A Delayed Coker (DCU) is an established candidate for the application of Advanced Process Control (APC) with an attractive return on investment. The process involves a semi continuous coking operation where vaporized heater-charge is fed to the large delayed coker drums, which provide residence time for completion of thermal cracking reactions. The coke is deposited inside the drum and the overhead vapor flows into the fractionator where it is separated into wet gas, unstabilized naphtha, light coker gasoil (LCGO), heavy coker gasoil (HCGO), and recycle oil. Once the drum is filled with coke the vaporized heater-charge is diverted to another empty drum; to provide continuous operation, one or more pairs of coke drums are used. While one of the drums in each pair is on-stream, the off-stream drum is being cleaned, steamed, cooled and decoked, and finally, the empty drum is warmed up and made ready for switchover - simultaneously, in the same time interval. Figure 1 is a simplified block diagram of a typical Delayed Coker with two pairs of coke drums.

The disturbances that arise due to the semi-continuous operation pose unique challenges in Advanced Process Control (APC), The implementation skills, as well as the capability of the Multi-variable Predictive Control (MPC) technology in handling unmeasured disturbances, are put to test.

APC Design Challenges

The challenge in Delayed Coker-APC lies in handling the big disturbances that emerge from semi-batch coking operation. The composition, enthalpy, and flow rates of overhead vapor leaving each pair of coke drums vary continuously during coking cycle. The major disturbances occur during the drum-switchover and vapor-heating events, when the vapor-enthalpy feed to the main fractionator is suddenly reduced. Timely actions need to be taken on several control-loops to minimize the effect of sudden cooling of the column, which leads to large variations in Coker Gas Oil quality, as well as flow rates. The exact occurrence of these events is unmeasured, and the conventional control system is inadequate in handling these big disturbances. The resulting economic losses due to quality give away and off-spec product generation are substantial, not only in a Delayed Coker but also in downstream units where the disturbances are propagated. Hence, continuous operator attention is required for managing these events.

The on-stream time for APC, and the extent of which the potential benefits are realized, is determined by effective disturbance rejection. This involves several design challenges:

Automatic Detection of Drum Switch Events

The effective disturbance handling with APC requires DCS-logic for unambiguous detection of various events, that can lead to major disturbances in downstream fractionation sections. The inferred discrete events are then used for artificial generation of continuous disturbance functions which, in turn, are utilized for multi-variable modeling, as well as predictive feed forward control, honoring multiple constraints. Figure 2 shows the variation of HCGO draw temperature during 24 hours for a Delayed Coker, having two pairs of coke drums (A-B and C-D), along with the artificially generated disturbance variables calculated using inferred drum-switch and vapor-heating events.

Step Test and Model Identification

With 2 pairs of coke drums, the coke drums are typically switched at an interval of every 12 hours. There is continuous disturbance in the plant due to the various events that come with two on-stream coke drums. Consequently, the plant is in a relatively stable condition for only 3-4 hours in a day, thus requiring excessively long step test time. Also as the operating conditions keep changing during different modes of coker operation, the dynamic models are also expected to be different. Hence, for robust design it is essential to account for disturbances as inputs for modeling and obtain the ‘average-models’ which would be valid for all the stages of coking cycle. This is achieved by including the inferred drum switch and vapor heating events as inputs for Multiple Input Single Output (MISO) model identification. Figure 3 shows step test on HCGO Circulating Reflux (HCGO-CR) with response of HCGO draw temperature. It can be seen that the variation in HCGO draw temperature that is completely dominated by disturbance variables could be satisfactorily represented with MISO model.

Conflicting constraints and Change in control priorities during Drum Switch

During normal and relatively stable operation, the steady state material and heat balances are ensured with level and temperature (or quality) control strategies. However, during a drum switch, the equipment safety and material balance takes precedence over heat balance. Therefore, the control priorities must be changed and the APC design should ensure safe operation with minimum process upset.

Unmeasured Disturbances

The process dynamics in a Delayed Coker is dominated by unmeasured disturbances mainly due to drum switch events and feedstock changes. The effects of disturbances due to drum switch events are partly addressed by inferring these events, and using the artificial disturbance variables for feed forward actions. However, since manual actions in field are also involved in these events, the extent of the unmeasured disturbances vary in each coking cycle. The Intermediate variable and state observer (Kalman filter) concepts are utilized for robust control in the wake of these unmeasured disturbances.

DISTURBANCE REJECTION WITH APC

The inferred drum switch events are used as disturbance variables for predictive feed forward control. The onset of these disturbances predict the wide variation in column temperature profile and product qualities, thereby providing feed-forward compensation by making aggressive moves on Circulating refluxes and product draw flows. As seen in Figure 1, the HCGO draw temperature control significantly improved (Standard deviation reduced by 45%) due to timely and aggressive actions with APC. The draw temperature is maximized by reduction in heat removal from HCGO CR circuit, thereby increasing LCGO yield (by about 2-3%).

As seen in Figure 2 before APC, due to excess cooling of the column, LCGO was condensing on the HCGO section resulting in loss of LCGO tray-level and internal reflux. With APC preemptive actions are taken on HCGO-CR and LCGO draw flows which ensures that the level is stable, with sufficient internal reflux which ensures better fractionation between LCGO and HCGO products with maximum yield of valuable LCGO product.

Conclusion

Advanced Process Control systems have been successfully implemented in Delayed Coker's and, in turn, have enjoyed very high uptime (exceeding 95%). Automatic and unambiguous detection of various types of events involved in drum switch, and reliable handling of these disturbances, are the main challenges in APC design and implementation. The process stabilization with effective disturbance rejection while honoring conflicting constraints shows substantially tangible results and benefits from product up-gradation and throughput maximization.