An embedded‐agent technique for industrial control environments where process modelling is difficult

Most automatic devices designed to control even moderately complex systems are based on feedback. The concept of feedback control is so intuitively straightforward that many people assume it must be quite simple to combine such simple mechanisms to create relatively sophisticated automatic control systems. However, the reality is that to design a truly effective automatic controller is often the greatest challenge an engineer may face. A major cause of the difficulty is time lag, which means that the system is constantly trying to correct for conditions that existed earlier. Conventional control theory seems to work best when attempting to design control devices for a process that can be well approximated by a model with linear and otherwise straightforward relationships between just a few variables. However, for complex and non‐linear processes, where it is difficult (if not impossible) to develop a mathematical model for the system to be controlled, conventional control fails and the control is left to the human operator. Examples of such systems appear in process control industries such as cement industry, water treatment processes … etc. In addition, conventionally designed automatic control devices often have relatively narrow performance bands, as many types of real world processes and systems can be well approximated by linear models when variables are limited to narrow ranges. If a variable goes beyond this range then the system will become unstable and the control device may no longer be able to make appropriate adjustments. Until recently, certain types of processes were still controlled by human operators simply because it proved to be difficult to design an automatic control device by the conventional methods. This was not due to the lack of sensors or any other hardware problem, but simply because of non‐linearities or other complexities in the process.