By Todd Sommer, ME, CEO,
Promecon USA, Inc.
Smart sensors, smart transmitters, smart controllers, smart tools, etc., seem to be everywhere. Advanced control technology for combustion systems is increasingly providing cost-effective boiler performance optimization and emissions reduction. As the technology matures, the benefits can be multiplied by optimizing not only individual operating systems, but entire plants and even extensive generating and distribution networks.
Now that most firing equipment has been retrofit with distributed control systems (DCS), there are numerous opportunities to add hardware and software products that can be easily applied and result in improved operation. In addition, performance monitoring can now be achieved at a level of detail never before possible. The products available include advanced sensors and optimization software.
Advanced sensors are currently available to provide detailed information about the combustion process. Data that could only be gathered by laborious field testing methods in the past can now be monitored in real time. Sensors for monitoring pulverized coal flow, gas temperatures, gas composition, and unburned carbon in fly ash can be applied to provide valuable information for combustion process control and optimization. Other sensors are being developed to provide operators and control systems information that can be used in a feedback control and optimization scenario.
The use of “self-learning” neural network (NN) software for optimizing the combustion process in large fossil-fired boilers started in the early 1990s and has grown to become an accepted technology for improving unit operation (reduced heat rate and emissions). The technology is being deployed worldwide, particularly in response to NOx emission regulations. Optimization software has been installed successfully on tangential, turbo, and wall-fired boilers by all the combustion optimization vendors. In general, the results have fallen into two improvement ranges. The wall-fired units tend to achieve a 5-15 percent reduction in NOx emissions with a 0.25-0.75 percent improvement in heat rate. The tangentially-fired units tend to get better results in the range of 10-25 percent reduction in NOx emissions and 0.25-1 percent improvement in heat rate. Turbo-fired units appear to achieve similar results with perhaps slightly higher heat rate improvements. The most impressive improvements seem to have been achieved on units at larger coal-fired facilities equipped with newer digital control systems and on-line performance monitoring systems.
The industry has limited experience with roof-fired, cell burner and cyclone-fired units, but we should see these units being addressed in the next few years as additional NOx reductions are required and the cost of operating SCR facilities becomes better understood. The boiler heat release rate is also important in understanding the improvements to be gained through NN optimization. Older units designed with high heat release rates can be optimized for improved heat rate but are less amenable to additional NOx reduction. The industry has also applied combustion process optimization to gas-fired and oil-fired plants with similar results depending on the boiler design.
The benefits achieved in almost all installations have been achieved with the existing plant instrumentation and controllable variables (automation of individual overfire air ports has been included on some units) and the addition of some type of carbon monoxide (CO) monitoring system. Attempts over the years have been made to use multiple CO monitors at the economizer outlet in the same location as the excess oxygen probes. This has had very limited success due to the maintenance and reliability involved with analyzers installed in this location.
A number of new sensor developments are being made and products deployed to provide both CO and carbon-in-ash data on a real-time, or near real-time, basis. These new devices have only become available in the last couple of years. They are currently being retrofit to some of the combustion optimization systems and the results should be available in the near future. In addition, units are currently being equipped with additional air and fuel flow monitoring sensors to get a more direct indication of individual burner firing conditions. A number of demonstration programs, designed to quantify the benefits of combining advanced sensors and process optimization software, are currently underway and results should be available to the industry later this year. New sensors will continue to become available in the future providing more direct feedback capability versus using intuitive techniques.
Combustion optimization has matured in a couple of areas over the past few years. One involves neural network ability to adapt to changes over time while continuing to provide improvements and dealing with abnormal plant conditions such as a poorly performing pulverizer or an oscillating furnace draft control system. There are currently more than 100 systems successfully installed with adaptive learning techniques. The other area is that of the application of combustion optimization on units that continually ramp up and down in load on automatic dispatch. Most of the units that have installed neural network combustion optimization operate under automatic dispatch. The majority of the industry’s experience has been on units with high load factors so the ramping rate tends to be in the 3-5 MW/min range, although some are operating on units with higher ramp rates up to 10 MW/min. As mentioned earlier, combustion optimization systems have also been successfully installed on oil and gas units, which tend to change load at even faster rates. One of the industry’s current challenges is to ramp units faster while staying within the constraints of NOx, CO, LOI and equipment capability.
The industry is currently at a crossroads between three competing optimization technologies: first principle models; neural network optimization, and advance control techniques such as multi-variable predictive controllers. The technology used to achieve these improvements usually depends on the roots of the vendor you choose or from the direction from which you approach the application. Looking at it from the top down (from the asset manager’s perspective), the solution is more based on first principle models, which run faster than real time so you can predict where the demand will be and how to deploy your assets to meet that demand, thereby maximizing profitability. This solution also appeals to the long-term/strategic asset managers for deciding which units to refurbish/retire, which units to build, etc.
A neural net optimization software provider will recommend empirical model-based optimization since this reflects real operating conditions, not original design based conditions, which may be no longer relevant. These vendors will also recommend their solution because they believe that, in order to do combustion optimization, you have to handle the entire combustion process together, and not as a separate solution. The controls vendors seem to favor multi-variable predictive technology which tends to push much of the optimization down into the combustion control system to take advantage of their installed base and to maintain product control.
In the end, the solution will probably be a combination of all three of these technologies as the deregulated industry matures and continues to embrace digitization. As new sensor technology is added, the significant economic value of optimization software will be realized. In addition, as the boundaries between operations, maintenance and long-term asset management erode, optimization can truly begin to occur on a plant and even system-wide basis.