Expert systems model processes by representing knowledge in the form of situation-action rules.
A rule engine of this type is what allows SootOpt® to model sootblowing processes opportunistically, taking advantage of all the things that various experts know and whatever information is available, even when that knowledge isn’t comprehensive and is instead based more on available instrumentation, historical experience, and tested habits.

Like a super-vigilant operator
SootOpt works in conjunction with existing sootblowing controls and instrumentation – whether the plant has PLC-based controls or an advanced ISB system. SootOpt overlays these controls with a rules-based model and an optimization search algorithm that relates boiler cleaning to operational constraints and goals as well as global performance objectives. Like an operator, SootOpt looks at current data for temperatures, pressures, sprays, load and the time since last cleaning. It collects the possible relevant actions given current conditions and uses a sense of priority to decide what, if anything, to do. In cases where there is no advanced instrumentation – such as flux, FEGT, or strain gauges – SootOpt can still emulate the thought and decision making process of a super-vigilant operator who is looking at whatever information is available. Where advanced instrumentation exists, the rules-based model and decision making machinery has that much more information to go on. At either end of the spectrum a major benefit of using an expert system is that any available relevant information can be utilized to help make decisions, and the rules governing that utilization need not be based on fixed assumptions or constrained to a narrow set of principles.

Freedom of information
The information SootOpt can utilize is unlimited. It can include spray-flows, steam and exit gas temps, tube metal thermocouples, pressure taps, combustion control settings, media supply indicators (even from other units that share that supply), load, etc. It can also include data from more advanced instrumentation such as strain gauges, ash-loading measurements, flux instruments, measured and/or calculated FEGT, PerformanceOpt® cleanliness-factors, fuel analysis, operator entered data recording clinker buildup severity, and indications taken from SootOpt’s neural network-based scenario mill. Any variable that can be electronically converted can be considered as an input to a rule. And SootOpt can use any rule that can be invented using those conditions, in its decision making process. To invent a rule the human machine manager just needs to be able to describe how one or more detectable conditions should be used to suggest taking (or prohibiting) one or more of the available cleaning actions.

Transparent decision machinery
A rules-based model is an opportunistically defined picture of how things work based on whatever is known about how a process responds, rather than a complete first principles knowledge of exactly why the process responds the way it does. A fundamental challenge to this way of knowing and controlling things is “emergent” behavior. The interaction of all the relatively simple rules as they respond to a process that is not 100 percent understood can be surprising and dynamic. Having the rules model and optimization engine buried inside a black-box just doesn’t work. The AI needs to be able to explain itself and verify to the human manager what its instructions are and how it is carrying them out. The manager needs to be able to easily understand the knowledge being relayed about how the process works and adjust the AIs decision-making process based on new process feedback. All of this needs to be possible under real-conditions, in real time, and with minimal effort.

To accomplish this critical task SootOpt makes the decision machinery transparent. It explains each of its decisions succinctly and provides quick access to the branches of supporting justifications. This provides the machine manager with valuable situational awareness - not just about how the currently relevant rules and actions are defined but also how they are playing out as they respond to the process. This transparency makes the proposition of using significantly large rules-based models to make decisions about complex problems realistic. In short this transparency of function and dispatch helps give AI real-world utility.