Electricity is required to keep process equipment, including compressors and pumps, running. Steam is needed to drive reactions. Loss of either resource can cause a costly, unplanned shutdown. Managing these resources in a large chemical plant is a complicated undertaking that can impact the profitability of the plant.
The Dow Chemical Co.&rsquos Seadrift, Texas, petrochemicals plant is saving $1.25 million annually due to reduced overall energy demand, including electricity and natural gas resources. These savings were made possible by a combination closed-loop optimizer and the G2 expert system. The closed-loop optimizer is linked directly to the plant&rsquos control system and automatically makes changes. G2, an object-oriented expert system software platform from Gensym Corp., Burlington, Mass., captures operations expertise in the form of rules, procedures and models to infer production conditions and make supervisory control decisions. G2 reduces the programming effort required to run the optimizer for the many possible operating conditions found in a large petrochemical plant that produces a variety of products.
Last year the Seadrift plant ran in closed-loop mode 98% of the time, which is remarkable for a plant of its complexity. The savings that have been achieved at this location are the first step. Dow believes that it can more than double these savings by expanding the optimizer and expert system implementation to other parts of the Seadrift facility, as well as to other locations.
Optimizer alters behavior
The Seadrift plant has a number of gas turbines. The energy they generate can be used in the plant or sold to the grid. The turbines, as well as other systems in the plant, also produce waste heat that is passed through a boiler to make steam.
When the various power-generating functions were manually controlled, operators rarely had time to weigh the cost and value of power when making decisions, such as at what level to run a gas turbine. An operator&rsquos main objective is to keep the plant running âas long as the plant keeps running, supervisors and peers are unlikely to notice if he or she is also saving a bit of money. So the operators would run the plant in a way that wouldn&rsquot cause any shutdowns or disruptions rather than also try to optimize energy expenses and revenues.
In an attempt to optimize operations, Dow installed a linear program. It would suggest new operating conditions after operators had typed in the current data, but it was rarely used.
Next, Dow engineers developed an open-loop optimizer that captured data from sensors and continuously calculated the most efficient equipment settings. Operators often did not have time to implement these settings because of their other duties, and, in other cases, did not do so because they disagreed with them. Dow started to see substantial savings, however.
Some of these savings were a result of the operators taking advice from the optimizer, while others were because they learned new behaviors by using the optimizer. For example, operators would run the gas turbines at partial load in the evening hours because they understood that the prices received for any power that was produced rarely covered the costs of generating it. Although this was a step in the right direction, the full potential of the open-loop optimizer was never realized since it required the constant attention of the operators, whose other duties didn&rsquot allow them to respond to constantly changing energy prices.
Dow engineers decided it would be best to switch to a closed-loop optimizer that would not rely on the operators to make the necessary changes and would deal with the nonlinear behavior of steam. So they put together a request for proposal and received five serious bids.
Dow engineers noticed that three of the bids included G2 for managing the optimizer in real time. G2, developed by Gensym, is a comprehensive, object-oriented environment for building and deploying real-time expert system applications. The bids that included G2 appeared easier to build and maintain since the other two proposals would have required writing a lot of code to manage the optimizer. The engineers had already learned from using the open-loop optimizer that changes are continuously made, so any system that relied on hard code would have needed constant maintenance, which would take a lot of time and would be prone to errors.
In addition to the dynamic environment inherent to plant operations, other complexities complicate the modeling of energy system operation. Sensor readings may be inconsistent or &ldquonoisy.&rdquo Equipment behavior varies with time. An upset in operating conditions must be dealt with before the process can be optimized. A simple optimizer model cannot directly handle all of these intricacies. An expert system, such as G2, can manage these complexities and work with the optimization model to determine the best operating conditions.
G2 is a practical means for managing a closed-loop optimizer since it allows you to work at a higher level of abstraction. The user defines inputs, such as sensors, and outputs, such setpoints, as objects. Typical optimization tasks, such as averaging a meter reading over time, checking a meter reading against a mass balance, or confirming a relief-valve position before changing a setpoint, can be made parametrically without having to recode the algorithms. With G2, you simply change the properties of or linkages between objects to implement the rule logic for these tasks. Variables are time-stamped for keeping data and event histories, and for reasoning about behavior changes over time.
Building confidence
G2 enhances the process optimizer by functioning as a resident &ldquoexpert.&rdquo If a process value is fluctuating in a narrow range, the optimizer, run on its own, may recommend process changes that have very little impact since, in theory, it doesn&rsquot distinguish between changes that save $100 or 10 cents. But in the real world, there is a cost associated with every change, and G2 makes it easy to determine the amount saved by the change and evaluate whether it is worthwhile.
The Seadrift plant has achieved savings in the operation of heat-recovery steam generators (HRSG). These generators create steam from the hot exhaust of a gas turbine for plant heating or to generate electricity via steam turbines. In manual-control mode, operators tend to run at a level that allows them to comfortably handle any fluctuations in steam demand. As a result, the electrical load on the steam turbines varies since the steam load changes inversely with changes in plant steam demand.
With the aid of the real-time expert system, the optimizer will automatically adjust the load to achieve the best savings. The optimizer quantitatively evaluates continually fluctuating electricity prices and steam demands. It can then determine the appropriate level at which to run the HRSG and best utilize the steam generators. The optimizer and expert system change operating setpoints as often as every five minutes, whereas operators typically made changes only once or twice per day.
G2 also has a built-in simulator that allows users to run cases offline and measure the results. These simulated data can run concurrently with real system data to improve or compare models in the test environment to the conditions in the actual plant. The simulator can be used for testing application logic throughout the development cycle, for monitoring actual-to-ideal performance in the finished application, and for what-if analyses to help identify optimal operating conditions and designs.
Dow engineers frequently use this tool to evaluate potential changes to the optimizer and to assist in operator training. The operators play a critical role in the use of a closed-loop system â they not only need to feel comfortable with its operation, but their invaluable knowledge helps improve it.
For example, when the closed-loop optimizer was first implemented, operators often turned it off. The optimizer would recommend running the turbines at partial load, but the operators were concerned that the hydrogen-rich fuel fed to the turbines might flash back and force a shutdown. When this problem was brought to their attention, Dow engineers used G2 to implement guidelines to the optimizer on the relationship between fuel richness and the limitation on partial loading of the gas turbine.
The optimizer provides additional savings by giving Dow a better understanding of plant operations. For example, the operators were controlling a header pressure by venting. Based on output from the optimizer, the header controls were changed to allow the operators to maintain a reliable steam supply to the units, which resulted in additional energy savings. Dow has validated the operation of the optimizer and expert system at Seadrift. The operators&rsquo confidence in the system is high enough so they rely on it in all but the most unusual of circumstances.
Share the wealth
Cost savings of $120 per hour have been attributed to the combined use of the closed-loop optimizer and the expert system. The success of Dow engineers in increasing the operators&rsquo comfort level is demonstrated by the fact that last year the Seadrift plant ran nearly continuously in closed-loop mode. The operators now view the optimizer as being an important part of their success.
Dow believes that substantially greater savings can be captured by extending the application to other areas of the plant&rsquos operation besides energy usage. Dow is also ready to take this concept to other plants. The use of closed-loop optimization in combination with a real-time expert system clearly provides Dow with important improvements to plant performance and the company intends to take full advantage of its capabilities.
James F. Sturnfield is a senior specialist for Union Carbide Corp., a subsidiary of The Dow Chemical Co., South Charleston, W.Va. Sturnfield has a Ph.D. in mathematics and specializes in simulation, optimization and expert systems. E-mail him at [email protected].