drone. Additionally, some of this information will need to be transmitted to automation platforms, while other data will be directed to asset management systems. A platform is necessary to effectively integrate and manage all this infor- mation (see Figure 5 ). Example of autonomous operations Conventional control technology has been around for decades and has significantly contributed to production efficiency and product quality. Undoubtedly, proportional- integral-derivative (PID) control is the workhorse of down- stream operations. However, it still requires human inter - vention due to changes in operating state or rapid changes in disturbances. Similarly, advanced process control (APC) has demonstrated great value but needs human interven- tion to update and tune the model parameters to account for asset performance changes. Despite the extensive use of PID and APC, there are numerous processes where these techniques do not per - form well, resulting in significant human intervention and/ or manual control. Over the past few years, an autonomous AI-reinforced learning controller has been deployed to con- trol processes challenging conventional techniques. One of the AI controller’s strengths is that it can deal with conflicting targets such as balancing quality with energy savings.
Product
Distilland
Manual valve-A
Manual valve-B
Steam
Solvent
Solvent
H/E
H/E
H/E
By-product
Figure 6 Distillation process using AI control
The controller operates the manipulated variable (MV) and controls it directly. The reinforced learning technique requires only 30 learning trials. It can handle strongly non - linear processes and multiple disturbances. It integrates with a distributed control system and interlocking functions for safe, stable operation and accident prevention. It can learn independently without past data, handling new cir- cumstances to some extent. The controller has proven itself by continuously operating a distillation column for more than two years (see Figure 6 ). The AI solution successfully deals with the complex con- ditions (including changing feed quality, changing product
Figure 7 8D digital twin
69
PTQ Q2 2025
www.digitalrefining.com
Powered by FlippingBook