The need to enhance their plant`s competitive position has led many power plant operators to turn to new information and control systems that provide critical management information as well as plant control.
In today`s deregulated environment, improving plant competitiveness is the driving force behind most power plant control system replacements. New control systems not only control plant operations, but also facilitate operator and engineering staff access to diagnostic information that can help improve efficiency and reliability and, above all, cost-effectiveness. This is quite a change from the past, when control systems were typically replaced only after they became so old that replacement parts were either unavailable or outrageously expensive.
Information is Key
Because of the key role new information technology plays in nearly all new plant control systems, some in the industry have replaced the old term “instrument and control system” (I&CS) with “information and control system.” The new I&CS integrates the entire plant instrumentation and controls, plant computer and information systems, along with communications and diagnostic capabilities, into what is now termed an enterprise management system. Terry Bogard1 of Westinghouse Electric defines enterprise management as a “plant infrastructure merging information technology (IT) and instrumentation and controls (I&C). It is a unified information and control system (I&CS) with a shared common information repository. It is a management style focused on the advantages of I&CS technologies as the enabling tool to reduce plant O&M cost.” Enterprise management includes not only operating the plant and tracking omponent status, but also using financial models that allow testing various “what-if” scenarios for the impact of control and maintenance actions on the plant`s financial bottom line. The traditional boundaries between instrumentation systems and information systems are gone.
The new information technology enables I&CS systems to be built on non-proprietary networks. Controllers are commercially available PC platforms. Typically, user interfaces are familiar, commercially available operating systems that are totally platform-independent. Because of the significant levels of electromagnetic interference in power plants, high speed (100 megabits/sec) fiber distributed data interface (FDDI) networks substitute for much of the standard instrument cabling. These FDDI networks, too, are completely off-the-shelf items with wide application and a proven track record. Through them, the data for making both operating decisions and business decisions come together in the familiar PC environment.
Data vs. Information
With all this capability to communicate large quantities of data in a short time, data overload becomes a potential problem. Fortunately, the two-way communications with plant equipment, as facilitated by FDDI networks, and the power of the control PC`s software, together, can create useful operator information from plant data. One good example of this software capability to turn data into information is valve diagnostics.
Control valves make many small movements each day to control plant parameters such as pressure, temperature or flow rate to an optimum level for efficient plant operation. Diagnostic tools typically measure valve and valve actuator characteristics to determine whether or not such control valves are functioning properly or require maintenance. For example, analyzing the valve actuator pressure vs. valve travel can give information regarding such things as packing friction, actuator spring compression, galling, etc. This information can be important to maintaining plant performance and, hence, competitiveness, but with hundreds of valves and thousands of data points, the necessary manpower to interrogate each valve and to perform the analysis could be extremely manpower-intensive.
Fortunately, the Fieldbus standard communications protocol is capable of two-way communications. Using Fieldbus, a smart valve positioner can send back to the controller information on its actual final position, using the same wires as the outgoing signal. Feedback can include other information such as air pressure on each side of the cylinder, stroke time, etc. When these data are coupled with the appropriate analysis software, the operator can identify many specific valve problems, such as sticktion and hysteresis.
Sensors throughout the plant also provide an opportunity for significant operational improvements. It is common for a large plant to have one or two technicians devoted full-time to instrument calibration. Advanced calibration monitoring software uses a mathematical technique to predict an expected value, e.g. temperature or pressure, in process real-time. The software then computes the statistical variation from the predicted value and checks to see if the actual performance meets some predetermined standard, e.g., is within three standard deviations of the norm. The software then keeps track of the amount of time each sensor is beyond the preset limit. Ultimately, the software generates a report that flags those devices most in need of maintenance attention. In addition to maintaining plant efficiency by ensuring that process variables are actually within the proper range, this system improves labor utilization. Donald Frerichs2 of the Bailey Power Group observes, “The plant can make it truly predictive…trend the diagnostic values, sound alarms when things get bad enough to require attention. Believe it or not, this is the true performance benefit of all the Fieldbus fanfare.”
A new generation of valve controllers, called digital valve controllers, contains the valve diagnostic capabilities as a permanent part of the positioner. This is the next step in distributed controls. Microprocessors are now available with such low power consumption that these digital valve controllers can operate on the 4 mA minimum power available from the standard 4 – 20 mA instrument wiring. The HART communications protocol superimposes a digital signal on top of the 4 – 20 mA analog signal, allowing the control loop to operate as usual, but with the added feature of two-way communications with the valve controller. According to Tom Podhajsky3 of Fisher Controls International, “PC and Windows based software can communicate with the instrument to gather the basic valve diagnostic variables as the valve is stroked a complete open/closed cycle. The data can be plotted to determine the health of the overall valve assembly, the health of the digital valve controller alone, or the health of the valve/actuator combination.” This can alert the plant operator to potential valve functionality problems.
One big advantage of having the diagnostic capability on the valve is that the plant system can diagnose more valves in a given period of time. This can translate to either quicker response or a reduction in required computing power.
Besides linking plant operations with the control of plant maintenance activities and their supporting spare parts, a further step in enterprise management is system modeling and the resulting ability to do “what if” analyses. Unfortunately, complex power plants with many variable parameters and many complex interactions between those variables are difficult to model. Recently companies have turned to neural network models, which are excellent for dealing with such complex situations. As power plant operators gain confidence in this relatively new technology, they are even beginning to use it for real-time plant control.
Petrochemical and refining industries have used predictive control technology, referred to as model predictive control (MPC) for a more than 20 years, using linear modeling techniques. The increasing complexity of plants has led to development of nonlinear models. An alternative to these complex, nonlinear models, is a linear model combined with the “learning” capability of neural networks that helps cope with the real-world nonlinearity of a power generation facility.4 A neural network controller suffers from the fact that it does not know how to treat data that are outside the range of data collected in its “training” or open loop mode. The hybrid addresses this issue by providing an underlying linear system that controls when any of the parameters are outside the neural network`s original excitation range. A power plant near Warsaw, Poland, recently installed and tested such a hybrid system (See Neural Nets below).
The Ostroleka power plant near Warsaw, Poland, is one of the first to apply a hybrid neural network MPC system, an Aspen Target MPC controller using a nonlinear, state-space model integrated with a neural network model. The plant has three wall-fired boilers, each producing 650 t/hr steam and 200 MW of electricity. The plant uses pulverized coal from 4 pulverizer mills per boiler. Six sets of low NOx burners are installed in each of four elevations.
After a period of open-loop (training) operations, the MPC was placed in full closed-loop operation with the following early results:
1T. Bogard, et. al., “Plant Information & Control Systems to Improve Nuclear Power Plant Performance,” Westinghouse Electric Co., presented at Power-Gen International `98, Orlando, Fla., December 1998.
2D. Frerichs, “Predictive Maintenance Now Available for Controls and Instrumentation,” Bailey Power Group, presented at American Power Conference, Chicago, Ill, April 1999.
3T. Podhajsky, “The Evolution and Future of Control Valve Diagnostics,” Fisher Controls International Inc., presented at Power-Gen International `98, Orlando, Fla., December 1998.
4R. Neelakantan, et. al., “Hybrid Neural Network Based Control of a Coal Fired Boiler,” Aspen Technology, presented at Power-Gen International `98, Orlando Fla., December 1998.