The real-time monitoring of cracking furnaces in olefin plants aims to produce real-time information on key performance indicators (KPIs) such as product yields and coking that are important for controlling the furnaces and for optimising their operation over the run length.
In this paper, we present a rigorous tool for the above task that makes use of a first-principles model describing both the cracking and coking reactions taking place within the furnace, as well as the important heat transfer and pressure drop phenomena. The tool receives all available measurements from plant sensors and processes them in real time using a state estimation algorithm based on an Extended Kalman Filter algorithm. By combining model predictions with measured information, and taking account of the uncertainty inherent in both, the tool produces optimal estimates of the current state of coking in each coil and of the product yields and other KPIs.
Probably the weakest point of most first-principles models for the prediction of the cracking furnace behaviour is the characterisation of the coking reactions. This is a reflection of the complexity of the coking phenomena, the many factors that affect them, and the variability of these factors from one furnace to another. Accordingly, a key feature of the proposed tool is that, simultaneously with the state estimation mentioned above, it performs a continual and automatic recalibration of some aspects of the coking model.
The performance of the above tool has already been tested and verified in a full-scale industrial application.
See more of this Group/Topical: Computing and Systems Technology Division