about a year’s worth of pre-control data. The reduction in variance achieved upon activation is readily apparent. After a few days of testing, the setpoint of sulfur content control was raised to 6.0 ppm against a specification limit of 10 ppm. For the next two months, product sulfur content remained within a band of
After two months of operation, analyzer data showed a sustained 70% decrease in variation when compared to recent pre-control data. At the same time, average sulfur content increased from 4.7 to 6.1 ppm. Although such a move sounds quite modest, this increase corresponds to a 14 °F decrease in average reactor temperature. When sustained over the life of the catalyst, this change corresponds to an increase in catalyst life of about 25%. For a catalyst that needs to be replaced after three years at a cost of $2 million, this represents a cost savings of $167,000 per year. Equally important for day-to-day operating is the transformation of this irritating problem into just another control loop to be monitored and supervised.
First, when solving complex operating and control problems, detailed process analysis of unit operation is essential. While software tools for data analysis and APC implementation abound, it’s the experience of process engineers who have a fundamental understanding of unit operation, both dynamically and steady-state, that guides development of the control design.
Second, there is no “cookbook” or standard solution for complex control problems, nor is there always a “best” technology. This problem would not have been solved by a standard application of off-the-shelf APC software. The best solution in this case makes use of APC techniques that have been in use for the last 30 or 40 years, among them:
Variable tuning of model parameters as a function of reactor space velocity (feed rate);
Variable dead-time compensation of process measurements to eliminate dynamic control variable error due to process lag;
Adaptive, incremental feedforward control action;
Use of log transforms to linearize inherently non-linear relationships between dependent and independent variables; and
Smart PID feedback control action, adjusting WABT on feedback from the sulfur analyzer. [Note: This control algorithm was developed specifically to handle feedback loops with long dead-time and lag. It determined P, I, and D control action independently based upon the recent history of the PV’s trajectory and proximity to the setpoint.]
APC in this case was implemented on a modern Foxboro DCS using only standard function blocks and calculation blocks.
The best solution to a control problem has three key features: It accomplishes the control objective, is the lowest cost, and has long-term sustainability. If you apply this philosophy consistently, you will find that nothing succeeds like success, even in the process control world.
James R. (Jim) Ford, Ph.D., P.E., is business leader, advanced process control for Maverick Technologies. Robert K. (Bob) Poag, Ph.D., P.E., P.A., is an APC consultant.